USF Libraries
USF Digital Collections

The influence of pre-existing memories on retrieval-induced forgetting

MISSING IMAGE

Material Information

Title:
The influence of pre-existing memories on retrieval-induced forgetting
Physical Description:
Book
Language:
English
Creator:
Goodmon, Leilani B
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Memory
Forgetting
Suppression
Integration
Interference
Dissertations, Academic -- Psychology -- Doctoral -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: When people form episodic connections between memories that share a common retrieval cue, the tendency for those memories to interfere in later retrieval is often eliminated, and forgetting of the interfering information is reduced. For example, episodic integration protects memories from retrieval-induced forgetting (RIF), a phenomenon in which practicing retrieving some associates of a cue leads to the suppression of others that interfere with retrieval (Anderson, Green, and McCulloch, 2000). The purpose of this study was to determine whether semantic integration, as a result of pre-existing associations between practiced items and their unpracticed competitors, also moderates RIF. This research was motivated by the existence of many pre-existing associations between the practiced and unpracticed sets in one study that failed to replicate the RIF effect with item specific cues (Butler, William, Zacks, and Maki, 2001).It was hypothesized that pre-existing associations increase the implicit, semantic integration among the items, insulating them from inhibitory effects. The results were consistent with this hypothesis: when associations between the practiced and non-practiced sets were maximized, no forgetting was observed, however when such associations were minimized, there were reliable levels of RIF. The benefits of semantic integration were replicated across four experiments including one that used Butlers original materials and design. Furthermore, when Butlers items were simply re-arranged in order to minimize the associations and thus reduce semantic integration, the typical RIF effect was observed. Additional results revealed that the moderating effects of semantic integration are not mediated by explicit integration strategies.
Thesis:
Thesis (Ph.D.)--University of South Florida, 2005.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Leilani B. Goodmon.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 169 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001670330
oclc - 62276846
usfldc doi - E14-SFE0001191
usfldc handle - e14.1191
System ID:
SFS0025512:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001670330
003 fts
005 20051216093253.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 051115s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001191
035
(OCoLC)62276846
SFE0001191
040
FHM
c FHM
049
FHMM
090
BF121 (Online)
1 100
Goodmon, Leilani B.
4 245
The influence of pre-existing memories on retrieval-induced forgetting
h [electronic resource] /
by Leilani B. Goodmon.
260
[Tampa, Fla.] :
b University of South Florida,
2005.
502
Thesis (Ph.D.)--University of South Florida, 2005.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 169 pages.
Includes vita.
520
ABSTRACT: When people form episodic connections between memories that share a common retrieval cue, the tendency for those memories to interfere in later retrieval is often eliminated, and forgetting of the interfering information is reduced. For example, episodic integration protects memories from retrieval-induced forgetting (RIF), a phenomenon in which practicing retrieving some associates of a cue leads to the suppression of others that interfere with retrieval (Anderson, Green, and McCulloch, 2000). The purpose of this study was to determine whether semantic integration, as a result of pre-existing associations between practiced items and their unpracticed competitors, also moderates RIF. This research was motivated by the existence of many pre-existing associations between the practiced and unpracticed sets in one study that failed to replicate the RIF effect with item specific cues (Butler, William, Zacks, and Maki, 2001).It was hypothesized that pre-existing associations increase the implicit, semantic integration among the items, insulating them from inhibitory effects. The results were consistent with this hypothesis: when associations between the practiced and non-practiced sets were maximized, no forgetting was observed, however when such associations were minimized, there were reliable levels of RIF. The benefits of semantic integration were replicated across four experiments including one that used Butlers original materials and design. Furthermore, when Butlers items were simply re-arranged in order to minimize the associations and thus reduce semantic integration, the typical RIF effect was observed. Additional results revealed that the moderating effects of semantic integration are not mediated by explicit integration strategies.
590
Adviser: Michael Anderson.
Co-adviser: Lili Sahakyan
653
Memory.
Forgetting.
Suppression.
Integration.
Interference.
0 690
Dissertations, Academic
z USF
x Psychology
Doctoral.
773
t USF Electronic Theses and Dissertations.
856
u http://digital.lib.usf.edu/?e14.1191



PAGE 1

The Influence of Pre-existing Memori es on Retrieval-Induced Forgetting by Leilani B. Goodmon A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Psychology College of Arts and Sciences University of South Florida Co-Major Professor: Michael Anderson, Ph.D. Co-Major Professor: Lili Sahakyan, Ph.D. Dewey Rundus, Ph.D. Mark Goldman, Ph.D. Cheryl Kirstein, Ph.D. Date of Approval: July 18, 2005 Keywords: memory, forgetting, suppre ssion, integration, interference Copyright 2005 Leilani B. Goodmon

PAGE 2

i Table of Contents List of Tables iv List of Figures vi Abstract ix Introduction Concept of Inhibition and Inhibi tory Control as an Adaptive Mechanism of Memory Retrieval-Induced Forgetting and the Retrieval Practice Paradigm Potential Problems for the Inhibition Accounts of RetrievalInduced Forgetting: Failure to Replicate with Item Specific Cues Possible Reasons for Failure to Repl icate with Item Specific Cues: Findings: Integration as B oundary Conditions of RIF Semantic Integration as a Moderato r of RIF: Predictions from the Semantic Generalization Hypothesis 1 1 3 7 11 12

PAGE 3

ii Experiment 1: Investigating the m oderating effects of pre-existing associations using a list with strong pre-existing associations Methods Design and Participants Materials Procedure Results Discussion 23 26 26 26 30 36 42 Experiment 2: Investigating the m oderating effects of pre-existing associations with lists that failed to exhibit Retrieval-Induced Forgetting. Methods Design and Participants Materials Procedure Results Discussion 44 45 45 45 50 52 57 Experiment 3: Investigating whether th e moderating effects of pre-existing associations are due to the implicit activation of the pre-existing associations or are mediated by exp licit integration of the exemplars during study. Methods Design and Participants Materials Procedure Results Discussion 59 62 62 62 62 63 76 Experiment 4: Investigating whether th e moderating effects of pre-existing associations can also be due to associations formed through explicit integration strategies likely to be employed when study time is increased. Methods Design and Participants Materials Procedure Results Discussion 78 81 81 81 81 82 95

PAGE 4

iii Meta-analysis Retrieval-Induced Forgetting Benefits of Retrieval Pr actice on Practiced Items Output Interference Episodic Integration Discussion 96 96 98 98 99 101 General Discussion Support for the Semantic Generalization Hypothesis The importance of establishing boundary conditions An additional factor that may have contributed to the failure to replicate with item specific cues Evidence against interf erence accounts of RIF Alternatives to the inhi bitory account of RIF Concluding remarks 107 107 108 111 112 113 119 References Cited 120 Bibliography 127 Appendices Appendix A: Activation Pr edictions made by PIER2 Appe ndix B: Experimental Lists Appendix C: Experime ntal Lists: BetweenSet Associations Appendix D: Experimental Questionnaires 132 133 137 140 152 About the Author End Page

PAGE 5

iv List of Tables Table 1. Associative analysis of practiced and unpracticed competitor sets across several studies that obtained significant RIF (top), and seve ral studies that did not obtain significant RIF (bottom). 15 Table 2. Between-Set and Within -Set strength ch aracteristics, and other variables in the Inter-set (strong) condition. 29 Table 3. Between-Set and Within -Set strength ch aracteristics, and other variables in the Intra-set (weak) condition. 30 Table 4. Between-Triad and Within-Triad strength characteristics, and ot her variables in the Inter-set (strong) condition of Experiment 2. 47 Table 5. Between-Triad and Within-Triad strength characteristics, and other vari ables in Butler et al.s (2001) original list. 48 Table 6. Between-Triad and Within-Triad strength characteristics, and ot her variables in the Intra-set (weak) condition of Experiment 2. 49 Table 7. RIF effect as a function of type of learning and strength of association between the Rp + and Rpitems. 66 Table 8. Retrieval practice benefit e ffect as a function of type of learning and strength of asso ciation between the Rp+ and Rpitems. 68 Table 9. Output interference effect as a function of type of learning and strength of association between the Rp+ and Rpitems. 70

PAGE 6

v Table 10. RIF effect as a function of type of learning and strength of association between the Rp+ a nd Rpitems, and level of episodic integration. 71 Table 11. RIF effect in the Inci dental Learning condition as a function of strength of association between the Rp+ and Rpitems and level of co mparative integration. 73 Table 12. RIF effect in the Inci dental Learning condition as a function of strength of association between the Rp+ and Rpitems and level of compos ite integration (i.e., the average of the episodic and comparative integration scores). 74 Table 13. RIF effect as a function of length of study and the strength of associations between the Rp+ and Rpitems. 87 Table 14. Retrieval practice benefit e ffect as a function of length of study time and strength of a ssociation between the Rp+ and Rpitems. 89 Table 15. Output interference effect as a function of length of study and strength of association between the Rp+ and Rpitems. 92 Table 16. RIF effect as a function of length of study, strength of associations between the Rp+ and Rpitems, and level of episodic integration. 93 Table 17. RIF effect as a function of study time, retrieval practice pattern, and episodic integrati on in the top quartile (high integrators) and bottom quartile (low integrators). 93 Table 18. Results of Meta-analysis: RIF effect as a function of strength of association be tween the Rp+ and Rpitems and level of episodic integration. 100

PAGE 7

vi List of Figures Figure 1. Typical results adapte d from Anderson, Bjork & Bjork, Experiment 3 (1994) using the re trieval practice paradigm. 5 Figure 2. Illustration of how high target -competitor similarity (top panel) and low target-competitor similarity (bottom panel) influence inhibition. 17 Figure 3. Illustration of how high target -competitor similarity (top panel) and low target-competitor similarity (bottom panel) influence inhibition. 19 Figure 4. Illustration of how high co mpetitor-competitor similarity (top panel) and low competitor-competitor similarity (bottom panel) influence inhibition. 21 Figure 5. Two categories used in Experiment 1. 23 Figure 6. Inter-set (strong) condition design. 24 Figure 7. Intra-set (weak) condition design. 25 Figure 8. Schematic of the RIF procedure. 32 Figure 9. Results of Experiment 1: Significant retrieval-induced forgetting of Rpitems only in the Intra-set (weak) condition, but significant strengthening of Rp+ items in both Intra-set (weak) and Inter-set (strong) conditions. 37 Figure 10. Output Interference Effect in Experiment 1: Significant output interference in the Intra-set (weak) condition, but not in the Interset (strong) condition. 39 Figure 11. Results of Experiment 2: Significant RIF effect only in the Intraset (weak) condition (top panel), but significant retrieval practice benefits in both the Intra-set (weak) and Inter-set (strong) conditions (bottom panel). 53

PAGE 8

vii Figure 12. Output Interference Effect in Experiment 2: Significant output interference in the Intra-set (weak), and marginally significant output interference in the Inter-set (strong) condition. 55 Figure 13. Results of semantic integration in Experiment 3: Significant RIF effect only in the Intra-set (weak) condition (top panel), but retrieval practice benefits in both the Intra-set (weak) and Interset (strong) conditions (bottom panel). 64 Figure 14. Results of episodic integrat ion in Experiment 3: Marginally significant RIF in the Intentiona l Learning and significant RIF in the Incidental Learning condition (top panel), and retrieval practice benefits in both learni ng conditions (bottom panel). 65 Figure 15. Output Interference Effect in Experiment 3: Significant output interference in the Intra-set (weak) condition, but not in the Interset (strong) condition. 69 Figure 16. Results of semantic integration in Experiment 4: Significant RIF only in the Intra-set (weak) condition (top panel), but retrieval practice benefits in both Intra-set (weak) and Inter-set (strong) conditions (bottom panel). 83 Figure 17. Results of episodic integrati on in Experiment 4: Significant RIF only in the 5s study condition (top panel), but retr ieval practice benefits in both the 5s and 10s study conditions (bottom panel). 85 Figure 18. Output interference effect as a result of seman tic integration in Experiment 4: Significant output interference in the Intra-set (weak) condition, but not in the Inter-set (strong) condition. 90 Figure 19. Output interference effect as a result of episodic integration in Experiment 4: Marginally signifi cant output interference in the 5s study condition, but no significant ou tput interference in the 10s study condition. 91 Figure 20. Results of the Meta-analyses: Significant RIF effect only in the Intra-set (weak) condition (top panel), but significant retrieval practice benefits in both the Intra-set (weak) and Inter-set (strong) conditions (bottom panel). 97

PAGE 9

viii Figure 21. Meta-analysis of the output interference effect: Significant output interference in the Intra-set (weak) condition, but not in the Inter-set (strong) condition. 99 Figure 22. Retrieval dynamics in the baseline of the Inter-set (strong) condition: Explicit and implicit activation, and in terference on subsequent items in first ha lf of testing sequence. 102 Figure 23. Retrieval dynamics in the baseline of the Intra-set (weak) condition: Explicit and implicit activation, and in terference on subsequent items in first ha lf of testing sequence. 103 Figure 24. Typical results adapted fr om Anderson and Spellman (1995) using the independent probe tec hnique and the retrieval practice paradigm. 117

PAGE 10

ix The Influence of Pre-existing Memori es on Retrieval-Induced Forgetting Leilani B. Goodmon ABSTRACT When people form episodic connections between memories that share a common retrieval cue, the tendency for those memories to interfere in later retrieval is often eliminated, and forgetting of the interferi ng information is reduced. For example, episodic integration protects memories from retrieval-induced forgetting (RIF), a phenomenon in which practicing retrieving so me associates of a cue leads to the suppression of others that interfere w ith retrieval (Anderson, Green, & McCulloch, 2000). The purpose of this study was to dete rmine whether semantic integration, as a result of pre-existing asso ciations between practiced items and their unpracticed competitors, also moderates RIF. This research was motivated by the existence of many pre-existing associations between the pract iced and unpracticed se ts in one study that failed to replicate the RIF effect with item sp ecific cues (Butler, William, Zacks, & Maki, 2001). It was hypothesized that pre-existing as sociations increase th e implicit, semantic integration among the items, insulating them fr om inhibitory effects. The results were consistent with this hypothesis: when associations between the practiced and nonpracticed sets were maximized, no forgetting was observed, however when such associations were minimized, there were reliable levels of RIF. The benefits of semantic integration were replicated across four experiments incl uding one that used Butlers original materials and design. Furthermore, when Butlers items were simply re-arranged in order to minimize the associations and t hus reduce semantic integration, the typical RIF effect was observed. Additional results revealed that the moderating effects of semantic integration are not mediated by explic it integration strategies. Participants who received incidental learning instructions and so reported very low levels of episodic integration, still exhibited the sa me benefits of semantic inte gration. Finally, it was also shown that increasing the use of explicit integration strate gies by increasing the study time, also reduced the RIF effect. The results of the current set of studies reveal that failure to control for pre-existing associa tions may account for variability in the RIF phenomenon. The results also s uggest that the memory system is adaptive to the needs of the organism, in that it operates to keep re lated memories that are necessary for cognition active, but suppresses interfering memories.

PAGE 11

1 Introduction: Concept of Inhibition and Inhibitory Control as an Adaptive Mechanism The need to stop or redirect unwanted actio ns or even to resi st distraction is crucial to normal functioning in every day life. Without these abilities we would not be able to adapt our behavior according to changes in our goals, or changes in the environment. For example, imagine a situ ation in which you were startled by someone from behind. Your reflexive response may be to lash out to protect yourself from the grasp of some stranger. As your hand extends out to fight off any danger, you realize that it is your significant other. To their relie f you are able to stop the movement of your hand before it smacks them. In this situa tion, you stopped the propone nt reaction to fight or flight from a perceived threat, once the threat was appropriate ly re-assessed. How might this be accomplished? One explanation is that we have inhibitory mechanisms that suppress unwanted responses or memories. In hibition refers to the active reduction in the level of activation for the representation of an action or memory thereby reducing its tendency to interfere with the current goals of the system. The use of inhibition to achieve control over actions or memories serves an ad aptive function. In th e situation described earlier, inhibitory control may save you from expending highly valued energy, damaging a valued relationship, or causing unn ecessary injury to your own body. In a similar fashion, inhibitory control can be utilized to resist distracting information. For example, during conversation it is necessary not only to maintain in working memory the contents of the current discussion, but also to prevent information that interferes with the current mainte nance goals from entering into conscious awareness. Imagine if you had no such inhi bitory control, such as someone with Tourettes Syndrome. Individuals with Touret tes and other disorders such as obsessivecompulsive disorder (O-CD) are believed to su ffer from an inhibitory inefficiency. This inhibitory inefficiency results in uncontrol lable speech and body movements in the case of Tourettes, and invasive a nd disturbing thoughts in the case of O-CD. In fact, research indicates an important link between these diso rders and measures of inhibition. Enrich and Beech (1993), for instance, showed that individuals with O-CD suffer from more interference from both external and internal stimuli. These st udies highlight the important role of inhibitory control, or th e use of inhibition to achieve control, in regulating and controlling thought and behavior (Bjorklund & Harnishfeger, 1995). Inhibitory control may also help us in ot her ways including re gulating our emotional state, maintaining attention on current events maximizing available resources to improve task performance, or minimizing compet ition or interferen ce during learning or remembering.

PAGE 12

2 The view that inhibitory mechanisms are involved in human information processing has been the subject of much discus sion in areas ranging fr om neural research to research in the area of development (Har nishfeger, 1985) and cognitive aging (Hasher, Stolzfus, Zacks, & Rypma, 1991; Lustig, Hasher, & Toney, 2001; McDowd, OseasKreger, & Kilion, 1995; Holley, & McEvoy, 1996). Mu ch of this evidence is consistent with the existence of inhib itory processes, though there is not universal agreement on how these inhibitory processes operate, what tasks provide pu re indices of inhibition, or even how the inhibitory account fares agains t non-inhibitory explanations (Hamilton & Martin, 2005; Friedman & Miyake, 2 004; MacLeod, Dodd, Sheard, Wilson, & Bibi, 2003). Nevertheless, recent research suggests that theories of memory may need to incorporate inhibitory mechanisms. One important question for theories of memory is how the system distinguishes related, competing information from related in formation that is important for the current needs of the system. That is, while it is ad aptive to resist interference from certain types of information during retrieval, it may not be adaptive to prevent other types of information from becoming active, because that information may be integral to understanding. For example, trying to remember that horses gallop should not make us forget that ponies can also gallop Also reading a passage about horses should make it easier to process the word saddle not make it worse. These examples highlight the need for some parts of a representation to be resi stant to inhibitory processes.

PAGE 13

3 Retrieval-Induced Forgetting and the Retrieval Practice Paradigm Some evidence for the role of inhibitory processes in memory retrieval has come in part from research on the retrieval-induced forgetting effect (her einafter abbreviated the RIF effect or RIF). Much of this wo rk has used the retrie val practice paradigm, whereby the retrieval practice of certain items makes other items less accessible (Anderson, Bjork, & Bjork, 1994; Anderson & Spellman, 1995; see Anderson 2003, for a review). RIF refers to the finding that reme mbering can cause forgetting of information that interferes with what is being retrieved. It has been s hown that the repeated retrieval of a given item results in the loss of retrie val access to other items that interfere with retrieval of the target item According to Anderson and colleagues (Anderson et al., 1994; Bjork, 1989) inhibition is a mechanism of forgetting that influences memories related to the cue so that they do not interfer e with remembering the target. They assume that this inhibitory mechanism renders non-target memories less accessible for subsequent recall, and thus facili tates the recall of the target. In order to test this assumption, they developed the retrieval-practice paradigm. There are four phases in a typical design: (1 ) a study phase, (2) a re trieval-practice phase, (3) a retention interval phase, and (4) a final test phase. First, part icipants study categoryEXEMPLAR pairs (e.g., Weapon-SWORD, Weapon-RIFLE, Profession -DENTIST and Profession -PLUMBER). After studying the cate gory-EXEMPLAR pairs, participants practice retrieving half of the exemplars from half of the categories. In our example, participants would practice re trieving SWORD to the cue, Weapon SW_______. These practiced items are denoted as Rp+ items, because they are assumed to receive some additional activation and so are more strongly represented and more accessible in longterm memory. Unpracticed items from the practiced categories, such as RIFLE, are denoted as Rpitems, because they are assume d to be inhibited in long-term memory. Unpracticed items from unpr acticed categories, such as DENTIST and PLUMBER, provide a baseline measure of recall performance, and are denoted as Nrp items. Nrp items provide baseline measures because th ere is no retrieval practice on any of its exemplars and so there is no interference as a result of retrieval and therefore no need to suppress any interfering information. After the retrieval practice phase, there is a 20 minute retention interval phase in which participants typi cally complete a series of reasoning problems. The reten tion interval phase is include d to demonstrate that the effects of inhibition are somewhat long lasting. In the final phase, there is a memory test for all exemplars presented in the study phase. Recall can be tested in a number of ways including cate gory cued recall or category-plus-letter-stem cued recall. In cat egory cued recall participants are given the categories ( Weapon Profession ) one by one and are asked to recall as many exemplars as they can remember from each category. However, with this type of test there is no way to control for output interfer ence on unpracticed items. Output interference refers to the finding that the probability of recalling an ite m declines as a function of the number of items already elicited ((Rundus, 1973; Smith, 1971) Various theories (e.g., Raajmakers & Shiffrin, 1981; Rundus, 1973) suggest that retr ieval of an item duri ng test strengthens it representation and this strengthened item then interferes with subsequent recall of other

PAGE 14

4 study items. Typically in cat egory cued recall, the practiced items are recalled first because they were strengthened during the re trieval practice phase. Recalling them first in the testing sequence strengthens them even more causing them to become even stronger responses, and increasing the likelihood that they will interf ere with subsequent recall. Thus category cued recall increases output interf erence on unpracticed or weak items making them even less likely to be recalle d. In this fashion, forgetting also occurs during the test phase and there is no way to discern if the forget ting is due to inhibition as a result of retrieval practi ce or output interfer ence as a result of recalling strong items first in the testing sequence. In order to control for output interference and to help es tablish that the effect is long-term resulting from retrieval practice, a category-plus-letter stem cue can be used in which the category is presented along with the first letter of one member from that category (e.g. Weapon S_______). With this type of test, the weak items (i.e. unpracticed exemplars) can be tested first in the testing sequen ce, eliminating the problem of output interference. In categor y-plus-stem cued recall, the categories are typically presented in a blocke d presentation, and all weak or unpracticed items are tested first followed by the strong or practiced items. Rpitems are always tested first in their category and are always compared to Nrp items (i.e. baseline items) that are tested first in their category, ensuring that Rpand Nrp items have the same average serial position in the final test phase. Typical findings us ing the category-plus-letter stem cue are illustrated in Figure 1.

PAGE 15

Practiced Category Unpracticed Category 5 Weapon RIFLE .57 SWORD .80 Profession PLUMBER .65 DENTIST .65 Practiced Unpracticed Unpracticed Unpracticed EXEMPLAR EXEMPLAR EXEMPLAR EXEMPLAR Rp+ item Rpitem Nrp item Nrp item Figure 1. Typical results adapted from Anderson, Bjork & Bjork, Experiment 3 (1994) using the retrieval practice paradigm. Retrieval-induced forgetting effect: Rpitem recall (.57) minus Nrp baseline recall (.65), yielding a forgetting effect of 8%. In Figure 1, the numbers in each node indicate the percentage of items correctly recalled on the final recall test. In this example, only Weapon-SWORD received practice, as indicated by the bold line. The amount of RIF is calculated by comparing recall of the unpracticed exemplars from practiced categories (e.g., RIFLE (Rp-)) with recall in the baseline condition (Nrp). As shown in Figure 1, RIFLE (Rp-) was recalled at a lower rate to the cue Weapon S_____ in the final test phase (.57), compared to the recall of DENTIST or PLUMBER to the cues Profession D____ (.65) or Profession P____ (.65). Because the item-specific cues control for output order, output interference is not a viable explanation of the observed forgetting effects. The interpretation offered by Anderson and colleagues (1994) instead is that retrieval practice of SWORD to Weapon SW______ inhibits other weapon responses, such as RIFLE, that interfere with the recall of SWORD. Thus, when confronted with recalling all Weapon items at test, the response RIFLE is less accessible compared to unpracticed category-exemplars, such as DENTIST and PLUMBER, because those latter items did not suffer from inhibition during the retrieval practice phase. It is important to note this argument was premised, to a large extent, on the interaction of this effect with taxonomic frequency or the strength of the exemplar with its category, not just the presence of the effect when output order was controlled. Specifically, RIF effects are typically not found when the Rpitem is a weak member of its category (see Williams & Zacks (2001), for an exception). This predicted interaction was based on the assumption that only interfering information needs to be suppressed. It was assumed that weak members do not interfere with retrieval practice because they are not likely to be activated by the category cue. Strong members, however, are highly

PAGE 16

6 likely to be activated by the category cue and so should interfere with retrieval practice. Subsequently, strong exemplars will be suppres sed in order to successfully retrieve the desired exemplar from memory. This in terference dependence property of retrievalinduced forgetting highlights the adaptive functi on of an inhibitory mechanism to exert control over the contents of memory. In order for forge tting to be observed, there has to be some need to overcome interference. Without interference there is no need to suppress information and so no forgetting is evident. The RIF effect is well documented a nd has been found in several different laboratory settings and with a variety of stimuli (see Anderson, 2003, for a review), including colored objects (Ciranni & Shimamura, 1999), propositions (Anderson & Bell, 2001), performed actions (Koustaal, Schact er, Johnson, & Galluccio, 1999), imagined actions (Macrea & Roseveare, 2002), crime scene details (MacLe od, 2002; Shaw, Bjork, & Handel, 1995), event information (Saunde rs & MacLeod, 2002), personality traits (Dunn & Spellman, 2003; Macrae & MacLeod, 1999; White, 2003), emotional and unemotional autobiographical memories (Barnier, Hung, & Conway, 2004), false memories (Starns & Hicks, 2004), and even mathematical representations (Phenix & Campbell, 2003). The RIF effect also appears to be quite robust in that it is also found on tests of recognition memory (Hicks, & Starns, 2004) an d on tests of conceptual implicit memory (Perfect, Moulin, Conway, & Perry, 2002; Camp, Pecher & Schmidt, in press). It also is not limited to college-age individuals, for it has been demonstrated in those with Alzheimers disease (Moulin, Perfect, Conw ay, North, Jones, & James, 2002) and even children (Zellner & Bauml, in press, Fo rd, Keating, & Patel, 2004). Finally, the inhibitory effect also manifests itself the area of semantic memory (Bauml, 2002; Blaxton & Neely, 1983; Johnson & Anderson, 2004; Carter, 2002). In summary, the RIF effect is generalizable to various types of stimulus classes and manifests its effects in a variety of different psychological phenomena.

PAGE 17

7 Potential Problems for Inhibitory Acc ounts of RetrievalInduced Forgetting: Failure to Replicate with Item Specific Cues Despite the fact that the RIF effect has been replicated across a wide range of laboratory settings and domains, Butler, Williams, Zacks, and Maki (2001) found a much reduced RIF effect with category cued recall (e.g. bird ), and actually failed to find the item specific cueing effect using several different item specific cues, including categoryplus-two-letter-stem cues (e.g. bird SP________ Note: cue for SPARROW), categoryplus-fragment cues (e.g., bird _P_R_OW), fragment cues (e.g., _P_R_OW), and implicit fragment completion cues (e.g., _P_R_OW). In their study, participants studie d 60 category-exemplar pairs (e.g. bird SPARROW, drink ALE). Participants were told that they would have eight seconds to study each pair and to spend this time relating the exemplar to its category name. After the study phase, participants we re informed that their memory would be tested with the category name and the first two letters of one exemplar that had been paired with that category in the study phase (e.g., bird SP_______). The experimenter explained that the first two letters of the exemplar were give n to provide a hint, and they were given 10 seconds to write in the remainder of the exem plar. After completing the retrieval-practice phase, participants worked on two retention inte rval tasks, the Shipley Institute of Living ScaleVocabulary Test (Shipley, 1940) a nd a picture fragment completion task (Snodgrass & Vanderwart, 1980; Irwin & Schi ppits, 1979;, De Graef, Christiaens, & dYewalles, 1990). Participants only comple ted 20 minutes of the retention interval tasks, in four of the five different types of final recall tests used in the experiment. However, when participants received the implicit fragment completion task in the final test phase, they had to identify all 90 pictur e fragments in the retention interval before going on to the final test phase of the experiment, even if this took more than 20 minutes. No mention was made about how much longer th e retention interval was for participants who received implicit fragment completion in the fi nal test phase. In the final test phase, participants received one of five types of tests; category cued recall (e.g. bird ), categoryplus-two-letter-stem cued recall (e.g. bird SP________), category-plus-fragment cued recall (e.g., bird _P_R_OW), fragment cued recall (e .g., _P_R_OW), or implicit fragment completion (e.g., _P_R_OW). Butler and colleagues found a significant (o ne-tailed) RIF effect of 5.2 %, for participants who received categ ory cued recall in the final test phase. This forgetting effect is smaller than what has typically b een found in other RIF studies using category cues that have found forgetting effects in the range of 10% to 25% (Anderson, 2003). The observation of such a small effect is surprising from the standpoint of both the inhibition theory and interference accounts of RIF, because significant forgetting is expected, if only because of output interfer ence from the stronger practiced items during the final test (a point which I will return to later). As mentioned previously, in category cued recall, strong items (i.e. Rp+ items) are typically recalled first in the testing sequence and so they interfere with recall of weak items (i.e. Rpitems). Proponents of interferen ce explanations of RIF argue that in order for the RIF effect to be due to an inhibitory mechanism, the effect must also be observed

PAGE 18

8 with item specific cues that control for out put order and so eliminate the influence of output interference. However, Butler and colleagues failed to find the RIF effect with all four different types of item specific cues. With the category-plus-two-letter-stem cued recall test, they actually found facilitation of Rpitems ( 3.0%), though the facilitatory effect was not significant. They failed to find significant forgetting on the category-plusfragment cued recall test (-1.3%), the frag ment cued recall test (-1.6%), and on the implicit fragment completion test (-1.3%). Because they found RIF on the category cued recall test where output interference was pres ent, but failed to fi nd it on all four item specific cueing tests, where output interference was absent, Butler and colleagues concluded that forgetting in the retrieval pr actice paradigm may not be a reliable outcome when item-specific cued recall tests are used. Proponents of interfer ence explanations of RIF take Butlers failure to find the item sp ecific cuing effect as evidence that RIF is unlikely to involve inhibition and the observed forgetting in the category cued recall test is more likely due to interference or blocking as a result of recalling strong items (i.e. practiced items) first in th e final testing sequence. The findings of Butler and colleagues are im portant because they suggest that the RIF effect may not be a reliable outcome when the final recall test uses item specific cues. Also because they failed to find the RIF effect with item specific cues, proponents of interference accounts of RIF take this as evidence that the RIF effect is better explained in terms of interference rather th an inhibition. However, there are several reasons to doubt these conclusi ons. For example, there are several studies that found the RIF effect with item specific cues (Ande rson et al., 1994; Anderson & McCulloch, 1999; Anderson, Bjork, & Bjork, 2000; Anderson, Green, & McCulloch, 2000; Anderson & Bell, 2001). For example, Anderson and coll eagues (1994) found a 7% RIF effect with category-plus-two letter stem cues (e.g., Fruit Or_______) (Experiment 2), and they found a 10% RIF effect with categor y-plus-1st letter stem cues (e.g., Fruit O_______) (Experiment 3). Anderson and McCulloch (1999 ) found a 7% RIF effect in experimental conditions in which participants were less likel y to integrate or form episodic associations between the items during the study phase (E xperiment 3). Anderson, Bjork, and Bjork (2000) found a 10% RIF effect with categ ory-plus-1st letter stem cues (e.g., Fruit O_______), when retrieval practice was competitive ( Fruit Or_______) compared to when retrieval practice was non-compet itive (e.g., Orange Fr________). Another especially noteworthy study that found RIF effects with item specific cues was a study by Anderson, Green, and McCull och (2000). They found a 7% to 10 % RIF effect with category-plus 1st letter stem cues (e.g., Food C____) in two experiments when there was a high degree of similarity be tween unpracticed exemplars. In this study, the category cue in the final test phase wa s actually a novel category that had not been studied in the encoding phase but was sema ntically related to the tobe-retrieved exemplar. This study was extremely important for inhibitory accounts of RIF because it helped establish that not only is the RIF effect evident with item specific cues but also that the forgetting effect is cue independent (Anderson & Spellman, 1995). That is, the inhibition acts on the specific representati on in memory so any cue that normally produces it is less likely to do so, a finding that is difficult for interference theories to explain. Interference accounts suggest that forgetting of the non-practiced item occurs

PAGE 19

9 because the cue that is used strongly activat es the practiced items and so the practiced items simply block access to the weaker, non-pr acticed items. Therefore, interference accounts predict that when an independent cue is used that does not activate the practiced item, no forgetting of the non-practiced items should be found. However, the results clearly establish that when the non-practiced items are cued w ith an independent cue, the typical forgetting effect is observed. Anderson and Bell (2001) also found the item specific cueing RIF effect with a different type of stimuli, namely propositions. They found significant forgetting of facts such as The actor is looking at the violin , when related propositions were retrieval practiced, such as The actor is looking at the tulip. This forgetting effect was observed despite the fact that a very specific cue was used in the test phase, such as The actor is playing the ________. In many of the experime nts, they also provided a letter stem, such as The actor is playing the v_______. In fact, there are seven experiments in this paper that found RIF effects using various kinds of item specific cues. In addition to the studies of Anderson and colleagues, many others have found RIF with item specific cues (Bauml, 2002; Bauml & Hartinger, 2002; Ciranni & Shimamura, 1999). For example, Bauml (2002) found an 8% RIF e ffect with categorysubcategory-plus-1st letter stem cued recall (e.g., Four-legged Animal Hoofed H______). In two experiments, Bauml and Hartinger ( 2002) also showed a 7.4% to 10% RIF effect with category-plus 1st lette r stem cued recall (e.g., Animal T_____), when the practiced and unpracticed items were dissimilar, that is they belonged to the same category (e.g., Animals ) but belonged to differe nt subcategories (e.g., Hoofed vs. Predators ). In several experiments, Ciranni and Shimamura (1999) also found RIF with colored objects using item specific cues (e.g., the objects). In addition to cued recall, the RIF effect has also been found with an even more item specific test, the recognition test. Hick s and Starns (2004) demonstrated in two experiments that RIF occurs in tests of item recognition and Ford, Keating, and Patel (2004) found RIF with a recognition test in Experiment 2. Recently, Veling and van Knippenburg (2004) used a cue independent re cognition test to examine RIF on reaction time measures. Their cues were item specifi c because the exemplars were presented to the participants for recognition, and they we re independent because the category label was not presented along with the exemplars. They found that Rpitems were recognized more slowly than baseline items, and were responded to more slowly in a lexical decision task, suggesting that the items were inhibite d resulting in them being less accessible to the recognition and lexical decision task. Alt hough all Rpitems were not tested first in the testing sequence, all items were presented randomly and exemplars did not have to be overtly reported, so it is unlikely that output interference was causing slowing of responses to Rpitems. It is also unlike ly that blocking caused slowing of responses because the effect was found with independent cues. The results of Veling et al. (2004) therefore extend the item specific cueing eff ect and cue independent property of RIF to tests of item accessibility. In summary, the studies described here overwhelmingly reinforce the conclusion that RIF effects do occur on item specific tests. Furthermore, the results of these studies support inhibitory accounts of RIF by pr oviding evidence against interference

PAGE 20

10 explanations. First, interfere nce is not a viable explanat ion of the observed forgetting because the item specific cues that were us ed controlled for the effects of output interference. Second, the studi es established that the RIF effect is cue independent, a property of RIF that interf erence accounts have a difficult time reconciling.

PAGE 21

11 Possible Reasons for Failure to Re plicate with Item Specific Cues: Integration as Boundary Conditions on RIF Episodic Integration as a Boundary Condition on RIF Given the vast array of st udies that found the RIF effect with item specific cues that controlled for output interference, conclusions made by proponents of interference accounts purporting that the RIF effect is due to interferen ce or blocking, seem unlikely to be correct. Nevertheless, Butler and colle agues findings raise im portant questions as to why they did not observe RIF effects with item specific cues. There is evidence that certain experimental factors can influence th e magnitude of the RIF effect. One major factor that has been shown to moderate RIF, is the use of explicit integration strategies during the study phase. Anderson and McCulloch (1999) showed that the amount of RIF depends on how well integrated the to-be-retrieved memories are with the practiced competitors. Integration refers to interconne ctions formed either on the basis of preexisting associations, on episodic associati ons developed during the experiment, or on both. It is assumed that integration serv es as a moderator of the RIF effect by insulating some related items from inhi bition. In support of this assumption, they found that participants who we re explicitly instructed to rehearse and inter-relate the items during encoding showed a reduced RIF effect. In addition, other studies have shown that some participants, without being asked to, spontaneously integrate the items by rehearsing them together during encoding, as measured by a post-experimental que stionnaire (Anderson & McCulloch, 1999; Anderson & Bell, 2001). Participants who inte grated the items in this fashion also exhibited far less forgetting. The moderating effects of episodic integration are not surprising given the evidence for its protective effects in other domains of forgetting. For example, in a study of fan effects, Radvansky (1999) found a similar pattern of less forgetting when participants integrated propositional knowle dge. By integrating facts into more cohesive representations, particip ants appear to experience less interference between related facts, require less inhibiti on, and are thus protected from impairment (Anderson & Bell, 2001; Smith, Adams, & Schorr, 1978). Another very important study by A nderson, Green, and McCulloch (2000) revealed that in order for episodic integrat ion to protect Rpitems from forgetting, the integration has to be formed between the Rp+ and Rpitems. They varied the specific pattern of episodic integrati on between the items within a category by varying whether participants explicitly encoded similarities or differenc es between them. They found that the specific pattern of episodic integration that occurs really matters. When participants formed episodic associations between the Rp + and Rpitems by encoding similarities between them (i.e. high target-competitor si milarity), not only were the Rpitems protected from forgetting, they were actually recalled at a significantly higher rate than baseline items (+7% facilitation). Howe ver, when participants formed episodic associations between the Rpand other Rpit ems by encoding similarities between them (i.e. high competitor-competitor similarity) Rp items were not protected from forgetting and a reliable RIF effect was observed (10% RI F effect). This study provides additional evidence of the benefits of episodic integr ation and reveals that the benefits are

PAGE 22

12 dependent on the specific and critical inte gration between the Rp+ and the Rpitems. Episodic integration between Rpitems has no beneficial effect. The findings of Anderson et al. (2000) reveal that explicit in structions to integr ate the exemplars by encoding similarities during the study phase creates more cohesive representations that are readily formed on the basis of epis odic associations, and these integrated representations appear to be protected from inhibition when a representation with shared features (i.e. a similar representation) is strengthened through re trieval practice. Semantic Integration as a Possible Boundary Condition on RIF As mentioned previously, in tegration can be formed on the basis of episodic associations developed during the experiment or on the basis of pre-existing knowledge or pre-existing associations. In fact, th e inhibitory model developed by Anderson and colleagues to explain the benefits of episodic integration on RIF, the Distributed Representation Approach (Anderson and Sp ellman, 1995; Anderson et al., 2000) makes no distinction between associa tions formed episodically or semantically. According to the Semantic Generalization hypothesis of this inhibitory theory, the strengthening effects of retrieval practice can generalize implicitly to semantically related competitors (i.e. Rpitems), thereby increasing their implicit ac tivation and saving th em from inhibitory effects. Anderson, Bjork, and Bjork (1994) an ticipated that pre-ex isting associations could serve as a boundary condi tion of RIF by increasing the likelihood that semantically related items would be implicitly integrated during retrieval practi ce. They increased their chances of obtaining significant RIF eff ects because they ensured that there were minimal inter-set associations between practiced items and th eir unpracticed competitors. Although its specific effects on RIF have not been tested empirically, there are several reasons to assume that semantic integr ation based on pre-existing associations can reduce the RIF effect, and so in order for RI F effects to be observed, the associations between the practiced and non-practiced sets must be redu ced to the greatest extent possible. First, pre-existing associations have been shown to play an important role in retrieval and the retrieval practice paradigm has a significant retrieval practice component (Nelson & Goodmon, 2002; Goodmon & Nelson, 2004). Second, as described previously, associations reduced RIF when participants were explicitly instructed to focus on and encode these associations (Ande rson, Green, & McCulloch, 2000). Finally, semantic integration has been shown to have a protective effect in other inhibitory domains such as directed forgetting even when participants are not given explicit instructions to integrate. For exampl e, Conway, Harries, Noyes, Racsmany, and Frankish (2000) found a reduced directed forg etting effect when five items on the to-beforgotten list were strongly associated to fi ve other items on the to -be-remembered list. Recent unpublished work by Sahakyan and Goodmon (in prep), revealed similar moderating effects of pre-existi ng associations on directed forgetting. When there were strong associations, as measured by Nelsons free association norms, between the to-beforgotten list and the to-be-remembered list, there was significantly less directed forgetting (-.03) compared to when the lists were completely unrelated (-.10). In the studies by Conway et al. and in those by Sahaykan and Goodmon, no mention was made to the participants about the relationships betwee n the lists, nor were they told to integrate

PAGE 23

13 the lists, and nor were they told to find si milarities among the lists. These results reveal that semantic integration can serve as a moderator of forgetting at least within the directed forgetting domain, and th is moderation occurs without explicit cues to integrate the items. Given the important influence of epis odic integration on RIF and the possible influence of semantic integration on RIF, one reason that Butler and colleagues failed to find RIF with item specific cues is that they may have increased semantic integration by not controlling the critical inter-set associati ons or associations be tween the practiced and unpracticed sets of exemplars. In fact, an analysis of Butler and colleagues list using the Nelson free association norms revealed th ere were many associations between the practiced (Rp+) and unpract iced sets (Rp-) (Nelson, McEvoy, & Schreiber, 1999). Specifically, there were 39 direct associations between practic ed and unpracticed sets. Of the direct associations, 24 were moderate-to-strong associations, with an average forward and backward strength of .07 and .07, respectively, between exemplars that were retrieval practiced and exemplars from th e same category that were not retrieval practiced. In addition, there were approximately 332 indirect associations (i.e. shared associates) between practiced items and their unpracticed competitors from the same category. From this analysis, it appears that th eir experiment failed to control for associations between exemplars that were re trieval practiced and those that were not retrieval practiced. The pattern of retrie val practice they employed maximized the critical, pre-existing interrelationships betw een the practiced and unpracticed sets and may have contributed to an at tenuated RIF effect. In fact, their design is comparable to the target-competitor si milarity condition in Anderson et al. (2000), where no RIF was found. However, instead of integrating on the basis of associations formed episodically during encoding, the items may have been integrated on the basis of pre-existing associations activated implicitly during re trieval practice. Because Anderson and colleagues (2000) only varied epis odic similarity and did not act ually vary pre-existing or semantic similarity in their study, it is unc lear whether the exis tence of pre-existing associations moderates RIF, independent of explicit instructions to integrate the exemplars during encoding. A major aim of th is dissertation was to examine the effects of semantic integration on RIF. In doing so, we evaluated whether pre-existing associations are an important boundary condi tion on RIF, and whether this may account for Butlers findings. In the case of RIF, im plicitly activated associations between the practiced and unpracticed sets may insulat e the unpracticed items from inhibition and thus serve as an important mode rator of the forgetting effect. To date, there is only one study that experimentally manipulated a priori similarity between the practiced and unpracticed sets, to determ ine its effects on RIF. In agreement with the aforementioned pred ictions, Bauml and Ha rtinger (2002) found reduced RIF when there was a high degree of similarity between the practiced and unpracticed exemplars. In their experiment, Bauml and Hartinger varied the similarity between the practiced sets and their unpracticed counterparts, and they did not explicitly instruct their participants to integrate the items in the study phase. However, they made similarity among the items explicit by presenti ng a subcategory label in addition to the

PAGE 24

14 category label (e.g., tree conifer PINE, tree deciduous BIRCH). They manipulated similarity between the practi ced and unpracticed exemplars by varying whether or not the unpracticed exemplars (e.g. tree conifer PINE) were drawn from the same subcategory (e.g. conifers ) as the practiced items (e.g. tree conifer CYPRESS), or a different subcategory (e.g. deciduous, such as tree deciduous BIRCH). Because subcategories were explicitly presented to the particip ants, their study does not provide sufficient evidence that incidental similarity is sufficient to reduce the RIF effect. Also, given that they explicitly presen ted a linking concept, they might have gotten similar results if instead of presenting tree conifer CYPRESS and tree conifer PINE, they presented tree one CYPRESS and trees one BIRCH. In the latter case, CYPRESS and BIRCH are from different subcategories but may be integrated episodically by the verbal mediator, one. If they found similar resu lts with the latter case despite the fact that CYPRESS a nd BIRCH are the members of different subcategories, it would suggest that it was si mply the presence of a verbal mediator and not incidental similarity that moderated the RIF effect (for reviews, see Horton & Kjeldergaard, 1961; Jenkins, 1963; Kjelderg aard, 1968; Postman, 1971). The purpose of the present set of experiments was to eliminat e all such explicit cu es to organization to see if implicitly activated associations ar e sufficient to influence the RIF effect. Using Nelsons free association database, th e associative structure of the materials from several RIF studies were examined to de termine if variability in the RIF effect was related to variability in association strengths between the practiced and unpracticed sets (Anderson, Bjork & Bjork, 1994; Anderson & McCulloch, 1999; Anderson, Green & McCulloch, 2000; Bauml & Hartinger, 2000; Butler, Williams, Zacks & Maki, 2001; Williams & Zacks, 2001). Studies were included if they used item specific cues such as category-plus-stem cued recall (e.g. Animal HO______) or category plus 1st letter stem cued recall (e.g. Animal H________), because these types of tests are more diagnostic of inhibition. In addition, this an alysis only included those studie s that used standard study instructions in which no mention was made to episodically integrate or to encode similarities among the items, because episodic integration has been shown to reduce the RIF effect (Anderson & McCulloch, 1999; Anderson, Green & McCulloch, 2000). The specific studies that were in cluded, length of study time, nu mber of associative links, mean associative strength, and the amount of RIF obtained are reported in Table 1. Results reveal that differences in the RIF effect coincide with differences in the number of associative links as well differences in the mean associative strength between the practiced items and their unpracticed competitors.

PAGE 25

15 Table 1. Associative analysis of practiced and unp racticed competitor sets acr oss several studies that obtained significant RIF (top), and several studies that did not obtain significant RIF (bottom). Studies that DID obtain significant RIF Study Time Test Type Number of Associative Links Mean Associative Strength RIF Effect Anderson, Bjork, & Bjork (1994), Exp 2 5s Category + 2 letter stem 9 .04 -7.60* Anderson, Bjork, & Bjork (1994), Exp 3 5s Category + 1 letter stem 14 .01 -9.40* Anderson & McCulloch (1999), Exp 3 4s Category + 1 letter stem 14 .04 -7.00* Anderson, Green, & McCulloch (2000), Exp 1 3s Category + 1 letter stem 1 .02 -10.00* Bauml & Hartinger (2000), Exp 1 6s Ca tegory + 1 letter stem 3 .005 -8.28* Mean: 3.20 .023 -8.23 Studies that DID NOT obtain significant RIF Bauml & Hartinger (2000), Exp 1 6s Ca tegory + 1 letter stem 15 .14 -1.80 Butler, Williams, Zacks, & Maki (2001), Exp 1 8s Category + 2 le tter stem 24 .07 3.00 Mean: 19.50 .11 .60 Note Number of Associative links refers to th e total number of links, both forward and backward, between the practiced and unpracticed sets across all categories. Mean Associa tive Strength refers to the average strength, both forward and ba ckward, of the total number of asso ciative links. Numbers in the last column indicate amount of RIF, negative numbers refl ect amount of forgetting of Rpitems relative to baseline, and positive numbers reflect facilitation of Rprelative to baseline. Note indicates significance at p <.05. As shown in top portion of Table 1, the pres ence of an effect (i.e. larger negative numbers in the last column) coincided with fewer and weaker associative links between all of the practiced items and their unpracticed competitors, compared to studies that did not obtain significant RIF effects with item specific cues. As shown in the bottom portion of Table 1, there were greater and stronger associative links in the studies that failed to obtained RIF effects (i.e. smaller negative numbers or larger positive numbers in the last column). The average forgetting eff ect across the studies th at obtained significant RIF was -8.23, and these studies had relatively few associative links ( M = 3.20) that were weaker on average ( M = .023). In contrast, the average forgetting effect across the studies that did not obtain si gnificant RIF was .60, and these st udies had relatively greater numbers of associative links (M = 19.50) that were stronger on average ( M = .11). While there are certainly many differences ac ross the studies that were included in this examination including length of study time, the results are striking in the sense that they highlight a possible relationship between the absence of forgetting and the presence of pre-existing associations. What is especi ally striking is that the presence of preexisting associations between the practiced and unpracticed sets appears to moderate the RIF effect in absence of any instructions to episodically integrate the items. These findings are consistent with empirical evidence from other inhibitory domains. As mentioned previously, the protective effects of semantic in tegration have been observed in studies of directed forgetting. Because dir ected forgetting was attenuated in absence of explicit cues to integrate the items, the resu lts suggest that pre-existing associations can insulate items from inhibitory effects and that this insulation effect can occur when

PAGE 26

16 associations are activated implicitly. Howeve r, because the directed forgetting paradigm includes intentionality on the part of the part icipant to forget, it is unclear whether this insulation effect would genera lize to a retrieval-induced fo rgetting situation in which participants are not inte ntionally trying to inhibit. We al so do not know if the participants noticed the associative relationships betw een the to-be-forgotten list and the to-beremembered list and then used these links durin g the test phase to help them recover the forgotten items. The purpose of the current set of studies was to examine whether semantic integration as a result of th e implicit activation of strong pre-existing associations between the practi ced and their unpracticed competitors, as measured by free association moderates retrieval-induced fo rgetting independent of any intentional forgetting or explicit recovery strategies on the part of the participant. It is also important to note that th e possible moderating factor described and investigated here is very diffe rent from the moderation of i nhibition due to the strength of the association between the exemplar and the category. We were not interested in the moderating effects of taxonomic frequency or the strength between the category and the exemplar on RIF. Instead, we were interest ed in the moderating effects of semantic integration or the strength between the exem plars. In other words, the relationships addressed here were those between the exemplars, and not between the exemplars and the category. It has already been shown that strongly related ex emplars of the category (i.e. high taxonomic frequency members) are more like ly to be inhibited than weakly related exemplars, because strong exemplars are more likely to intrude and compete for recall during retrieval practice (A nderson et al., 1994; Johnson & Anderson, 2004). Weak members of the category are less likely to be activated by the category and so they are less likely to interfere with re trieval of other exemplars from that category. Because they are less likely to interfere, there is little need to inhibit those weak items. This interference dependence property of RIF was demonstrated in previous studies that minimized the associations between the pr acticed and unpracticed exemplars. The purpose of the current set of experiments was to show that even strong taxonomic members that are normally inhibited when pre-existing associations are minimized, can nevertheless be saved from inhibition, when pre-existing associations between the practiced and non-practiced sets are maximized (i.e. when a semantically related associate of the high taxonomic exemplar is retrieval practiced).

PAGE 27

Semantic Integration as a Moderator of RIF: Predictions from the Semantic Generalization Hypothesis The predictions regarding the effects of semantic integration were derived from the same theoretical model of inhibition used by Anderson et al. (2000) to explain when and why episodic integration has a beneficial effect. According to the model, a particular item is represented by various features that are distributed throughout the semantic network. As shown in the top panel of Figure 2, the more semantically or episodically related two items are, the greater the feature overlap between the two representations. As shown in the bottom panel, the less semantically or episodically related the two representations are, the lower the feature overlap. High Target-Competitor Episodic or Semantic Similarity RpRp+ X X X Low Target-Competitor Episodic or Semantic Similarity RpRp+ X X X X X X X X Figure 2. Illustration of how high target-competitor similarity (top panel) and low target-competitor similarity (bottom panel) influence inhibition. Large circles represent memory items. Small circles, squares, and triangles represent various semantic features. Strengthened features are darkened and inhibited features are X-ed. According to the model, when an item is retrieval practiced, all of its features are strengthened (i.e. darkened objects), including those that happened to be shared with other, potentially competing items; however, features that are unique to the competing items are inhibited (i.e. X-ed objects) to the extent that they interfere with retrieval of the target. As shown in the top panel of Figure 2, when there is more feature overlap between the practiced items and their non-practiced competitors, such as when one encodes similarities between them or when they are semantically related, fewer unique features of the competitor are inhibited resulting in reduced inhibition (i.e. fewer X-ed objects). However, when there less feature overlap between the practiced items and their competitor, such as when one encodes differences between them or when they are not semantically related, more unique features of the competitor are inhibited resulting in reliable inhibitory effects (i.e. more X-ed objects). 17

PAGE 28

18 As mentioned previously, the Distribu ted Representation Approach does not adequately differentiate between episodic sim ilarity and similarity based on pre-existing knowledge. In fact, it predicts that either episodically based integration or semantically based integration can insulate items from inhi bitory effects as long as the integration occurs between the practiced items and their unpracticed competitors (i.e. Rpitems). According to the Semantic Generalization Hypothesis, RIF effects should be reduced when the target and competitor are actually semantically similar based on pre-existing knowledge, because the strengthening effects of retrieval practice should generalize to semantically related items in memory, thereby strengthening the related competitor in addition to the retrieval practi ced target. The specific de tails regarding the different predictive outcomes as a result of the specific pattern of associations are presented below. High Target-Competitor Similarity or Inter-set Associations Condition According to the model, the inhibitory e ffects of retrieval pr actice should affect features that are unique to the competitors (i.e. Rpitems). For example, retrieval practice of HORSE (i.e. Rp+ item or target ), results in strengthening of HORSEs features and inhibition of features that are un ique to its unpracticed competitors (i.e. Rpitems). For example, as shown in Figure 3, retrieval practice of HORSE results in strengthening of HORSEs feat ures and inhibition of features that are unique to its unpracticed competitors. As shown in the t op panel, when there is a high degree of similarity or a strong association between the target and its competitor, such as between HORSE and PONY, strengthening th e same number of features (i.e. ten features) has a greater impact because retrieval practice also strengthens the features that are shared by both representations. In this case, strengthening the 10 features of HORSE also results in strengthening of the seve n features that PONY shares with HORSE. The result is that out of 10 features for each representation, only th ree unique features of PONY are inhibited and the rest are strengthened because the strengthening effects of retrieval practice generalize to features that HORSE and PONY share. The model assumes that an items recall probability should be related to the summed activity of all of is feature units. Because a high proportion of features in both representations are strengthened when there is a high degree of semantic similarity or associations between the target and its competitors, the model predicts that practic ing shared features should protect against inhibition of the competitor resulting in attenuated forgetting of the competitor. The model makes very different predictions however, when there is a low degree of similarity or weak association between th e target and the competitor, such as between HORSE and LION. As shown in the bottom pa nel of Figure 3, when similarity between the target and its competitor is low, strength ening the same number of features (i.e. ten features) has a smaller strengthening impact on the competitor (Rpitem) because it only shares a small number of features with the ta rget (Rp+ item), resulting in inhibition of LIONs eight unique features presumably becau se those unique features interfere with retrieval of the target. The model assumes that an items recall probability should be related to the summed activity of all of is feature units. Assuming there is constant facilitation of practiced feat ures across both high (top panel) and low similarity cases (bottom panel), the probability of recalli ng the competitor should be reduced when similarity between the target and the compe titor is low (bottom panel) because a higher

PAGE 29

proportion of the competitor features are inhibited. For example, when target-competitor similarity is high (top), only 30% of the Rpitems representation is inhibited, whereas when target-competitor similarity is low (bottom), 80% of the Rpitems representation is inhibited. The model therefore predicts significant forgetting of the competitor when the target and competitor are semantically dissimilar, but reduced forgetting when they are very semantically similar. 19 Category CueAnimal RpPONY Rp+ HORSE High Target-Competitor Similarity or Strong Inter-set Association Low Target-Competitor Similarity or Weak Inter-set Association Figure 3. Illustration of how high target-competitor similarity (top panel) and low target-competitor similarity (bottom panel) influence inhibition. According to the Semantic Generalization Hypothesis, when there is a high degree of semantic similarity or strong associations between the target (i.e. Rp+) and the competitor (i.e. Rp-) the strengthening effects of retrieval practice are implicitly exaggerated because strengthening occur on the features shared by the target and competitor, resulting in attenuated RIF. When there is a low degree of similarity or weak associations between the target (i.e. the Rp+) and the competitor (i.e. the Rpitem) there are more unique features of the competitor to inhibit, resulting in significant RIF. X X X Category Cue RpLION Rp+ HORSE X X X X X X X X Result: because there are onsmuni Significant RIF of Rpitem, ly a all # of shared features to be strengthened along with the Rp+ item. A large # of q ue featu res are inhibited. Result: No RIF of Rpitem, because its large # of shared features are implicitly strengthened along with the Rp+ item. Only a small # of unique features are inhibited.

PAGE 30

20 High Competitor-Competitor Similarity or Intra-set Associations Condition The model makes very different predictions when one varies the pre-existing similarity between the competitors. As me ntioned previously, Anderson et al. (2000) found significant forgetting when participan ts encoded similarities between the competitors. The model also predicts that RIF effects should be observed when the competitors are actually similar in the sense that they are semantically related, not just when one encodes similarities between them. According to the model, the inhibitory effects of retrieval practice s hould affect features that are unique to the competitors (i.e. Rpitems). The model assumes that retrieval practice of LION (i.e. Rp+ item or target), results in strengthening of LION s features and inhibition of f eatures that are unique to its unpracticed competitors (i.e. Rpitems). As shown in top panel of Figure 4, when there is a high degree of similarity between the competitors, such as between HORSE and PONY, inhibiting the same number of features (i.e. seven fe atures) has a greater impact because the inhibition affects th e features shared by both compet itors representations. In this case, inhibiting six features of HORSE also results in inhibition of the five features that PONY shares with HORSE. The result is that 60% of HORSEs representation and 60% of PONYs representation is inhibited. That is, out of 10 features for each representation, two features that HORSE a nd PONY share with LION are strengthened, but seven features are inhibi ted, six on each representation (i .e. five shared features between HORSE and PONY, one unique featur e of HORSE, and one unique feature of PONY). This example, illustrates how wh en competitors are hi ghly similar, a large proportion of their represen tations are inhibited. When competitors are highly similar, the proportion of the representations that are inhibited is relatively larger compared to the proportion of inhibiti on that is manifested when competitors are not very similar. As shown in the bottom panel of Figure 4, a higher proportion of competitor features remain unaffected when there is a low degree of similarity between the competitors, such as between HORSE and WALRUS. In this case, inhibiting the same number of features (i.e. seven) has a smaller inhibitory impact on both representations because there are fewe r shared features between the competitor representations. In this cas e, the same number of features are inhibited as when competitor-competitor similarity is high (i.e two shared features between HORSE and WALRUS, three unique featur es of HORSE, and two unique features of WALRUS). However, because HORSE and WALRUS only share two competitor features, the result is that only 50% of HORSEs representati on and only 40% of WALRUSs representation is inhibited. Compare this to the predicted 60% inhibition of the representations when competitor similarity is high. Therefore, ev en though the same number of features are inhibited as when competitor similarity high, th e chance for the features to be inhibited is less when competitor similarity is low. The model therefore predicts that when competitor similarity is low, there will be significant RIF because inhibition of unique, competing features is still predicted to occur, however the RIF effect will not be as large as when competitor similarity is high.

PAGE 31

High Competitor-Competitor Similarity or Strong Intra-set Association 21 Low Competitor-Competitor Similarity or Weak Intra-set Association Category Cue RpPONYRp+ LION RpHORSE X X X X X X X X X X X X X X Rp+ LION Category Cue RpWALRUS RpHORSE Result: Smaller RIF ofbecause coshare a large # of unique features. More comfeatures remain unaffected. Rpitem, mpetitors do not petitor Result: Large RIF of Rpitem, because competitors share a large # of unique features that are inhibited. Fewer competitor features remain unaffected. Figure 4. Illustration of how high competitor-competitor similarity (top panel) and low competitor-competitor similarity (bottom panel) influence inhibition. According to the Semantic Generalization Hypothesis of the Distributed Representation Approach, when there is a high degree of similarity or strong associations between the competitors (i.e. Rp-), the effects of inhibition are exaggerated because the inhibition affects the features shared by the two competitors. When there is a low degree of similarity or weak associations between the competitors (i.e. the Rpitems), the effects of inhibition are not as exaggerated. To summarize, the Semantic Generalization hypothesis of the Distributed Representation Approach assumes that when there is a high degree of pre-existing semantic similarity between the practiced (i.e. targets) and unpracticed items (i.e. competitors), those items may be integrated on a semantic level so that explicit practice of one results in implicit practice of the features that it shares with the other. Therefore,

PAGE 32

22 to the extent that semantically related items are highly associated in memory, the model predicts that when there are strong inter-set associations between the practiced and unpracticed sets (high target-competitor simila rity), the unpracticed items will be saved from inhibitory processes because they are implicitly retrieval pr acticed along with the explicit retrieval practice of practiced items, re sulting in attenuated RIF. However, when there are strong associations between th e competitors (high competitor-competitor similarity or intra-set associations), the model predicts that inhibitory effects will be exaggerated resulting in even greater degrees of RIF.

PAGE 33

Experiment 1: Investigating the effects of pre-existing associations on RIF with a list that contains strong semantic associations The purpose of Experiment 1 was to test the semantic integration predictions of the Semantic Generalization hypothesis of the Distributed Representation Approach. Because of the relationship found between the amount of forgetting and the association strengths between practiced and unpracticed sets in the foregoing analysis of studies that did and did not obtain significant RIF effects, we experimentally manipulated the strength of associations between sets as an index of similarity to determine its effects on the magnitude of RIF. Using Nelsons free association database, a list of four categories with eight exemplars in each was developed. The categories were constructed so that there were four pairs of two strongly related items. Exemplars were selected such that the members of each pair were strongly associated according to the Nelsons free association norms (Nelson, McEvoy, & Schreiber, 1999). However, between pairs, there were little to no associations. The complete list is shown in Appendix A, but an example of the category composition is shown in Figure 5. Category: Animal Pair 1: HORSE PONY Pair 2: LION TIGER Pair 3: WALRUS SEAL Pair 4: DUCK GOOSE 23 Category: Profession Pair 1: JUDGE LAWYER Pair 2: NURSE DOCTOR Pair 3: CHEF BAKER Pair 4: SENATOR PRESIDENT Figure 5. Two categories used in Experiment 1. Note that within each pair, the two exemplars of each pair are strongly associated, whereas there are few associations between pairs.

PAGE 34

In order to test the hypothesis that certain exemplars will be saved from inhibition because a strongly related exemplar is retrieval practiced, the experiment manipulated which exemplars received retrieval practice, between subjects. In the Inter-set (strong) practice condition, there were strong associations between the set that was retrieval practiced and the set that was not practiced, with weak-to-no associations within the sets. So, for instance, participants would perform retrieval practice on a single item from each of the four pairs illustrated previously. This retrieval pattern introduces a high degree of semantic similarity between the targets and the competitors (high target-competitor similarity in top panel of Figure 3), and a low degree of similarity among the competitors (low competitor-competitor similarity in bottom panel of Figure 4). The design of the Inter-set (strong) condition is illustrated in Figure 6. Practiced Category: Animal Retrieval Practiced (Rp+) NOT Retrieval Practiced (Rp-) HORSE PONY LION TIGER WALRUS SEAL DUCK GOOSE Baseline Category: Profession NOT Retrieval Practiced (Nrp) NOT Retrieval Practiced (Nrp) JUDGE LAWYER NURSE DOCTOR CHEF BAKER SENATOR PRESIDENT Figure 6. Inter-set (strong) condition design. Note that one item from each pair (e.g., HORSE, LION) is given practice, so that everything that is practiced, has an associated exemplar in the unpracticed set. Because of the strong associations between the set that was retrieval practiced and the set that was not retrieval practiced, it was hypothesized that the RIF effect would be reduced. The results of the Inter-set (strong) condition were compared to the Intra-set (weak) condition, where there were strong associations within the set that was retrieval practiced and within set that was not retrieval-practiced, but there were weak-to-no associations between the sets that were retrieval practiced and those that were not. This retrieval pattern introduces a high degree of semantic similarity between the competitors (high competitor-competitor similarity in top panel of Figure 4) and a low degree of similarity between the targets and competitors (low target-competitor similarity in bottom panel of Figure 3). The design of the Intra-set (weak) condition is illustrated in Figure 7. 24

PAGE 35

Practiced Category: Animal Retrieval Practiced (Rp+) NOT Retrieval Practiced (Rp-) HORSE PONY LION TIGER WALRUS SEAL DUCK GOOSE Baseline Category: Profession NOT Retrieval Practiced (Nrp) NOT Retrieval Practiced (Nrp) JUDGE LAWYER CHEF BAKER NURSE DOCTOR SENATOR PRESIDENT Figure 7. Intra-set (weak) condition design. Note that both items from pair (e.g., HORSE and PONY, WALRUS and SEAL) are given practice, so that everything that is practiced, is not associated to the unpracticed set. Because of the weak-to-no associations between the set that was retrieval practiced and the set that was not retrieval practiced, it was hypothesized that there would be a typical RIF effect in the Intra-set (weak) condition. Note that this was predicted to occur, despite the fact that across participants, every item participated equally often in every condition. Finally, because Butler and colleagues failed to obtain RIF with item specific cues, we tested all exemplars with item specific cues to control for output order and eliminate forgetting due to output interference. Specifically, we used category plus 1st letter stem cues blocked by category. Rpitems were cued in the first half of each category block and Rp+ items were cued in the second half of each block. Because we tested for the Rpitems first, any forgetting of Rpitems would have to be due to inhibitory processes that occurred as a result of retrieval practice and not to output interference as a result of strong items being recalled first in the test phase. 25

PAGE 36

26 Methods Design and Participants The experimental design formed a 2 X 8 X 4 mixed-subjects factorial with retrieval practice pattern ( Inter-set (strong) Intra-set (weak)) and counterbalancing (eight counterbalancing conditions) manipulated between subjects, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) ma nipulated within subjects. Half of the categories (two categories) received retrieval practice (i.e., Rp or retrieval practice categories), and half of the categories received no retrieval practice (i.e., Nrp or no retrie val practice). Of the Rp categories, half of the exemplars w ithin each (i.e., four of the eight) received retrieval practice (i.e. the Rp+ items) and the remaining four did not receive retrieval practice (i.e., the Rpitems). Nrp ite ms were divided into two subgroups for counterbalancing purposes and served as baselines against which to measure the detrimental effects of retrieval practice on Rpitems, and the facilitory effects of retrieval practice on Rp+ items. The dependent variable was the percentage of items correctly recalled on a category-plus-stem-cued recall test (e.g., A nimal H______). On this test, the Rpitems were always tested in the first four positions of the eight-item category test. These were compared against Nrp items from the corresponding posit ions (denoted Nrp1 for the1st half of baseline in testing sequence). Rp + items were always tested in the last four positions of the eight-item category test and were compared against Nrp items from the corresponding positions (denoted Nrp2 for the 2nd half of baseline items in testing sequence). Nrp1 and Nrp2 items served as baselines against wh ich to measure the detrimental effects of retrieval practice on Rp items (Nrp1 verses Rp-) and the positive effects of retrieval practice on Rp+ items (Nrp2 versus Rp+). Sixty-four undergraduate psychology students from the University of South Florida served as participants in exchange for extra credit. Participants ranged in age from 18 to 35 years, with a median age of 21. The participant pool was 94% female and 6% male, and the ethnic make-up of the participant pool was 61.1 % Caucasian, 11.6% African American, 15.5% Hispanic, 5% Asian, and 2% other. The average Shipley vocabulary scores was 28.61 (SD = 3.63) (rang e: 0 to 40). The average MEQ score was 45.03 (SD = 9.48), indicating that on average the circadian rhythm of the participant pool fell into the neutral range between moderately morning and moderately evening. The average Cognitive Failures score was 40 (SD = 13.8) (range: 0 to 100, where 100 indicates high cognitive failures), and the average reasoning score was 79% ( SD = 16) out of 100. Materials The list for this experiment is shown in Appendix B. Eight exemplars from four different categories (i.e. Animal Profession, Appliance, Weapon) were selected from several category norms (Battig & M ontague, 1969, McEvoy & Nelson, 1982; Overschelde, Rawson, & Dunlosky, 2004; Yoon, Feinberg, Hu, Gutchess, Hedden, Chen, Jing, Yao, & Park, 2003). Unrelated categories were chosen in order to ensure that measures of recall performan ce between the categories were independent. According to Nelsons free association norms, there were no direct relationships between the categories

PAGE 37

27 and members from other categories. The aver age word frequency of the category labels was low to moderate, with an average of 30.75 occurrences per million, with a range of 5 to 68 (Kucera & Francis, 1967). In addition to the four experimental categories, two filler categories ( City and River ) were constructed with eigh t exemplars each. No attempt was made to select strongly related pairs of exemplars for the filler categories. The filler categories and their exemplars ar e shown in Appendix A. Each exemplar within a category had a unique first letter, because in the final test phase the items were cued with a category-plus-1st-letter stem cue. Versatility was set to moderate levels because low versatility item s have relatively few correct responses and so are easily guessed. For example, usi ng the low versatility stem cue Um__________, is a problem because there are only five po ssible solutions and most people think of Umbrella. Because there are so few responses that fit this letter cue, the subject has a strong chance of simply gue ssing the correct response wit hout recalling the response from memory. Because it is relatively easy to circumvent the episodic representation and rely more on the orthographic representation of the exemplar when versatility is low, it was set to moderate levels and averaged 249.78 ( SD = 26.45) according to Solso and Juel (1980). The word frequency of the exemplars was low to moderate and averaged 41.83 ( SD = 31.14) occurrences per million. All four categories were normed by the Michigan Category Norms (Yoon, Feinberg, Hu, Gu tchess, Hedden, Chen, Jing, Yao, & Park, 2003), however appliance was not normed by Battig and Montague (1969). The average typicality of the exemplars was strong to moderate according to these two sources. Typicality refers to the average position in a list rank ordered by frequency of report. The average typicality was 20.54 ( SD = 4.99) according to the Battig and Montague norms, and 18.36 ( SD = 4.97) according to the Michigan norms for young Americans. These frequencies are somewhat weaker th an Anderson, Bjork, and Bjorks (1994) strong exemplar condition (M = 8), but stronger than their w eak condition in Experiments 1 and 2 ( M = 33), and much stronger than their weak condition in Experiment 3 (M = 50). The normative association strengths, bot h direct and indirect, between the exemplars are shown in Appendix C. Exem plars were selected so that they were strongly related to one other exemplar fr om the same category (e.g. HORSE PONY), but relatively unrelated to others from th e same category (LION, WALRUS, etc). There were only five un-wanted direct associa tions with other members from the same category, and they were relatively weak, with direct streng th averaging .06 ( SD = .07) and probability of recovery averaging .19 ( SD = .23). Eleven of the 16 strongly related pairs were bi-directionally related and had an average forward strength, backward strength, and probability of recovery value of .26 ( SD = .24), .22 ( SD = .19), and .68 ( SD = .28), respectively. Direct strength for the re maining five uni-direc tionally related pairs averaged .12 ( SD = .19) and probability of recovery averaged .17 ( SD = .26). The pairs also shared a total of 43 associat es with an average number of 2.69 (SD = 1.74) shared associates and an average shar ed associate strength of .03 ( SD = .05) between them. There was a total of 59 mediators with an average number of 1.85 ( SD = 2.14) mediators and an average mediator strength of .01 (SD = .02) between them. Other variables that have been shown to affect cued recall including concreteness, set size, connectivity, resonance, and frequency of the exemplars we re also examined. Exemplar concreteness

PAGE 38

28 was high and averaged 5.93 ( SD = .53) on a scale of 1 to 7. Associative set size for each exemplar was low-to-moderate and averaged 13.80 ( SD = 5.17) associates, connectivity was low and averaged 1.62 ( SD = .76), the probability of a resonant connection from the exemplars associates back to the exemplar was moderate and averaged .40 ( SD = .23), and the frequency of the exemplars was moderate to high and averaged 41.19 ( SD = 26.45). As shown in Appendix B, the exemplars w ithin each category were divided into two sets of four exemplars each (Set A and B) The exemplars were assigned to sets so that in the Inter-set (strong) condition, association strengths were strong between the sets, but weak within the sets. The average direct strength and probability of recovery was .23 (S D = .23) and .59 ( SD = .33), respectively between the sets. There was total of 82 shared associates with an average strength of .02 ( SD = .04) between the sets. There was a total of 95 mediators with an average strength of .01 ( SD = .01) between the sets. As shown in top portion of Table 2, the sets were equated on Between-Set strength characteristics such as the predic ted probability of recovery (Eq3), forward strength (FSG), backward strength (BSG), total number of shared associates (OLAPS), shared associate strength (OSG), total num ber of mediators (MEDS), and mediator strength (MSG). For example, when Set A was used to activate or cue Set B in the model, the predicted probability of recovery of Set B was .65 ( SD = .31) and when Set B was used to activate Set A in the model, the pr edicted probability of recovery of Set A was .53 ( SD = .35). As shown in the middle portion of Table 2, the sets were also equated on WithinSet strength measures, so that the likelihood of exemplars within either set activating other exemplars in their sets was very low. Finally, as shown in the bottom portion of Table 2, the sets were equated on several other variables that have b een shown to affect cued recall including concreteness (CONCR) set size (MSS), connectivity (CONNM), probability of resonance (P_RESO), and frequency (FREQ). Unlike the Inter-set (strong) condition, the associati on strengths in the Intra-set (weak) condition were relatively weak between sets but relatively str ong within sets. All Between-Set strength measures were very weak, with direct strength averaging .002 ( SD = .015), predicted probability of recovery averaging .002 ( SD = .015). In addition, there were only a small number of shared associates between the sets (42) with an average strength of .01 (SD = .03). The total number of mediators was also low (31), with an average mediator strength of .01 ( SD = .01).

PAGE 39

29 Table 2. Between-Set and Within-Set streng th characteristics and other variables in the Inter-set (strong) condition. Between-Set Strength Characteristics EQ3 FSG BSG OLAPS OSG MEDS MSG Mean .65 .22 .22 82 .02 32 .01 Between Sets: From Set A to B SD .31 .21 .23 .04 .01 Mean .53 .22 .22 82 .02 63 .01 Between Sets: From Set B to A SD .35 .23 .21 .04 .02 Within-Set Strength Characteristics EQ3 FSG BSG OLAPS OSG MEDS MSG Mean .00 .00 .00 19 .01 26 .01 Within Set A SD .00 .00 .00 .02 .02 Mean .00 .00 .00 16 .01 8 .004 Within Set B SD .00 .00 .00 .01 .01 Other Variables CONCR MSS CONNM P_RESO FREQ Mean 5.79 13.13 1.43 .43 53.75 Set A SD .47 4.57 .81 .26 98.19 Mean 6.06 14.47 1.80 .37 28.63 Set B SD .59 5.77 .70 .20 30.75 Note EQ3(#) = PIER2s probability of recovery (number of connections); FSG = forward strength; BSG = backward strength; OLAPS = number of shared associates; OSG = shared associate strength; MEDS = number of mediators; MSG = mediated strength; CONCR = concreteness; MSS = set size; CONNM = connectivity; P_RESO = probability of resonance; FREQ = frequency. As shown in the top portion of Table 3, the sets were equated on Between-Set strength characteristics so that the likeli hood of exemplars activat ing each other between the sets in either direction (from Set A to B or from Set B to A) was very low. All Within-Set strength measures in the Intra-set (weak) condition were very strong. Within the sets, direct strength averaged .23 ( SD = .23), predicted probability of recovery averaged .58 ( SD = .33), and there was total of 75 sh ared associates with an average strength of .01 ( SD = .03). There were 98 mediators with an average strength of .007 ( SD = .02). As shown in the middle portion of Table 3, the sets were also equated on WithinSets strength characteristics including the predicted probability of recovery, forward strength, backward strength, total number of shared associates, shared associate strength, total number of mediators, and mediator st rength. For example, when the exemplars within Set A were used to activate or cue the others in Set A in the model, the predicted probability of recovery averaged .67 (SD = .26), and when the exemplars within Set B

PAGE 40

were used to activate or cue the others Set B in the model, the predicted probability of recovery averaged .48 (SD = .39). Finally, as shown in the bottom of Table 3, the sets were equated on other variables including conc reteness, set size, connectivity, resonance, and frequency. Table 3. Between-Set and Within-Set streng th characteristics and other variables in the Intra-set (weak) condition. Note EQ3(#) = PIER2s probability of recovery (num ber of connections); FSG = forward strength; BSG = backward strength; OLAPS = number of shared associates; OSG = shared associate strength; MEDS = number of mediators; MSG = mediated strength; CONCR = concreteness; MSS = set size; CONNM = connectivity; P_RESO = probability of resonance; FREQ = frequency. Between-Set Strength Characteristics EQ3 FSG BSG OLAPS OSG MEDS MSG Mean .004 .004 .003 42 .01 12 .01 Between Sets: From Set A to B SD .03 .03 .02 .03 .02 Mean .00 .00 .00 42 .01 19 .01 Between Sets: From Set B to A SD .00 .00 .00 .03 .01 Within-Set Strength Characteristics EQ3 FSG BSG OLAPS OSG MEDS MSG Mean .67 .24 .24 45 .01 61 .01 Within Set A SD .26 .22 .22 .04 .02 Mean .48 .21 .21 30 .01 37 .004 Within Set B SD .39 .24 .24 .03 .01 Other Variables CONCR MSS CONNM PRESO FREQ Mean 5.81 13.06 1.77 .13 53.75 Set A SD .51 4.55 .73 .11 98.19 Mean 6.04 14.53 2.17 .36 28.63 Set B SD .56 5.77 1.96 .24 30.75 Procedure Upon arrival, participants were randomly assigned to either the Inter-set (strong) or the Intra-set (weak) condition and to one of 16 count erbalancing conditions (two category status forms (Form 1, Form 2), two exem plar status sets (Set A, Set B), two Nrp baseline sets (NrpA, NrpB), and two fina l testing orders (Order 1, Order 2)). The two category status forms were cons tructed by assigning two categories to practiced status (Rp), and the remaining two categories to baseline status (Nrp). For example in Form 1, Animals and Professions served as retrieval pr acticed categories, and Appliances and Weapons served as baseline categories. However in Form 2, Appliances 30

PAGE 41

31 and Weapons served as practiced categories, and Animals and Professions served as baseline categories. The category status forms were counterbalanced so that all categories served equally often as practiced or baseline categories. As shown in Appendix A, the two exemplar status sets (Set A, Set B) were constructed by assigning four exemplars from each practiced category to Set A and the remaining four exemplars to Set B. For example, in the Inter-set (strong) condition, LION, HORSE, WALRUS, and DUCK were assigned to Set A and their related associates, TIGER, PONY, SEAL, and GOOSE were assigned to Set B. In the Intra-set (weak) condition, LION, TIGER, DUCK, and GOOSE were assigned to Set A, and their unrelated associates, HORSE, PONY, WALRUS, and SEAL were assigned to Set B. The exemplar status sets were counterbalanced so that all exemplars served equally often in both the retrieval practiced (Rp+ ) and unpracticed (Rp-) conditions. Two Nrp baseline sets were constructed by assigning four exemplars from each baseline categories (unpracti ced categories) to NrpA and the remaining four exemplars from the baseline categories to NrpB. The baseline sets were counterbalanced so all exemplars served equally often as base lines for RIF effect comparisons. Finally, two testing orders were develope d. In Order 1, a practiced category was tested first followed by a baseline category, a practiced category, and finally a baseline category. In Order 2, a baseline category wa s tested first, followed by a practiced category, a baseline category, and finally a pr acticed category. The testing orders were counterbalanced so that the practiced and baseline categories were roughly equated on serial position in the tes ting sequence. Data was collected in individual sessions There were four phases; a study phase, a retrieval practice phase, a retention interval phase, and a test phase. A schematic of the procedure is shown in Figure 8. In order to simplify the schema tic example, only two categories are depicted.

PAGE 42

Phase 1: Study Phase Ss study category-exemplar pairs: Animal HORSE Profession JUDGE Animal PONY Profession LAWYER 32 Phase 2: Retrieval Practice Phase Ss retrieval practice of the categories and of the members from that category: Animal HO_______ Phase 4: Test Phase Ss tested for all study items blocked by category with Rpitems tested first: Animal P________ Animal H________ Profession J_______ Profession L_________ Phase 3: Retention Interval Phase For 20 minutes, Ss complete: Shipley Vocabulary Test, Morningness-Eveningness Questionnaire, Cognitive Failures Questionnaire, & a booklet of 30 reasonin g p roblems Figure 8. Schematic of the RIF procedure. Note that only two categories are depicted in the schematic in order to simplify the example. Also note that in the Final Test Phase, (1) item specific cues are used (e.g., Animal P______), (2) all study items are blocked by category so that all the animals are tested together in the testing sequence, and (3) the Rpitems are tested first within each block.

PAGE 43

33 In the study phase, participants were seated in front of computer in a small testing room, and were handed a copy of the study instructions, that were read aloud by the experimenter. They were told that the e xperiment was about some of the processes involved in cognition and reasoni ng, that category-exemplars would appear in the center of the computer screen, and they would be given five seconds to study each category exemplar pair, so they should use the entire time relating the exemplar to its category name. The presentation order for the category-exemplar pairs was randomized in blocks, so that all the exemplars from a given cat egory were evenly di stributed throughout the study phase. Each block contained one exem plar from each category resulting in eight blocks of six items (each block containing f our items from the experimental categories and two items from the filler categories). The ordering within each block was random except that (1) in the first block, filler item s were presented first in order to minimize primacy effects, (2) in the last block, two fille rs were presented last in order to minimize recency effects, (3) no two categories appear ed in sequence more than once, and (4) exemplars within the same category were space d so that there was an average of seven other items in between them. Also the av erage distance between the Rp+ and the Rpitems for a given category was kept consta nt across the inter and Intra-set (weak) conditions, and Rp+ and Rpitems within each category appeared distributed throughout the study list. The computer presented a di fferent category-exemplar pair every five seconds in the order determined by the aforem entioned restrictions, un til the screen read Finish. In the retrieval practice phase, participan ts were given a copy of the retrieval practice instructions that were read aloud by th e experimenter. They were told that their memory would be tested for the information studied earlier, and on the computer screen, they would see a category name and the firs t two letters of one member from that particular category. They were told to recall the previously studied category member that fits the letters and say the cat egory and the member out loud. They were also told that they would have 10 seconds to recall each ex emplar, and that some of the words would be tested more than once, and to reca ll them as they would any other. The presentation order of the categoryplus-two letter stem cues was pseudorandomized in the following way; (1) The first th ree and last three cues were filler items to acquaint participants with the task and to control for primacy and recency effects; (2) Items were tested three times on an expandi ng schedule with an average of 3.5 trials between the 1st and 2nd presentations and 6.5 trials between the 2nd and 3rd presentation; (3) No two category members were retrieval practiced in succession, and (4) the use of filler items helped ensure that no two pairs appeared consecutively more than once. Using these constraints two retrieval practi ce orders were developed. The computer presented each category-plus-two-letter-ste m-cue every 10 seconds, and this process continued until the screen read Finish. During the retention interval phase, particip ants were asked to complete a series of questionnaires and reasoning problems for 20 minutes. A 20 minute retention interval was used in order to show that the inhibito ry effects are not short-lived. Anderson and colleagues typically use a 20 minut e retention interval in which the participants complete

PAGE 44

34 a series of causal reasoni ng problems. In addition to solving 30 reasoning problems, participants in this experiment also comp leted the Shipley Vocabulary Test (Zachary, 1991), the Morningness-Eveningness Questionn aire (Horne & Ostberg,1976), and the Cognitive Failures Questionnaire (Broadbent Cooper, FitzGerald, & Parkes, 1982) because these questionnaires were used by Butler et al. (2002) in their retention interval phase. Because of cost issues, the picture fragment completion task used by Butler and colleagues in their experiments was not used in the present experiment. Instead, participants solved reasoning problems in addition to the three questionnaires. The questionnaires are shown in Appendix D. Two of the distracter items on the Shipley test were changed because they were associatively related to items on the experimental list. At the end of the 20 minutes, the booklet s were collected by the experimenter. In the test phase, participants were give n a surprise memory test for all of the category-exemplar pairs studied in the study phase. Participants were given a copy of the final test instructions that we re read aloud by the experimenter. They were told that their memory would be tested for the information studied earlier during th e first phase of the experiment, and on the computer screen, they would see a category name and the first letters of one member from that particular category. They were told to recall the previously studied category member that fits the letter and say the category and the member out loud. They were also told that they would have 10 s econds to recall each exemplar. The presentation order of the category-plus-1st letter stem cues was pseudorandomized in blocks in the following way: (1) All the members of a category were tested together in a block; (2) The two filler categories were tested first; (3) In order to control for output interference from stronger practiced items (Rp+ items) in practiced categories, Rpitems were tested first in th e testing sequence within each block, followed by the Rp+ items (4) Practiced category blocks were strictly alternated in the testing sequence with unpracticed baseli ne categories so that the se rial position in the testing sequence was similar between practiced and base line categories. Using these constraints two testing orders were deve loped (Order 1, Order 2). In testing Order 1, practiced category #1 was tested first followed by base line #1, practiced category #2, and finally baseline #2. In testing Order 2, baseline #2 was tested first, followed by practiced category #2, baseline #1, and finally practiced category #1. Th e two testing orders also differed in the order in which the individual exemplars were tested. For example, in testing Order 1, exemplars were tested in the following sequence; exemplars 2, 4, 6 & 8, followed by 1, 3, 5, & 7. In testing Order 2, the orders were reversed and presented in the following sequence; 8, 6, 4, & 2, followed by 7, 5, 3, & 1. The computer presented each category-plus-1st-letter-stem-cue every 10 seconds, a nd this process continued until the screen read Finish.

PAGE 45

35 Finally, participants answered some que stions concerning the experiment. The post-experimental integration questionnaire was designed to assess the degree to which participants episodically inte grated the items during the st udy phase (see Appendix B). For each category, participants indicated how often on a 5 point scal e ( being none of the time, and being all of the time) duri ng the study phase, did they intentionally think back to previously seen category members when they encountered a new exemplar, and rehearse the exemplars together. Participants had 3 minutes to complete the integration questionnaire. Following completion of the questionnaire, participants were debriefed. They were told that the purpose of the study was to investigate how remembering causes forgetting.

PAGE 46

36 Results A series of mixed subjects ANOVAs were conducted, with retrieval practice pattern ( Intra-set (weak) Inter-set (strong) ) and counterbalancing (eight counterbalancing conditions) as between subjec ts factors, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) as within subjects factors. Because the items used in Experiment 1 were not pre-tested to ensure comparable item memorability across categories, there may have been differences in overall item memora bility that contribute d to variability in the inhibition effect, thus weakening stat istical power to find inhibition, and to find differences in inhibition across condition. Th erefore, counterbalanc ing was included as a between-subjects factor in or der to reduce counterbalancing-related variance from the error term, and to increase the chance of finding signifi cant levels of inhibition. However, because item-related variability is not relevant to the research question, its specific effects will not be discussed. Th is convention regarding counterbalancing will be adopted for all subsequent studies. Retrieval Induced Forgetting The overall inhibition (Baseline (Nrp1) vs. Rp-) collapsed across the retrieval practice pattern manipulation was significant, and revealed that probability of recalling Rpitems was lower than the probability of recalling baseline items (7% effect), F (1, 48) = 5.46, Mse = .026, p < .05. The current findings replicate previous work showing that the RIF effect occurs with it em specific cues, providing additional evidence that Butler et al.s conclusion that the RIF e ffect may not be a reliable outcome with item specific cues, is unlikely to be correct. The item specific cueing results also provide additional evidence that the forgetting effect obs erved in studies of RIF is not simply due to output interference. More importantly for the purposes of this experiment, the interaction between the inhibition effect (Baseline vs Rp-) and the retrieval practice pattern was significant, F (1, 48) = 5.36, Mse = .026, p < .05, indicating that there wa s a difference in the amount of RIF between the Intra-set (weak) and Inter-set (strong) conditions. As shown in the top panel of Figure 9, the hypot hesis that there would be significant RIF in the Intra-set (weak) condition was supported, F (1, 24) = 12.64, Mse = .023, p < .01. The hypothesis that there would be reduced RIF in the Inter-set (strong) condition was also supported, as no RIF was observed, F < 1. In the Intra-set (weak) condition, where associations between the practiced (Rp+) a nd unpracticed sets (Rp-) were minimized, there was a 13% decrement in recall between Rpitems and base line items in the final test phase, however in the Inter-set (strong) condition, where associations between the practiced and unpracticed sets were present, there was no difference in recall pe rformance between Rpitems and baseline items in the final test phase (0%). The results of this experiment thus support the assumption that strong associa tions between practiced items and their unpracticed competitors, moderate the inhibitory effect and that this moderation occurs when associations are implicitly activated. Wh en pre-existing associations are minimized between the sets, the typical inhibitory effect is observed.

PAGE 47

.66.68.55.66.40.50.60.70.80.901.00Intra-setInter-setPercentage Items Recalled Baseline (Nrp) RpRetrieval Practice Pattern.58 .86d 37 .65s .81.90lle .40.50.601.00Intra-setInter-setPercentage I Baseline (Nrp) .80 Re ca .70tem Rp+ Figure 9. Results of Experiment 1: Significant retrieval-induced forgetting of Rpitems only in the Intra-set (weak) condition, but significant strengthening of Rp+ items in both the Intra-set (weak) and Inter-set (strong) conditions. Top panel: RIF effect: Mean percentage of Baseline and Rpitems recalled as a function of retrieval practice pattern (Intra-set (weak), Inter-set (strong)). Bottom panel: Retrieval practice benefit effect: Mean percentage of Baseline and Rp+ items recalled as a function of retrieval practice pattern.

PAGE 48

38 Baseline Tests Because the amount of RIF is determined by comparing recall of Rpitems to recall of baseline items, it wa s important to compare the baselines between the Intra-set (weak) (.68) and Inter-set (strong) conditions (.66). For example, it could be argued that no forgetting was found in the strong condition because the baseline was significantly deflated in that condition. Had the baseline been higher, reliable levels of forgetting may have emerged in that condition. It could also be argue d that reliable levels of RIF emerged in the Intra-set (weak) condition, because the baseline was exaggerated or inflated in that condition. Recall for Rpitems in the weak condition may have been equivalent to the strong condition, but because of an inflated baseline in the weak condition, it may have appeared th at there were reliable levels of RIF, when in fact no RIF occurred. However, results revealed that there was no significant differences between the baselines, F < 1. Furthermore, there was a significant difference between recall for the Rpitems in the weak condition (.55) compared to the strong condition (.66), F (1, 48) = 5.58, Mse = .026, p <.05. As shown in the top panel of Figure 9, Rprecall in the weak was 11% lower than Rprecall in the strong condition Because the baselines were statistically e quivalent, and recall of Rpitems was reliably lower in the weak condition, the reliable RIF found in the weak condition was not due to an inflated baseline, and the lack of reliable RIF found in the strong was not due to a deflated baseline. Benefits of Retrieval Practice on Practiced Items As predicted by the inhibitory account of RIF, the amount of retrieval-induced forgetting was independent of the amount of strengthening of the practiced items. The overall benefit (Baseline(Nrp2) versus Rp+) in recall for practiced items (Rp+) across the retrieval practice pattern conditions was significant, F (1, 48) = 84.30, Mse = .019, p < .0001. Twenty-two percent more practiced items were recalled in the final test phase ( M = .84, SD = .15) compared to baseline items, ( M = .62, SD = .20). According to blocking explanations, the amount of forgetti ng of unpracticed items should be dependent on the amount of strengthening of pract iced items (McGeoch, 1942; Melton & Irwin, 1940). Specifically, the recall of practiced items should alwa ys be negatively related to recall of unpracticed items from the same category. Howeve r, results rev eal that there was a non-significant, positive correlation between the benefits and the amount of RIF, r = .13, p = .31. There was significant facilita tion of Rp+ items in the condition where forgetting was present, F (1, 24) = 48.36, Mse = .017, p < .0001, and in the condition where forgetting was absent, F (1, 24) = 36.92, Mse = .02, p < .0001. As shown in the bottom panel of Figure 9, the interaction betw een facilitation of Rp+ items and retrieval practice pattern was not significant, F < 1, revealing that even though forgetting was only apparent in the Intra-set (weak) condition, retrieval practi ce produced a significant benefit for practiced items rela tive to baseline in both the Intra-set (weak) (23%) and Inter-set (strong) (21%) conditions. Interference acc ounts of RIF have a hard time reconciling the fact that forget ting of unpracticed items in the Inter-set (strong) condition did not occur despite considerable strengthening of practiced items.

PAGE 49

Output Interference Anderson and colleagues consider output interference to be another form of RIF, whereby retrieving items from memory in the initial portion of the testing sequence causes retrieval-induced forgetting of other items that are cued later in the testing sequence, so that later items are less likely to be recalled. Essentially, retrieval during the test phase can be likened to retrieval practice in the retrieval practice phase. The more items that are retrieved or retrieval practiced, the greater the forgetting of those interfering items. Results were consistent with this assumption and reveal that there was significant output interference (Baseline-Nrp1 vs. Baseline-Nrp2) within the baseline condition, F(1, 48) = 4.37, Mse = .024, p = .04. As shown in Figure 10, baseline items cued in the last half of the category block (i.e. Nrp2) were recalled at a lower rate (M = .62, SD = .20) than baseline items that were cued in the first half (i.e. Nrp1) of the category block (M = .67, SD = .20). 39 Baseline 1st 1.00 Baseline 2nd .90 d Retrieval Practice Pattern.58.40.50.60.70Intra-setInter-setPercentage Items R .68.66.65e .80calle Figure 10. Output Interference Effect in Experiment 1: Significant output interference in the Intra-set (weak) condition, but not in the Inter-set (strong) condition. Mean percentage of baseline items recalled in the 1st half of the category block and the 2nd half of the category block as a function of retrieval practice pattern. If output interference is a form of RIF then pre-existing associations between items in the initial portion of the testing sequence and items later in the testing sequenceshould also moderate forgetting of items in the later portion of the test phase, just as they moderate forgetting when there are strong associations between practiced and unpracticedsets in the retrieval practice phase. For example, if one successfully retrieves HORSE tothe item specific cue, Animal HO_______ towards the beginning of the testing sequencethis may also activate the Animal PONY because of its strong association to HORSE. PONY will thus receive a boost in activation level along with HORSE and will not suffer from forgetting even when it is cued towards the end of the testing sequence. However,

PAGE 50

40 SE, the tions between e first and the second half. Although the inter action between output interference and rn was not significant, F (1, 48) = 2.24, Mse = .024, p = .14, the pattern a ect. These results s uggest that semantic integration can moderate the effe e time, and being all of the time) during th e study phase, did they intentionally think back to previously seen category members when they encountered a new exemplar, and rehearse the exemplars together. The amount of integration is measur ed by averaging the ratings across the four categories (i.e. 4), th erefore the largest possi ble integration score was 5 and the smallest possible integration sc ore was 1). Higher scor es indicate greater incidences of episodic integra tion and lower scores indicate fewer incidences of episodic integration. The average integration score in the Inter-set (strong) condition ( M = 3.27, SD = .69) was similar to the aver age integration score in the Intra-set (weak) condition ( M = 3.34, SD = .70). A median split by post-experimental integration score revealed that there was no significant interaction between overall i nhibition and the amount of integration, F < 1. For the Inter-set (strong) condition, there was no interact ion between inhibition and the amount of integration indicating th ere was no added benefit of episodic integration, F < 1. In other words, high ( M = 3.89, SD = .86) and low episodic integrators ( M = 2.64, SD = .53) showed equivalent lack of RIF e ffects (2% and -2%, respectively). These if one retrieves HORSE towards the beginni ng of the testing sequence, LION is less likely to be implicitly retrieved along with it because of its weak association to HOR and so LION will suffer from inhibitory processes when cued towards the end of testing sequence. This s uggests that there should le ss output interference in the Inter-set (strong) condition where there are strong associati ons between baseline items in the first half of the testing se quence and baseline items in the s econd half of the testing sequence, than in the Intra-set (weak) condition where there are weak -to-no associa th retrieval practice patte was at least moderately consistent with the idea that retrieving an exemplar in the first half of the testing sequence should insulate related items in the second half of the testing block from forgetting, but should not in sulate unrelated items later in the testing sequence. In fact, separate an alyses revealed that in the Inter-set (strong) condition where semantic integration was higher (i.e. stro ng associations between the first half (Rp ) and the second half (Rp+) of the category bl ock), output interferen ce was not significant (only 1% effect), F < 1. However, in the Intra-set (weak) condition where semantic integration was lower (i.e. weak associati ons between the first and the second half), output interference was significant at 10%, F (1, 24) = 5.15, Mse = .03, p = .03. Therefore the pattern of results was consistent with the idea that if out put interference is form of RIF, then semantic integration should reduce the output interference effect, just as is reduces the RIF eff cts of output interf erence just as it moderates the effects of RIF. Episodic Integration The amount of episodic integration that participants engaged in during the study phase was measured by a self-report post-e xperimental integration questionnaire. For each category, participants indi cated how often on a 5 point scale ( being none of th

PAGE 51

41 results suggest that when there were st rong pre-existing associ ations between the practiced and unpracticed sets, semantic inte gration between the sets was sufficient to reduce the effect, and episodic in tegration was not necessary. In the Intra-set (weak) condition, high integrators ( M = 3.89, SD = .86) exhibited less forgetting (-11% RIF effect) than low integrators (-16% RIF effect) ( M = 2.64, SD = .53). Although this 5% benefit appears to replicate previous work on the moderating effects of episodic integration by Anderson and colleagues, the interaction between inhibition and the amount of integration wa s not significant, indicating there was no significant benefit of episodic integration, F < 1. Although the episodic integration benefit observed here is weaker than in prev ious studies, it is generally consistent with prior findings (Anderson et al., 1999; Anderson et al., 2000). However, there are several reason why we may have observed a weaker ep isodic integration benefit. These reasons are discussed in the follow ing discussion section.

PAGE 52

42 Discussion To summarize the findings of Experiment 1, pre-existing associations appear to be an important moderator of the RIF effect wh en associations between the practiced and unpracticed sets are strong. The moderating effects of semantic in tegration based on preexisting associations are also evident in an output inference situation, where there are strong associations between items cued toward s the beginning of the testing sequence and items cued towards the end. These findings highlight the importance of controlling for these associations and may account for Butler and colleagues (2001) failure to find the RIF effect. They further confirm and exte nd the notion that inte gration is a boundary condition on RIF, by establishing that integrat ion need not occur explicitly, but may be based on pre-experimental associations activat ed implicitly during study. The results also clearly demonstrate that pre-existing associations in and of themselves do not necessarily moderate forgetting. It is the specific pa ttern of associations that matters. When associations between the pract iced and unpracticed sets are strong, forgetting is reduced, but when associations within the practiced sets and within the unpracticed sets are strong, forgetting is apparent. The benefits regarding semantic integrat ion in Experiment 1 were quite clear. However, the evidence in favor of episodic integration benefits was not as promising. That is, although we found episodic integration be nefits in the range of what has typically been found, the benefits were not statistical ly reliable. However, there may be two important reasons why the benefits of episodic integration were not replicated in the present experiment. First, the post-experimental questionn aire designed to measure the rate of episodic integration was limited in th at it did not focus on the specific episodic links that were formed. As mentioned previously, links between the Rp+ and Rpitems are crucial in order for integration benefits to occur. It could be th at participants were exclusively linking the Rpitem s with other Rpitems, or exclusively linking Rp+ items with other Rp+ items. If the episodic integr ation occurred in this way, and not between the Rp+ and Rpitems, then this could have been why no episodic integration benefits were found. The second reason that episodic integration benefits were not replicated may be because the strong pre-existing associations in our experiment biased participants to focus their integration efforts along those preexisting links, because they so easily came to mind. One of main differences between th e experiments presented here and previous studies by Anderson and colleagues (1994) was that Anderson made great efforts to minimize the pre-existing associations between exemplars, whereas we maximized those associations. The retrieval pathways between related exemplars may have been so strong in the present experiment that participants spontaneous efforts at episodic integration were largely driven by these a ssociations. According to this Spontaneous Integration Bias Hypothesis, the presence of pr e-existing associations in this study biased participants to episodically integrate the items along those pr e-existing links because they so strongly came to mind. For example, when particip ants studied HORSE, its strong, previously studied associate PONY was hi ghly likely to be brought to mind and rehearsed along with HORSE. Therefore, any episodic integration was highly likely to have occurred on

PAGE 53

43 the pre-existing link between strongly rela ted exemplars like HORSE and PONY, instead of between unrelated exemplars. This tenden cy to form episodic links between strongly associated pairs was confirmed by participants self reports of their rehearsal strategies during the study phase. When participants we re asked to elaborat e on their integration behavior they were more likely to report rehearsing the str ongly related exemplars together than unrelated exemplars. According to the Spontaneous Integration Bias Hypothesis in the Intra-set (weak) condition, the a priori associations among the Rp+ items and among the Rpitems, biased participants to episodically in tegrate the Rp+ items together and/or the Rpitems together. One of the consequences of th is bias was that direct ed all of integration energy towards non-useful integration am ong Rp+ items or among Rpitems, thereby directing integration energy away from the cr itical integration between the Rp+ and Rpitems. As a result, very little episodic integration was formed between Rp+ and Rpitems and so no episodic integration benefits were observed. Therefore in the present sets of experiments, the presence of pre-existing associations within the Rp+ sets and within the Rpsets biased them away from forming the critical episodic links between the sets, thus eliminating any potential for episodic bene fits to occur. Without this bias towards episodically integrating already associated items (as in prior studies by Anderson and colleagues), participants have to find their own connections, and so they may have equally formed episodic connections between unrelated Rp+ and Rpitems, between Rp+ items, and between Rpitems. The evidence for this assumption comes from previous studies showing the reliable and consistent benefits of episodic integration when preexisting associations were minimized (A nderson et al., 1999; Anderson et al., 2000). When all a priori associations were eliminated in this previous rese arch on RIF, the bias was eliminated and the benefits of episodic integration emerged. According to the Spontaneous Integration Bias Hypothesis, the reason that there was no added benefit of episodic integration above the benefit of semantic integration (i.e. Inter-set (strong )) was that the pre-existing associ ations between Rp+ and Rpitems biased participants to form the critical ep isodic links necessary for the benefits of episodic integration to emerge. The result however was no appreciable added benefit of episodic integration because the Rpitems were already implicitly strengthened through semantic integration. This al ternative hypothesis highlights the importance of minimizing pre-existing associations in order to observ e spontaneous episodic integration benefits, and provides the most compelling reason for w hy we failed to observed reliable benefits in the current experiment. Because of the Spontaneous Integration Bias it is unlikely that episodic integration benef its will emerge when there are pre-existing associations in the experimental lists.

PAGE 54

44 Experiment 2: Investigating the moderating effects of pre-exis ting associations with lists that failed to exhibit retrieval-induced forgetting effects Because Experiment 1 established that pr e-existing associations can influence the magnitude of RIF, failure to control for th ese associations may explain why Butler and colleagues (2001) failed to find RIF. As me ntioned previously, there were 39 direct associations and approximately 332 indirect associations between items that were retrieval-practiced and items that were not retrieval-practiced in their study. Their original experiment was constructed so that there were many associations between the exemplars that were practiced (Rp+) and unpracticed exemplars (Rp-) from the same category. In other words, their design was very similar to the design implemented in the Inter-set (strong) condition of Experiment 1, where the levels of RIF were not reliable. The purpose of Experiment 2 was to exam ine whether this feature of the Butler stimulus sets contributed to their failure to observe RIF with item specific cues. To examine this possibility, we mimicked the manipulation of Inter-set (strong) and Intraset (weak) associations performed in Experi ment 1 using only the pre-existing associations of Butler et al.s ex act stimulus set. Investigation of their list revealed that it was possible to re-arrange thei r assignment of exemplars to retrieval practice conditions, in order to minimize or maximize the asso ciations between pract iced and unpracticed exemplars (see Appendix B). It was predicted that when items were configured to maximize Inter-set (strong) associations, as they were in their design, no RIF would be observed; however, when the sets were configured to minimize Inter-set (strong) associations, significant RIF would be found. If this pa ttern is observed, it would indicate that Butler and colleagues failure to control inter-item associations between practiced and unpracticed sets contributed to their failure to replicate the RIF effect on tests using item specific cues. Because Butler et al.s stimulus set had fewer and weaker associations than those used in Experiment 1, it would further show that the effects of Experiment 1 are not limited to the presen t, specially designed stimulus set.

PAGE 55

45 Methods Design and Participants The experimental design formed a 2 X 8 X 4 mixed-subjects factorial with retrieval practice pattern ( Inter-set (strong) Intra-set (weak)) and counterbalancing (eight counterbalancing conditions) manipulated between subjects, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) ma nipulated within subjects. Sixty four undergraduate psychology students from the University of South Florida served as participants in the study in exchange for extra credit. Participants ranged in age from 18 to 52 years, with a median age of 21. The participant pool was 83% female and 17% male, and the ethnic make-up of the participant pool was 64% Caucasian, 25% African American, 5% Hispanic 5% Asian, and 1% other. The average Shipley score was 29.75 ( SD = 3.39). The average MEQ score was 45.98 ( SD = 9.62), indicating that on average the ci rcadian rhythm of the particip ant pool fell into the neutral range between moderately morning and moderately evening. Approximately 88% of participants were tested during a time a day th at was consistent with their peak circadian rhythm. The average Cognitive Failures score was 38.55 ( SD = 13.72) and the average reasoning score was 75% ( SD = 23) out of 100. Materials The list for this experiment is shown in Appendix B. The following eight experimental categories were selected from the 12 used in Butler and colleagues (2001) study; Bird, Drink, Fruit, Furniture Insect, Metal, Profession, and Sport Four of their categories were excluded (i.e. fish, flower, ship, and tool ), because many of the exemplars from these categories were not included in Nelsons free association norms, and so the degree of strength between practiced and unprac ticed sets could not be determined. Of the nine exemplars that they used for each category, only six were included in this experiment. Items were excluded if they shared a first letter with another exemplar from the same category because our final memory test cued participants w ith a category plus a single letter stem cue. Butler and colleagues (2001) created thei r practiced and unpracticed conditions by dividing the categories into triads of three exemplars each. So for example, if items from Triad A were retrieval practiced then items from Triad B were not retrieval practiced. A third triad was included to serve as fillers in the implicit testing condition. Because no implicit memory test will be employed, only two of their original triads for each category were included. For three categories, one ite m was replaced with an item from another triad, because the items had versatility values (i.e., the number of words in the English language that can complete the words two le tter stem) of less than 100 (Solso & Juel, 1980), and as mentioned in the methods secti on of Experiment 1, low versatility items may be easy to guess based on orthographic repr esentations. According to the Battig and Montague (1969) category norms, the triads within each category were equated on the typicality or strength of their exemplars as members of the category. The average typicality of the triads was 13.3. Butler et al.s two original filler categories ( Clothing and Disease ) were also used as fillers in the current experiment.

PAGE 56

46 The normative association strengths, bot h direct and indirect, between the exemplars are shown in Appendix C. The Inter-set (strong) condition was very similar to Butler et al.s original design. The association strengths between the triads were moderate to strong, but the associations within triads were relatively weak. According to Nelsons free association norms, there were many associations between the triads used for the practiced and unpracticed conditions. There were 24 direct associations (13 forward links, eleven backward links), and a total of 112 shared associates between triads. The average direct strength and the predicted probability of recovery (PIER2s Equation 3) was .08 ( SD = .05) and .23 ( SD = .30), respectively between the triads. This means that according to the norms, triads we re 8% likely to activate other triads on average, and they were 23% likely to successfully retrieve each other when used as cues on average. There was an average of 3.20 ( SD = 1.73) shared associates with an average strength of .06 ( SD = .10) between the triads. In contrast, there were fewer associati ons within the triads and they were relatively weak in the Inter-set (strong) condition. There were only six forward links, two backward links, and 84 shared associates within the triads. The average direct strength and probability of recovery was .035 and .43, respectively within the triads, and the average shared associate st rength was .037 within the triads. It is important to note that although the average probabi lity of target recovery seem s very high, it is only based on eight pair-wise associations (six forward links and two backward links) compared to 24 pair-wise associations between the triads. When all 9 possible pair-wise connections (3*3 = 9) were include d in the analysis of the Inter-set (strong) condition, results reveal that ev en though the strength values were smaller, there was a similar associative pattern in that associ ative strengths were relatively strong between triads but were relatively weak within triads. Between the triads, the average direct strength and probability of recovery was .02 ( SD = .07) and .08 ( SD = .21), respectively, and there was an average of 1.58 ( SD = 2.01) shared associates with an average strength of .03 ( SD = .08). Within the triads, the average direct strength and probability of recovery as indexed by PIER2s Equation 3 was only .001 and .007, respectively and there were an average of only .07 shared associates with an average strength of .003. As shown in Appendix B, the exemplars w ithin each category were divided into two sets of 3 exemplars each (Triad A and B). The exemplars were assigned to triads so that in the Inter-set (strong) condition, association strengths were strong between the sets, but weak within the sets. As shown in top portion of Table 4, the triads were equated on Between-Triad strength characteristics such as the predicted probability of recovery (Eq3), number of forward links (# FSG), forward strength (FSG), number of backward links (# BSG), backward strength (BSG), to tal number of shared associates (OLAPS), and shared associate strength (OSG). For exam ple, when Triad A was used to activate or cue Triad B in the model, the predicted probabil ity of recovery of Triad B was .22 ( SD = .30) and when Triad B was used to ac tivate Triad in the model, the predicted probability of recovery of Triad A was .23 ( SD = .30).

PAGE 57

Table 4. Between-Triad and Within-Triad stre ngth characteristics and other variables in the Inter-set (strong) condition of Experiment 2. Note EQ3(#) = PIER2s probability of recovery (number of connections); # FSG = number of forward strength connections; FSG = forward strength; # BSG = number of backward connections; BSG = backward strength; OLAPS = number of shared asso ciates; OSG = shared associate strength; CONCR = concreteness; MSS = set size; CONNM = connectivity ; P_RESO = probability of resonance; FREQ = frequency. Between-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean .22 (24) 13 .062 11 .068 112 .064 Between Triads: From Triad A to B SD .30 .06 .15 .10 Mean .23 (24) 11 .068 13 .062 112 .064 Between Triads: From Triad B to A SD .30 .15 .06 .10 Within-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean .39 (5) 3 .03 2 .02 48 .014 Within Triad A SD .19 .017 .015 .02 Mean .47 (3) 3 .04 0 0 36 .061 Within Triad B SD .11 .035 115 Other Variables CONCR MSS CONNM P_RESO FREQ Mean 5.20 14.57 2.04 .37 16 Triad A SD 2.23 6.60 1.18 .23 24.02 Mean 6.00 12.63 1.95 .34 37.44 Triad B SD .45 5.18 1.14 .26 94.12 As shown in the middle portion of Table 4, the Triads were also equated on Within-Triad strength measures, so that the likelihood of exemplars within either triad activating other exemplars in their triads was very low. Finally, as shown in the bottom portion of Table 4, the triads were equated on several other variables that have been shown to affect cued recall includin g concreteness (CONCR), set size (MSS), connectivity (CONNM), probability of resonance (P_RESO), and frequency (FREQ). It is important to note that the betw een-triad and within-triad strength characteristics of list used for the Inter-set (strong) condition in this experiment are similar to the strength characteristics of Butler and colleagues original list shown in Table 5. That is, even though we only used ei ght of their 12 categor ies and two of their three triads per category, the integrity of th e strength characteristics of their list remained. There were many associations between the triads but few associations with the triads in Butlers original lists and in the lis t used in the current experiment. 47

PAGE 58

48 Table 5. Between-Triad and Within-Triad streng th c haracteristics and other variables in Butler et al.s (2001) original lists. Between-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean 39 (39) 18 .067 21 .094 332 .026 Between Triads SD .27 .036 .156 .073 Within-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean .56 (1) 1 .08 0 .00 28 .018 Within Triad A SD .00 .00 .025 Mean .50 (4) 3 .07 1 .05 39 .029 Within Triad B SD .03 .00 .08 Mean .13 (2) 2 .015 0 .00 42 .007 Within Triad C SD .01 .00 .016 Other Variables CONCR MSS CONNM P_RESO FREQ Mean 5.47 12.55 1.99 .34 33.85 Triad A SD 1.92 5.20 .97 .24 86.27 Mean 5.14 12.95 1.76 .34 19.24 Triad B SD 2.22 5.51 1.25 .22 32.35 Mean 5.28 14.84 1.97 .32 20.68 Triad B SD 1.92 6.40 1.06 .26 21.15 Note EQ3(#) = PIER2s probability of recovery (number of connections); # FSG = number of forward strength connections; FSG = forward strength; # BSG = number of backward connections; BSG = backward strength; OLAPS = number of shared associates; OSG = shared associate strength; CONCR = concreteness; MSS = set size; CONNM = co nnectivity; P_RESO = probability of resonance; FREQ = frequency. For the Intra-set (weak) condition, the triads were rearranged in order to m inimize the between triad associations, so the association strengths in the Intra-set (weak) condition were relatively weak between triads but were relatively strong within the triads. According to the norms, there wa s only three forward links, three backward links, 95 shared associates between the triads for all eight ca tegories. The average direct strength and probability of recovery was .06 ( SD = .02) and .50 ( SD = .08) respectively. There was an average of only 2.79 (SD = 1.23) shared associates with an average strength of .036 ( SD = .04). In contrast, there were many more associations within the triads. There were 16 forward links, 9 backward links and 114 shared associates across all the categories. The average direct strength and probability of recovery was .06 and .36, respectively, and there was an average shared associate strength of .06.

PAGE 59

49 When all 9 possible pair-wise associations were included in the analysis of the Intra-set (w eak) condition, results reveal that although the strength values were much smaller, there was a similar associative pa ttern in that associative strengths were relatively weak between triads but were relatively strong within triads. Between the triads, the average direct strength and proba bility of recovery as indexed by Equation 3 was only .01 ( SD = .01) and .01 ( SD = .08), respectively, there was an average of only 1.60 ( SD = 1.67) shared associates with an average strength of .021 ( SD = .04), and there was an average of only .31 ( SD = .53) mediators, with an average strength of .001 ( SD = .003). However within the triads the average direct strength and probability of recovery was .09 and .38, respectively within the triads and there was an average of 3.38 shared associates with an average stre ngth of .05 within the triads. As shown in the top portion of Table 4, the triads in the Intra -set (weak) condition were equated on Between-Triad stre ngth characteristics so that the likelihood of exemplars activating each othe r between the sets in either direction (from Triad A to B or from Triad B to A) was very low as th ere were very few associations between the triads. Table 6. Between-Triad and Within-Triad stre ngth characteristics and other variables in the Intraset (weak) condition in Experiment 2. Between-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean .47 (1) 1 .02 2 .095 95 .036 Between Triads: From Triad A to B SD .00 .00 .078 .039 Mean .53 (2) 2 .095 1 .02 95 .036 Between Triads: From Triad B to A SD .084 .078 .00 .039 Within-Triad Strength Characteristics EQ3 (#) # FSG FSG # BSG BSG OLAPS OSG Mean .46 (24) 14 .12 7 .043 81 .077 Within Triad A SD .32 .18 .019 .0153 Mean .26 (4) 2 .036 2 .015 33 .024 Within Triad B SD .08 .04 .01 .038 Other Variables CONCR MSS CONNM PRESO FREQ Mean 5.43 12.95 2.20 .37 30.14 Triad A SD 1.80 5.14 1.28 .24 80.01 Mean 5.69 14.93 1.59 .36 18.50 Triad B SD 1.73 7.44 .80 .25 31.41 Note EQ3(#) = PIER2s probability of recovery (number of connections); # FSG = number of forward strength connections; FSG = forward strength; # BSG = number of backward connections; BSG = backward strength; OLAPS = number of shared asso ciates; OSG = shared associate strength; CONCR = concreteness; MSS = set size; CONNM = connectivity ; P_RESO = probability of resonance; FREQ = frequency.

PAGE 60

50 As shown in the middle portion of Table 6, th e triads were not as similar in their Within-Triad strength characteristics. A lthough there were many more associations within the triads in the Intra-set (weak) condition, most of them we re in Triad A. For example, when the exemplars within Triad A we re used to activate or cue the others in Triad A in the model, the predicted probabi lity of recovery averaged .46 (SD = .32), based on 24 pair-wise associations. However, when the exemplars within Triad B were used to activate or cue the others Triad B in the model, the predicted probability of recovery averaged .26 (SD = .08), and was only ba sed on four pair-wise associations. However, this was not considered a problem because the main purpose of the Intra-set (weak) condition was to minimize the between tria d associations and not to maximize the associations within the triads. As shown in the top portion of Tabl e 6, the associations between triads were equally minimized in bot h directions from Triad A to B and from Triad B to A. Finally, as shown in the bottom of Table 6, the triads were equated on other variables including concre teness, set size, connectivity, resonance, and frequency. Procedure Upon arrival, participants were randomly assigned to either the Inter-set (strong) or the Intra-set (weak) condition and to one of 16 c ounterbalancing conditions; two category status forms (Form 1, Form 2), two ex emplar status triads (Triad A, Triad B), two Nrp baseline Triads (NrpA, NrpB), and two final testing orders (O rder 1, Order 2). The two category status forms were constructe d by assigning four cat egories to practiced status (Rp), and the remaining four categories to baseline status (Nrp). For example in Form 1, Birds, Fruits, Metals, and Professions served as retrieval practiced categories, and Drinks, Furniture, Insects, and Sports served as baseline categories. However in Form 2, Drinks, Furniture Insects, and Sports served as practiced categories, and Birds, Fruits, Metals, and Professions served as baseline categories. The category status forms were counterbalanced so that all categories se rved equally often as practiced or baseline categories. As shown in Appendix B, the two exemplar status triads (Triad A, Triad B) were constructed by assigning three exemplars from each practiced category to Triad A and the remaining three exemplars to Triad B. For example, in the Inter-set (strong) condition, ROBIN, OSTRICH, and FALCON were assigned to Triad A and their more strongly related associates, SPARROW, PIGEON, and VU LTURE were assigned to Triad B. In the Intra-set (weak) condition, ROBIN, SPARROW, a nd PIGEON were assigned to Triad A, and their more weakly related associates, OSTRICH, FALCON, and VULTURE were assigned to Triad B. The exemplar st atus sets were counter balanced so that all exemplars served equally often in both the retrieval practiced (Rp+) and unpracticed (Rp-) conditions. Two Nrp baseline triads were constructe d by assigning three exemplars from each baseline categories (unpracti ced categories) to NrpA, and the remaining three exemplars from the baseline categories to NrpB. The ba seline triads were counterbalanced so all exemplars served equally often as base lines for RIF effect comparisons.

PAGE 61

51 Finally, two testing orders were develope d. In Order 1, a practiced category was tested first followed by a baseline category, a practiced category, a baseline category and so on for all eight categories. In Order 2, a baseline category was tested first, followed by a practiced category, a baseline category, a practiced category, and so on for all eight categories. The testing orders were counterbalanced so th at the practiced and baseline categories were roughly equated on serial pos ition in the testing sequence. Data was collected in individual sessions. As in Experiment 1, there were four phases; a study phase, a retrieva l practice phase, a retention interval phase, and a test phase. The procedures for the study phase were the same as those used in Experiment 1. The presentation order followed the same criteri a as the Experiment 1 so that items were pseudo-randomized in blocks so that each block contained one exemplar from each category resulting in 6 blocks of 10 items (each block containing eight items from the experimental categories and two items from the filler categories). The ordering restrictions for the study phase, retrieval practice phase, and th e test phase were the same as those implemented in the Experiment 1. All other aspects of th e procedure, including the retention interval phase, the post-experimental questionna ire, and debriefing were the same as those used in Experiment 1.

PAGE 62

52 Results A series of mixed subjects ANOVAs were conducted, with retrieval practice pattern (Intra-set ( weak), Inter-set (strong) ) and counterbalancing (eight counterbalancing conditions) as between subjec ts factors, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) as a w ithin subjects factor. Retrieval-induced Forgetting The overall inhibition (Baseline vs. Rp-) collapsed across the retrieval practice pattern manipulation was significant, and reve aled that probability of recalling Rpitems was lower than the probability of recalling baseline items (7% effect), F (1, 48) = 16.10, Mse = .009, p < .001. The current findings again repl icate previous work showing that the RIF effect occurs with item specific cues. This provides additional evidence that Butlers conclusion that the RIF effect may not be reliable with item specific cues, is unlikely to be correct. The item specific cuei ng results also provide additional evidence that RIF is not simply due to output interference. More importantly for the purposes of this experiment, the interaction between the inhibition effect (Baseline vs Rp-) and the retrieval practice pattern was significant, F (1, 48) = 12.60, Mse = .009, p < .001, indicating that there was a difference in the amount of RIF between the Intra-set (weak) and Inter-set (strong) conditions. As shown in the top panel of Figure 11, the hypothesis that there woul d be significant RIF in the Intra-set (weak) condition was supported, F (1, 24) = 20.41, Mse = .013, p < .0001. The hypothesis that there would be reduced RIF in the Inter-set (strong) condition was also supported, as no reliable RIF was observed, F < 1. In the Intra-set (weak) condition, where associations between the practiced (Rp+) and unpracticed sets (Rp-) were minimized, th ere was a 13% decrement in recall between Rpitems and baseline items in the final test phase. However, in the Inter-set (strong) condition, where associations between the practiced and unpracticed se ts were present, there was no difference in recall performan ce between Rpitems and baseline items in the final test phase (1% decrement). The resu lts of this experiment strongly support the assumption that associations between practi ced items and their unpracticed competitors moderate the inhibitory effect and that this moderation occurs when associations are implicitly activated. When pre-existing associ ations are minimized between the sets, the typical inhibitory effect is observed. This experiment also generalizes the results observed to Experiment 1 to a completely new stimulus set.

PAGE 63

.72.70.69.59.40.50.60.70.80.90Intra-set (weak)Inter-set (strong)Percentage Items Recalled Baseline (Nrp) Rp53 .64.64.85 .89 Baseline (Nrp) 1.00 .40.50.60.70.80Intra-set (weak)Inter-set (strong)Retrieval Practice PatternPercentage Items Recalle .90d Rp+ Figure 11. Results of Experiment 2: Significant RIF effect only in the Intra-set (weak) condition (top panel), but significant retrieval practice benefit in both the Intra-set (weak) and Inter-set (strong) conditions (bottom panel).

PAGE 64

54 Baseline Tests Because the amount of RIF is determined by comparing recall of Rpitems to recall of baseline items, it wa s important to compare the baselines between the Intra-set (weak) (.72) and Inter-set (strong) conditions (.70). As in Experiment 1, there was no significant difference between the baselines, F < 1. Furthermore, there was a significant difference between recall for Rpitems in the weak condition (.59) compared to the strong condition (.69), F (1, 48) = 6.78, Mse = .025, p <.05. As shown in the top panel of Figure 11, recall for Rpitems in the weak condition was 10% lower than recall in the strong condition Because the baselines were statistically equivalent, and the recall of Rpitems was reliably lower in the weak condition, the reliable RIF found in the weak condition was unlikely due to an inflated baseline, and the lack of reliable RIF found in the strong was unlikely due to a deflated baseline. Benefits of Retrieval Practice on Practiced Items As predicted by the inhibitory account of RIF, the amount of retrieval-induced forgetting was independent of the amount of strengthening of the practiced items. The overall benefit (Baseline vs. Rp +) in recall for practiced items (Rp+) (collapsed across the retrieval practi ce pattern conditions) was significant, F (1, 48) = 132.63, Mse = .012, p < .0001. Twenty-three percent more practice d items were recalled in the final test phase ( M = .87, SD = .12) compared to baseline items (M = .64, SD = .17). According to blocking explanations, the amount of forgetti ng of unpracticed items should be dependent on the amount of strengthening of practiced items, so the amount of forgetting should be negatively correlated with the amount of strengthening (McGeoch, 1942; Melton & Irwin, 1940). However, inconsistent with bl ocking accounts, there was a non-significant, positive correlations between facilitation of pr acticed items and forgetting of unpracticed items, r = .23, p = .07. As shown in the bottom panel of Figure 11, the interaction between facilitation of Rp+ items and retr ieval practice pattern was not significant, F < 1. Thus even though forgetting was only apparent in the Intra-set (weak) condition, retrieval practice produced a sign ificant benefit for pr acticed items relative to baseline in both the Intra-set (weak) condition, F (1, 24) = 47.64, Mse = .014, p < .0001, and the Inter-set (strong) condition, F (1, 24) = 92.34, Mse = .01, p < .0001. There was a 21% benefit effect for Rp+ items in the Intra-set (weak) condition, where forgetting was present, and 25% benefit effect in the Inter-set (strong) condition, where forgetting was absent These results are inconsistent w ith the blocking hypothesis of RIF Output Interference Consistent with the results of Experime nt 1, there was also a significant output interference effect with in the baseline condition, F (1, 48) = 11.10, Mse = .012, p = .002. As shown in Figure 12, baseline items cued in the last half of th e category block were recalled at a lower rate ( M = .64, SD = .17) than baseline items th at were cued in the first half of the category block ( M = .71, SD = .16). As in Experiment 1, it was predicted that there would be less output interference in the Inter-set (strong) condition, because this condition should benefit from strong associations between the first half and last ha lf of the baseline category. For example,

PAGE 65

retrieving an exemplar in the first half of the testing sequence should insulate related items in the second half of the testing block from forgetting, but should not insulate unrelated items later in the testing sequence. .70.72.64.64.40.50.60.70.80.90Intra-set (weak)Inter-set (strong)Retrieval Practice PatternPercentage Items Recalled Baseline 1st Baseline 2nd Figure 12. Output Interference Effect in Experiment 2: Significant output interference in the Intra-set (weak), and marginally significant output interference in the Inter-set (strong) condition. As in Experiment 1, the interaction between output interference and retrieval practice pattern was not significant, F < 1. However, unlike Experiment 1, the pattern of output interference was only weakly consistent with the prediction that there would be more output interference in Intra-set (weak) condition. Although, separate analyses revealed that the output interference effect was significant in the Intra-set (weak) condition, F(1, 24) = 7.40, Mse = .012, p = .012, but only marginally significant in the Inter-set (strong) condition, F(1, 24) = 3.99, Mse = .012, p = .06, the differences in output interference were quite small (8% vs. 6%). Unlike Experiment 1, the present results only weakly suggest that associations can also moderate output interference effects just as they moderate RIF. Possible reasons for why we failed to observe the pattern of results are presented in the discussion section. Episodic Integration As in Experiment 1, the amount of episodic integration that participants engaged in during the study phase was measured by a self-report post-experimental integration questionnaire. The average integration score in the Inter-set (strong) condition (Mean rating = 3.51, SD = .95) was similar to the average integration score in the Intra-set (weak) condition (M = 3.12, SD = 1.05), F(1, 48) = 3.014, Mse = .81, p = .089. As in Experiment 1, a median split by post-experimental integration score revealed that there was no significant interaction between overall inhibition and the amount of integration, F < 1. For the Inter-set (strong) condition, there was no 55

PAGE 66

56 interaction between inhibition and the amount of integration indicating that there was no added benefit of explicit, episodic integra tion above the benefit afforded by semantic integration based on pre-existing associations that were implicitly activated during the study phase, F < 1. Results revealed that both high (M = 4.09, SD = .67) and low episodic integrators ( M = 2.94, SD = .85) benefited from pre-exis ting associations between the Rp+ and Rpitems in that there was no forgetting of Rpitems for either level of integration (0% and -2%, respectively). These results suggest that when there are preexisting associations between the practiced and unpracticed sets, semantic integration between the sets is sufficient to reduce th e effect, and episodic integration may not provide an additional benefit. As in Experiment 1, the effects of epis odic integration were different in the Intraset (weak) condition. High integrators ( M = 3.82, SD = .80) exhibited less forgetting than low integrators ( M = 2.42, SD = .80) (-10% and -15%, resp ectively), however this 5% benefit was not reliable as the interaction be tween inhibition and the amount of integration was not significant, F < 1. Although the episodic integration benefit observed here is weaker than in previ ous studies, it is numerically c onsistent with prior findings and Experiment 1. Also the finding of a weak er benefit is not surprising given the strong likelihood that the pre-ex isting associations biased partic ipants to episod ically link the items along those semantic pathways. According to the S pontaneous Integration Bias introduced in the discussion section of Experiment 1, in the case of the Intra-set (weak) condition, there was a Bias to form useless episodic links among the related Rp+ items or among the related Rpitems. Because the Bias steered them away from forming the critical links between the Rp+ and Rpitems, no episodic integration benefit emerged. In the Inter-set (condition) the Bias was towards forming useful links between the related Rp+ and Rpitems, but because such links were already strengthened as a result of semantic integration, there was no adde d benefit of episodic integration.

PAGE 67

57 Discussion The current results replicate and extend th e notion that integration need not occur explicitly, but may occur implicitly, based on pre-experimental associations. As in Experiment 1, pre-existing associations prove d to be an important moderator of RIF: when associations between the practiced it ems and their unpracticed competitors were maximized, RIF was eliminated. Of great impor tance was that the benefits of semantic integration were replicated with a different set of materials that were used by Butler and colleagues (2000) in one study that failed to re plicate the RIF effect with item specific cues. When their stimulus sets were simply re-arranged to minimize the associations between the Rp+ and Rpitems, reliable levels of RIF emerged. The results suggest that their failure to replicate may have been due to a failure to control for pre-existing associations between the practiced items and their unpracticed competitors. The results also suggest that the pre-exis ting associations do not necessarily have to be very strong according to free association measures of strength. That is, even though the associations in the present experiment we re not as strong those in Experiment 1, they still reduced forgetting. For example, in Experiment 2, there were 24 inter-item associative connections with an average direct strength of .06; however in Experiment 1, there were 33 inter-item associ ative connections that were mu ch stronger averaging .26 in direct strength. It could be that free association va lues simply underestimate the likelihood that associations will have an in fluence on performance, under circumstances in which participants encode many differe nt exemplars from the same category-potentially highlighting inter-item linkages. It could also be that even the presence of weak associations will increase semantic integration as long as there are many connections present. Therefore even many weak to moderate associations may be enough to produce semantic integration that will influence the strength of RIF effects. Although the protectiv e effects of semantic integration on RIF were replicated in the current experiment, the benefits of se mantic integration on out put interference in the baselines were not replicated. Although th ere was more output interference in the weak baseline condition, the pattern of results was only weakly consistent with the notion that if output interference is a form of RIF, then semantic integrat ion should also reduce the output interference effect just as it reduces the RIF effect. Howe ver, it could be that the presence of many weak a ssociations was enough to modera te the RIF effect but not enough to moderate the effects of output in terference, given that activation levels, priming, and interference were certainly di fferent between the practiced and baseline categories. For example, r ecall of baseline item s only benefited from the pre-existing associations that were activated during the test phase, whereas the practiced categories benefited from priming effects in both the retrie val practice phase and the final test phase. Given that the baseline categ ories do not benefit from im mediate reactivation during the retrieval practice phase and ar e not reactivated (cued for) unt il 24 minutes after the study phase, certain weaker associations may not be as easily accessible as so there is less priming of related associates that are cued for in the 2nd half of the baseline. Because items in the 2nd half are not adequately primed by items in the first half, they are less likely to be saved from the inhibitory eff ects of recalling competitors. Therefore when

PAGE 68

58 associations are relatively wea k, between the first and second half of the baseline, the benefits of semantic integration may be less likely to emerge compared to when there are strong associations. To summarize, the current results suggest that Butler et al.s (2001) failure to replicate the RIF effect with item specific cu es was most likely due to moderating effects of semantic integration. When their stimulus sets were simply re-arranged to minimize semantic integration, reliable levels of RI F emerged. This is additional evidence that Butler et al.s (2001) findings are not repres entative, for the RIF effect was once again replicated with item specific cues when semantic integration was controlled.

PAGE 69

59 Experiment 3: Investigating whether the moderating effects of pre-existi ng associations are due to the implicit activation of the pre-existing associations or are mediated by explicit integration of the exem plars during study The results of the previous studies reveal that semantic integration as a result of pre-existing associations is an important moderator of RIF. One important question for the present set of studies is the extent to wh ich the associative effects that were observed in the previous experiments were based on preexisting associations that were implicitly activated, or on new episodic associations fo rmed during the intentional study phase, that were triggered by these pre-existing associa tions. In other words, was the forgetting effect reduced by the explicit a nd intentional integr ation of these pai r-wise associations? Anderson and colleagues have shown that the amount of RIF depends on how well integrated the to-be-retrieved memories ar e with the practiced competitors (Anderson & McCulloch, 1999). Participants instructed to rehearse and inter-r elate the items during encoding and participants who spontaneously and intentionally integrated the items in this explicit fashion exhibited significantly le ss RIF. Episodic integration strategies are believed to serve as a moderator of the RIF effect by insulating some related items from inhibition. Although we did not ask our pa rticipants to intentionally integrate the items in the previous experiments, directions to integrate may not be not needed when there are strong associations among the list items because th e pair-wise relationships may be more noticeable. When strong associations are noticeable, partic ipants may be more likely to spontaneously integrate the items (i .e., without instructi ons to intentionally integrate them), and it may be this explicit in tegration strategy that leads to a reduced forgetting effect, not the associations by themselves. If explicit integration strategies mediated the semantic integration effect in the Inter-set (strong) conditions of Experiments 1 and 2, then one would expect to have found reduced RIF only when participants repo rted high amounts of intentional episodic integration. When participants reported low amounts of episodic integration then they should have exhibited reliable levels of RIF. Howeve r in both of the previous experiments, both high and low integrat ors failed to show forgetting in the Inter-set (strong) condition. That is, even participants wh o did not explicitly integrate the items during study still did not exhi bit forgetting when there were strong associations between the practiced and non-practiced sets. From these results, it appears that explicit integration is not needed in order for the bene fits of pre-existing associations to emerge. The results are therefore more consistent w ith the assumption that the moderating effects of semantic integration observed in Experi ments 1 and 2 and in Butlers study are based

PAGE 70

60 on pre-existing semantic associations activat ed implicitly during study between practiced and non-practiced items and these moderati ng effects are not dependent on episodic integration. Despite this evidence in favor of the idea that semantic integration effects were not dependent on episodic inte gration, a much cleaner demons tration of the independent nature of the semantic integration effect wa s warranted. For example, it could be argued that although there were reliable differences in episodic integration scores between those categorized as high integrators and those categorized as low integrators, the integration scores may have been high enough ove rall so that even low integrators were actually using intentional integration strategies at a high rate (just not as high as high integrators). One could therefore argue that because integration scores were generally high, then it was episodic integration and not semantic integration that reduced the RIF effect. Therefore, the purpose of Experi ment 3 was to minimize the use of episodic integration to the greatest extent possible in order to clearly demonstrate that the protective effects of pre-existing associations are due to im plicit activation and occur in absence of episodic integration strategies. This was accomplished through manipulating the intentionality of the lear ning episode. Some participan ts intentionally studied the items for a later memory test (as in the pr evious experiments). Others incidentally encode the exemplars, by judging how well th e each exemplar fit as member of its category. In this Incidental Learning condition, no menti on was made of learning the items for a later memory test. Our hope wa s that by employing the Incidental Learning task in the study phase, participants would focus their attention on the relation between each exemplar and its category, and they would therefore be unlikely to intentionally seek inter-exemplar relationships for purposes of e nhancing memory. It was hypothesized that there would be relatively small amounts of self-reported explicit integration in the Incidental Learning condition compared to the Intentional Learning condition, but that the protective effects of semantic integration as a result of pre-existing associations would still emerge. Because the purpose of the Incidental Learning condition was to reduce the amount of episodic integration to the greatest extent possible, two additional controls were introduced in order to ensure that the items were not episodica lly integrated during the judgment (study) phase. First, an additional question was included on the postexperiment questionnaire that measured the extent to which participants episodically linked the items while making their judgments For example, in order to make their judgments, some participants may have t hought back to the way they judged other members from the same category, and then judged the items in relation to previous judgments. In this process of making comp arative judgments, the two items could have become episodically integrate d. For example, some participants may have thought back to the way they judged PONYs fit to the Animal category in order to make their judgment of how well other Animals such as HORSE fit the Animal category. Because HORSE and PONY are very similar, these pa rticipants may have tried to give them similar ratings by thinking back to previously judged items, the result being high amounts of episodic integration be tween the members of the Animal category. Therefore, it could be argued that particip ants, who incidentally encoded the items and so reported very little

PAGE 71

61 episodic integration in orde r to improve their memory, could have nevertheless episodically linked the items in order to ma ke their judgments. Because intentional integration scores may have captured only a part of Incidental Learning participants overall episodic integration behavior, the amount of episodic integrati on in the process of making comparative judgments needed to be taken into account as well. Therefore a comparative judgment question was added that m easured the extent to which participants thought back to previously judged categor y members when they encountered a new member, and made their judgments in relati on to judgments of other members from the same category. In order to ensure that the items were not episodically integrated during the judgment (study) phase of the Incidental Learning condition, anothe r question was added to post-experiment questionnaire that measured the extent to which participants expected that their memory would be tested. The a ssumption was that those that expected a memory test would be more likely to believe that the experiment was about memory and so they would be more likely to episodically integrate the items in order to improve their memory. This question simply asked participan ts if they expected their memory of the judgment items would be tested later on after the judgment phase. Because the use of spontaneous episodic integration strategies usua lly increase when people expect that their memory will be tested, it was important to en sure that participants in the Incidental Learning condition did not expect a later memory test, and to ensure that participants were truly incidental in the sense that they were not adopt ing strategies in order to improve their memory. Therefore, the purpose of the current experi ment was to replicate the results of the previous experiments using the same proce dure and stimuli as Experiment 1, but to manipulate the intentionality of the lear ning episode by having participants either intentionally or incidentally encode the ite ms. We replicated the Intentional Learning condition in order to measure the rate of spontaneous explic it integration during the study phase, and to directly compare recall and integration performance between the two learning conditions. We wanted to compar e the amount of RIF directly, and more importantly we wanted to discover if the sa me pattern of results related to retrieval practice pattern (i.e. semantic integration) was still evident when episodic integration was low as a result of receiving incidental learni ng instructions. It was hypothesized that the integration scores in the incidental learni ng condition would be reliably lower in the Incidental Learning condition compared to the Intentional Learning condition. It was also hypothesized that the patt ern of RIF related to sema ntic integration that was observed in Experiments 1 and 2 would also be observed in both learning conditions of Experiment 3. If the same pattern of resu lts is found between the two types of learning conditions and there is relative ly little explicit integration in the Incidental Learning condition, then it can be concluded that the absence of forgetting in the Inter-set (strong) condition is due, in some degree, to the imp licit activation of preexisting associations, that semantic integration protects memories from forgetting, and that this protection is not mediated by explicit, episodic integration st rategies. It would also provide additional confirmation that intention to learn is not a necessary component in finding the RIF effect (Anderson & Bell, 2001).

PAGE 72

62 Methods Design and Participants The experimental design formed a 2 X 2 X 8 X 4 mixed-subjects factorial with type of learning (intentional, incide ntal), retrieval practice pattern ( Inter-set (strong) Intra-set (weak) ), and counterbalancing (eight count erbalancing conditions) manipulated between subjects, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) manipulated within subjects. One hundred and forty-seven undergraduate psychology students from the University of South Florida served as partic ipants in the study in exchange for extra credit. Participants ranged in age from 18 to 52 years, with a median age of 21. The participant pool was 78% female and 22% male, and the ethnic make-up of the participant pool was 60% Caucasian, 20% Af rican American, 14% Hispanic, 3% Asian, and 3% other. The average Shipley score was 29.56 ( SD = 3.97). The average MEQ score was 45 ( SD = 9.55), indicating that on averag e the circadian rhythm of the participant pool fell into the neutral range between moderately mo rning and moderately evening. Approximately 80% of participants were tested during a time a day that was consistent with their peak circadian rhythm. The average Cognitive Failures score was 44.07 ( SD = 12.58) and the average reasoning score was 77% ( SD = 21) out of 100. Materials The materials and the list and for this experiment were the same used in Experiment 1 and are shown in Appendix B. Procedure Upon arrival, participants were randomly assigned to either the Inter-set (strong) intentional learning, Inter-set (strong) incidental learning, Intra-set (weak) intentional learning, or the Intra-set (weak) incidental learning condition. The procedures were identical to those used in the Experiment 1 except that a study instruction manipulation was adde d. Half of the participants were given the intentional learning instructions used in Experiment 1, and the other partic ipants were given incidental learning instructi ons. Those who received inci dental learning instructions were told that the experiment involved judgment and reasoning, that category-exemplars would appear in the center of the computer sc reen, and they would have five seconds to judge how well the exemplar fit the categor y on a scale of 1 to 5, where 1 equals no fit and 5 equals best fit. All other procedures for the s tudy phase, retrieval practice phase, distracter phase, and te st phase were identical to t hose used in Experiment 1.

PAGE 73

63 Results A series of mixed subjects ANOVAs were conducted, with retrieval practice pattern ( Intra-set (weak) Inter-set (strong) ), type of learning (int entional, incidental), and counterbalancing (eight counterbalancing conditions) as between subjects factors, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) as a within subjects factor. Because the incidental learning instruc tions were intended to reduce attempts to explicitly study the exemplars, participants in the incidental learning c ondition who reported that they expected a later memory test ( n = 19) were excluded from further analysis, leaving 128 subjects ( n = 32 in each condition). The exclusion of these participants does not alter the conclusions drawn from this study. Retrieval-Induced Forgetting The overall inhibition (Baseline vs. Rp-) collapsed across our retrieval practice pattern manipulation, and across type of learning was signifi cant, and revealed that the probability of recalling Rpitems was lower th at the probability of recalling baseline items (-5% RIF effect), F (1, 96) = 9.00, Mse = .02, p = .003. The current findings replicate previous work and Experiments 1 and 2, demonstrating that the RIF effect occurs with item specific cues. Retrieval-Induced Forgetting as a Function of Retrieval Practice Pattern. More importantly, the interaction between the inhib ition effect and the retrieval practice pattern was significant, F (1, 96) = 16.90, Mse = .02, p < .0001, indicating that there was a difference in the amount of RIF between the Intra-set (weak) and Inter-set (strong) conditions. As shown in the top panel of Figure 13, the results supported the hypothesis; there was significant RIF in the Intra-set (weak) condition, F (1, 56) = 18.74, Mse = .027, p < .0001, but no RIF in the Inter-set (strong) condition, F < 1. In the Intra-set (weak) condition, where associations between the pr acticed (Rp+) and unpracticed sets (Rp-) were minimized, there was a 13% decrement in recall between Rpitems and baseline items in the final test phase; however in the Inter-set (strong) condition, where strong associations between the pr acticed and unpracticed sets were present, there was facilitation in recall of Rpitems compared to baseline items in the final test phase (+2%), though this facilitation was not reliable. Baseline Tests. Results revealed that the diffe rence between the baselines only approached significance, F (1, 96) = 3.84, Mse = .029, p =.06. There was, however, a significant difference between recall for Rpitems in the weak condition (.57) compared to the strong condition (.66), F (1, 48) = 6.98, Mse = .034, p <.01. As shown in the top panel of Figure 13, recall for Rpitems in the weak condition was 9% lower than recall in the strong condition These results suggest that ther e was no statistical difference between the baselines, but that recall for Rpitems was reliably lower in the weak condition, thus indicating that the reliable RIF found in the weak condition was not due to an inflated baseline, and the lack of RIF found in the strong condition was not due to a deflated baseline.

PAGE 74

.70.64.66.57.40.50.60.70.80.90Intra-set (Weak)Inter-set (Strong)Retrieval Practice PatternPercentage Items Recalled Baseline (Nrp) Rp64 .62.62.80 .83.90 Baseline (Nrp) 1.00 Rp+ .40Intra-set (Weak)Inter-set (Strong)Retrieval Practice PatternPerc .50.60.70.80entage Items Reca lled Figure 13. Results of semantic integratio n in Experiment 3: Significant RIF effect only in the Intra-set (weak) condition (top panel), but retrievalpractice benefit in both the Intra-set (weak) and Inter-set (strong) conditions (bottom panel).

PAGE 75

65 Retrieval-Induced Forgetting as a Function of Type of Learning. Results revealed that the interaction between the inhibition effect and the learning manipulation was not significant, F < 1. As shown in the top panel of Figure 14, the RIF effect approached significance in the Intentional Learning condition, F(1, 48) = 3.63, Mse = .023, p = .06, and there was a reliable RIF effect in the Incidental Learning condition, F(1, 48) = 5.75, Ms = .017, p = .02. e .67.67.61.62.40 .50rc .60.70.80.90IntentionalIncidentalage Items Recalled Baseline (Nrp) Rpent Pe .83.80.901.00ed Baseline (Nrp) Rp+ .61.63.40.50.60IntentionalPercentage Item .70.80s Recall Incidental Type of Learning Figure 14. Results of episodic integration in Experiment 3: Marginally significant RIF in the Intentional Learning and significant RIF in the Incidental Learning condition (top panel), and retrieval-practice benefits in both learning conditions (bottom panel).

PAGE 76

66 elfSD = 1.07). Despite this sizeable difference in intentional integration, the magnitude of the RIF effect in th Inten s w there was no significan diffe basel iti wao signi etwee in thnalning ion (.62) compIncide .61), 1. These results suggest that there was no statistical difference be conditions, cating th f RIF f in bornin conditions were not due to in Semantic Integration Effects on RIF as a Function of Type of Learning As shown in Table 7, the three-way interaction between the inhibition effect, retrieval practice pattern, and type of learning was not significant. This indicates that the two-way interaction between the inhibition effect and retrieval pr actice pattern did not di ffer as a function of type of whether participants engaged in Intentional or Incidental learning, F (1, 96) = 1.94, Mse = .02, p = .25. As hypothesized, participants who received Incidental Learning instructions exhibited a modula tion of RIF as a function of the strength of association between Rp+ and Rpitems, F (1, 48) = 16.70, Mse = .017, p < .0001. For those who received Intentional Learning instructions, the two-way interaction between inhibition and retrieval practi ce pattern approached significan ce revealing that they also exhibited a modulation of the RIF as a function of retrieval pr actice pattern as those in the Incidental condition, F (1, 48) = 3.63 Mse = .023, p = .063. Table 7. RIF effect as a function of type of learning and strength of association between the Rp+ and Rpitems. Note ** indicates significance at p < .05. Type of Learning Strength of Association between Rp+ and RpBaseline RpRIF Effect An examination of the post-experimental episodic integration rating scores revealed that the learning manipulation produced significantl y different levels of s reported episodic integration, F (1, 64) = 225.25, Mse = .64, p < .0001. The episodic integration scores in the Incidental Learning cond ition were significantly lowe r ( M = 1.53, SD = 1.01) than in the Inten tional Learning condition ( M = 3.60, e Incidental Learning condition (-6% RIF) was similar to that found in the tional Learning cond ition (-5% RIF effect). Baseline Tests. A ith retrieval practice pattern nd t s also n rence between the ficant difference b ine s of the learning co ons, F < 1 e Itio There L n recall for Rpitems nten ear condit ared o the t ntal Learning condition ( tween the baselines or between the RpF < thus indi at the reliable levels o ound th lea g flated baselines. Intra-set (weak) (weak) .70 .60 -.10** Intentional Inter-set (strong) (strong) .64 .64 .00 Intra-set (weak) (weak) .70 .55 -.15** Incidental Inter-set (strong) (strong) .64 .68 +.04

PAGE 77

67 Evidence that Semantic Integration Be nefits are not dependent on Episodic Integration. In the Intra-set (weak) condition, incidental learni ng participants exhibited similar amounts of forgetting of Rpitems, F (1, 24) = 14.84, Mse = .024, p < .001, as those who received Intenti onal Learning instructions, F (1, 24) = 5.76, Mse = .029, p = .024, (-15% and -10% RIF, respectively). In the Inter-set (strong) condition, incidental learning participants exhi bited a similar lack of forgetting, F (1, 24) = 2.76, Mse = .01, p = .13, as those who received Intentional Learning instructions, F < 1 (+4% facilitation and 0%, respectively). As mentioned previously, an examination of the post-experimental episodic integration scor es revealed that the learning manipulation produced significantly different le vels of episodic integration, F (1, 64) = 225.25, Mse = .64, p < .0001. The integration scores in th e Incidental Learning condition were significantly lower ( M = 1.53, SD = 1.01) than in the Intentio nal Learning condition ( M = 3.60, SD = 1.07). The present results thus indicate that the moderating effects of semantic integration were not mediated by in tentional episodic integration strategies for there was reduced forgetting in the Inter-set (strong) condition even for participants whose episodic integration scores were low as a result of receivi ng incidental learning instructions. There remains the possib ility, however, that unintentional episodic integration arising from comparative judgments in the incidental learning condition may have produced the integration benefits observed in that condition. I shall return to this point in our later discussi on of the data, broken out by comparative reports. Baseline Tests. In the Intentional Learning condition, there was no significant difference between the baseline in the Intra-set (weak) condition (.70) and the baseline in the Inter-set (strong) condition (.64), F (1, 48) = 1.94, Mse = .028, p = .17. In the Incidental Learning condition, there was al so no significant difference between the baseline in the Intra-set (weak) condition (.70) and the baseline in the Inter-set (strong) condition (.64), F (1, 48) = 1.87, Mse = .029, p = .18. These results indicate that the reliable levels of RIF found in the Intra-set (weak) condition of both learning conditions was not due to an inflated baseline and the lack of RIF found in the Inter-set ( strong) condition of both learning conditions was not due to a deflated baseline. Benefits of Retrieval Practice on Practiced Items The overall benefit (Baseline vs. Rp+) in recall for practiced items (Rp+) collapsed across our manipulations of retrie val practice pattern and type of learning was significant, F (1, 96) = 96.59, Mse = .024, p < .0001. Participants re called 18% more of the practiced items in the final test phase ( M = .81, SD = .17) than baseline items ( M = .63, SD = .21). As shown in bottom panel of Figure 13 and Figure 14, the amount of RIF was independent of the amount of strengthening of the practiced items, thereby replicating the strength independence property of RIF. In contrast to blocking accounts that predict that the amount of strengthening of practiced items should be negatively correlated with the amount of forgetting of unpracticed items, ther e was a non-significant, negative correlation between th e facilitation of practiced items and the forgetting of unpracticed items, r = -.05, p = .60. As in the previous experiments, the interaction between facilitation of Rp+ items and retr ieval practice pattern was not significant, F < 1. There was significant facilitation of Rp+ items in the Intra-set (weak) condition where

PAGE 78

68 forgetting was present, F (1, 56) = 32.18, Mse = .031, p < .0001, and in the Inter-set (strong) condition where forgetting was absent, F (1, 56) = 70.48, Mse = .019, p < .0001. In other words, although forge tting was only apparent in the Intra-set (weak) condition, retrieval practice produced a sign ificant benefit for pr acticed items relative to baseline in both the Intra-set (weak) and Inter-set (strong) conditions (+17% and +21%, respectively). As shown in Table 8, the lack of an interaction between retrieval practice benefit for practiced items and retrieval practice patt ern did not vary as a function of type of learning, as the three-way inter action between benefit, pattern, and type of learning was not significant, F (1, 96) = 2.77, Mse = .024, p = .10. Participants in the Incidental Learning condition exhibited equivalent benef its for practiced items across the retrieval practice patterns conditions (+18% facilitation) as those in the Intentional Learning condition (+19% facilitation). Table 8. Retrieval practice benefit effect as a function of type of learning and strength of association between the Rp+ and Rpitems. No te ** indicates significance at p < .05. Type of Learning Strength of Association between Rp+ and RpBaseline Rp+ Benefit Effect Intra-set (weak) .64 .85 +.21** Intentional Inter-set (strong) .63 .80 +.17** Intra-set (weak) .61 .74 +.13** Incidental Inter-set (strong) .62 .85 +.23** These findings replicate previous work by and Anderson and colleagues (Anderson et al., 1994; Anderson & McCu lloch, 1999) and Experiments 1 and 2, showing that RIF effects are not dependent on the amount of streng thening of practiced items as interference accounts would predict (e.g. McGeoch, 1942; Melton & Irwin, 1940). The strength independent nature of the forgetting effect is therefore more consistent with an inhibitory account of RIF. The current results extend previous work by showing that the strength independence prop erty of RIF also manifests itself in incidental learning tasks in which ther e is relatively sma ll amounts of episodic integration. Output Interference Results revealed that the overall output in terference effect (Nr p1 baseline vs Nrp2 baseline) within the baselin e condition was significant, F (1, 96) = 5.43, Mse = .025, p = .02. Overall, baseline items cued in the last half of the category block (i.e. Nrp2) were recalled at a lower rate ( M = .62, SD = .21) than baseline items cued in the first half of the category block (i.e. Nrp1) ( M = .67, SD = .17). As in Experiments 1 and 2, it was hypothesized that if output interference is a fo rm of RIF, then preexisting associations should moderate the effects of output interference just as they moderate RIF when there are strong associations between practiced and unp racticed sets. It was therefore predicted that there would be significant output interference in the Intra-set (weak) condition

PAGE 79

where there were weak-to-no associations between the first half of the testing sequence (e.g. HORSE) and second half of the testing sequence (e.g. LION), but that there would less output interference in the Inter-set (strong) condition where there were strong associations between baseline items in the first half (e.g. HORSE) and baseline items in the second half (e.g. PONY). 69 .70 Baseline 1st .90 Baseline 2nd .64.62.62.50.60.70Percentage Items Recall .40e .80d Intra-set (Weak)Inter-set (Strong)Retrieval Practice Pattern Figure 15. Output Interference Effect in Experiment 3: Significant output interference in the Intra-set (weak) condition, but not in the Inter-set (strong) condition. As in Experiments 1 and 2, the interaction between output interference and retrieval practice pattern was not significant, F(1, 96) = 2.36, Mse = .025, p = .13. However, as in Experiment 1, the pattern of output interference was at least moderatelconsistent with the idea that retrieving an exemplar in the first half of the testing sequence should insulate related items in the second half from forgetting (Figure 15). Separate analyses revealed that in the weak condition, the output interference effect was significant at -8%, F(1, 56) = 7.58, Mse = .025, p < .01. However, in strong condition, the output interference effect was not significant (only a -2% output interference effect), F < 1. s show y n in Table 9, the three-way interaction between output interference, retrhypions exhibited a similar pattern of output interference as a function of the strengths of associations between Rp+ and Rpitems, as those who received Intentional Learning instructions. As in Experimeleanotioterence effect. Separatevealra-set (weakditionse in theental Learning condition exhibited significant levels of output interference (-9% output A ieval practice pattern, and type of learning was not significant, F < 1. As othesized, participants who received Incidental Learning instruct ent 1, thn that semantic in pattern on results was at ast moder tely consisut nterfe tent with the gration should also moderate the outp i, tho analses re y ed that in the Int ) con Incid

PAGE 80

70 interference effect), F (1 = .018, p = .01, and those in the Intentional Learning conition exhi s of output ference% outpu terference effect); however th e effect for Intentional Learni ng did not reach statistical = 1.80, Ms = .035, p = .19. In the Inter-set (strong) condition, those in nce , 24) = 7.25, Mse d bited similar level inter (-6 t in significance, F (1, 24) e the Incidental Learning condition exhi bited a similar lack of output interfere as those in the Intentional Learning condition, F < 1, in both cases (-2% and -1% respectively). Table 9. Output interference effect as a function of type of learning and strength of association between the Rp+ and Rpitems. Note ** indicates significance at p < .05. Type of Learning Strength of Association between Rp+ and RpBaseline 1st half Baseline 2nd half Output Interference Effect Intra-set (weak) .70 .64 -.06 Intentional Inter-set (strong) .64 .63 -.01 Intra-set (weak) .70 .61 -.09** Incidental Inter-set (strong) .64 .62 -.02 Episodic Integration As in prior experiments, the amount of episodic integration that participants engaged in during the study/judgment pha se was measured by a self-report postexperimental integration questionnaire. This was measured for both learning conditions. The learning manipulation produ ced significantly different leve ls of episodic integr F (1, 96) = 130.78, Mse = 1.037, p < .0001. The integration sc ores in the Intentional Learning condition were greater ( M = 3.60, SD = 1.07) than in the Incidental Learning condition ( M = 1.53, SD = 1.01). For the Intentional Learning condition, the average integration score in the Inter-set (strong) condition ( M = 3.51, SD = 1.25) was simila average integration score in the Intra-set (weak) condition ( M = 3.64, SD = .72), F < 1 For the Incidental Learning condition, th e average integration score in the Inter-set (strong) condition ( M = 1.46, SD = 1.18) was similar to the av erage integration score in the Intra-set (weak) condition ( M = 1.60, SD = .97), F < 1. As in Experiments 1 and 2, the median split by post-experimental integration score revealed that there was no significant intera ation, r to ction between overall inhibition and the amount of episodic integration, F < 1. Regardless of the type of learning, there was no inte con ready affo F < 1 in both learning cases. As shown in Table 10, in the Intentional Learning condition, when semantic integration was high, both high ( M = 4.31, SD = .70) and low episodic integrators ( M equiva le In Learning hen tion was high ( ter-set (str both high ( M = 1.92, SD = 1.24) and low episodic integrators ( M = 1.00, SD = .00) also exhibited equivalent lack of reliable RIF effects (+ 5% and +4%, respectively). These results replicate the results of Experiments 1 and 2, suggesting that when th ere were strong raction between inhibition and the amount of integration in the Inter-set ( strong) dition, indicating no aded benefit of episod ic integration above the benefit al d rded by semantic integration = 2.75, SD = respectively). 1 25) exhibited lent lack of reliab RIF effects (2% and 2% the Incidental condition w semantic integra i.e. In ong) ),

PAGE 81

71 pre-existinions ticed and unpr sets, semtegration between the se was suff e effect, and episodic in tegration did not add any significant benefit. The lack of episodi c integration benefits are not surprising given the hig ink s g associat between the prac acticed antic in ts icient to reduce th h likelihood that there was a Spontaneous Integration Bias to episodically l items that are already semantically associated. Table 10. RIF effect as a function of type of l earning, strength of asso ciation between the Rp+ and Rpitems, and level of episodic integra tion. Note ** indicates significance at p < .05. indicates significance at p < .10 Type of Learning Strength of association between Rp+ and RpRIF effect for High Episodic Integrators RIF effect for Low Episodic Integrator Intra-set (weak) -.12 -.09* Intentional Inter-set (strong) +.02 -.02 Intra-set (weak) -.17** -.13** Incidental Inter-set (strong) +.05 +.04 The effects of episodic integration were not very different when semantic integration was low (i.e. Intra-set (weak) ). As shown in Table 10, high episodic integrators did not exhibit the typical 5% episodic integrati on benefit as they did in the Intra-set (weak) conditions of Experiments 1 and 2. In fact, the episodic integration benefits did not emerge in either learning condition (F < 1, for both learning conditions). In the Intentional Learning condition when semantic integration was low, high ( M = 4.27, SD = .54) and low episodic integrators ( M = 3.03, SD = .57) exhibited similar levels of RIF (-12% and -9%, respectively), F < 1. In the Incidental Learning condition when semantic integration was low, both high ( M = 1.88, SD = .73) and low episodic integrators ( M = 1.00, SD = .00) also exhibited similar levels of RIF (-17% and -13%, respectively), F < 1. From these results it appears that epis odic integration did not produce an added benefit above the benefit of semantic integration in the Inter-set (strong) condition, nor did it insulate items from forgetting in the Intra-set (weak) condition. However, given a Spontaneous Integration Bias to episodically link semantically related items together in the present set of experiments, it is not surpri sing that the benefits of episodic integration failed to emerge. Because semantically related items were already strengthened by semantic integration effects, episodic integration on those related li nks was expected to have little added benefit.

PAGE 82

72 More Evidence that Semantic Integration Benefits are not dependent on Episodic Integration. It is important to note that the main purpose of introducing the Incidental Learning instructions in this experiment was to reduce epis odic integration scores to determine if semantic integration benefits continue to emerge even when episodic integration is at the lowest level. Th erefore, the lack of inhibition in the Inter-set (strong) condition is especially noteworthy for the lo w episodic integrators in the Incidental Learning group because the average rating of in tentional, episodic integration for this group was 1.00. This is the lowest score that is possible on our inte gration scale. In other words, these participants claimed that they never episodically integrated the items during the study phase, yet they benefited from semantic integration just as those who reported higher levels of episodic integration. This is clear evidence that the semantic integration benefits are not dependent on episodic integr ation strategies. However, it could be argued that ep isodic integration measured by our postexperimental questionnaire, di d not capture the full intenti onal integration behavior of participants in the Incidental Learning condition. This is because the episodic integration question measured the extent to which participants rehearsed the items together for the purposes of improving their memory. Incidental participants were not aware that their memory would be tested and so they were not likely to engage in intentional integration strategies for the purposes of improving memory. However, as mentioned previously, these participants could have intentionally integrated in order to make their judgments of how well the exemplar fit the category. Theref ore, it could be argued that integration as measured by the comparative judgment ques tionnaire would be a better index of Incidental participants epis odic integration behavior, be cause intentional, episodic integration scores only captured a part of their overall epis odic integration behavior. If low episodic integrators, as measured by co mparative judgments, still exhibit the same benefits of semantic integration as high episodi c integrators, then this would provide even more convincing evidence that the insulati ng effects of semantic integration are not dependent on the use of exp licit integration strategies. The results on comparative integration are presented and discussed in the following section. Additional Evidence that Semantic Inte gration Benefits are not Dependent on Explicit Integration Strategi es: Episodic Integration in the form of Comparative Judgments. For the incidental lear ning condition, the amount of episodic integration engaged in during the judgment/study phase was also measured by a self-report postexperimental integration questionnaire. The comparative judgment questionnaire measured the extent to which participants thought back to previously judged category members when they encountered a new member and made their judgments in relation to judgments of other members from the same category. For example, it measured how likely they were to think back to th e way they judged PONYs fit to the Animal category in order to make their judgme nt of how well HORSE fit the Animal category. Because HORSE and PONY are very similar participan ts may have tried to give them similar ratings by thinking to previously judged items in this way. For each category, participants indicated how of ten on a 5-point scale ( being none of the time, and

PAGE 83

73 being all of the time) they made these comp arative judgments. The average comparative integration score in the strong condition ( M = 2.65, SD = 1.03) was similar to the average integration score in the weak condition ( M = 2.81, SD = 1.23), F < 1. Because of the Spontaneous Integration Bias to episodically link semantically related items together, it was no surprise that as with episodic integration, comparative integration did not produce any added benefit. Like the median split on episodic integration, the median split by post-experimental comparative integration rating scores revealed no significant interaction betw een overall inhibition and the amount of comparative integration in the incidental lear ning condition, F < 1. There was no interaction between inhibiti on and the amount of comparative integration in the strong condition, indicating no added benefit of episod ic integration above the benefit already afforded by semantic integration, F < 1. As shown in Table 11, when semantic integration was high, both high ( M = 3.43, SD = .69) and low compara tive integrators ( M = 1.92, SD = .64) did not exhibit reli able RIF effects (+2% and +5%, respectively), F < 1 in both cases. There was also no inter action between inhibiti on and the amount of integration in the weak condition, indicating no benef it of episodic integration, F < 1. As shown in Table 11, when semantic integration was low, both high ( M = 3.86, SD = .82) ( F (1, 8) = 11.77, Mse = .033, p < .01), and low compar ative integrators ( M = 2.19, SD = .83) exhibited reliable RIF effect s (-16% and -14%, respectively), F (1, 8) = 5.77, Mse = .055, p = .043. Table 11. RIF effect in the Incidental Learning condition as a function of strength of association between the Rp+ and Rpitems and level of comparative integration. Note ** indicates significance at p < .05. Strength of association between Rp+ and RpRIFeffect for High Comparative Integrators RIFeffect for Low Comparative Integrators Intra-set (weak) -.14** -.16** Inter-set (strong) +.02 +.05 These results also indicate that episodic integration in the form of comparative judgments did not produce any added benefit above the benefit of sema ntic integration in the Inter-set (strong) condition, nor did it help insula te items from forgetting in the Intraset (weak) condition. These results a ppear inconsistent with previous studies and the results of Experiments 1 and 2 showing that ep isodic integration produc es a 5% benefit. However, the present form of incidental in tegration (via comparative judgment) has not generally been measured in prior studies, so it also remains possible that incidental integration of this sort may behave differently than intentional integration. In addition, as mentioned previously, the lack of episodic in tegration benefits was expected given the presence of a Spontaneous Integration Bias to comparatively judged items in relation to semantically related items because those related items came so easily to mind.

PAGE 84

74 The main purpose of the current investig ation of comparative integration was to examine whether low comparative integrator s also exhibited the same benefits of episodic integration as high comparative inte grators. The results were completely consistent with the hypothesis that the benefits of sema ntic integration would be completely independent of any type of epis odic integration strategy, whether the strategy was to rehearse items together in order to im prove memory or to recall previously judged items in order to make comparative judgment s. As shown in Table 11, when semantic integration was low, low comparative integrator s exhibited similar reliable levels of RIF as high comparative integrators. In addition, both low and high comparative integrators were saved from RIF effects when semantic integration was high. These results provide even more convincing evidence that the benefici al effects of semantic integration are not dependent on explicit, integration strategies More Evidence that Semantic Integration Benefits are not Dependent on Explicit Integration Strategies: A Final Critical Test. In this final analysis, we sought to evaluate, under the most stringent possible conditions, wh ether episodic integration was necessary for the benefits of semantic integration to emerge. In order to do this, we created a composite integration score based on participants intentional episodic integration and their comparative judgment sc ores. The reason for examining composite integration scores was that it may have been possibl e for the low-integrat or groups in the two separate analyses (the one for episodic and the one for comparative) to have contained high integrators who were high integrated ba sed on the other measure. Therefore in the present analysis, we constructed a group of in cidental learning part icipants who had low intentional episodic integrati on scores and low comparative integration scores. We did this by averaging the episodic and compara tive integration scores and performing the median split on that composite measure of integration. Table 12. RIF effect in the Incidental Learning condition as a function of strength of association between the Rp+ and Rpitems and level of composite integration (i.e., the average of the episodic an d comparative integration scores). Note ** indicates significance at p < .05. Strength of association between Rp+ and RpRIF effect for High Composite Integrators RIF effect for Low Composite Integrators Intra-set (weak) -.17** -.13** Inter-set (strong) +.03 +.04 As with episodic integration and compar ative integration, the findings clearly demonstrate that semantic integration moderated the forgetting effect, and that this moderation was not dependent on episodic in tegration strategies. Because the main purpose of present experiment was to reduce ep isodic integration to the greatest extent possible to determine if the moderating effects of semantic integration were still evident, it is especially telling that there was a signi ficant interaction between inhibition and the retrieval practice pattern for participants whose composite integration scores were

PAGE 85

75 extremely low, based on the strictest measur e of episodic integration possible. The lowest integrators in the experiment exhibited the same interaction between inhibition and the retrieval practice pattern as high integrators, F (1, 16) = 9.62, Mse = .013, p < .01. As shown in Table 12, low composite integrators ( M = 1.68, SD = .43) exhibited a reliable RIF effect of -13% in the Intra-set (weak) condition, F (1, 8) = 12.57, Mse = .022, p < .01. Similar to high integrators, low composite integrators ( M = 1.58, SD = .31) actually exhibited a facilitati on of Rpitems (+4%) in the Inter-set (strong) condition, however this facilitation was not reliable, F (1, 8) = 1.13, Mse = .031, p = .32. Despite the extremely low levels of integration, Rpitems were still protected from forgetting when semantic integration was strong. The evidence presented here clearl y demonstrates that the moderating effects of semantic integrati on are not dependent on the use of explicit, integration strategies.

PAGE 86

76 Discussion The purpose of Experiment 3 was to provide additional support for the assumption that the moderating effects of sema ntic integration observed in Experiments 1 and 2 were not dependent on e xplicit integration strategies The idea was to eliminate episodic integration strategies to the greatest extent possi ble in order to see if the moderating effects of semantic integration we re still observable. In order to reduce episodic integration strategies during the study phase, Incidental Learning instructions were implemented. Although participants in the Incidental Learning condition reported very little episodic integrati on, the same pattern of results was found in the Incidental learning condition as in the Intentional Learni ng condition. That is, even when there was very little episodic integrati on, the Rpitems were still protected from forgetting when they were strongly related to a practiced item (i.e. Inter-set (strong) condition). From these results, it appears that the moderating e ffects of pre-existing associations are most likely due, in part, to semantic integrati on or the implicit activation of a priori associations. Furthermore, the protective e ffects of pre-existing associations are not mediated by explicit awareness of those associatio ns on the part of the pa rticipant. If the moderating effects of pre-existing associations were dependent on the explicit integration, then items would have been saved from forgetting only when participants explicitly integrated the item s during study. However, because low integrators were also saved from forgetting in the condition where th ere were strong associations between the practiced and non-practiced sets (i.e. Inter-set (strong) ), it appears that the associations do not have to be explicitly noticed to observe the semantic integration benefits based in a priori associations. The current findings also replicate the re sults of Experiment 1 on the moderating effects of semantic integration on output in ference within the baseline. Although the interaction between semantic integration and the output interference effect was not significant, the pattern of results was consiste nt with the idea that semantic integration moderates the effects of output interference just as it moderates the e ffects of RIF. When semantic integration was high (i.e. strong associ ations between items cued in the first half and second half of the baseline), the output interference effect was reduced, just as the RIF effect. The findings also support the sugge stion that the associations have to be somewhat strong in order for output interference effects to be reduced. That is, when the associations were stronger as in the curre nt experiment and Experiment 1, the output interference effect was reduced when semantic integration was high. However, when the associations were weaker as in Experime nt 2, the output interference effect was not reduced. These results suppor t the assumption that seman tic integration benefits on output interference may depend on adequate priming of the items cued in the second half of the baseline. In conclusion, the current findings replicate the results of Experiments 1 and 2 and strongly support the assump tion that associations betw een practiced and unpracticed competitors moderate the RIF effect. More importantly, this experiment clarifies and confirms that associations that are purely semantic in nature can have this moderating effect, even in apparent absence of episodi c integration. Once agai n the results support

PAGE 87

77 the assumption that semantic integration is an important boundary condition on RIF and that integration need not occur explicitl y, but may occur implicitly, based on preexperimental associations.

PAGE 88

Experiment 4: Investigating whether the moderating effects of pre-existing associations can also be due to associations formed through explicit integration strategies likely to be employed when study time is increased. Thus far, the results of the current experiments suggest that semantic integration is a very clear factor in moderating RIF. The evidence for semantic integration as an important boundary condition on RIF comes from the following three findings. First, low episodic integrators exhibited similar semantic integration benefits as high integrators in the Intentional Learning conditions of Experiments 1-3. Second, when explicit integration was greatly reduced through Incidental Learning instructions, the benefits of semantic integration remained. Finally, even when the Incidental Learning group was restricted to only those participants who claimed to have never integrated either intentionally or incidentally, the benefits of semantic integration remained. By contrast the results thus far only provide modest evidence that explicit integration is an important moderator of the RIF effect at least when there are many associations among the list items. For example, there was only a very small additional 5% benefit of episodic integration for high integrators in Experiments 1 and 2, and no additional benefit of episodic integration in Experiment 3. These results are inconsistent with the results of previous studies that have consistently shown that episodic integration is an important moderator of the RIF effect, with less forgetting associated with greater self-reported intentional integration on the part of the participant (Anderson et al., 1999; Anderson et al., 2000). However, as mentioned previously, one of the main differences between past research on episodic integration and the current set of studies was that the pre-existing associations in our studies may have biased participants to episodically link the items on the pre-existing, associative links. Because of this Spontaneous Integration Bias to episodically link items that were already benefiting from semantic integration, no episodic integration effects were likely to emerge in the present studies. However, it could be argued that the benefits of episodic integration may emerge if participants are given the opportunity to go beyond the semantic links in order to link unrelated Rp+ and Rpitems together. For example, Anderson and McCulloch (1999) found that increasing the study time from five to 10 seconds over two repetitions of the material, increased episodic integration scores and decreased the RIF effect. Although their materials were unrelated and associations among the list items were minimized, perhaps the increased study time allowed more opportunities to link additional items within the category. If additional study time has this effect, then increasing the study time on the current set of materials may allow participants more time to go beyond the semantic links that are present and create the additionally, critical links between other, 78

PAGE 89

79 unrelated Rp+ and Rpitems. In the weak condition, this would translate into greater opportunities to episodically link the Rp+ items (e.g. HOR SE) with their unrelated competitors (Rpitems, e.g. DUCK, GOOSE WALRUS, SEAL). The same episodic linking is predicted to occur in the strong condition as a result of the pre-existing associations between Rp+ and Rpitems, but instead of just episodically linking the Rp+ items (e.g. HORSE) with their related Rpite ms (e.g. PONY), the extra study time should also allow for greater opport unities to episodically link the Rp+ items (e.g. HORSE) with the other unrelated competitors (Rpitems, e.g. TIGER, WALRUS, GOOSE). In addition, we have already shown in E xperiments 1 and 2 that although episodic integration did not produce a benefit when sema ntic integration was high, it did produce a 5% benefit when semantic integration was low. These results, however weak, suggest that either semantic or episodic integra tion can moderate the RIF effect, and that increasing either type of integration should decrease the amount of forgetting. Perhaps the inconsistencies between past research and the results of the current studies, regarding the role of episodic integration can theref ore be reconciled by an experimental manipulation designed to increase the use of spontaneous integration strategies to determine if independent benefits of epis odic integration can emerge when semantic integration is low (i.e., the Intra-set (weak) condition). Another important point is that the pr evious experiments presented in this manuscript failed to examine an alternative to semantic integration as an explanation as to why Butler and colleagues failed to find RIF with item specific cues. For example, not only did Butlers experimental design increase semantic integration by increasing preexisting associations between the practiced a nd non-practiced sets, their design may have also increased episodic inte gration. Their experimental design may have introduced greater levels of episodic integration because they used an eight seconds study exposure, instead of the standard four or five sec ond study exposure. As mentioned previously it has been found that increasing the study time from five to 10 seconds over two repetitions of the material increases episodic integration decreases RIF (Anderson & McCulloch, 1999; Anderson, 2003). Perhaps th e lengthy study time that Butler provided allowed participants more time to go beyond the semantic links and engage in additional, useful integration between the Rp+ and Rpitems. The purpose of the final experiment was to provide more precise insight as to why Butler and colleagues failed to replicate a nd to more precisely examine the specific influence of semantic and episodic integration on RIF. This was accomplished by crossing retrieval practice pattern ( Inter-set (strong), Intra-set (weak) ), as a manipulation of semantic integration, with study time (f ive, 10 s), as a manipulation of episodic integration. As in the previous experime nts, it was hypothesized that the RIF effect would be reduced in the Inter-set (strong) condition compared to the Intra-set (weak) condition, as a result of semantic integrat ion. However, because more study time should allow for more useful episodic integrat ion beyond the links provided by semantic associations, it was hypothesized that there w ould be a significant three-way interaction between inhibition, retrieval practice pattern, and length of study time. Specifically, in the five seconds study condition, we expected to replicate the pattern observed in Experiments 1, 2 and 3. However, when particip ants have twice as long to study as in the

PAGE 90

80 10 seconds study condition, it was predicted that there would be reduced RIF, regardless of the retrieval practice patte rn manipulation. This is because a longer study time should allow participants more opportunities to epis odically integrate be yond the links provided by pre-existing associations, insulating unpractic ed items from forgetting in either the Intra-set (weak) or the Inter-set (strong) condition.

PAGE 91

81 Methods Design and Participants The experimental design formed a 2 X 8 X 2 X 4 mixed-subjects factorial with study time (5s, 10s) and counterbalanci ng (eight counterba lancing conditions) manipulated between subjects, a nd retrieval practice pattern ( Inter-set (strong) Intra-set (weak) ), and retrieval practice status (Rp+ Rp-, Nrp1, Nrp2) manipulated within subjects. Sixty-four undergraduate psychology students from the University of South Florida served as participants in the study in exchange for extra credit. Participants ranged in age from 18 to 52 years, with a median age of 21. The participant pool was 83% female and 17% male, and the ethnic make-up of the participant pool was 61% Caucasian, 17% African American, 11% Hisp anic, 1% Asian, and 10% other. The average Shipley score was 29.08 ( SD = 3.71). The average MEQ score was 43.66 ( SD = 8.42), indicating that on av erage the circadian rhyt hm of the participan t pool fell into the neutral range between moderately morning a nd moderately evening. Approximately 89% of participants were tested during a time a day that was consis tent with their optimal time of day. The average Cognitive Failures score was 41.97 ( SD = 11.83) and the average reasoning score was 79% ( SD = 16) out of 100. Materials The materials for this experiment were th e same as those used in Experiment 1. Procedure Upon arrival, participants were randomly assigned to either the 5s or the 10s study condition. The procedures were identical to those used in Experiment 1, except that retrieval practice pattern ( Inter-set (strong) Intra-set (weak)) was manipulated within subjects, and there was an added betweensubjects study time manipulation. Intentional learning instructions were used during the st udy phase and half of the participants had five seconds to study each category exemplar pair, and the other half had 10 seconds. All other procedures for the study, re trieval practice, re tention interval, a nd test phases were identical to those used in Experiment 1.

PAGE 92

82 Results A series of mixed-subjects ANOVAs were conducted with study time (5s, 10s) and counterbalancing (eight counterbalanci ng conditions) as between-subjects factors, and retrieval practice pattern ( Inter-set Intra-set ) and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) and as within-subject factors. Retrieval-induced Forgetting The overall inhibition (Baseline vs. Rp -) collapsed across, retrieval practice pattern, and length of stu dy approached significance, F (1, 32) = 3.25, Mse = .04, p = .08. Although it only approached sign ificance, the % RIF effect obtained was similar the significant RIF effects found in Experime nts 1, 2, and 3 (-7%, -7%, and -5%, respectively). Thus, the result s of the current study replicat e previous research with the retrieval practice paradigm s howing that the probability of recalling Rpitems is lower than the probability of recalling baselin e items, even with item specific cues. Overall Retrieval-Induced Forgetting as a Function of Retrieval Practice Pattern. More importantly, as in the previous experime nts, the interaction between the inhibition effect and the retrieval practice pattern was significant, F (1, 32) = 5.14, Mse = .025, p = .03, indicating that there was a differe nce in the amount of RIF between the Intra-set (weak) and Inter-set (strong) conditions. As shown in the top panel of Figure 16, the results supported the hypothesis; there was significant RIF in the Intra-set (weak) condition, F (1, 32) = 7.67, Mse = .034, p = .01, but no RIF in the Inter-set (strong) condition, F < 1. In the Intra-set (weak) condition, where associations between the practiced (Rp+) and unpracticed sets (Rp-) were minimized, there was a -9% decrement in recall between Rpitems and baseline items in the final test phase. However in the Inter-set (strong) condition, where there were strong a ssociations between the practiced and unpracticed sets, there was no reliable differe nce in recall of Rpitems compared to baseline items (0%).

PAGE 93

.75.67.67.66.40.50.60.70.80.90Intra-set (Weak)Inter-set (Strong)Percentage Items Recalled Baseline (Nrp) Rp83 .65.65 Baseline (Nrp) 1.00 .88 .80.40.50.60.70.80Intra-set (Weak)Inter-set (Strong)Retrieval Practice PatternPercentage Items Recalle .90 d Rp+ Figure 16. Results of semantic integration in Experiment 4: Significant RIF only in the Intra-set (weak) condition (top panel), but retrieval-practice benefits in both the Intra-set (weak) and the Inter-set (strong) conditions (bottom panel).

PAGE 94

84 Baseline Tests Unlike the previous experime nts, there was a significant difference between the baseline in the weak condition (.75) and the baseline in the strong condition (.67), F (1, 48) = 5.44, Mse = .032, p <.05. Unlike the previous experiments, there was no significant difference be tween recall for Rpitems in the weak condition (.66) compared to the strong condition (.67), F < 1. These results suggest that the baselines were not statistically equivalent, and that recall fo r Rpitems was not reliably lower in the weak condition. Unlike the previous expe riments, the baseline in the weak condition appears to be inflated co mpared to the baseline in the strong condition. Some could argue that elevated baseli nes may have produced significan t, yet artificial levels of RIF in the weak condition. However, it is important to note that there are differences in the retrieval and associative dynamics between the different baselines of the retrieval practice pattern conditions, a nd so the baselines used for comparisons matched the practiced categories with respect to thes e different dynamics. Therefore the same mechanisms that inflated recall in the baseline conditions of the weak condition should have also inflated recall in the practiced categories in that condition, and the same mechanisms that deflated the baseline in the strong condition should have also deflated recall in the practiced category of that condition. Despite the fact that the both the weak baseline and practiced categories should have received similar boosts in recall, we still observed significant difference between the two, indicating that there was significant forgetting of Rpitems in the weak condition. Likewise, despite the fact that both the strong baseline and practiced cat egories should have experi enced similar declines in recall, we still observed no significant difference between th e two, indicating that there was no reliable forgetting of Rpitems in the strong condition. We will return to this point and provide a more detailed description of the applicable mechanisms in the MetaAnalysis of all four experiments.

PAGE 95

Overall Retrieval-Induced Forgetting as a Function of Length of Study. As hypothesized, the two-way interaction between inhibition and the length of study was significant, F(1, 32) = 5.16, Mse = .04, p = .03. As shown in the top panel of Figure 17, participants who only had five seconds to study exhibited a significant -10% decrement in recall of Rpitems compared to baseline items, F(1, 16) = 8.67, Mse = .021, p = .01. However, participants who had 10 seconds to study, did not exhibit a reliable decrement in recall of Rpitems (-1%), F < 1. .70.72.62.71.40.50.60.70.80.905 seconds10 secondsPercentage Items Recalled Baseline (Nrp) Rp85 .67.64.40.50.60.70.80Percentage Items Recalled .83.85.90 Baseline (Nrp) 1.00 Rp+ 5 seconds10 secondsLength of Study Time Figure 17. Results of episodic integration in Experiment 4: Significant RIF only in the 5s second study condition (top panel), but retrieval-practice benefits in both the 5s and 10s study conditions (bottom panel).

PAGE 96

86 els of self-r eported episodic integration, F (1, 32) = 4.21, Ms = .45, p = .048. As hypothesized, participants who had 10 seconds to study o call of Rpitems was reliably lower in the 5s S t y condition). As shown in Table 13, when participants only had five s econds to study, they exhibited a significant 14% RIF effect in the Intra-set (weak) condition, F (1, 16) = 10.13, Mse = .031, p < .01, but exhibited a non-significan = 2 gration moderated the RIF effect. For exam pa tudodic ratiore reliably y did ble RIF effects in the Intt (we tion (-4%), F < 1. Not only did they fail to show forgetting in the Inter (stron on, they actually exhibited filitation of Rpitems in the Iet conion (+7%), how this facil not = 1.78, Mse = .035, p = .20. The present results suggest that both seman tegration ca rate the RIF effect because forgetting was found only in the condition wher e semantic integration was low (i.e. the Intra-s An examination of self-reported episodic integration re vealed that length of study manipulation produced significantly different lev e reported significantly higher levels of episodic integration ( M = 3.75, SD = 1.01) compared to those who had 5 seconds to study ( M = 3.41, SD = 1.08). These elevated levels of episodic integration in the 10s st udy condition coincided with reduced levels of forgetting in the 10s study condition, and are consistent with previous studies showing that increased study time leads to increased le vels of episodic integration that leads t reduced levels of forgetting. These results ar e consistent with the notion that additional study time allows for more useful integrati on between the Rp+ and Rpitems. Baseline Tests There was no significant difference between the baseline in the 5s Study condition (.72) and the baseline in the 10s Study condition (.70), F < 1. The difference in recall for Rpitems in th e 5s Study condition (.62) and the 10s Study condition (.71) approached significance, F (1, 48) = 3.01, Mse = .078, p = .09. Because the baselines were statistically equivalent, a nd re tudy condition, the reliable RIF effect found in the 5s Study condition was unlikely due to an inflated baseline, and th e lack of reliable RIF found in the 10s Study condition was unlikely to a deflated baseline. Semantic Integration Effects on RIF as a Function of Length of Study. The threeway interaction between the inhibition effect, retrieval practice pattern, and the length of study was not significant, F < 1. Although the three-way inte raction was not significant, the pattern of results was consistent with th e prediction that there would be independent effects of semantic and episodic integration. For example, the results revealed tha forgetting was reduced when either semantic integration was high (i.e. Inter-set (strong) condition) or when episodic integration was high (i.e. 10s stud t -6% RIF effect in the Inter-set (strong) condition, F (1, 16) g effects of semantic .29, Mse = .027, p = .15. These results replicate th e moderatin ous experiments. integration found in the previ The results also revealed that episodic in te ple, when partici nts had 10 seconds to s y (i.e. epis in teg n score we greater) the not exhibit relia ra-se ak) condi -set g) conditi ac nter-s dit ever itation was reliable, F (1,16) tic and episodic in n mode et ) and episodic integration was low (i.e. 5s study). When either semantic integration was high (i.e. Inter-set ) or episodic integration was high (i.e. 10s study), the unpracticed items (i.e. Rp-) were protected from forgetting.

PAGE 97

87 RpRIF Effect Table 13. RIF effect as a function of length of st udy and the strength of associations between the Rp+ and Rpitems. Note ** indicates significant forgetting at p < .05. Strength of Association Length of Study between Rp+ and RpBaseline Intra-set (weak) .77 .63 -.14** 5 seconds Inter-set (strong) .68 .62 -.06 Intra-set (weak) .73 .69 -.04 10 seconds Inter-set (strong) .66 .73 +.07 Baseline Tests. In the 5s Study condition, the di fference between the baseline in the Intra-set (weak) condition (.77) and the baseline in the Inter-set (strong) condition (.68), only approached significance, F (1, 31) = 3.76, Mse = .031, p = .06. In the 10s Study condition, there was also no significant difference between the baseline in the Intra-set (weak) condition (.73) and the baseline in the Inter-set (strong) condition (.66), F (1, 31) = 1.29, Mse = .048, p = .26. These results indicate that the reliable RIF found in the Intra-set (weak) condition of the 5s Study condi tion was unlikely due inflated baseline, and the lack of RIF effects found in the Inter-set ( strong) conditions of both Study conditions was unlikely due to deflated baselines. Separate Effects of Semantic and Episodic Integration. In order to ascertain the benefits of semantic integration above the benefits afforded by episodic integration, effect to an RIF effects were compared at both retrieval-practice pattern as n was %, f udy lengths within each semantic integration condition. When semantic integration was low (i.e. Intra-set ), efit of 10%. For example, when conditions within each study length condition. When episodic integration w low (i.e. 5s study), it appears that semantic in tegration produced a benefit of 8%. That is, when semantic integration was high (i.e. Inter-set ), there was a non-sign ificant RIF effect of only -6%, compared to a significant RIF eff ect of -14% when semantic integratio low (i.e. Intra-set ). Therefore, semantic integrati on reduced the RIF effect by 8% when episodic integration was lower (i.e. 5s study). When episodic integration was higher (i.e 10s study), it appears that semantic integrati on produced a benefit of 11%. That is, when semantic integration was high (i.e. Inter-set), there was facilitation of the Rpitems (7 however this facilatory effect was not significant). Compare this +7% to a nonsignificant RIF effect of -4% when semantic integration was low (i.e. Intra-set ). Therefore, semantic integration produced an 11% benefit when epis odic integration was higher. When the benefits of semantic integration were compared across the levels o high and low episodic integrati on, it appears that episodic in tegration had a beneficial effect of about 3% above the benefit afforded by semantic integration (11%, the semantic benefit when episodic integration was hi gh, minus 8%, the semantic benefit when episodic integration was low). In order to ascertain the benefits of episodic integration above the benefits afforded by semantic integration, RIF effects were compared at both st it appears that episodic inte gration produced a ben

PAGE 98

88 episodi isodic ant RIF cross f the practiced items in the fina l t of een tion f Rp+ items in the Inter-set (strong) condition where forgetting was ab gth of rences (1,16) = 28.93, Mse = .05, p < .0001, and in the 10s study condition where forgetting was absent, F (1,16) = 56.23, Mse = .055, p < .0001. In other words, although c integration was high (i.e.10s study), there was a non-significa nt RIF effect of only -4%, compared to a signi ficant RIF effect of -14% when episodic integration was low (i.e. 5s study). Therefore, episodic in tegration reduced the RIF effect by 10% when semantic integration was low. When semantic integration was high (i.e. Inter-set ), it appears that episodic integrati on produced a benefit of 13%. For example, when ep integration was higher (i.e.10s study), ther e was facilitation of the Rpitems (7%, however this facilitation was not significant) Compare this +7% to a non-signific effect of -6% when episodic integration was lower (i.e. 5s study). The result is that episodic integration produced a 13% benefit in recall above the benefit already afforded by semantic integration when it was high. When the benefits of episodic integration were compared across the levels of semantic integr ation, it appears that semantic integration had a beneficial effect of about 3% above the benefit afforded by episodic integration (13%, the episodic integrati on benefit when semantic in tegration was high, minus 10%, the episodic integration benefit when semantic integration was low). Benefits of Retrieval Practice on Practiced Items The overall benefit (Baseline vs. Rp+) in recall for practiced items (Rp+) a the retrieval practi ce pattern and length of study conditions was significant, F (1, 32) = 75.58, Mse = .052, p < .0001. Participants recalled 15% more o l test phase ( M = .84, SD = .21) than baseline items (M = .65, SD = .29). As in al of the previous experiments, the amount of RIF was independent of the amount of strengthening of the practiced items, ther eby replicating the strength independent property of RIF. In contra st to blocking accounts that predict that the amount of strengthening of practiced items should be negatively correlated with the amoun forgetting of unpracticed items, there was a non-significant, positive correlation betw the facilitation of practiced items a nd the forgetting of unpracticed items, r = .12, p = .16. As shown in the bottom panel of Figure 16, the interaction between the facilitation of Rp+ items and retrieval practi ce pattern was not significant, F (1, 32) = 1.79, Mse = .052, p = .19. There was significant f acilitation of Rp+ items in the Intra-set (weak) condi where forgetting was present, F (1,32) = 49.47, Mse = .033, p < .0001. There was also significant facilitation o sent, F (1,32) = 14.61, Mse = .049, p < .0001. In other words, although forgetting was only apparent in the Intra-set (weak) condition, retrieval practice produced a significant benefit for practiced items relative to baseline in both the Intra-set (weak) and Inter-set (strong) conditions (+23% and +15%, respectively). This strength independent property of RIF was also ev ident across the len study conditions that produced different amounts of episodic integration and diffe in the amount of RIF found. As shown in the bottom panel of Figur e 17, the interaction between the facilitation of Rp+ items and th e length of study time was not significant, revealing that the amount of RIF (top panel) was independent of the amount of strengthening of the practi ced items (bottom panel), F < 1. There was a significant retrieval practice benefit for Rp+ items in 5s study condition where forgetting was present, F

PAGE 99

89 forge a h the 5s and 10s study conditions (+20% and 17%, respectively). e t retrieval practice benefits regardless of the re a dy, ee-wntera between retice practice patt and lenof st not significant, F (1, 32) = 1.58, ethere was a significant retr nefit of +27% in the Intt (we condition, 0.42 .0001. Theas alsoargin significant retrieval practice benefit of +12% in the Inter-set (strong) condition, (1,16) = 4.09, M al ieval practice benefit effect as a function of length of study time and strength een the Rp+ and Rpitems. No te ** indicates significance at p < .05. tting was only apparent in the 5s study condition, retrieval practice produced significant benefit for practiced items in bot + As shown in Tabl trieval practice p 1 n 4, there were significa h of stu tte rn or the lengt as thr the ay i ction s trieval prac benefit, retrieval Ms = .052, p = .22. When participants only had 5s to study, er n, gth udy wa ieval practice be ra-se ak) F (1, 16) = 3 Mse = .037, p < re w a m ally F se = .054, p = .06. The results were similar when participants had 10 seconds to study. There was a signifi cant retrieval practice be nefit of +18% in the Intra-set (weak) condition, F (1, 16) = 19.20, Mse = .029, p = .000. There was also a significant retriev practice benefit of +16% in the Inter-set (strong) condition, F (1, 16) = 11.91, Mse = .045, p < .01. Table 14. Retr of association betw indicates signific ance at p < .10. Length of Study Strength of Association between Rp+ and RpBaseline Rp+ Benefit Effect Intra-set (weak) .61 .88 +.27** 5 seconds Inter-set (strong) .66 .78 +.12* Intra-set (weak) .70 .88 +.18** 10 seconds Inter-set (strong) .66 .82 +.16** The current findings on benefits of retrie val practice, replicate previous work and Anderson and colleagues (1994; 1999) a nd Experiments 1 3, showing that RIF effects are not dependent on the amount of strengthening of practiced items as interference accounts would predict (e.g. McGeoch, 1942; Melton & Irwin, 1940). Th strength independent nature of the forgetting e ffect is more consistent with an inhibitory account of RIF. Output Interference Results revealed that the overall output in terference effect (N rp1 baseline vs Nrp2 baseline) within the baselin e condition was significant, F (1, 32) = 5.10, Mse = .052, p = .03. Overall, baseline items cued in the last half of the category block (i.e. Nrp2) were recalled at a lower rate ( M = .67, SD = .29) than baseline items cued in the first half of the category block (i.e. Nrp1) ( M = .73, SD = .23). As in the previous experiments, it was pr edicted that there would be sign output interference in the Intra-set (weak) condition where there were weak-to-no associations between by is ificant the firs t half of the testing sequen ce (e.g. HORSE) and second half of the testing sequence (e.g. LION), but th at there would be no significant output

PAGE 100

90 interference in the Inter-set (strong) condition where there were strong associations between baseline items in the first half (e.g. HORSE) and baseline items in the second half (e.g. PONY). Here again, although the interaction between output interference and retrieval practice pattern was not significant, F(1,32) = 1.70, Mse = .052, p = .20, the pattern of output interference was consistent with the idea that retrieving an exemplar in the first half of the testing sequence (e.g. HORSE) should insulate related items (e.g. PONY) in the second half from forgetting. As shown in Figure 18, in the Intra-set (weak) condition, where there were weak associations between the first half and second half of the category block, the output interference effect was significant at -10%, F(1, 32) = 5.43, Mse = .052, p = .03. However in Inter-set (strong) condition, where there were strong associations between the first and the second half, the output interference effect was not significant (only a -2% output interference effect), F < 1. These results suggest that semantic integration can also moderate the output interference effect. When semantic integration was low (i.e. weak), there was significant levels of output interference, but when semantic integration was high (i.e. strong) there was no output interference effect. .69.67.67.40.50.60.70.80Intra-set (Weak)Inter-set (Strong)Retrieval Practice PatternPercentage Items Recalled .77 .90 Baseline 1st Baseline 2nderence in the Intra-set (weak) condition, but not in the Inter-set (strong) condition. The reshe output interferce efn output interfer s Figure 18. Output Interference Effect as a result of semantic integration in Experiment 4: Significant output iterf n ults also suggest that just as semantic integration can moderat e tfect so can episodic integration. Although the interaction betwee en ence and length of study time was not significant, F(1,32) = 1.70, Mse = .052, p = .20, the pattern of output interference was consistent with the idea that retrieving anexemplar in the first half of the testing sequence should insulate episodically linked itemsin the second half from forgetting, but no protection should be given to items that were not episodically linked. As shown in Figure 19, separate analyses revealed that in the 10

PAGE 101

91 ect of -8% pproached significance, F(1, 16) = 4.10, Mse = .038, p = .06. These results suggest that episodic integration can also moderate the output interference effect. When episodic integration was low (i.e. Intra-set), output interference effects approached significance, but when episodic integration was high (i.e. Inter-set) the output interference effect was not significant, nor did it approach significance. study condition, where episodic integration was higher, the output interference eff-3% was not significant, F(1, 16) = 1.04, Mse = .07, p = .32. However in the 5s study condition, where episodic integration was lower, the output interference effect of a .72.70.67.64.40.50.60.70.80.905 seconds10 secondsLength of StudyPercentage Items Recalled Baseline 1st Baseline 2nd Figure 19. Output Interference Effect as a result of episodic integration in Experiment 4: Marginally significant output interference in the 5s study condition, but no significant output interference in the 10s study condition. As shown in Table 15, the three-way interaction between output interference, retrieval practice pattern, and length of study was not significant, F(1, 32) = 1.36, Mse = .052, p = .25. Separate analyses revealed that when participants had five seconds to study, they exhibited significant output interference effect of -16% in the Intra-set (weak) condition, F(1, 16) = 6.25, Mse = .063, p = .02. However, there was no reliable output interference effect in the Inter-set (strong) condition (-2%), F <1. When participants had 10 seconds to study, they did not exhibit significant output interference, in either the Intra-set or Inter-set condition, F <1 (-3% and -2%, respectively).

PAGE 102

92 Table 15. Output interference effect as a function of length of study and strength of association between the Rp+ and Rpitems. Note indicates significance at p < .05. Length of Study Strength of Association between Rp+ and RpBaseline 1st half Baseline 2nd half Output Interference Effect Intra-set (weak) .77 .61 -.16* 5 seconds Inter-set (strong) .68 .66 -.02 Intra-set (weak) .73 .70 -.03 10 seconds Inter-set (strong) .66 .64 -.02 Episodic Integration As in the previous experiments, th e amount of episodic integration that participants engaged in du ring the study phase was meas ured by a self-report postexperimental integration questionnaire. For the 5s study condition, the average integration score in the Inter-set (strong) condition ( M = 3.36, SD = 1.16) was similar to average integration score in the Intra-set (weak) condition ( M = 3.35, SD = 1.25), F < 1. For the 10s study condition, the aver age integration score in the Inter-set (strong) condition ( M = 3.75, SD = 1.13) was similar to the aver age integration score in the Intraset (weak) condition ( M = 3.67, SD = 1.05), F < 1. As mentioned previously, th e many pre-existing associati ons in our materials may have created a S pontaneous Integration Bias that masked the be nefits of episodic integration in our current experiments. It was hypothesized that th e increased study time would allow participan ts more opportunities to go beyond the pre-existing links in order to link Rp+ items with unrelated Rpitems. This assumption was supported by the fact that those in the 10s Study condition reported significantl y greater levels of episodic integration ( M = 3.75, SD = 1.01), than those who had 5s to study ( M = 3.41, SD = 1.08), F (1, 32) = 4.21, Mse = .45, p = .048. Furthermore, the increased levels of episodic integration in the 10s study condition coincided with much reduced levels of RIF. These results are consistent with previous studies showing that increased study time leads to increased levels of episodic integration that l eads to reduced levels of forgetting. The key assumption was that participants overcame the Spontaneous Integration Bias because the extra study time increased their episodic integration beyond the semantic integration that already occurred. Ho wever, one potential fl aw of this argument was that the median split analysis of the episodic integration scores did not predict forgetting in the 5s Study condition ( F < 1), nor did it predict fo rgetting in the 10s Study condition, where integrat ion scores increased, F < 1. As shown in Table 16, regardless of the length of study, there was no interaction between inhibition and the amount of integration in the weak or the strong conditions, indicating no a dded benefit of episodic integration, F < 1 for both 2-way interactions.

PAGE 103

93 Table 16. RIF effect as a function of length of study, strength of association between the Rp+ and Rpitems, and level of episodic integration. Note ** indicates significance at p < .05 *indicates significance at p < .10. Length of Study Strength of association between Rp+ and RpRIF effect for High Integrators RIF effect for Low Integrators Episodic Integration Benefit Intra-set (weak) -.14** -.14** .00 5 seconds Inter-set (strong) -.06 -.05 -.01 Intra-set (weak) -.04 -.03 -.01 10 seconds Inter-set (strong) +.04 +.08 -.04 If participants had enough study time to do useful integration, th en the integration split should have revealed differences in forgetting in the 10s Study condition between high and low integrators. Specifically, high integrators in this condition should have benefited more from episodic integration, pr esumably because high integrators in this condition were more likely to engage in usef ul, episodic integration beyond the semantic links. However, one reason that the median split analysis did not predict performance was that it may not have adequately distinguis hed high integrators from low integrators. Therefore, in order to create a more dis tinctive difference between the high and low integration groups, the data was divided into quartiles by integrat ion score and the low integrator group was operationally defined as the bottom 25% and the high integrator group was operationally defi ned as the top 25%. Table 17. RIF as a function of study time, retrieval practice pattern and episodic integration in the top quartile (high integrators) and bottom quartile (low integrators). Length of Study Strength of Association between Rp+ and RpRIF effect for High Integrators Top 1/4 RIF effect for Low Integrators Bottom 1/4 Episodic Integration Benefit Intra-set (weak) -.22 -.22 .00 5 seconds Inter-set (strong) -.10 -.09 -.01 Intra-set (weak) +.03 -.09 +.12 10 seconds Inter-set (strong) +.16 +.07 +.09 As shown in the right-hand column of the Table 17, when there is a greater difference in integration scores between the high and low integrators, it appears that there were episodic integration benefits for thos e who had 10s to study, but not for those who had only 5s, although the interaction between RIF, study time, retrieval practice pattern, and episodic integration group was not significant, F < 1. As shown the bottom panel of Figure 18, for those who had more time to st udy and therefore more time to do useful integration, it appears that there were episodic integration be nefits in both the Intra-set (weak) condition (+12% benefit) and in the Inter-set (strong) condition (+9% benefit). Subsequent analyses of the 10s study condi tion, revealed that the interaction between

PAGE 104

94 inhibition and integration group was not relia ble in either retrieval practice pattern condition. Although the interactio ns in the quartile split analysis were not significant, it is important to note that thes e analyses were based on only eight subjects per cell instead of the standard 16 per cell in the median split analysis, thus re ndering it without enough power to detect significant effects. Furt hermore, the pattern of data supports the assumption that more study time allowed for mo re useful, episodic integration beyond the integration provided by semantic associations.

PAGE 105

95 Discussion The purpose of Experiment 4, was to provi de more precise insight as to why Butler and colleagues failed to replicate. Their experimental design may have increased semantic integration, episodic integration, or both types of inte grations so it was important to show the moderating effects of both types of integration. Another purpose of Experiment 4 was to more precisely ex amine the specific influence of episodic integration on RIF because of the less persuasive evidence in favor of episodic integration as a moderator of the RIF effect in Experiments 1-3. This was accomplished by crossing a semantic integration ( Inter-set (strong), Intra-set (weak) ), with episodic integration (5s study time, 10s study time). Th e idea was that giving participants more time to study should increase their use of spontaneous episodic integration strategies. As mentioned previously increasing the study time from five to 10 seconds over two repetitions of the material was previous ly shown to increase episodic integration scores and decrease the amount of RIF (Anderson & McCulloch, 1999; Anderson, 2003). The study manipulation had the same effect on integration scores as it did in previous studies. In fact, when participants had 10s to study, they reported significantly greater levels of episodic integration than particip ants who had only 5 seconds. Furthermore, forgetting was reduced when episodic integr ation was high as a re sult of increasing the study time from 5 to 10 seconds. Although the me dian and quartile sp lit analyses of the post-experimental integration scores did not reveal a reliable benefit of episodic integration between high and low episodic inte grators in the 10s c ondition, the pattern of data supports our assumption that extra study ti me allows more opportunity to inter-relate useful, episodic connections beyond those alr eady provided by pre-existing links. For example, in the quartile split analysis, high integrators exhibited benefits of episodic integration in the range of 9 to 12%, wher eas low integrators exhibit no benefits. Furthermore, even though the ra te of explicit integration clea rly went up in the 10s study condition, and even though the rate of usef ul integration MUST have gone up, given a reduced RIF effect, the episodic integrati on measure may not have been a good measure of useful integration, because it blended Intra-set and Inter-set rates of episodic integration. That is, even in the 10s c ondition, a person with a very high rate of integration, who fell into th e high episodic integration gr oup, may have actually had a lower amount of useful integr ation than a person with a lowe r rate of integration, who fell into the low episodic integration group. Because of this noise, the effects of spontaneous integration may not have emerged even in the 10s Study condition. The results of the current experiment repl icate and confirm the results of previous studies showing that that epis odic integration is also an im portant factor in moderating RIF. The current results extend previous research by showing that the study time does not have to be broken up over two repetitions Simply increasing the time given during a single study episode, increases ep isodic integration scores an d decreases the size of the retrieval-induced forgetting effect. These findings are especially important because Butler and colleagues failure to replicate may have been partially caused by the fact they used quite a lengthy study time which may have given participants more time to spontaneously integrate the items during study.

PAGE 106

96 Meta-analysis In order to more completely explore the effects of semantic integration and to evaluate the regularity of various effects, th e data from all of the experiments presented in this study were combined and a meta-ana lysis was performed on th at combined set.. Because the experiments in this study em ployed similar designs it was possible to combine the data in this way. Another pur pose of performing a meta-analysis was that some weaker effects found in the individua l experiments, may become stronger when more enough power is added. Therefore a series of mixed subjects ANOVAs were conducted on the combined database with retrieval practice pattern (Intra-set ( weak), Inter-set (strong) ), counterbalancing (eight counterba lancing conditions), and experiment (Experiments 1-4) as between subjects factor s, and retrieval practice status (Rp+, Rp-, Nrp1, Nrp2) as a within subjects factor. Retrieval-induced Forgetting The results of the meta-analysis revealed the overall RIF effect collapsed across all four experiments and the retrieval practice pa ttern manipulations, was consistent and reliable, F (1, 320) = 22.78, Mse = .025, p < .0001. The probability of recalling Rpitems was lower than the probability of recalling baseline items (-6% RIF effect). Results support the assumption that seman tic integration moderates the RIF effect for the overall interaction between RIF and th e retrieval practice pattern was significant, F (1, 320) = 24.97, Mse = 025, p < .0001. As shown in the top panel of Figure 20, there was a significant RIF effect in the Intra-set (weak ) condition, where semantic integration was low (i.e., no associations betwee n practiced and non-practiced sets), F (1, 192) = 37.86, Mse = .026, p < .0001. However, in the Inter-set (strong) condition, where semantic integration was high (i.e., associa tions between practiced and non-practiced sets), no reliable RIF was observed, F < 1. As shown in Figure 20, in the Intra-set (weak) condition, where associations between the practiced (Rp+) a nd unpracticed sets (Rp-) were minimized, there was a 12% decrement in recall between Rpitems and baseline items in the final test phase. However, in the Inter-set (strong) condition, where associations between the practiced and unpracticed sets were pres ent, there was no difference in recall performance between Rpitems and baseline items in the final test phase (0%). The results of this experiment strongly support the assumpti on that associations between practiced items and their unpracticed competitors moderate the inhi bitory effect, because when semantic integration was low the typical RIF effect was obtained, but when semantic integration was high, there was no RIF effect.

PAGE 107

.71.67.67.59.40.50.60.70.80.90Intra-set (weak)Inter-set (strong)Percentage Items Recalled Baseline (Nrp) Rp97 .64.63.40.50.60.70Intra-set (weak)Inter-set (strong)Retrieval Practice PatternPercentage Items Re .84.84.90d Baseline (Nrp) 1.00 Rp+ .80cal le Figure 20. Results of Meta-analyses: Significant RIF effect only in the Intra-set (weak) condition (top panel), but significant retrieval practice benefit in both the Intra-set (weak) and Inter-set (strong) conditions (bottom panel). Baseline Tests. Unlike the individual Experiments 1, 2 and 3, there was a ignificant difference between the ba selines in the Meta-Analysis, F(1, 369) = 5.97, t that the baselines were not s s Mse = .037, p <.05. Although these results suggesstatistic ally equivalent, there was only a 4% difference between the baselines (.71 vs. .67). Not only were the baseline differences quite small, when Rprecall was compared between the retrieval practice pattern conditions, results revealed that recall of Rpitemwas significantly lower in weak condition (.59) compared to the strong condition (.67),

PAGE 108

98 ble ss items s ignificant, F (1, 320) = 286.01, Mse = .025, p < .0001. Overall, 21% more practiced n the final test phase ( M = .84, SD = .17) compared to baseline items ( M = .6 e he results of the meta-analyses reveal that significant strengt hening occurred in both retrie val-practice pattern conditions despite the fact that forgetting was only apparent in the Intra-set (weak) condition. Retrieval practice produced a significant bene fit of 21% for practiced items relative to baseline in the Intra-set (weak) condition where forgetting was present, F (1, 192) = 174.99, Mse = .024, p < .0001. As inhibitory accounts of RIF would predict, retrieval practice also produced benefit of 20% for practiced items in the Inter-set (strong) condition, where forgetting was absent, F (1, 192) = 132.58, Mse = .027, p < .0001. O utput Interference The results of the meta-analysis revealed that the overall output interference effect within the baseline conditions collapsed across all four e xperiments and the retrieval practice pattern manipulations was significant, F (1, 320) = 19.21, Mse = .028, p < .0001. As shown in Figure 21, baseline items cued in the last half of th e category block were recalled at a -6% lower rate ( M = .63, SD = .24) than baseline items that were cued in the first half of the category block ( M = .69, SD = .20). F (1, 48) = 11.17, Mse = .043, p <.01. As shown in the top panel of Figure 20, recall for Rpitems in the weak was actually 8% lower than Rprecall in the strong condition Therefore, even if the weak baseline was more in line with the strong baseline, relia levels of RIF would have still emerged. Because the baseline differences were quite small and the recall for Rpitems was reliably lower in the weak condition compared to recall for Rpitems in the strong condition, the reliable RIF effect in the weak condition was more likely to due forgetting of Rpitems instead of artificial levels of RIF as a result of an inflated baseline. In additi on, the lack of a reliable RIF effect in the strong condition was more likely do to the moderating effects of semantic integration and le likely to a deflated baseline. Benefits of Retrieval Practice on Practiced Items The results of the meta-analysis of the overall benefit in recall of practiced collapsed across all four experiments and th e retrieval practice pattern conditions, wa s items were recalled i 3, SD = .22). As predicted by the inhibi tory account of RIF, the amount of retrieval-induced forgetting was independent of the amount of strengthening of the practice. There was a non-significant, positive correlation between the facilitation of practiced items and the forg etting of unpracticed items, r = .09, p = .09. As shown in th bottom panel of Figure 20, the overall intera ction between facilita tion of Rp+ items and retrieval practice pattern was not significant, F < 1. T

PAGE 109

.71.67.64.63.60.70.80.90centage Items Recalled Baseline 1st Baseline 2nd .50er P .40Intra-set (weak)Inter-set (strong)Retrieval Practice Pattern Figure 21. Meta-analysis of Output Interference Effect: Significant output interference in the Intra-set (weak), but not in the Inter-set (strong) condition. Although the interaction between output interference and retrieval practice patterwas not significant in any of the individual experiments, F < 1, the results of the meta-analysis revealed that when the data was collapsed across the experiments, there was a significant interaction between output interference and the retrieval practice pattern manipulation, F(1, 320) = 5.29, Mse = .028, p = .022. The results support the notionsemantic integration also moderated the output interference effect just like it moderated the RIF effect. As shown in Figure 21, there was significant a output interference effect in the weak condition, where there were weak associations between the first and last half of the category block, F(1, 192) = 19.38, Mse = .03, p < .0001. However in the strong n that ondition, where there were strong associations between the first and last half of the category F(1, 192)nce ffect in ttput interference effect in the strong condition. Episodic IRe-analysis revat a median split by the post-experimental integratioacross all foueriments and the retrieval practice pattern anipund g c block, the output interference effect was only marginally significant, = 3.99, Mse = .025, p = .055. There was a reliable -8% output interferehe weak condition, but only a non-reliable -3% ou e ntegration sults of the meta ealed th n score collapsed r exp m lations resulted in no reliable interaction between overall inhibition and the amount of integration, F < 1. Separate analyses revealed that when semantic integration was high (i.e., Inter-set (strong)), there was no reliable interaction between inhibition athe amount of integration indicating that there was no added benefit of explicit, episodic integration above the benefit of semantic integration based on pre-existing associations that were implicitly activated during the study phase, F < 1. As shown in the bottom portion of Table 17, low episodic integrators (M = 2.63, SD = .95) benefited from stron 99

PAGE 110

100 .97) also benefited from semantic integration for they showed no forgetting of Rpitems (0% RIF effect), F < 1. These results again support the conclusion that semantic integration is sufficient to reduce the forgetting effect, and episodic integration may not provide an additional benefit. Table 18. Results of Meta-analysis: RIF eff ect as a function strength of association between the Rp+ and Rpitems and level of episodic integration. Note ** indicates significance at p < .05, indicates significance at p < .10. Strength of association between Rp+ and RpRIFeffect for High Integrators RIFeffect for Low Integrators Episodic Integration Benefit associations between the Rp+ a nd Rpitems, as can be seen from the fact that they did not exhibit a significant RIF effect, (0% RIF effect) F < 1. As shown in the bottom portion of Table 18, high episodic integrators ( M = 3.86, SD = Intra-set (weak) -.11** -.13** .02 Inter-set (strong) .00 .00 .00 Separate analyses revealed that when semantic integration was low (i.e., Intra-set (weak) ), there was no reliable interaction between inhibition and the amount of integration indicating that there was no added benefit of explicit, episodic integration even when semantic integration was low, F < 1. As shown in the top portion of Table 17, high integrators (M = 3.90 SD = .89) exhibited reliable levels of RIF (-11%), F (1, 64) = 19.56, Mse = .053, p < .0001. Low integrators ( M = 2.71, SD = .95) also exhibited reliable levels of RIF (-13%), F (1, 64) = 24.60, Mse = .028, p < .0001 Although high integrators did not exhibit as much forgetti ng as low integrators, subsequent analyses revealed that the 2% benefit of episodic in tegration was not sufficient to produce reliable differences in RIF between high and low integrators, F < 1.

PAGE 111

101 Discussion The results of the meta-analysis confirm th at semantic integration is an important moderator of the retrieval-i nduced forgetting effect. Desp ite all of the experimental manipulations designed to moderate the RIF effect (i.e. semantic and episodic integration), overall there was a reliable RIF e ffect of -6%. Because the RIF effect was consistently obtained with items specific cu es, this is strong evidence that Butlers conclusion that the RIF effect may not be relia ble with item specific cues, is unlikely to be correct. Furthermore, because the RIF effect was consistently found with item specific cues that controlled for output inte rference, it can be concluded that the RIF effect was not simply due to interference. More importantly, because the main purpose of the present study was to show that semantic integration in the form of pre-ex isting associations betw een the practiced and non-practiced sets moderates forgetting, it wa s important that a reliable interaction between inhibition and retrieval practice pa ttern was found in each of the individual experiments and in the meta-analysis. Acro ss all of the experiments, when semantic integration was low there was a reliable RIF effect of -12%, however when semantic integration was high, no reliable RIF effect was observed (0%). The results of the experiments presented in this study support the assumption that associations between practiced items and their unpracticed competitors moderate the inhibitory effect and that this moderation occurs when associations are implicitly activated. When pre-existing associations are minimized between the sets, the typical inhibitory effect is observed. The results of this study provide evidence that Butler et als (2001) failure to replicate was due, in part, to the presence of pre-exis ting associations between the practiced items and their non-practiced competitors. More im portantly, the current findings highlight the importance of minimizing pre-existing a ssociations in studies of RIF. One potential problem for interpretations of the present findings, is that the baseline in the Inter set (strong) condition appears to be lower than the baseline in the Intra-set (weak) condition. Because RIF is determined by comparing recall of Rpitems to recall of baseline items, some may argue that an inflated baseline in one condition could potentially produce reliable, yet artificial levels of RI F. For example, it could be argued that if the strong baseline was more in line with the weak baseline, significant levels of forgetting may have emerged in the strong condition as well as the weak condition. Therefore, a deflated baseline ma y have artificially produced a lack of RIF in the strong condition, and an inflated baseline may have artificially produced RIF in the weak condition. It is important to note that there may ha ve been factors that di d in fact contribute to a slightly inflated baseline in the weak condition and a slightly deflated baseline in the strong condition. By the construction of the strong and weak retrieval practice conditions, there were clear differences in th e retrieval and associ ative dynamics between the conditions that may have increased or decreased recall performance. For example, in the strong baseline, items in the first half were re lated to items in the second half, but all the items in the 1st half were unrelated to each other, and all items in the 2nd half were unrelated to each other. These particular retrieval and associative dynamics in the strong

PAGE 112

condition may have caused a deflation of its baseline. Specifically, items in the strong baseline may have been at higher levels of activation (i.e. similar to the weak baseline, e.g. .71) than what was found, but some items were blocked by activation of highly semantically similar items in the 1st half of the testing sequence, thus reducing baseline (e.g. .67). For example, as shown in Figure 22, assume that HORSE is the first item that is tested in the 1st half of the testing sequence. If HORSE is successfully retrieved then this may increase the implicit activation or prime its related associate, PONY. Now because HORSE and PONY are at a heightened level of activation they then block access to other Animals that are subsequently tested. By the time the 3rd exemplar in the testing sequence is cued for (e.g. WALRUS), two explicitly activated exemplars (e.g. HORSE and LION) and two implicitly activated exemplars (e.g. PONY and TIGER) may be blocking access to WALRUS in the testing sequence. As a result of all of this interference, the recall of subsequent baseline items (e.g. WALRUS, GOOSE) may be reduced. Thus, the baseline in the strong condition may be deflated because the associative structure of its testing sequence created high degrees of interference compared to the weak condition. Figure 22. Retrieval dynamics in the baseline of the Inter-set (strong) condition: Explicit and implicit activation, and interference on subsequent items in first half of testing sequence. Retrieval Dynamics in the Inter-set (strong) Condition Baseline (Nrp1): Cue(Response) Explicit & Implicit Activation & Interference on Subsequent items 1. Animal H____ (HORSE) if recall of HORSE is successful PONY is primed 2. Animal L ____(LION) Recall of LION is blocked by competition from two representations (HORSE & PONY). However, if recall of LION is successful primes TIGER 3. Animal W____(WALRUS) Recall of WALRUS is blocked by competition from four representations (HORSE, PONY, LION & TIGER. However, if recall of WALRUS is successful primes SEAL 4. Animal G____(GOOSE) Recall of GOOSE is blocked by competition from six representations (HORSE, PONY, LION, TIGER, WALRUS, & SEAL). However, if recall of GOOSE is successful primes D UC K Because the level of RIF is determined by comparing recall of Rpitems with baseline items, then it would appear that there is no forgetting of items in the strong condition (.67 .67 = .00). However it is important to note, that even one assumes that the baseline was deflated, using an increased baseline (such as the weak baseline), still results in very little forgetting. In fact, it only results in 4% forgetting (e.g. .71 .67 = .04). Contrast the retrieval dynamics in the strong condition with those in the weak condition. In the weak baseline, the 1st half and the 2nd half were unrelated, but there were relationships among items in the 1st half and among items in the 2nd half. Specifically, there were two related pairs tested in the 1st half and two related pairs tested in the 2nd half. The related pairs were therefore separated by only one tested item in the 102

PAGE 113

testing sequence. Unlike the strong baseline, the particular retrieval and associative dynamics in the weak condition may have inflated its baseline. Specifically, items in the weak baseline may have been at lower levels of activation (i.e. similar to the strong baseline, e.g. .67) but the activation of some items was increased because they were primed by a related item that was recently recalled. Retrieval Dynamics in the Intra-set (weak) Condition Baseline (Nrp1) Cue (Response) Explicit & Implicit Activation & Interference on Subsequent items 1. Weapon A____(ARROW) if recall of ARROW is successful primes BOW 2. Animal R ____(RIFLE) RIFLE is blocked by competition from two representations (ARROW & BOW). However, if recall of RIFLE is successful primes GUN 3. Animal B____(BOW) BOW is blocked by competition from two representations (RIFLE & GUN), however BOW is already at a heightened level of activation because it was recently implicitly activated (primed) by ARROW, so it does not suffer as much from the interference from RIFLE & GUN. If recall of BOW successful re-activates ARROW 4. Animal G____(GUN) GUN is blocked by competition from two representations (BOW & ARROW), however GUN is already at a heightened level of activation because it was recently primed by RIFLE, so it does not suffer as much from the interference. Figure 23. Retrieval dynamics in the baseline of the Intra-set (weak) condition: Explicit and implicit activation, and interference on subsequent items in first half of testing sequence. For example, as shown in Figure 23, assume that ARROW is the first item that is tested in the 1st half of the testing sequence. If ARROW is successfully retrieved then this may increase the implicit activation or prime its related associate, BOW. Now because ARROW and BOW are at a heightened level of activation they then block access to other Weapons that are subsequently tested, such as RIFLE. However if recall of RIFLE is nonetheless successful, its related associate GUN will be primed. Now RIFLE and GUN will interfere with subsequent recall of the third Weapon in the testing sequence (i.e. BOW). However, BOW does not suffer from as much interference because it was already primed by its recently tested associate ARROW. Unlike the 3rd item in the strong condition which potentially suffers from interference from up to four representations, the 3rd item in the weak condition may only suffer from interference from up to two representations (i.e. explicitly activated RIFLE and implicitly activated GUN). Thus, in the weak condition, the retrieval and associative dynamics create a situation where subsequent items are highly likely to be recalled because they experience an increase in activation as a result of being primed by a related associate. Because the amount of RIF is determined by comparing recall of Rpitems with baseline items then it would appear that there is significant forgetting of items in the weak condition (.71 .59 = .12). However it is important to note that, even if one assumes that the weak baseline was 103

PAGE 114

104 slightly elevated, using a smaller baseline (such as the strong baseline) would still result in more forgetting than in the strong condition (i.e. 0%). In fact it would result in an RIF effect of 9% (e.g. .67 .59 = .09), a level that is comparable to reliable levels of RIF found in past research (Anderson, 2003). Furt hermore, even under the most stringent conditions one would still obser ved more forgetting in the weak condition. For example as mentioned previously, if the higher baseline of the weak condition was used to determine levels of RIF in the strong condition, there would only be a 4% RIF effect. This RIF effect is still smaller compared to when the lower baseline of the strong condition is used to determ ine levels of RIF in the weak condition (i.e. 9%). Because there may have been a slightly inflated baseline in the weak condition and a slightly deflated baseline in the strong condition, perhaps a more appropriate baseline for both conditions would be so mewhere in between the two baselines. However, even if one averages the two base lines (.69), and then uses this value as a baseline for both conditions, th e results would be the same as when the appropriate baselines were utilized. Specifically, there would be an interaction between inhibition and the retrieval practice pattern. In other words, the RIF effect in the weak condition would be much larger at 10% (.69 59 = .10), than the RIF effect in the strong condition (69 .67 = .02). Although averaging the baselines seems like an appropriate way to handle the problems associated with inflated and deflated baselines, it is not appropriate to average the two baselines and use this average to determine the am ount of RIF in both retrieval practice pattern conditions. As mentioned previously, the baselines are matched with their practiced categories on re trieval dynamics and associative relatedness. Therefore the same retrieval dynamics that differen tially affect the am ount of priming and interference between the base lines, should also differen tially affect the practiced categories. In other words that same rules of priming and interfer ence should also apply to the practiced categories that are equa ted with the baselines on these presumed dynamics. This would translate into less in terference as a result of higher levels of priming in the practiced categories of the weak condition, and more interference in the practiced categories of the strong condition. Thus, both the baseline and the practiced category in the weak condition should benefit from similar rates of inflation, and both the baseline and practi ced category in the strong condition should suffer from similar rates of deflation. Therefore, it is not appropriate to use the baseline from the other condition or an average of the baselines because the presumed retrieval dynamics are not equated. Furthermore, because the practiced cate gories receive additional activation as a result of retrieval practice, the affects of associative relatedness should be even more exaggerated on the practiced cate gories. In other words, the weak practiced category should benefit more from priming because its items are also primed during retrieval practice, resulting in greater in flation of Rp items compared to baseline. This suggests that even if the baseline and practiced categ ory were both inflated, the practiced category would be inflated even more, resulting in smaller recall differences between the two (i.e. smaller levels of RIF). Despite the retrie val dynamics that may have reduced the amount of RIF, we still observed reli able levels of RIF in the weak condition.

PAGE 115

105 The affects of associative relatedness should be exaggerated on the practiced categories in the strong condition as well. That is, the strong practiced category should suffer from more interference be cause the interfering informa tion is also activated during retrieval practice, resulting in greater impairment of Rp items compared to baseline. This suggests that even if the baseline a nd practiced category were both deflated, the practiced category would be de flated even more, resulting in larger recall differences between the two (i.e. larger le vels of RIF). Despite the retrieval dynamics that may have increased the amount of RIF, we still failed to observe reliable levels of RIF in this strong condition. In conclusion, because of the differences between the strong and weak conditions, the most appropriate strategy in determining the level of RIF is to use the baseline that matches the practiced conditi on in associative relatedness. Of substantial importance for inhibitory accounts of RIF wa s that the strength independent property of RIF also emerged in the meta-a nalysis. According to interference accounts of RIF, the amount of forgetting of unpracticed items should be dependent on the amount of strengthening of practiced items (McGeoch, 1942; Melton & Irwin, 1940). That is, anytime there is significa nt strengthening of Rp+ items, this should block access to Rpitems causing them to be recalled at a lowe r rate than baseline items. Therefore because significant strengthening oc curred in both retrieval practice pattern conditions, blocking accounts w ould have predicted signif icant forgetting in both retrieval practice pattern conditi ons. However, this was consis tently and reliably not the case across all four experiments in the Inter-set (strong) condition, where there was significant strengthening of Rp+ items but a total absence of forg etting of Rpitems. Furthermore, there was no correlation between the amount of strengthening of practiced items and the amount of forgetting of unpract iced items from practiced categories. Blocking explanations therefore have a difficult time reconciling th e strength independent property of RIF that was consis tently found a repeatedly repli cated in the current set of experiments. As mentioned in the previous experime nts, Anderson and colleagues consider output interference to be a form of RIF whereby practicing retrieving items in the beginning of the testing sequence causes retrie val induced forgetting items of items that are tested later. It was assumed that if output interference is a form of RIF then it should be moderated by the same mechanisms that mode rate RIF. It was therefore predicted that semantic integration would reduce the output in terference effect, just as it reduced the RIF effect. Although the interaction between output interference and retrieval practice pattern was not significant in any of the indi vidual experiments, the results of the metaanalysis revealed that when we had more power to detect significant differences, a significant interaction between output interf erence and the retrieval practice pattern manipulation emerged. The pattern of the interaction between output interference and the retrieval practice pattern manipul ation was highly similar to the pattern of the interaction found between the RIF effect and the retrieval practice pattern manipulation. When semantic integration was low, the typical output interference and RIF effect was observed. When semantic integration was hi gh, both the output interference effect and the RIF effect was reduced. These results str ongly suggest that semantic integration as a result of pre-existing associations can modera te output interference effects just as they

PAGE 116

106 moderate RIF effects. The results also suppor t the notion that output interference may be a form of RIF, because a factor that in fluenced RIF was shown to have a similar influence on output interference. The results of the meta-analysis regard ing the moderating effects of episodic integration were less promising. Although semantic integration had clear and consistent benefits, episodic integration did not pr oduce any added benefit when semantic integration was high, nor did it produce a benefit when semantic integration was low. As mentioned previously the lack of episodic integration benefits is inconsistent with past research showing consistent and reliable benefits of episodic integr ation (Anderson et al., 1994; Anderson and McCulloch, 1999). However, a major difference between past research and the current set of studies was that our studies maximized the pre-existing associations within the lists. Our failur e to find episodic integration benefits was expected given the assumption that th e pre-existing associations incited a Spontaneous Integration Bias to episodically link already sema ntically related items. Because semantically related items were already benefiting from the protective effects of semantic integration, the episodic integration benefits were unlikely to emerge. It was therefore assumed that increasing the study time in Experiment 4, should allow participants more opportunities to link additional items beyond the semantically links that were already present. The result of Experiment 4, confirmed this assumption. When participants were given more time to study as they were in Butler et al.s (2001) study, their episodic integration scores increase d and forgetting decreased. Furthermore, a quartile split by episodic integration scores re vealed that high episodic integrations who were given more time to study exhibited le ss forgetting than low episodic integrators (however a word of caution is warranted when interpreting these results because we did not have enough power to detect significant e ffects in the quartile sp lit analysis). Because all but one of the experiments included in th e meta-analysis used a short study time, the Spontaneous Integration Bias was likely to prevent episodic integration benefits overall. To summarize, the current set of experi ments establishes that integration, both semantic and episodic is as an important boundary condition on th e retrieval-induced forgetting effect, and that they benefits of semantic integration are not dependent on the benefits of episodic integration. The results su ggest that Butler et al.s (2001) failure to replicate the item specific cueing effect was due to an experimental design (i.e. strong associations and increased study time) that in creased the occurrence of either semantic integration, episodic integration, or both type s of integration. Ther efore, experimental designs that increase the occurrence of either type of integration will result in reduced levels of forgetting.

PAGE 117

107 General Discussion Support for the Semantic Generalization Hypothesis The main purpose of the present set of experiments was to examine whether the Semantic Generalization Hypothesis of the Distributed Re presentation Approach also applies to semantic integration. This model was introduced by Anderson Green & Mcculloch (2000) to explain how episodic integration between the practiced and nonpracticed sets protects memories from inhi bitory processes. A nderson et al.(2000) found that when episodic associations were form ed between the practiced and non-practiced sets, no RIF effect was observed, but when ep isodic associations were formed between the non-practiced items, reliable levels of RIF were found. Although their study did not actually vary the semantic similarity between the sets, the model makes similar predictions regarding the role of semantic inte gration in moderating inhibitory processes. As mentioned in the introduc tion, the model makes no distin ction between episodically driven integration and integration driven by th e presence of pre-existing knowledge. Just as with episodic integrati on, the model makes certain predictions depending on the specific pattern of associations that exist. According to the Semantic Generalization Hypothesis, in order for integration to mani fest it beneficial e ffects the pre-existing associations have to be between the pr acticed and non-practiced sets (i.e. the Inter-set (strong) ). When the pre-existing associations are among the Rp+ items or among the Rpitems (i.e. Intra-set (weak)), the model predicts no such pr otective effects. In fact, it predicts that when there are Intra-set associations, the effects of RIF should be exaggerated. The results of the present experiments support the predictions of the Semantic Generalization Hypothesis and confirm that semantic integration also serves as a moderator of the RIF effect just as does episodic integration. When there were inter-set associations (i.e. strong a ssociations between the Rp+ and Rpitems, but weak associations among the Rpitems and among th e Rp+ items) the to be inhibited items were saved from forgetting. According to th e model, retrieval practice strengthened the features of the practiced items and this strengt hening generalized to th e features that they shared with semantically related competitor s, thereby strengthening the competitors in addition to the retrieval practiced targets. The Inter-set associations simply increased the implicit activation of the non-practiced items thereby savi ng them from inhibitory processes. The results were very different when there were intra-set associations (i.e. weak association between Rp+ and Rpitems but strong associations among Rpitems and among Rp+ items). Here the to be inhibited items were not saved from forgetting. Instead as the model predicted, RIF effects were slightly magnified. Typically, RIF

PAGE 118

108 are observed in the range of -8 to -10%, but in the intra-set conditions of the present experiments, forgetting fell in the range of -12 to -14%. According to the model, retrieval practice strengthened th e features of the practiced it ems and caused inhibition of features that were unique to the competitor s. When the competitors were strongly associated, the inhibitory effects on one repres entation generalized to the features that it shared with other semantically related comp etitors. Therefore when competitors were semantically associated, the effects of inhibition occurred simultaneously on more than one representation, thereby ma gnifying the RIF effect. The Importance of the Establishing Boundary Conditions The studies presented here are important for a number of reasons. First, they clearly establish an additi onal boundary condition of the RI F effect, namely semantic integration. The moderating effects of seman tic integration were replicated in five conditions across four experiments, with two different sets of materials, and under two types of encoding instructions. The benefits of semantic in tegration have already been established in other domains of forgetting such as directed forgetting, but it was unclear if the benefits would emerge in an unintentional forgetting paradigm. It was important to establish the benefits of semantic integrat ion within the retrieval practice paradigm because it may explain why Butler et al. (2001) failed to replicate the RIF effect with item specific cues. This study was actually motivated by an examination of their materials that revealed many pre-existing asso ciations between their Rp+ and Rpitems. We concluded that semantic in tegration may have contributed to their failure to replicate because when we simply rearranged their lis ts to reduce these associations and thus reduce the level of semantic integration, we found reliab le levels of RIF. Second, the present findings replicate a nd confirm the importance of episodic integration as an important moderator of forgetting. We replicated the effects of study time on RIF, confirming the prediction that explicit integration also influences the magnitude of the RIF effect. This interpretati on is bolstered by the significant increase in integration reports with increasing study time. When study time was increased from five to 10 seconds, episodic inte gration increased, and not onl y was RIF reduced, it was completely eliminated. In previous st udies, episodic integration was enhanced by increasing the study time over two repetitions of the materials. Current results suggest that simply increasing the study time in one repetition is enough to increase episodic integration strategies and thus reduce the RIF effect. This was important to show because Butler et al. (2001) used a lengthy study time of 8 seconds in one repetition of the material instead of the standard four to fi ve seconds. The use of such a lengthy study time may have allowed participants greater oppor tunities to spontane ously interrelate the items together beyond the semantic links. We have no idea if th e study time factor actually increased the us e of spontaneous integration stra tegies in their experiment, but current findings on the benefits of episodic in tegration as a result of increased study time provide evidence that episodic integration could have also contributed to their failure to replicate.

PAGE 119

109 The evidence favoring explicit integration as a moderating factor is complicated however by the failure to find consistent evidence for the beneficial effects of spontaneous explicit integration, as measured by the post-experimental questionnaires. Although the episodic in tegration benefits were not replicated in the median split analyses of the spontaneous integration ra tings, there may be three important reasons why. First, the post-experimental questionnair e was not designed to measure the rate of episodic integration between specific items a nd so there was no way to discern if the critical episodic links between the Rp+ and Rpitems were formed. In other words, even if a participant scored high on the post-experiment integration measure, there was no way to tell if the integration was useful (i.e. li nks between Rp+ and Rpitems) or whether the integration non-useful (i.e. links among the Rp+ or among the Rpitems). Second, even if participants formed the critical Rp+ Rp associations and so scored high on the integration measure, we cannot ascertain what proportion of that spontaneous integration was devoted towards useful integration and what proportion was devoted towards useless integration. This ambiguity makes our measur e of integration very noisy when it comes to rank ordering participants in terms of usef ul integration. For in stance, a participant with an integration ra ting of may have primarily integrated within-set (i.e. useless integration), but another partic ipant with a rating of ma y have integrated entirely between set (i.e. useful integration). In this case, our median split would have put these two participants in the wrong integration ca tegory. Because his integration score was higher, the useless integrator would have been labeled as a high integrator, whereas the useful integrator would have been labele d as a low integrator. However, based on the more useful pattern of integration he employed, the low integrator should have been labeled as a high integrator. The re sults of this study and the previous study by Anderson, Green, and McCulloch (2000) clearly support the idea that the pattern of episodic and semantic associations matters a great deal. Because there is no way to discern the pattern of integration that was employed or what proportion was devoted towards useful integration, our median split may have failed to appropriately categorize high and low integrators. The final and the most likely reason th at we did not replicate the episodic integration benefits in the median split analys es, was that we engineered the experiment in a way that may have biased participants to focus their spontaneous integration efforts along the strong pre-existing links that were present. As me ntioned previously, one of main differences between the current experiments and previous studies (Anderson et al., 1994) was that Anderson made great efforts to minimize the pre-existing associations between exemplars, whereas we maximized the associations. A nderson and colleagues (1994) found reliable and consis tent benefits of episodic integration when pre-existing associations were minimized because without the bias, participants were just as likely to form the critical episodic connections betw een the Rp+ and Rpitems as they were to form non-useful episodic connections between Rp+ items or between Rpitems. Here the retrieval pathways between related exem plars were so strong there may have been little opportunity to retrieve and rehearse additional, unrelated exemplars together (beyond the ones that would be integrated based on our established associations) in order to improve memory. According to the Spontaneous Integration Bias Hypothesis in the

PAGE 120

110 Intra-set (weak) condition, there was a bias to episodically integrate the related Rp+ items together and/or the related Rpitems together, thereby directing all of the integration energy towards useless integra tion among Rp+ items or among Rpitems, and away from the critical integration between the Rp+ and Rpitems, resulting in little benefit. In the Inter-set (strong) condition, there was no a dded benefit of episodic integration above the benefit of semantic integration (i.e. Inter-set (strong )) because there was a bias to episodically integrate the related Rp+ and Rpitems. The result was no appreciable added benefit of episodic integration because the Rpitems were already implicitly strengthened through semantic integrati on. If this bias was in fact present in our experiments, then our failure to find epis odic integration effects in the median split analyses is not inconsistent with the claim that episodic in tegration reduces inhibition. The only potential problem for the argument that a spontaneous integration bias masked the benefits of episodi c integration, was that the me dian split analysis of the integration scores did not predict forget ting even in the 10s study condition, where integration scores increased. It was assume d that in the 10s study condition, the benefits of episodic integration emerged because the increased study time allowed participants more opportunities to create more useful epis odic links in addition to the semantic links that were already present. If participants had enough study time to do useful integration, then the median split of integration ratings s hould have revealed differences in forgetting between high and low integrators. However, integration ratings failed to predict recall performance. A quartile split analysis was also conducted in order to create a more distinctive difference between high and low integr ators. The pattern of data that emerged from this analysis clearly supports the assu mption that more study time allows for more useful episodic integration so that the benefits of episodic integration can emerge. Those who had less study time exhibited no episodi c integration benefits, but those who had more time to study exhibited benefits in the area of 9 to 12% across the retrieval practice pattern conditions, however the interaction between inhibition and episodic integration was not reliable. It is impor tant to note that although th e interaction was not reliable there was not enough statistical power to detect reliable eff ects. Given more power, the statistical analysis would have most likely confirmed reliable diffe rences between high and low integrators in the 10s study condi tion as the means already indicated. Furthermore, even though the ra te of explicit integration incr eased, and the rate of useful integration must have also increased, given a reduced RIF effect, the integration measure did not provide an adequate measure of useful integration because it blended Intra-set and Inter-set ratings of integration. Because usef ul and useless integrators may have been mislabeled as high or low integrators based on their rate of total integration, the benefits of spontaneous integr ation may not have emerged. Therefore, our failure to find reliable ep isodic integration bene fits in the median and quartile split analyses, was not detrimenta l given the problems associated with the Spontaneous Integration Bias and the inability to disti nguish useful from useless integrators. Because of these problems, the results may not be inconsistent with episodic integration as a moderator of RIF. Future studies will eliminate the problem associated with the spontaneous integration bias, by elim inating all inter-item associations. If the benefits of episodic integration emerge when the bias is eliminated, then this would

PAGE 121

111 provide evidence that inter-item associations simply masked the effects of episodic integration in the current set of studies. In conclusion, the episodic integration results are not necessarily inconsistent with past re search supporting episodic integration as a moderator of RIF. The moderating effects of episodic integration clearly emerged in the 10s study condition of Experiment 4. When participants had more time to study, their integration scores increased and the forgetti ng effect was completely eliminated. The results support the notion that the use of such a lengthy st udy time allowed participants greater opportunities to spontan eously interrelate the item s together during the study phase, thus saving them from inhibition. Gi ven the results on semantic and episodic integration, Butler et al.s (2001) failure to replicate the item speci fic cueing effect was most likely due to an experimental design (i .e. strong associations and increased study time) that increased the occurr ence of either semantic integr ation, episodic integration, or both types of integration. Therefore, experime ntal designs that increase the occurrence of either type of integration should result in reduced levels of retrieval-induced forgetting. A third important finding was that semantic integration benefits were independent of episodic integration strategies, clearly estab lishing that semantic integration is a unique moderator of RIF. Bauml and Hartinger (2002) showed that similarity among the practiced and non-practiced sets reduces the forgetting effect However, their study did not provide sufficient evidence that incident al similarity is sufficient to reduce RIF because they used subcategory cues that were explicitly presented. The presence of these subcategory cues may have incited participan ts to episodically integrate the items along those subcategory links. The purpose of this st udy was to eliminate all such explicit cues to organization to see if implicitly activate d associations are sufficient to influence the RIF effect. As shown in Experiment 3, when episodic integration scores were extremely low as a result of incidentally encoding th e items during study, there was no forgetting when semantic integration was high. The findings therefore extend the work by Bauml and Hartinger (2002) on the benefits of seman tic integration by showing that the benefits of semantic integration are independent of explicit integration strategies. The current set of experiments strongly establish that integration, both semantic and episodic, is as an im portant boundary condition on th e RIF effect, and that the benefits of semantic integration are not depe ndent on the benefits of episodic integration. The results strongly suggest that Butler and colleagues failure to replicate the item specific cueing effect was due to an experi mental design (i.e. st rong associations and increased study time) that increased the occurrence of either type of integration or both types. Therefore, experimental designs that increase the occurrenc e of either type of integration should result in re duced levels of forgetting. An Additional Factor that may have Contributed to the Failure to Replicate with Item Specific Cues Although this study established semantic in tegration as a moderator of RIF and reinforced the conclusion that episodic inte gration can reduce the RIF effect, it does not provide conclusive evidence that integration solely contributed to Butler and colleagues failure to replicate. In fact, there is anot her factor that may have contributed to their failure to replicate; th e use of category-plus-two -letter-stem cues (i.e. Bird SP____) in the

PAGE 122

112 final test phase. It is impor tant to use category-plus-stem-cu ed recall during the final test phase because this method controls for output interference by ensuring that weak items (i.e. Rpitems) are tested first. However, they used category-plus-stem-cued recall with two -letter stems that may have i nduced participants to use orthographic representations to complete the categoryplus letter stems, instead of the episodic or semantic representations that were inhibited during re trieval practice. Typi cal RIF studies use a one-letter stem (i.e. Bird S_____) to avoid this problem. For example, participants can produce a number of different responses to the cue Bird S______, such as SEAGULL, STORK, SWALLOW, or SPARROW However, cueing with Bird SP_____ limits the possible responses to a single response, SPARR OW. This single response can be easily guessed based on the strong orthographic cue provided by the first two letters of the target word. Another reason that the use of the category-plustwo -letter stems may have made it difficult to observe forgetting, is that strong cues can override any forgetting that may have occurred by reinstating the inhibite d items semantic network or contextual representation (Goodmon and Nelson, 2003; Sahakyan and Goodmon, in prep). An inhibitory mechanism could have been em ployed to overcome interfering information during the retrieval practice phase of their experiment, however the e ffects of inhibition may have been masked by the use of a cue that may have activated different features than the ones that were inhibited or that may have re-activated the semantic or contextual representation. Evidence Against Interference Accounts of RIF Proponents of interference accounts of RIF, view Butler et al.s (2001) finding of RIF with category cues, but failure to replicate with item specific cues as an indication that the RIF effect is simply due to output interference. Th ey believe that the forgetting observed on category cue tests is simply due to the blocking of weaker, non-practiced items by stronger, practiced items that are re called initially in the testing sequence. Therefore, when item specific cues are us ed that control for output interference no forgetting should be observed. However, the current findings clearl y establish a number of different mechanisms that may have modera ted the overall inhibitory effect or simply masked its effects. If these factors would have been controlled in their experiment, perhaps they would have replicated the RIF effect with items specific cues. As mentioned in introducti on, there are several reasons to abandon interference accounts of RIF in favor of an inhibitory expl anation. First, there are copious amounts of studies that have replicated the effect with items speci fic cues (Anderson et al., 1994; Anderson & McCulloch, 1999; Anderson, Bjor k, & Bjork, 2000; Anderson, Green, & McCulloch, 2000; Anderson & Bell, 2001; Ba uml, 2002; Bauml & Hartinger, 2002; Ciranni & Shimamura, 1999; Hicks & Starns, 2004; Ford, Keating, & Patel, 2004; Veling & van Knippenburg, 2004). These studies either controlled for or manipulated some of the moderating and masking factors of RIF, a nd consistently found reliable RIF effects. Perhaps the most damaging evidence agai nst interference acc ounts was that all four experiments presented here replicated the strength independent property of RIF. That is, the rate of forgetting of Rpitems was completely and consistently independent of the rate of strengthening of the Rp+ items According to inte rference accounts, the

PAGE 123

113 amount of forgetting of non-practiced items should always be dependent on the amount of strengthening of practiced items because st rengthening of Rp+ items blocks access to Rpitems causing them to be recalled at a lower rate than baseline (McGeoch, 1942; Melton & Irwin, 1940). However, in the Inter-set (strong) condition, there was significant strengthening of Rp+ items but a to tal lack of forgetting of Rpitems. The replication of the strength independent property is strong evidence that the forgetting observed here was not simply due to blocki ng, but was actually due to an inhibitory mechanism designed to overcome the damagi ng effects of interfering information. As mentioned in the previous experime nts, Anderson and colleagues consider output interference to be a form of RIF whereby practicing retrieving items in the beginning of the testing sequence causes retrie val induced forgetting items of items that are tested later. It was assumed that if output interference is a form of RIF then it should be moderated by the same mechanisms that moderate RIF. The results of the metaanalyses support this assumption. When sema ntic integration was low, the typical output interference effect was observed, but when semantic integration was high, the output interference effect was reduced. The results support the notion that output interference may be a form of RIF, because a factor that influenced RIF was shown to have a similar influence on output interference. Acco rding to interferen ce accounts of output interference, items in the initial testing sequence are strengthened in memory causing them to block access to other items that ar e not yet recalled. Interference accounts predict forgetting of later items even if they are related to initially recalled items. Therefore, interference account s of output interference have a difficult time explaining why latter items were not forgotten when relate d items were tested at the beginning of the testing sequence. Alternatives to the Inhibitory Account of RIF It is important to note that the results of the current set of studies do not completely rule out interference or blocking explanations of the data. For example, it could be argued that in the strong condition, there were two opposing forces, activation and blocking, that cancelled each other out, re sulting in what appears to be no forgetting. That is, in the strong condition, practicing retrieving Rp + items spreads activation to related Rpitems. This increases the chance th at they will be recalled in the tested phase, but it also increases the chance that because of their recall success they will then block access to other unrelated Rp+ items. In a ddition, because Rp+ items also received a boost in activation as a result of retrieval pr actice, they can also block access to other unpracticed Animals For example, retrieving PONY (Rp+) during the retrieval practice phase, primes HORSE (Rp-), so that during the 1st half of the final testing sequence it is highly likely to be retrieved to the cue Animal H_____. If HORSE is successfully retrieved, it re-primes PONY, so that both HOR SE and PONY then in terfere with recall of subsequently tested items (e.g. LION, WALRUS, GOOSE). In other words, both activation and blocking are operating in the strong condition. No forgetting is observed because Rpitems are somewhat saved from blocking because they are at a heightened level of activation as a result of priming.

PAGE 124

114 It could be argued that the sp ecific retrieval dynamics in the weak condition create a very different situation, one that allo ws a forgetting like effect to emerge. For example, it could be argues that onl y blocking forces are operating in the weak condition, and so forgetting is due to interference and inhibition. This is because in the weak condition, there are no associations between Rp+ and Rpitems, and so the Rpitems do not receive a boost in activation during the retr ieval practice phase. In other words they are less likely to be retrieved and then block acc ess to other Rpitems. However, in this situation because they did not receive a boost in activation, they are more likely to suffer from blocking from strengthened items (Rp+). However, there are several reasons to doubt the blocking account of the data. First, many assumptions have to be adde d to explain how faci litation and blocking interact to produce no difference in recall between the baseline and Rpitems of the strong condition. For example, one would have to assume that of the four associates that are primed during retrieval practice, only one pr imed item serves a facilitory role (i.e. for itself) while the other three serve a bloc king role. This seems unlikely given the heightened explicit and implicit activation of all of the associates as a result of retrieval practice. A second reason to doubt blocking explanations is that blocking accounts of RIF rest on the assumption that interference occu rs because the category cue activates the strengthened items, which in turn block access to the non-strengthened items. This explanation does not hold up when one uses a cue that circumvents activation of the strengthened item. As mentioned in the in troduction, the RIF effect is cue independent, meaning that any cue that normally produces the Rpitem is less likely to do so. Although the present research di d not use independent cues, pa st research has shown that when a cue is used that does not activated the strengthened item, forgetting is still evident (Anderson & Spellman, 1995). Blocking accounts cannot explain how strengthened items can block access to non-strengthened items when the cue does not activate the strengthened items. The third reason to doubt blocking accounts of the data is that we used category plus-first-letter-stem cues that were very item specific. Compound cues such as the ones we used increase the chance of retrieving th e correct answer by reducing the number of competitors for that cue. For example, th e category cue BIRD activates a large number of possible responses, but the compound category cue BIRD S____ limits the possible responses to just a few. Blocking explanati ons have a hard time explaining how an item that is cued with a compound cue that reduces the activation of interfering information nevertheless suffers from interference. Because of these reasons, an inhibitory account of the data is more likely. Although these studies establis h semantic integration as an important factor in studies of forgetting, independent of any expl icit cues to integrate, they also do not completely rule out alternative, non-inhibito ry accounts of the da ta such as the cue change theory, the contextual change theory, or the transfer appropr iate theory of RIF (Perfect, Stark, Tree, Moulin, Ahmed, & Hutte r, 2004; for review, see Anderson & Bjork, 1994). According to these alternative account s, the design of RIF studies biases the category cues to elicit certain meanings of th e cue that were practiced during the retrieval

PAGE 125

115 practice phase. Specifically, the retrieval practic e of certain sub-categories biases the test cue towards this sub-category meaning because it was recently experienced and strengthened. For example, retrieval pr acticing LION, TIGER, WALRUS, and SEAL during the retrieval practice phase may have strengthened the feline and sea mammal sub-category meanings of the Animal category, so during the test phase the Animal cue was biased to strongly elicit feline and sea mammal exemplars of Animal because these meanings were more recently activat ed and strengthened during the retrieval practice phase. These accounts assume that other types of Animals such as equines and birds, are not inhibited. Instead, Animal is no longer an effective cue for Animals other than felines or sea mammals. Another way of conceptualizing the alte rnative accounts is to consider the detrimental effects of contextual mismatch between study and test. It is assumed that certain meanings of Animal are activated during the retr ieval practice phase and this strengthens a particular animal context. At test, Animal responses that are more closely linked to the Animal context previously activated are mo re likely to be retrieved. In our example, felines and sea mammals would be more likely to be elicited as responses than equines or birds because the former are more related to the recent contextual meaning of Animal Again, it is not that eq uine or bird representations are inhibited in memory, instead Animal is no longer an effective cue for items that are have a substantial contextual mismatch with th e original category cue. Th ese alternative accounts would therefore predict that because unpracticed items are not i nhibited and the forgetting effect is simply due to cue bias, then no interference should be observed with an unbiased cue. This is in contra st to inhibitory accounts that assume that unpracticed items are truly inhibited so any cu e (biased or unbiased) that no rmally produces the inhibited item should be less likely to do so. Cue/contextual change accounts are similar to interference or blocking explanations that suggest ther e is no inhibition of the spec ific representation of HORSE in memory. According to interference theory, practicing retrieving LION to Animal LI_____ makes LION so highly accessible that it interferes with, or blocks access to HORSE and results in reduced recall of HORSE (McGeoch, 194l; Anderson, 1983; Raaijmakers & Shiffrin, 1981; Rundus, 1973; Roediger & Neely, 1982; for review of non-inhibitory sources of memory impairment, see Anderson & Bjork, 1994 and Anderson, 2003). Other interference theories suggest that the connection between Animal and LION is strengthened so that LION is a stronger competitor for Animal compared to HORSE. Like cue/contextual change accounts, interf erence explanations assume that forgetting-like effects are co ntingent on the use of the original category cue. They predict no interference in a s ituation where the category does not cue items that interfere with recall of the unpracticed item. One way to discriminate between inhib itory and non-inhibitory accounts is to employ the independent probe technique, deve loped by Anderson and Spellman (1995) to provide specific support for an active suppression process that operates on specific representations in memory. This method us es an unbiased or non-interfering cue to determine if forgetting effects are still obser ved when interference is eliminated. With this technique, items are cued during the te st phase with novel cues that are uniquely

PAGE 126

116 related to the to-be-retrieved item. For example, instead of cueing HORSE with the original category cue Animal that may be biased to elicit strong exemplars that were retrieval practiced, it would be cued with a novel test cue such as Race because it is uniquely related to HORSE and is not directly related to LION or Animal The logic behind this technique is that if HORSE is actually suppressed as a result of retrieval practice of LION to Animal LI_____, then any other cue th at normally produces HORSE should be less likely to do so. Because Animal and LION are not related to race and are therefore not activated by Race during retrieval, they should not interfere with the recall of HORSE. As a result, any forgetting of HORSE would be due to inhibition of HORSEs specific representation in memory a nd not to interference from LION or from the Animal -LION association. The typi cal results of the indepe ndent probe technique are illustrated in Figure 24.

PAGE 127

Independent Cue Practiced Category Independent Cue 117 Practiced Unpracticed Sphinx Animal Race LION .75 HORSE .25 EXEMPLAR EXEMPLAR (Rp+) (Rp-) Independent Cue Unpracticed Category Independent Cue Robe Profession Needle JUDGE .35 NURSE .35 Unpracticed Unpracticed EXEMPLAR EXEMPLAR (Nrp) (Nrp) Figure 24. Typical results adapted from Anderson & Spellman (1995) using the independent probe technique and the retrieval practice paradigm. Note that recall of HORSE is impaired (.25), relative to performance on the baseline items JUDGE and NURSE (.35), which are members of the category that was not practiced. As shown in Figure 24, the retrieval practice of LION to Animal LI______ caused HORSE to be suppressed. Not only was HORSE produced at a lower rate to the cue Animal, it was also produced as at lower rate to the independent cue Race, compared to both baseline items, JUDGE and NURSE to their independent cues, Robe and Needle. Because the association between HORSE and Race was unlikely to be weakened by the strengthening of the association between Animal and LION, then race served as an independent cue for HORSE. The blocking explanation can be ruled out because the independent cue (i.e. Race) is unlikely to elicit the practiced item (i.e. LION) or the link

PAGE 128

118 between Animal and LION, eliminating the associat ive source of interference. They concluded that because RIF wa s found with independent cues inhibition acts specifically upon the related items in memory, and the effect is not due to interf erence or blocking. Many studies have obtained inhibition e ffects with independent probes using various types of stimuli (Anderson & Spe llman, 1995; Anderson & Bell, 2001; Anderson & Green, 2001; Anderson, Green, & Mc Culloch, 2000; Johnson & Anderson, 2004; Shivde & Anderson, 2001; Radvansky, 1999; Cart er, 2002; Camp, Pecher, & Schmidt, in press; Veling & van Knippenberg, 2004) however see Williams & Zacks, 2001 and Perfect, Stark, Tree, Moulin, Ahmed, & Hutte r, 2004), lending ev idence that the noninhibitory accounts are incorrect. However, there is some evidence in support of the cue change account. For example, Perfect and colleagues (2004) failed to find RIF with a second explicitly encoded word used as an independent probe. However, their findings are inconsistent with the s ubstantial array of studies that found RIF effects with independent probes suggesting the alternative, non-inhibitory account that they offer does not provide a general account of the data. These negative findings do suggest, however, important moderating factors that must be considered to achieve full understanding of this phenomenon Because many studies have used the indepe ndent probe technique to differentiate between inhibitory and noninhibitory accounts, future work with the paradigm developed in the present study might use this method to distinguish between the inhibitory account and the cue/contextual change account. Like other non-inhibitory accounts, the cue/contextual change accounts assumes that forgetting effects are contingent on the use of the original cue employed in the study and retrieval practice phases. With the independent probe techniqu e, the items are cued during the test phase with associates that are uniquely related to the to-be-retrieved item, circumventing any contextual bias that the original category ma y receive as a result of retrieval practice. If forgetting-like effects are observed onl y because the design of the experiment biases the original cue, then using a novel cue should eliminate such biases, and so no forgetting should be observed in either retrieval practice pattern condition. On the other hand, if forgetting effects are found in the Intra-set (weak) condition with independent cues, then this would provide convincing evid ence that RIF is due to inhibition of the specific items representation in memory a nd not to interference from blocking, cue change, or context mismatch. Future studies could use the indepe ndent probe technique with the list from Experiment 1 to determine if inhibition acts on sp ecific representations in memory. It is believed that semantic integration will reduce the RIF even with independent probes because Anderson, Gree n, & McCulloch (2000) performed a highly similar manipulation but with episodic, as opposed to semantic integration, and found reduced RIF in the Inter-set (strong) condition (i.e. high target-competitor similarity) and inhibition in the Intra-set (weak) condition (i.e. high competit or-competitor similarity) even with independent probes. These resu lts suggests that ap plying the independent probe method in the context of a semantic in tegration manipulation is likely to confirm the role of inhibitory processe s, and is likely to reveal that semantic integration reduces the effect, even on an independent cue test.

PAGE 129

119 Concluding Remarks The results of the current study highlight th e importance of the memory system to distinguish between related information that competes for cognitive resources and related information that does not compete. For example, some information is important because it supports the current needs a nd goals of the system. Sometimes that information is integral to understanding or ev en overlaps with other repres entations that reinforce that understanding. It is adaptive for the memory system to s uppress interfering information, but it is not helpful to prevent other thes e useful representations from becoming and staying active. For example, trying to remember that horses are part of the equine family should not make us forget that ponies and donkeys are also a part of the equine family. Because both the recalled and unr ecalled members of the equine family are integral to our conception of that sub-category and because the member representations are so integral to each other, our understanding of that fam ily should not be compromised by recalling specific highly related instances of that sub-category. The results of the current set of studies support the notion that the memory system is set up to prevent representations that are integral to understanding from becoming suppressed.

PAGE 130

120 References Cited Anderson, J. (1983). The Architecture of Cognition Cambridge, MA: Harvard University Press. Anderson, M.C. (2003). Rethinking interfer ence theory: Executiv e control and the mechanism of forgetting. Journal of Memory and Language,49, 415-445. Anderson, M. C., & Bell, T. (2001). Forgetting our facts: The role of inhibitory processes in the loss of propositional knowledge. Journal of Experimental Psychology: General, 130(3), 544-570. Anderson, M.C., & Bjork, R.A.(1994). Mechanisms of inhibition in long-term memory: A new taxonomy. In D. Daganbach, & T. Carr (Eds.), Inhibitory processes in attention, memory, and language (pp. 265-326). San Diego, CA: Academic Press. Anderson, M.C., Bjork, R.A., & Bjork, E.L. ( 1994). Remembering can cause forgetting: Retrieval Dynamics in long-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(5) 1063-1087. Anderson, M. C., Bjork, E. L., & Bjork, R. A. (2000). Retrieval-induced forgetting: Evidence for a recall-specific mechanism. Psychonomic Bulletin & Review, 7, 522-530. Anderson, M.C., Green, C., & McCulloh, K.C. ( 2000). Similarity and inhibition in longterm memory: Evidence for a two factor theory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(5) 1141-1159. Anderson, M.C., & Neely, J.H (1996). Interfer ence and inhibition in memory retrieval. In E.L. Bjork, & R.A. Bjork (Eds.), Memory. Handbook of perception and cognition (2nd ed.) (pp. 237-313). San Diego, CA: Academic Press. Anderson, M.C., & McCulloch, K.C. (1999). Integration as a general boundary condition on retrieval-induced forgetting. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(3), 608-629.

PAGE 131

121 Anderson, M.C., & Spellman, B.A. (1995). On th e status of inhibitory mechanisms in cognition: Memory retrieval as a model case. Psychological Review, 102(1), 68100. Bauml, K. (1998). Strong items get suppressed, weak items do not: The role of item strength in output interference. Psychonomic Bulletin & Review, 5 459-463. Bauml, K. (2002). Semantic genera tion can cause episodic forgetting. Psychological Science, 13(4), 356-360. Bauml, K.H., & Hartinger, A. (2002). On the ro le of item similarity in retrieval-induced forgetting. Memory, 10(3), 215-224. Barnier, A.J., Hung, L., & Conway, M.A. ( 2004). Retrieval-induced forgetting of emotional and unemotional autobiographical memories. Cognition & Emotion, 18(4) 457-477. Battig, W.F., & Montague, W.E. (1969). Category norms for verbal items in 56 categories: A replication and extension of the Connecticut category norms. Journal of Experimental Psychology: Monograph, 80(3-2) 1-46. Blaxton, T.A, & Neely, J.H. (1983). Inhibi tion from semantically related primes: Evidence of a category-specific inhibition. Memory & Cognition, 11(5), 500-510. Bjork, R.A. (1989). Retrieval inhibition as an adaptive mechanism in human memory. In H.L. Roediger III and F.I.M. Craik (Eds.), Varieties of memory and consciousness: Essays in honor of Endel Tulving (pp. 309-330). Hillsdale, NJ: Lawrence Erlbaum Associates. Bjorklund, D.F., & Harnishfeger, K.K. (1995). The evolution of inhibition mechanisms and their role in human cognition and be havior. In F.N. Dempster and C.J. Brainerd (Eds.), Interference and inhibition in cognition (pp. 141-204). San Diego, CA: Academic Press. Broadbent, D.E., Cooper, P.F., FitzGerald, P., & Parkes, K.R. (1982). The Cognitive Failures Questionnaire (CFQ) and its correlates. British Journal of Clinical Psychology, 21, 1-16. Butler, K. M., Williams, C. C., Zacks, R. T., & Maki, R. H. (2001). A limit on retrievalinduced forgetting. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 1314-1319.

PAGE 132

122 Camp, G., Pecher, D., & Schmidt, H.G. (submitted). Retrieval-induced forgetting in implicit memory tests: the role of test awareness. Psychonomic Bulletin & Review. Carter, K. L. (2002). Investigating semantic inhibition using a modifi ed independent cue Task (Doctoral Dissertation, Un iversity of Kansas, 2002). Dissertation Abstracts International, 51 417. Ciranni, M. A., & Shimamura, A. P. (1999). Retrieval-induced forgetting in episodic memory. Journal of Experimental Ps ychology: Learning, Memory, and Cognition, 25 1403-1414. Conway, M.A., Harries, K., Noyes, J., Racsmany, M, & Frankish, C. (2000). The disruption and dissolution of directed forgetting: Inhibitory control of memory. Journal of Memory and Language, 43 409-430. Dunn, E.W., & Spellman, B.A. (2003). Forgetting by remembering: Stereotype inhibition through rehearsal of al ternative aspects of identity. Journal of Experimental Social Psychology, 39 420-433. Enright, S.J., & Beech, A.R. (1993). Further evidence of cognitive inhibition in obsessive-compulsive disorder. Personality and Individual Differences, 14 387-395. Ford, R.M., Keating, S., & Patel, R. ( 2004). Retrieval-induced forgetting: A developmental study. British Journal of Developmental Psychology, 22 585-603. Golding, J.M., & MacLeod, C.M. (Eds.). Intentional forgetting: Interdisciplinary Approaches. Mahwah, NJ: Lawrence Erlbaum Associates. Goodmon, L.B., & Nelson, D.L. (2004). Streng thening the activati on of unconsciously activated memories. Memory & Cognition, 32(5), 804-818. Hasher, L., Stoltzfus, E.R., Zacks, R.T., & Rypma, B. (1991). Journal of Experimental Psychology: Learning, Memo ry, & Cognition, 17(1) 163-169. Hicks, J.L., & Starns, J.J. (2004). Retrieval-induced forgetting occurs in tests of item recognition. Psychonomic Bulletin & Review, 11(1), 125-130. Holley, P.E., & McEvoy, C.L. (1996). Aging and inhibition of unconsciously processed information. No apparent deficit. Applied Cognitive Psychology, 10 241-256.

PAGE 133

123 Horne, J.A., & Ostberg O. (1976). A self -assessment questionnaire to determine morningness-eveningness in human circadian rhythms. International Journal of Chronobiology, 4, 97-110. Horton, D.L., & Kjeldergaard, P.M. (1961). An experimental analysis of associative factors in mediated generalization. Psychological Monographs: General and Applied, 75, 1-26. Humphreys, M. (2002). Submitted research grant. Jenkins, J. (1963). Mediated a ssociations: Paradigms and s ituations. In C.N. Cofer & B.S. Musgrave (Eds.), Verbal behavior and learning: Problems and processes (pp 210-257). New York: Macgraw-Hill. Johnson, S.K., & Anderson, M.C. (2004). The role of inhibitory control in forgetting semantic knowledge. Psychological Science 15(7), 448-453. Kjeldergaard, P.M. (1968). Transfer and medi ation in verbal lear ning. In T.R. Dixon & D.L. Horton (Eds.), Verbal behavior and general behavior theory (pp. 10711075). Englewood Cliffs, NJ: Prentice Hall. Koustaal, W., Schacter, D.L., Johnson, M.K., & Galluccio, L. (1999). Facilitation and impairment of event memory produced by photograph review. Memory & Cognition, 27 478-493. Kucera, H., & Francis, W.N. (1967). Computational Analysis of present-day American English Providence, RI: Brown University Press. Lustig, C., Hasher, L., & Toney, S.T. (2001). Inhibitory control over the past and the present. European Journal of Cognitive Psychology, 13 107-122. Macrae, C. N., & MacLeod, M.D. (1999). On recollections lost: When practice makes imperfect. Journal of Personality and Social Psychology, 77(3), 463-473. Macrae, C. N., & Roseveare, T.A. (2002). I was always on my mind: The self and temporary forgetting. Psychonomic Bulletin & Review, 9, 611-614. MacLeod, M. (2002). Retrieval-induced forget ting in eyewitness memory: Forgetting as a consequence of remembering. Applied Cognitive Psychology, 16 135-149. McDowd, J.M., Oseas-Kreger, D.M., & Filion, D.L. (1995). Inhibition processes in cognition and aging. In F.N. Demp ster and C.J. Brainerd (Eds.), Interference and Inhibition in Cognition (pp. 363-400). San Diego, CA: Academic Press.

PAGE 134

124 McEvoy, C.L., & Nelson, D.L. (1982). Category name and instance norms for 106 categories of various sizes. American Journal of Psychology, 95(4) 581-634. McGeoch, J.A. (1942). The psychology of human learning: An introduction New York: Longmans. McNamara, D.S., & McDaniel, M.A. (2004). Suppressing irrelevant information: Knowledge activation or inhibition? Journal of Experi mental Psychology: Learning, Memory, and Cognition, 30(2) 465-482. Moulin, C.J.A., Perfect, T.J., Conway, M.A ., North, A.S., Jones, R.W., & James N. (2002). Retrieval-induced forgetting in Alzheimers disease. Neuropsychologia, 40, 862-867. Nelson, D.L., Bennett, D.J., Gee, N.R., Schr eiber, T.A., & McKinney, V.M. (1993). Implicit memory: Effects of network size and interconnectivity on cued recall. Journal of Experi mental Psychology: Learning, Memory, and Cognition, 19(4) 747-764. Nelson, D.L., & Goodmon, L.B. (2003). Disrupting attention: The need for retrieval cues in working memory theories. Memory & Cognition, 31(1), 65-76. Nelson, D.L., & Goodmon, L.B. (2002). Experiencing a word can prime its accessibility and its associative connections to related words. Memory & Cognition, 30(3), 380-398. Nelson, L.B., Dyrdal, G.M., & Goodmon, L.B. (in pr ess). What is strength? Predicting free association, similarity, and cued recall. Psychonomic Bulletin & Review Nelson, D.L., McEvoy, C.L. & Schreiber, T.A. (1999). The University of South Florida word association, rhyme, and word fragment norms. http://www.usf.edu/FreeAssociation/. Nelson, D.L., McKinney, V.M., Gee, N.R., & Janczura, G.A. (1998). Interpreting the influence of implicitly activated me mories on recall and recognition. Psychological Review, 105 299-324. Overshelde, J.P.V., Rawson, K.A., & Dunlos ky, J. (2004). Category norms: An updated and expanded version of the Battig and Montague (1969) norms. Journal of Memory and Language, 50, 289-335.

PAGE 135

125 Perfect, T.J., Moulin, C.J. A, Conway, M.A., & Perry, E. (2002). Assessing the inhibitory account of retrieval-induced forgetti ng with implicit-memory tests. Journal of Experimental Psychology: Learning, Memory, & Cognition, 28(6) 1111-1119. Perfect, T.J., Stark, L.J., Tree, J.J., Mouli n, C.J. A, Ahmed, L., & Hutter, R. (2004). Transfer appropriate forgetting: The cuedependent nature of retrieval-induced forgetting. Journal of Memory and Language, 51(3) 399-417. Phenix, T., & Campbell J.I.D. (2003). Lured from the integrated structures model by retrieval-induced forgetting. BBCS/SCCC 13th Annual Meeting Postman, L. (1971). Transfer, interference and forgetting. In J.W. Kling and L.A. Riggs, (Eds.) Woodworth and Schlosberg s experimental psychology (3rd ed.; pp 10191132). New York: Holt, Rinehart & Winston. Postman, L., Stark, K., & Fraser, J. (1968) Temporal changes in interference. Journal of Verbal Learning and Verbal Behavior, 7, 672-694. Raaijmakers, J.W., & Shiffrin, R.M. (1981). Search of associative memory. Psychological Review, 88, 93-134. Radvansky G.A. (1999). Memory retrieval and suppression: The inhi bition of situation models. Journal of Experimental Psychology: General, 128 563-579. Roediger, H.L. III, & Neely, J.H. (1982). Retrieval blocks in episodic and semantic memory. Canadian Journal of Psychology, 36 213-242. Rundus, D. (1973). Negative effects of using list items as retrieval cues. Journal of Verbal Learning and Verbal Behavior, 1 2, 43-50. Saunders, J., & MacLeod, M.D. (2002). Ne w evidence on the suggestibility of a memory: The role of retrieval-induced forgetting effects in misinformation effects. Journal of Experimental Psychology: Applied, 8 127-142. Shaw, J.S., Bjork, R.A., & Handal, A. (1995) Retrieval-induced forgetting in an eyewitness memory paradigm. Psychonomic Bulletin and Review, 2 249-253. Shivde, G., & Anderson, M. C. (2001). The role of inhibition in meaning selection: Insights from retrieval-induced forgetting. In Gorfein, David S. (Ed). (2001). On the consequences of meaning selecti on: Perspectives on resolving lexical ambiguity. Decade of behavior. (pp. 175-190). Washington, DC, US: APA. Smith, A.D. (1971). Output interference and organized recall from long-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 400-408.

PAGE 136

126 Smith, E.E., Adams, N., & Schorr, D. (1978). Fact retrieval a nd the paradox of interference. Cognitive Psychology, 10 438-464. Solso, R.L. (1980). Positional frequency and versatility of bigrams for twothrough nine-letter English words. Behavior Research Methods & Instrumentation, 12(3) 297-343. Starns, J.J., & Hicks, J.L. (2004). Ep isodic generation can cause semantic forgetting: retrieval-induced forgetting of false memories. Memory & Cognition, 32(4), 602-609. Tipper, S.P. (1985). The negative priming effec t: Inhibitory effects of ignored primes. Quarterly Journal of Ex perimental Psychology, 37A 571-590. Veling, H., & van Knippenberg, A. (2004). Remembering can cause inhibition: Retrieval-induced inhibition as cue independent process. Journal of Experimental Psychology: Learning, Memo ry, & Cognition, 30(2) 315-318. White, J.J (2003). Stereotypes and retrievalinduced forgetting: When elderly are stronger than their middle-aged counterpar ts. Thesis abstract, California State University. Williams, C.C., & Zacks, R.T. (2001). Is re trieval-induced forgetting an inhibitory process? American Journal of Psychology, 114(3), 329-354. Wright, D.B., Loftus, E.F., & Hall, M. (2001). Now you see it; Now you dont; Inhibiting recall and r ecognition of scenes. Applied Cognitive Psychology, 15, 471-482. Yoon, C., Feinberg, F., Hu, P., Gutchess, A.H, Hedden, T., Chen, H., Jing, Q., Cui, Y., & Park, D.C. (2004). Category norms as a function of culture and age: Comparisons of item responses to 10 5 categories by American and Chinese adults. Psychology and Aging, 19(3), 379-393. Zachary, R. A. (1991). The manual of the Shipley Institute of Living Scale Los Angeles, CA: Western Psychological Services. Zellner M., & Bauml K. (2003 and submitted). In tact retrieval inhibition in childrens episodic recall. X111 Meeting of the European Society for Cognitive Psychology.

PAGE 137

127 Bibliography Anderson, J.R., & Pirolli, P.L. (1994). Spread of activation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10 791-798. Anderson, M.C. (2001). Active forgetting: Evid ence for functional inhibition as a source of memory failure. Journal of Aggression, Maltreatment, and Trauma, 4(2), 185-210. Anderson, M.C., & Green, C. (2001). Suppr essing unwanted memories by executive control. Nature, 410(6826), 336-369. Anderson, M.C., Ochsner, K.N., Kuhl, B., C ooper, J., Robertson, E., Gabrieli, S.W., Glover, G.H., & Gabrieli, J.D.E. ( 2004). Neural systems underlying the suppression of unwanted memories. Science, 303, 232-235. Bauml, K.H., & Kuhbandner, C. (2003). Retrie val-induced forgetting and part-list cuing in associatively structured lists. Memory & Cognition, 31(8) 1188-1197. Bjork, R.A. (1989). Retrieval inhibition as an adaptive mechanism in human memory. In H.L. Roediger III and F.I.M. Craik (Eds.), Varieties of Memory and Consciousness: Essays in Honor of Endel Tulving (pp. 309-330). Hillsdale, NJ: Lawrence Erlbaum Associates. Brewin, C.R, & Andrews, B. (2000). Psychol ogical defense mechanisms: The example of repression. The Psychologist, 13(12) 615 617. Conway, M.A., & Fthenaki, A. (2003). Disruption of inhibitory c ontrol of memory following lesions to the front al and temporal lobes. Cortex, 39, 667-686. Craik, F.I.M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104 268-294. Deese, J. (1959). Influence of inter-item asso ciative strength upon imme diate free recall. Psychological Reports, 5 305-312. Deese, J. (1965). The structure of associations in language and thought. Baltimore, MD: Johns Hopkins University Press.

PAGE 138

128 Friedman, N.P., & Miyake, A. (2004). The re lations among inhibition and interference control functions: A latent variable analysis. Journal of Experimental Psychology: General, 133 101-135. Hamilton, A.C. & Martin, R.C. (in press). Dissociations among tasks involving inhibition: A single case study. Cognitive, Affective, and Behavioral Neuroscience Humphreys, M.S., & Galbraith, R.C. (1975). Fo rward and backward associations in cued recall: Predictions from the encoding specificity principle. Journal of Experimental Psychology: Human Learning & Memory, 1 702-710. Kahn, I., Wagner, A.D. (2004). Diminished medial temporal lobe activation with expanding retrieval practice. Department of Brain & Cognitive Sciences and Center for Learning & Memory, MIT; Mar tinos Center for Biomedical Imaging, MGH/MIT/HMS. Lane, S.M., Mather, M., Villa, D., & Morita, S.K. (2001). How events are reviewed matters: Effects of varied fo cus on eyewitness suggestibility. Memory & Cognition, 29(7) 940-947. Loftus, G.R., & Loftus, E.F. (1974). The influence of one memory retrieval on a subsequent memory retrieval. Memory & Cognition, 2 467-471. MacLeod, M.D., & Macrae, C. N (2001). Gone but not forgotten: The transient nature of retrieval-induced forgetting. Psychological Science, 12(2), 148-152. Neely, J.H., & Durgunoglu, A.Y. (1985). Dissoci ative episodic and semantic priming effects in episodic recognition and lexical decision tasks. Journal of Memory & Language, 24 (4) 466-489. Nelson, D.L., Bennett, D.J., & Leibert, T.W. (1997). One step is not enough: Making better use of association norms to predict cued recall. Memory & Cognition, 25 785-796. Nelson, D.L., Bennett, D.J., & Xu, J. (1997). Recollective and automatic uses of memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23(4) 872-885. Nelson, D.L., McEvoy, C.L., & Casanueva, D.M. (1982). Category size and free recall. American Journal of Psychology, 95, 235-249.

PAGE 139

129 Nelson, D.L., McEvoy, C.L., & Dennis, S. (2000). What is free association and what does it measure? Memory & Cognition, 28(6) 877-899. Nelson, D.L., McEvoy, C.L., & Schreiber, T.A. (1990). Encoding context and retrieval conditions as determinants of the effects of natural category size. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16 31-41. Nelson, D.L., Schreiber, T.A., & Holley, P.E. (1 992). The retrieval of controlled and automatic aspects of meaning on direct and indirect tests. Memory & Cognition, 20(6) 671-684. Nelson, D.L., Schreiber, T.A., & McEvoy, C.L. (1992). Processing implicit and explicit representations. Psychological Review, 99 322-348. Nelson, D.L., Schreiber, T.A., & Xu, J. (1999). Cue set size effects: Sampling activated associates or cross-target interference. Memory & Cognition, 27(3) 465-477. Nelson, D.L., & Xu, J. (1995). Effects of implic it memory on explicit recall: Set size and word frequency effects. Psychological Research, 57, 203-214. Nelson, D.L., & Zhang, N. (2000). The ties that bind what is known to the recall of what is new. Psychonomic Bulletin & Review, 7(4) 604-617. Nelson, D.L., Zhang, N, & McKinney, V.M. (2001). The ties that bind what is known to the recognition of what is new. Journal of Experimental Psychology: Learning Memory, and Cognition, 27(5) 1147-1159. MacLeod, M.D., & Macrae, C.N. (2001). Gone but not forgotten: The transient nature of retrieval-induced forgetting. Psychological Science, 12(2) 148-152. MacLeod, C.M., Dodd, M.D., Sheard, E.D., Wilson, D.E., & Bibi, U. (2003). In opposition to inhibition. The Psychology of Learning and Motivation, 43 163-214. McEvoy, C.L., Nelson, D.L., & Komatsu, T. ( 1999). What is the connection between true and false memories? The differential role of inter-item associations in recall and recognition. Journal of Experimental Ps ychology: Learning, Memory, and Cognition, 25(5) 1177-1194. McNamara, T.P. (1992). Theories of Prim ing: I. Associative Distance and Lag. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(6) 1173-1190.

PAGE 140

130 Oram, M.W., & MacLeod, M.D. (2001). Remembering to forget: Modeling inhibitory and competitive mechanisms in human memory. 23rd Annual Conference of Cognitive Science 738-743. Paivio, A., Yuille, J.C., & Madigan, S.A. (1968). Concreteness, imagery, and meaningfulness values for 925 nouns. Journal of Experime ntal Psychology: Monograph Supplement, 76(1-2) 1-25. Rubin, D.C. (1980). 51 properties of 125 words: A unit analysis of verbal behavior. Journal of Verbal Learning and Verbal Behavior, 19, 736-755. Sahakyan, L., & Kelley, C. (2002). A contextu al change account of the directed forgetting effect. Journal of Experimental Ps ychology: Learning, Memory, & Cognition, 28(6) 1064-1072. Schreiber, T.A. (1998). Effects of target se t size on feelings of know ing and cued recall: Implications for the cue effectivenes s and partial-retrieval hypotheses. Memory & Cognition, 26(3) 553-571. Shapiro, S.I., & Palermo, D.S. (1970). Concep tual organization and class membership: Normative data for representatives of 100 categories. Psychonomic Monograph Supplements, 3(11, whole No. 43) 107-127. Shiffrin, R.M., Huber, D.E., & Marinelli, K. (1995). Effects of category lengths and strength on familiarity in recognition. Journal of Experime ntal Psychology: Learning, Memory, and Cognition, 21(2) 267-287. Shiffrin, R.M., & Steyvers, M. (1997). A model for recognition memory: REMretrieving effectively from memory. Psychonomic Bulletin & Review, 4(2) 145-166. Sines, L. K. (1958). Intelligence tests correlates of Shipley -Hartford performance. Journal of Clinical Psychology, 14 399-404. Smith, R. E., & Hunt, R. R. (2000). The influe nce of distinctive processing on retrievalinduced forgetting. Memory & Cognition, 28 503-508. Steyvers, M. (submitted). Semantic spaces based in free association that predict memory performance. Manuscript submitted for publication. Storm, B.C., Bjork, E.L., & Bjork, R.A (submitted). On the role of retrieval-induced forgetting in maintaining and m odifying initial impressions. Special issue on metacognition in the Journal of Memory and Language

PAGE 141

131 Tulving, E., & Thompson, D.M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5) 352-373. Wixted, J.T. (2004). The psycholog y and neuroscience of forgetting. Annual Review of Psychology, 55, 235-269. Zeelenberg, R., Shiffrin, R.M., & Raaijmakers, J.G. (1999). Priming in a free association task as a function of association directionality. Memory & Cognition, 27 956-961.

PAGE 142

132 Appendices

PAGE 143

Appendix A: Predictions from Processing Implicit and Explicit Representations (PIER2) In order to demonstrate how PIER2 makes the aforementioned prediction, assume that HORSE activates a small set of five associates as shown below; RIDE, PONY, ANIMAL, COLT, and MULE with the probabilities of .26, .11, .08, .02, and .01 respectively. It is important to note that 10 associates were omitted from the example, in order to provide a more straightforward example. For the original associative database and the complete associative network of HORSE, visit the website http://www.usf.edu/FreeAssociation/ Simplified version of HORSEs associative network based on free association. HORSE A1 ( RIDE ) A2 ( PONY ) A3 ( ANIMAL ) A5 ( COLT ) A6 ( MULE ) HORSE 1.00 .26 .11 .08 .02 .01 A1(RIDE) .37 .02 A2(PONY) .75 .05 .02 A3 ( ANIMAL .02 A5(COLT) .65 .13 A6(MULE) .17 .03 Net Associate Strength 2.96 .31 .26 .11 .04 .01 The strength of the associations are based on free association probabilities or how likely the word HORSE produced the items in free association. According to PIER2, the associative network is implicitly and automatically activated in long-term working memory when the word HORSE is experienced, and some associates are activated to a greater degree than others. In this example and in the original associative matrix of HORSE, the word PONY is the second strongest associate of HORSE because it is produced with the second greatest probability of 11%. This means that out of 100 people, 11 produced the response PONY when free associating to the word HORSE. According the norms, HORSE is also a strong associate of PONY. In fact, it is the strongest associate of PONY being produced with a probability 75%. 133

PAGE 144

Appendix A (Continued) PIER2 provides a series of equations that mathematically describe the activation process. First, the pre-existing associations manifest their effects by increasing the episodic self-strength of the experienced word and the episodic strength of the associates of the experienced word. The self-strength for HORSE (Equation 1) and its associates (Equation 1a) are computed using free association probabilities, and the computational examples of the activation level of HORSE and one of its associates PONY, and a non-associate such as LION. Equation 1 and computational example of activation self-strength of HORSE: 134 ST,TSA,Tn ST iii1 nST,ASA,An j1jiji1 = [(1.00+1.96)]+[(.26+.05))+(.11+.15)+(.08+.03)+(.02+.02)+ (.01+.00)] = 3.69 s LION. Note a, target-to-associate strength; b, other associate strength: S(Ai) =n Equation 1a and computational example of activation strength of the associatePONY, and a non-associate of HORSE such as j1jSAi,Aji1 nST,A = Competitors are associates that are uniquely activated by either the target or the cue. S(A2(PONY)) = [(.11a)+(.15b)] = .26 S(Non-associate, e.g. LION) = [(.00a)+(.00b)] = .00 In the model, the target-cue intersection equation (Equation 2) is converted into probability of target recovery (Equation 3) by computing a signal to noise ratio that uses the strengths of competing associates activated by the target and by the test cue as noiseelements.

PAGE 145

Appendix A (Continued) Computational example of the probability of recovering HORSE when a strong associate (A2(PONY)) or non associate is experienced. Note a, target self-strength from Equation 1; b, primed associate strength derived from Equation 1a; c, forward strength indexed by free association norms; d, test associate nominal self-strength; e, competitor strength for target (t); f, competitor strength for the associate (q): Retrieval Matrix: Strong Cue (A2) = PONY Cue-Target Intersection Primed Target Word (ki) A2 (kj) Competitors (t) Competitors (q) Target(i) 3.69a .26b .38e -Cue(j) .75c 1.00d -.10e Equation 2: Q,T) S( ji SSn (*3.69) + jkikk = .75 (1.00*.26) = 3.03 Equation 3: n+ST,Atn Ps (Ti/Qj )S(Qj,Ti)S(Qj,Ti) SQ,Aq qt = Ps(HORSE/ PONY) = 3.03 / [(3.03) + (.38 +.10)] = .86 etrn Arget Prt Wi) A Comps (t) Competitors (q) R-Ta ieval Matrix: Uimrge related ssociate = LION Associate Intersection ed Tao rd (k n(kj) etitor Target(i) 3.48a .00b 1.00 -Asscd-1.00 ociate(j) .00 1.00 being a ere are stronger target and cue competitor strengths between them (1.00 and 1.00 between HORSE and LION, comeen HORSE and PONY). The Equation 2:S(Qj,Ti) = (.00*3.48) + (1.00*.00) = .00 Equation 3: Ps(HORSE/ LION) = .00 / [(.00) + (1.00 + 1.00)] = .00 The results of these computations suggest that HORSE stands a stronger chance ofactivated or cued when PONY is retrieval practiced and experienced (.86), than whenLION is retrieval practiced and experienced (.00). It is important to note that the probability of target recovery process takes into account the detrimental influence of competitors coming from the target and from the cue. In this case, not only is LIONnon-associate of HORSE, it also does not share many associates with HORSE. Thus, th pared to .38 and .10 betw 135

PAGE 146

136 be reduced when PONY is retrieval practiced and HORSE is not, but there would be normal levels of RIF of HO RSE when LION is retrieval-practiced and HORSE is not. Appendix A (Continued) model predicts that the stronger the competito rs strengths, the lower the probability of recovery. To the extent that strength of association is high ly negatively correlated with competitor strength, weaker associates are predicted to suffer more from the interfering influence of competitors. From PIER2s activ ation assumptions, it is predicted that the RIF effect would

PAGE 147

Appendix B: Experimental Lists Experiments 1, 3, and 4 Categories Exemplars Inter-set (strong) (strong) Intra-set (weak) (weak) Set A Set B Set A Set B Animal LION TIGER LION HORSE HORSE PONY TIGER PONY WALRUS SEAL DUCK WALRUS DUCK GOOSE GOOSE SEAL Profession LAWYER JUDGE LAWYER CHEF CHEF BAKER JUDGE BAKER NURSE DOCTOR PRESIDENT NURSE PRESIDENT SENATOR SENATOR DOCTOR Appliance TOASTER OVEN TOASTER BLENDER BLENDER MIXER OVEN MIXER RADIO STEREO WASHER RADIO WASHER DRYER DRYER STEREO Weapon GUN RIFLE GUN BOW BOW ARROW RIFLE ARROW POISON CHEMICAL DAGGER POISON DAGGER SWORD SWORD CHEMICAL Filler Categories Filler Exemplars Filler Exemplars Inter-set (strong) (strong) Intra-set (weak) (weak) Set A Set B Set A Set B City RENO MEMPHIS RENO DETROIT DETROIT LONDON MEMPHIS LONDON PARIS ATLANTA BERLIN PARIS BERLIN ORLANDO ORLANDO ATLANTA River CONGO THAMES CONGO AMAZON AMAZON POTOMAC THAMES POTOMAC HUDSON RHINE SUWANNE HUDSON SUWANNEE NILE NILE RHINE 137

PAGE 148

Appendix B (Continued) Experiment 2 Categories Exemplars Inter-set (strong) Intra-set (weak) Triad A Triad B Triad A Triad B Bird ROBIN SPARROW ROBIN OSTRICH OSTRICH PIGEON SPARROW FALCON FALCON VULTURE PIGEON VULTURE Drink SCOTCH WHISKEY SCOTCH ALE BOURBON CHAMPAGNE BOURBON CHAMPAGNE ALE DAIQUIRI WHISKEY DAIQUIRI Fruit ORANGE TANGERINE ORANGE BANANA BANANA APPLE TANGERINE STRAWBERRY STRAWBERRY CHERRY APPLE CHERRY Furniture BUREAU DRESSER BUREAU CABINET MIRROR HASSOCK DRESSER HASSOCK CABINET FOOTSTOOL MIRROR FOOTSTOOL Insect ROACH SPIDER ROACH CATERPILLAR CATERPILLAR BEETLE SPIDER BEETLE HORNET GRASSHOPPER HORNET GRASSHOPPER Metal NICKEL SILVER PLATINUM ALUMINUM ALUMINUM URANIUM SILVER URANIUM PLATINUM MERCURY NICKEL MERCURY Profession EXECUTIVE PRESIDENT EXECUTIVE FARMER FARMER ACCOUNTANT PRESIDENT TEACHER DOCTOR TEACHER ACCOUNTANT DOCTOR Sport VOLLEYBALL FOOTBALL VOLLEYBALL HOCKEY WRESTLING TENNIS FOOTBALL SKIING HOCKEY SKIING TENNIS WRESTLING 138

PAGE 149

Appendix B (Continued) Experiment 2 (Continued) Filler Categories Filler Exemplars Inter-set (strong) Intra-set (weak) Triad A Triad B Triad A Triad B Clothing HAT GLOVES HAT SKIRT BLOUSE JACKET GLOVES PANTS SKIRT PANTS JACKET BLOUSE Disease DIABETES POLIO MEASLES DIABETES MEASLES SMALLPOX SMALLPOX POLIO FLU TYPHOID FLU TYPHOID 139

PAGE 150

140 Appendix C: Direct and indirect associ ations between the practiced and unpracticed sets, according to Nelsons free association norms. Appendix C provides the between set associ ations for the two experimental lists that will be used in the studies. There is a different table for each experimental list, retrieval practice pattern condition ( Inter-set (strong) Intra-set (weak) ), and direction of association (i.e. from Set A to B and from Set B to A). The tables are arranged by categories and sets that will serve in the practiced (Rp+) and unpracticed conditions (Rp). The tables list pairs of exemplars that are directly or indirectly related according to Nelsons Free association norms. Bolded pairs are directly associated, that is there is a Forward Strength Association (Fsg) and a probability of recove ry value (Eq3) for that pair. Indirect connections are also listed, including the number of shared as sociates (# Olaps), shared associate strength (Osg), number of mediators, (# Meds), and mediator strength (Msg). It is important to note that consiste nt with the retrieval practice pattern manipulation, there are many direct and indir ect associations between the sets in the tables listing the between set associations in the Inter-set (strong) conditions, however there are not many direct associations and indi rect associations between the sets in the tables listing the between set associations in the Intra-set (weak) conditions.

PAGE 151

141 Appendix C (Continued) Between-Set Associations from Set A to Set B Experimental 1, 3, & 4 List: Inter-set (strong) Condition Inter-set (strong) Condition Set A Set B Between Set Normative Associations Category Members of Set A Members of Set B A B (Fsg) A B # of Indirect A B Indirect Strength B Recovery (Eq3) # Olaps # Meds Osg Msg Animal LION TIGER .36 3 6 .0156 .0047 .90 LION SEAL 1 .0078 LION GOOSE 1 .0012 HORSE PONY .11 2 2 .0149 .0090 .82 HORSE TIGER 1 .0024 HORSE SEAL 1 1 .0104 .0000 HORSE GOOSE 1 .0016 WALRUS SEAL .14 5 .0244 .59 WALRUS TIGER 1 .0048 WALRUS GOOSE 1 .0032 WALRUS BAKER 1 .0005 DUCK GOOSE .16 3 1 .0085 .0010 .49 DUCK TIGER 1 .0009 DUCK SEAL 2 .0051 DUCK DOCTOR 1 .0060 DUCK RIFLE 2 .0020 DUCK ARROW 1 .0003 Profession LAWYER JUDGE .10 2 4 .0469 .0424 .58 LAWYER DOCTOR .15 2 .0013 LAWYER SENATOR 1 .0004 CHEF BAKER .00 2 .0864 Missing CHEF OVEN 2 .1412 CHEF ARROW 1 .0005 NURSE DOCTOR .55 4 4 .0056 .0219 .93 NURSE TIGER 1 .0005 NURSE SEAL 1 .0010 PRESIDENT SENATOR .00 2 1 .0048 .0007 Missing PRESIDENT JUDGE 2 .0003 Appliance TOASTER OVEN .47 1 1 .0028 .0057 .64 TOASTER BAKER 2 .0507 BLENDER MIXER .11 Missing BLENDER SEAL 1 .0002 BLENDER BAKER 1 .0025 BLENDER OVEN 1 1 .0010 .0094 BLENDER CHEMICAL 1 .0002 RADIO STEREO .20 4 3 .0952 .0119 .03 RADIO SEAL 1 .0002 WASHER DRYER .76 2 .0134 .89 WASHER OVEN 1 .0004

PAGE 152

142 Appendix C (Continued) Between-Set Associations from Set A to Set B Experimental 1, 3, & 4 List: Inter-set (strong) Condition (Continued) Inter-set (strong) Condition Set A Set B Between Set Normative Associations Category Members of Set A Members of Set B A B (Fsg) A B # of Indirect A B Indirect Strength B Recovery (Eq3) # Olaps # Meds Osg Msg Weapon GUN RIFLE .06 5 1 .0358 .0010 .91 GUN DOCTOR 2 .0005 GUN ARROW 1 .0072 GUN SWORD 2 .0015 BOW ARROW .40 .74 BOW DRYER 1 1 .0080 .0004 POISON CHEMICAL .00 2 1 .0014 .0008 .03 POISON TIGER 1 .0005 POISON RIFLE 1 .0009 POISON SWORD 1 .0009 DAGGER SWORD .15 6 1 .1873 .0310 .87 DAGGER RIFLE 1 .0001

PAGE 153

143 Appendix C: (Continued) Between-Set Associations from Set B to Set A Experimental 1, 3, & 4 List: Inter-set (strong) Condition Inter-set (strong) Condition Set B Set A Between Set Normative Associations Category Members of Set A Members of Set B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Animal TIGER LION .31 3 4 .0156 .0147 .83 TIGER HORSE 1 1 .0024 .0006 TIGER WALRUS 1 .0048 TIGER DUCK 1 .0009 TIGER NURSE 1 .0005 TIGER PRESIDENT 1 .0002 TIGER POISON 1 .0005 PONY HORSE .75 2 2 .0149 .0362 .91 PONY GUN 1 .0004 SEAL WALRUS .03 5 .0244 .31 SEAL LION 1 2 .0078 .0028 SEAL HORSE 1 1 .0104 .0026 SEAL DUCK 2 .0051 SEAL NURSE 1 .0010 SEAL PRESIDENT 1 .0006 SEAL BLENDER 1 .0002 GOOSE DUCK .26 3 .0085 .58 GOOSE LION 1 1 .0012 .0004 GOOSE HORSE 1 1 .0016 .0004 GOOSE WALRUS 1 .0032 Profession JUDGE LAWYER .10 2 2 .0469 .0127 .60 JUDGE PRESIDENT 2 1 .0003 .0002 JUDGE GUN 1 .0002 BAKER CHEF .03 2 2 .0890 .0060 .10 BAKER WALRUS 1 .0005 BAKER TOASTER 2 .0507 BAKER BLENDER 1 .0025 DOCTOR NURSE .38 4 3 .0056 .0020 .89 DOCTOR DUCK 1 .0001 DOCTOR GUN 2 1 .0005 .0031 DOCTOR POISON 1 .0007 SENATOR PRESIDENT .06 2 6 .0046 .0340 .04 SENATOR LION 1 .0003 SENATOR LAWYER 1 1 .0004 .0006

PAGE 154

144 Appendix C (Continued) Between-Set Associations from Set B to Set A Experimental 1, 3, & 4 List: Inter-set (strong) Condition (Continued) Inter-set (strong) Condition Set B Set A Between Set Normative Associations Category Members of Set A Members of Set B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Appliance OVEN TOASTER .00 1 1 .0024 .0001 .30 OVEN CHEF 2 1 .1412 .0110 OVEN BLENDER 1 .0010 OVEN WASHER 1 .0004 OVEN POISON 1 .0002 MIXER BLENDER .00 0 0 .0000 .0000 Missing STEREO RADIO .13 4 3 .0952 .0203 .74 STEREO GUN 1 .0003 DRYER WASHER .43 2 1 .0134 .0020 .83 DRYER BOW 1 .0080 RIFLE GUN .55 5 9 .0358 .1024 .94 RIFLE DUCK 2 1 .0020 .0004 RIFLE POISON 1 .0009 RIFLE DAGGER 1 .0001 ARROW BOW .53 0 1 .0000 .0020 .00 ARROW HORSE 1 .0005 ARROW DUCK 1 .0003 ARROW CHEF 1 .0005 ARROW GUN 1 2 .0072 .0146 CHEMICAL POISON .04 2 0 .0018 .0000 .08 CHEMICAL BLENDER 1 .0002 CHEMICAL GUN 1 .0003 CHEMICAL BOW 1 .0012 SWORD DAGGER .03 6 0 .1873 .0000 .73 SWORD HORSE 1 .0120 SWORD GUN 2 3 .0015 .0150 SWORD POISON 1 .0009

PAGE 155

145 Appendix C (Continued) Between-Set Associations from Set A to Set B Experimental List 2: Intra-set (weak) Condition Intra-set (weak) Condition Set A Set B Between Set Normative Associations Category Members of Set A Members of Set B A B (Fsg) A B # of Indirect A B Indirect Strength B Recovery (Eq3) # Olaps # Meds Osg Msg Animal LION HORSE 1 1 .0048 .0012 LION WALRUS 1 .0096 LION SEAL 1 .0078 TIGER HORSE 1 1 .0024 .0006 TIGER WALRUS 1 .0048 TIGER SEAL 2 .0041 TIGER NURSE 1 .0005 TIGER POISON 1 .0005 DUCK HORSE 1 1 .0024 .0006 DUCK WALRUS 2 .0054 DUCK SEAL 2 .0051 DUCK DOCTOR 1 .0060 DUCK ARROW 1 .0003 GOOSE HORSE 1 1 .0016 .0004 GOOSE WALRUS 1 .0032 GOOSE SEAL 1 .0026 GOOSE ARROW 1 .0001 Profession LAWYER DOCTOR .15 2 .0013 LAWYER NURSE 1 1 .0825 .0570 JUDGE DOCTOR 1 1 .0100 .0150 Appliance TOASTER BAKER 2 .0507 OVEN CHEF 2 1 .1412 .0110 OVEN BAKER 2 .0274 OVEN BLENDER 1 .0010 DRYER BOW 1 .0080 Weapon GUN DOCTOR 2 .0005 GUN ARROW 1 .0072 GUN POISON 3 .0166 RIFLE ARROW 2 1 .0040 .0013 RIFLE POISON 1 .0009 DAGGER POISON 1 .0009 SWORD HORSE 1 .0120 SWORD ARROW 2 .0031 SWORD POISON 1 .0009

PAGE 156

146 Appendix C (Continued) Between-Set Associations from Set B to Set A Experimental List 2: Intra-set (weak) Condition Intra-set (weak) Condition Set B Set A Between Set Normative Associations Category Members of Set A Members of Set B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Animal HORSE LION 1 1 .0048 .0016 HORSE TIGER 1 .0024 HORSE DUCK 1 .0024 HORSE GOOSE 1 .0016 HORSE GUN 1 .0004 PONY GUN 1 .0004 WALRUS LION 1 2 .0096 .0044 WALRUS TIGER 1 .0048 WALRUS DUCK 2 .0054 WALRUS GOOSE 1 .0032 SEAL LION 1 2 .0078 .0028 SEAL TIGER 2 .0041 SEAL DUCK 2 .0051 SEAL GOOSE 1 .0026 SEAL PRESIDENT 1 .0006 CHEF OVEN 2 .1412 BAKER TOASTER 2 .0507 Profession BAKER OVEN 2 .0274 NURSE TIGER 1 .0005 NURSE LAWYER 1 1 .0825 .0550 Appliance DOCTOR DUCK 1 .0001 DOCTOR JUDGE 1 1 .0100 .0100 DOCTOR GUN 2 1 .0005 .0031 BLENDER OVEN 1 1 .0010 .0094 STEREO GUN 1 .0003

PAGE 157

147 Appendix C (Continued) Between-Set Associations from Set B to Set A Experimental List 2: Intra-set (weak) Condition (Continued) Intra-set (weak) Condition Set B Set A Between Set Normative Associations Category Members of Set A Members of Set B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Weapon BOW DRYER 1 1 .0080 .0004 ARROW DUCK 1 .0003 ARROW GOOSE 1 .0001 ARROW GUN 1 2 .0072 .0146 ARROW RIFLE 2 .0040 ARROW SWORD 2 .0031 POISON TIGER 1 .0005 POISON GUN 3 1 .0166 .0018 POISON RIFLE 1 .0009 POISON DAGGER 1 .0009 POISON SWORD 1 .0009 CHEMICAL GUN 1 .0003

PAGE 158

148 Appendix C (Continued) Between-Set Associations from Triad A to Triad B Experimental 2 List: Inter-set (strong) Condition Inter-set (strong) Condition Triad A Triad B Between Set Normative Associations Category Members of Triad A Members of Triad B A B (Fsg) A B # of Indirect A B Indirect Strength B Recovery (Eq3) # Olaps # Meds Osg Msg Bird ROBIN PIGEON 1 .4100 ROBIN SPARROW 2 .4700 Drink SCOTCH WHISKEY 14 7 6 .0200 .0060 .66 SCOTCH CHAMPAGNE 4 1 .0200 .0002 BOURBON WHISKEY .14 7 2 .0400 .0070 .81 BOURBON CHAMPAGNE 4 1 .0300 .0003 ALE CHAMPAGNE 3 .0200 ALE WHISKEY 3 .0300 Fruit ORANGE APPLE .08 3 4 .0400 .0600 .71 ORANGE TANGERINE .05 3 .0200 .72 BANANA APPLE .15 2 4 .0400 .0500 .59 BANANA CHERRY 4 1 .0500 .0005 BANANA TANGERINE 3 1 .0600 .0020 STRAWBERRY APPLE 4 4 .0600 .0500 STRAWBERRY CHERRY .05 2 3 .0600 .0020 .30 STRAWBERRY TANGERINE 3 .0300 Furniture BUREAU DRESSER .14 5 1 .0100 .0040 .59 MIRROR DRESSER 1 .0009 CABINET DRESSER 4 1 .0300 .0200 Insect ROACH SPIDER .01 2 2 .0500 .0300 .02 ROACH BEETLE 2 1 .2100 .0200 ROACH GRASSHOPPER 3 1 .0700 .0004 CATERPILLAR BEETLE 2 1 .1300 .0100 CATERPILLAR GRASSHOPPER 3 .0400 CATERPILLAR SPIDER 4 1 .0400 .0080 HORNET GRASSHOPPER 2 .0100 HORNET BEETLE 1 1 .0200 .0020 HORNET SPIDER 1 1 .0050 .0020 Metal ALUMINUM SILVER .02 4 .0200 .08 NICKEL SILVER .05 2 1 .0050 .0020 .21 Profession EXECUTIVE PRESIDENT .13 1 2 .0004 .0030 .00 Sport VOLLEYBALL FOOTBALL .04 9 8 .0300 .0600 .46 VOLLEYBALL TENNIS .01 4 4 .0200 .0200 .17 HOCKEY FOOTBALL 4 3 .0100 .0300 HOCKEY TENNIS 2 2 .0060 .0090 WRESTLING FOOTBALL 3 3 .0700 .0300 WRESTLING TENNIS 3 3 .0700 .0100

PAGE 159

149 Appendix C (Continued) Between-Set Associations from Triad B to Triad A Experimental 2 List: Inter-set (strong) Condition Inter-set (strong) Condition Triad B Triad A Between Set Normative Associations Category Members of Triad A Members of Triad B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Bird ROBIN PIGEON .02 1 1 .4100 .0100 .73 ROBIN SPARROW .05 2 2 .4700 .0100 .84 Drink SCOTCH WHISKEY .04 7 2 .0200 .0030 .58 SCOTCH CHAMPAGNE 4 1 .0200 .0008 BOURBON WHISKEY .05 7 1 .0400 .0010 .68 BOURBON CHAMPAGNE 4 1 .0300 .0010 ALE CHAMPAGNE 3 1 .0200 .0003 ALE WHISKEY 3 1 .0300 .0005 Fruit ORANGE APPLE .17 3 3 .0400 .0300 .69 ORANGE TANGERINE .73 3 2 .0200 .0200 .90 BANANA APPLE .02 2 1 .0400 .0100 .47 BANANA CHERRY 4 3 .0500 .0100 BANANA TANGERINE 3 2 .0600 .0070 STRAWBERRY APPLE 4 1 .0600 .0002 STRAWBERRY CHERRY .02 2 1 .0600 .0000 .34 STRAWBERRY TANGERINE 3 .0300 Furniture BUREAU DRESSER .02 5 1 .0100 .0100 .02 MIRROR DRESSER .03 .08 CABINET DRESSER .02 4 1 .0300 .0100 .14 Insect ROACH SPIDER 2 2 .0500 .0200 ROACH BEETLE 2 1 .2100 .0900 ROACH GRASSHOPPER 3 1 .0700 .0200 CATERPILLAR BEETLE 2 .1300 CATERPILLAR GRASSHOPPER 3 .0400 CATERPILLAR SPIDER 4 .0400 HORNET GRASSHOPPER 2 .0100 HORNET BEETLE 1 .0200 HORNET SPIDER 1 .0050 Metal ALUMINUM SILVER NICKEL SILVER 2 1 .0040 .0007 Profession EXECUTIVE PRESIDENT 1 .0004 Sport VOLLEYBALL FOOTBALL 9 .0300 VOLLEYBALL TENNIS 4 .0200 HOCKEY FOOTBALL 4 1 0100 .0010 HOCKEY TENNIS 2 .0060 WRESTLING FOOTBALL 3 .0100 WRESTLING TENNIS 3 .0070

PAGE 160

150 Appendix C (Continued) Between-Set Associations from Triad A to Triad B Experimental 2 List: Intra-set (weak) Condition Intra-set (weak) Condition Triad A Triad B Between Set Normative Associations Category Members of Triad A Members of Triad B A B (Fsg) A B # of Indirect A B Indirect Strength B recovery (Eq3) # Olaps # Meds Osg Msg Bird Drink SCOTCH ALE 2 1 .0056 .0002 SCOTCH CHAMPAGNE 4 1 .0200 .0002 BOURBON ALE 2 .0105 BOURBON CHAMPAGNE 4 1 .0300 .0003 WHISKEY ALE 3 1 .0300 .0005 WHISKEY CHAMPAGNE 7 1 .0227 .0003 Fruit ORANGE BANANA 3 .0600 ORANGE STRAWBERRY 3 .0460 ORANGE CHERRY 4 .0477 TANGERINE BANANA 3 2 .0564 .0070 TANGERINE STRAWBERRY 3 .0253 TANGERINE CHERRY 2 .0203 APPLE BANANA .02 2 1 .0398 .0090 .47 APPLE STRAWBERRY 4 1 .0564 .0002 APPLE CHERRY 4 1 .0851 .0056 Furniture BUREAU CABINET 2 2 .0051 .0030 DRESSER CABINET 4 1 .0338 .0086 Insect ROACH CATERPILLAR 2 .0741 ROACH BEETLE 2 1 .2086 .0165 ROACH GRASSHOPPER 3 .0668 SPIDER CATERPILLAR 4 .0363 SPIDER BEETLE 2 1 .0871 .0065 SPIDER GRASSHOPPER 3 .0428 HORNET CATERPILLAR 1 .0084 HORNET BEETLE 1 1 .0248 .0020 HORNET GRASSHOPPER 2 .0114 Metal NICKEL ALUMINUM 1 .0010 Profession PRESIDENT FARMER 1 .0001 PRESIDENT TEACHER 1 .0011 Sport VOLLEYBALL HOCKEY 2 .0141 VOLLEYBALL WRESTLING 3 .0128 FOOTBALL HOCKEY 4 1 .0114 .0012 FOOTBALL WRESTLING 3 .0107 TENNIS HOCKEY 2 .0059 TENNIS WRESTLING 3 .0065

PAGE 161

151 Appendix C (Continued) Between-Set Associations From Triad B to Triad A Experimental 2 List: Intra-set (weak) Condition Intra-set (weak) Condition Triad B Triad A Between Set Normative Associations Category Members of Triad A Members of Triad B B A (Fsg) B A # of Indirect B A Indirect Strength A Recovery (Eq3) # Olaps # Meds Osg Msg Bird Drink SCOTCH ALE 2 1 .0056 .0002 SCOTCH CHAMPAGNE 4 1 .0200 .0002 BOURBON ALE 2 .0105 BOURBON CHAMPAGNE 4 1 .0300 .0003 WHISKEY ALE 3 1 .0300 .0005 WHISKEY CHAMPAGNE 7 1 .0227 .0003 Fruit ORANGE BANANA .04 3 .47 ORANGE STRAWBERRY 3 .0460 ORANGE CHERRY 4 .0477 TANGERINE BANANA 3 2 .0564 .0070 TANGERINE STRAWBERRY 3 .0253 TANGERINE CHERRY 2 .0203 APPLE BANANA .15 2 1 .0398 .0090 .59 APPLE STRAWBERRY 4 1 .0564 .0002 APPLE CHERRY 4 1 .0851 .0056 Furniture BUREAU CABINET 2 2 .0051 .0030 DRESSER CABINET 4 1 .0338 .0086 Insect ROACH CATERPILLAR 2 .0741 ROACH BEETLE 2 1 .2086 .0165 ROACH GRASSHOPPER 3 .0668 SPIDER CATERPILLAR 4 .0363 SPIDER BEETLE 2 1 .0871 .0065 SPIDER GRASSHOPPER 3 .0428 HORNET CATERPILLAR 1 .0084 HORNET BEETLE 1 1 .0248 .0020 HORNET GRASSHOPPER 2 .0114 Metal NICKEL ALUMINUM 1 .0010 Profession PRESIDENT FARMER 1 .0001 PRESIDENT TEACHER 1 .0011 Sport VOLLEYBALL HOCKEY 2 .0141 VOLLEYBALL WRESTLING 3 .0128 FOOTBALL HOCKEY 4 1 .0114 .0012 FOOTBALL WRESTLING 3 .0107 TENNIS HOCKEY 2 .0059 TENNIS WRESTLING 3 .0065

PAGE 162

152 Appendix D: Distracter Phase and Post-Experiment Questionnaires Shipley Vocabulary Test from the Shipley Institute of Living Scale (Zachary, 1991) Vocabulary: In the test below, the first word in each line is printed in capital letters. Opposite are four other words. Draw a line under the one word which means the same thing, or most nearly the same thing, as the first word. A sample has been worked out for you. If you dont know, please guess. Be sure to underline the one word in each line that means the same thing as the first word. SAMPLE LARGE red big silent wet BEGIN HERE 1. TALK draw eat speak sleep 2. PERMIT allow sew cut drive 3. PARDON forgive pound divide tell 4. COUCH pin eraser sofa glass 5. REMEMBER swim recall number defy 6. TUMBLE drink dress fall think 7. HIDEOUS silvery tilted young dreadful 8. CORDIAL swift muddy leafy hearty 9. EVIDENT green obvious skeptical afraid 10. IMPOSTER conductor officer book pretender 11. MERIT deserve distrust fight separate 12. FASCINATE welcome fix stir enchant 13. INDICATE defy excite signify bicker 14. IGNORANT red sharp uninformed precise 15. FORTIFY submerge strengthen vent deaden 16. RENOWN length head fame loyalty 17. NARRATE yield buy associate tell 18. MASSIVE bright large speedy low 19. HILARITY laughter speed grace malice 20. SMIRCHED stolen pointed remade soiled 21. SQUANDER tease belittle cut waste 22. CAPTION drum ballast heading song 23. FACILITATE help turn strip bewilder 24. JOCOSE humorous paltry fervid plain 25. APPRISE reduce stew inform delight 26. RUE eat lament dominate cure 27. DENIZEN tile inhabitant town atom 28. DIVEST dispossess intrude rally pledge 29. AMULET charm orphan dinge pond 30. INEXORABLE untidy involatile rigid sparse 31. SERRATED dried notched armed blunt 32. LISSOME moldy loose supple convex 33. MOLLIFY mitigate direct pertain abuse 34. PLAGIARIZE appropriate intend revoke maintain 35. ORIFICE brush hole building lute 36. QUERULOUS maniacal curious devout complaining 37. PARIAH outcast tile lentil locker 38. ABET waken ensue incite placate 39. TEMERITY rashness timidity desire kindness 40. PRISTINE vain sound first level

PAGE 163

153 Appendix D (Continued) Morningness and Eveningness Questionnaire (Horne & Ostberg,1976) Instructions: 1. Please read each question very carefully before answering. 2. Answer ALL questions. 3. Answer questions in numerical order. 4. Each question should be answered independently of others. 5. Do NOT go back and check your answers. 6. All questions have a selection of answers. For each question place an X alongside ONE answer only. Some questions have a scale instead of a selection of answers. Place an X at the appropriate point along the scale. 7. Please answer each question as honestly as possible. Both your answers and the results will be kept IN STRICT CONFIDENCE. 8. Please feel free to make any comments in the section provided below each question. 1. Considering only your own feeling best rhythm, at what time would you get up if you were entirely free to plan your day? 2. Considering only your own feeling best rhythm, at what time would you go to bed if you were entirely free to plan your evening? 3.If there is a specific time at which you have to get up in the morning, to what extent are you dependent on being woken up by an alarm clock? _________ Not at all dependent _________ Slightly dependent _________ Fairly dependent _________ Very dependent 4.Assuming adequate environmental conditions, how easy do you find getting up in the mornings? __________ Not at all easy __________ Not very easy __________ Fairly easy __________ Very easy 5.How alert do you feel during the first half hour after having woken in the mornings? ___________ Not at all alert ___________ Not very alert ___________ Fairly alert ___________ Very alert 6.How is you appetite during the first half hour after having woken in the mornings? ___________ Very poor ___________ Fairly poor ___________ Fairly good ___________ Very good

PAGE 164

154 Appendix D (Continued) Morningness and Eveningness Questionnaire (continued) 7.During the first half hour after having woken in the morning, how tired do you feel? ___________ Very tired ___________ Fairly tired ___________ Fairly refreshed ___________ Very refreshed 8.When you have no commitments the next day, at what time do you go to bed compared to your usual bedtime? ___________ Seldom or never later ___________ Less than one hour later ___________ One to two hours later ___________ More than two hours later 9.You have decided to engage in some physical exercise. A friend suggests that you do this one hour twice a week and the best time for him/her is between 7 and 8 A.M. Bearing in mind nothing else but your own feeling best rhythm, how do you think you would perform? ____________ Would be on good form ____________ Would be on reasonable form ____________ Would find it difficult ____________ Would find it very difficult 10.At what time in the evening do you feel tired and, as a result, in need of sleep? 11.You wish to be at peak performance for a test which you know is going to be mentally exhausting and lasting for two hours. You are entirely free to plan your day and considering only your own feeling best rhythm, which ONE of the four testing times would you choose? ___________ 8 to 10 A.M. ___________ 11 A.M. to 1 P.M. ___________ 3 to 5 P.M. ___________ 7 to 9 P.M. 12.If you went to bed at 11 P.M., at what level of tiredness would you be? ____________ Not at all tired ____________ A little tired ____________ Fairly tired ____________ Very tired 13.For some reason you have gone to bed several hours later than usual, but there is no need to get up at any particular time the next morning. Which ONE of the following events are you most likely to experience? _________Will wake up at usual time and will NOT fall asleep _________Will wake up at usual time and will doze thereafter _________Will wake up at usual time but still fall asleep again _________Will NOT wake up until later than usual

PAGE 165

155 Appendix D (Continued) Morningness and Eveningness Questionnaire (continued) 14.One night you have to remain awake between 4 and 6 A.M. in order to carry out a night watch. You have no commitments the next day. Which ONE of the following alternatives will suit you best? ____________ Would NOT go to bed until watch was over ____________ Would take a nap before and sleep after ____________ Would take a good sleep before and nap after ____________ Would take ALL sleep before watch 15.You have to do two hours of hard physical work. You are entirely free to plan your day and considering only your own feeling best rhythm, which ONE of the following times would you choose? ____________ 8 to 10 A.M. ____________ 11 A.M. to 1 P.M. ____________ 3 to 5 P.M. ____________ 7 to 9 P.M. 16.You have decided to engage in hard physical exercise.A friend suggests that you do this one hour twice a week and the best time for him/her is between 10 and 11 P.M. Bearing in mind nothing else but your own feeling best rhythm, how well do you think you would perform? ____________ Would be on good form ____________ Would be on reasonable form ____________ Would find it difficult ____________ Would find it very difficult 17.Suppose that you can choose your own work hours. Assume that you work a FIVE hour day (including Breaks) and that your job was interesting and paid by results. Which FIVE CONSECUTIVE HOURS would you select? (mark them on the scale below) 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 MIDNIGHT NOON MIDNIGHT 18.At what time of day do you th ink that you reach your feeling best peak? (Please choose one hour only) 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 MIDNIGHT NOON MIDNIGHT 19.One hears about morning and evening types of people. Which ONE of these types do you consider yourself to be? ___________ Definitely a morning type ___________ Rather more a morning type than an evening type ___________ Rather more an evening type than a morning type ___________ Definitely an evening type

PAGE 166

156 Appendix D (Continued) The Cognitive Failures Questionnaire (Broadbe nt, Cooper, FitzGerald & Parkes, 1982) Instructions: The following questions are about minor mistakes which everyone makes from time to time, but some of which happen more often than others. We want to know how often these things have happened to your in the past 6 months. Pleas e circle the appropriate number. Very often Quite often Occasionally Very rarely Never 1. Do you read something and find you havent been thinking about it and must read it again? 4 3 2 1 0 2. Do you find you forget why you went from one part of the house to the other? 4 3 2 1 0 3. Do you fail to notice signposts on the road? 4 3 2 1 0 4. Do you find you confuse right and left when giving directions? 4 3 2 1 0 5. Do you bump into people? 4 3 2 1 0 6. Do you find you forget whether youve turned off a light or a fire or locked the door? 4 3 2 1 0 7. Do you fail to listen to peoples names when you are meeting them? 4 3 2 1 0 8. Do you say something and realize afterwards that it might be taken as insulting? 4 3 2 1 0 9. Do you fail to hear people speaking to you when you are doing something else? 4 3 2 1 0 10. Do you lost your temper and regret it? 4 3 2 1 0 11. Do you leave important letters unanswered for days? 4 3 2 1 0 12. Do you find you forget which way to turn on a road you know well but rarely use? 4 3 2 1 0 13. Do you fail to see what you want in a supermarket (although its there)? 4 3 2 1 0 14. Do you find yourself suddenly wondering whether youve used a word correctly? 4 3 2 1 0

PAGE 167

157 Appendix D (Continued) The Cognitive Failures Qu estionnaire (Continued) Very often Quite often Occasionally Very rarely Never 15. Do you have trouble making up your mind? 4 3 2 1 0 16. Do you find you forget appointments? 4 3 2 1 0 17. Do you forget where you put something like a newspaper or a book? 4 3 2 1 0 18. Do you find you accidentally throw away the thing you wa nt and keep what you meant to throw away as in the example of throwing away the matchbox and putting the used match in your pocket? 4 3 2 1 0 19. Do you daydream when you ought to be listening to something? 4 3 2 1 0 20. Do you find you forget peoples names? 4 3 2 1 0 21. Do you start doing one thing at home and get distracted into doing something else (unintentionally)? 4 3 2 1 0 22. Do you find you cant quite remember something although its on the tip of your tongue? 4 3 2 1 0 23. Do you find you forget what you came to the shops to buy? 4 3 2 1 0 24. Do you drop things? 4 3 2 1 0 25. Do you find you cant think of anything to say? 4 3 2 1 0

PAGE 168

158 Appendix D (Continued) Post-Experiment Integration Questionnaire Instructions: During the initial study phase of the experiment, when you encountered a word pair, how often, if ever, did you intentionally think back to previously seen category members? For example, if you first saw City Atlanta, and then saw City Orlando, did you intentionally rehearse Atlanta and Orlando (and other Cities you saw) together? It is important that you be accurate in describing how often and for which categories you did this. Please circle the number on the scal e corresponding to the proportion of the time you thought back to previously seen category members and intentionally rehearsed them together with the member you were studying. 1 being none of the time, and 5 being all of the time. Note: please try to use the whole scale (1-5). However, it is O.K. to say that you never rehearsed or that you always rehearsed. None of the time Some of the time All of the time City 1 2 3 4 5 River 1 2 3 4 5 Animal 1 2 3 4 5 Profession 1 2 3 4 5 Appliance 1 2 3 4 5 Weapon 1 2 3 4 5

PAGE 169

About the Author Leilani B. Goodmon is a native of Florid a. She received her B.A. from Stetson University in DeLand, Florida, where she gr aduated magna cum laude. As a graduate student in the Department of Psychology at the University of South Florida, she majored in cognitive science with an emphasis in human memory. Sh e enjoys outdoor activities including horseback riding and fishing. She is also an avid orchid enthusiast. She is married and resides in Lakeland, Florida.