|USFDC Home | Search all Groups | Natural Hazards Center Collection||| RSS|
This item is only available as the following downloads:
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 cam 2200313Ia 4500
controlfield tag 001 001985132
008 911121s1991 cou b s000 0 eng d
datafield ind1 8 ind2 024
subfield code a F57-00054
Norris, Fran H.
Reliability of delayed self-reports in disaster research /
by Fran H. Norris and Krzysztof Kaniasty.
Boulder, Colo. :
b Natural Hazards Research and Applications Information Center, University of Colorado,
21 p. ;
Quick response research report ;
"Institute of Behavioral Science #6."
Includes bibliographical references (p. 20-21).
Also issued online as part of a joint project with the Louis de la Parte Florida Mental Health Institute (FMHI) Research Librarys disaster mental health initiative.
x Social aspects.
z South Carolina
Hurricane Hugo, 1989.
University of Colorado, Boulder.
Natural Hazards Research and Applications Information Center.
" HAZARDHOUSE COpy(}f'RallAB' l\TYOf 'bEl"'i)),SE.lf Rf'ORTSttJH.tJ01tIl\SKRZ.')IsEo "STy of rVlstitLJe orJ Behavioro.lSc:..ieY1ceJ\ia.:r.a.r s Ke.se.o.V'ch0..VlclA-p"t' \ ic.a.4:io VIS IV\for"Wla.L011Ce.Y\.te...rl\aftOl1#4-3 /99/.
NaturalHazardsResearchandApplicationsInformationCenterCampusBox482UniversityofColoradoBoulder,Colorado80309-0482RELIABILITYOF DELAYEDSELF-REPORTS IN DISASTERRESEARCHByFranH.NorrisGeorgiastateUniversityandKrzysztofKaniastyIndianaUniversityofPennsylvaniaRunningHead:ReliabilityofDelayedSelf-Reports QUICK RESPONSE RESEARCH REPORT #43 1991This publication is partofthe Natural Research&Applications Information ongomg Quick Response Research Report Senes. http://wWIN.colorado.edu/hazards TheviewsexpressedinthisreportarethoseoftheauthorsandnotnecessarilythoseoftheNaturalHazardsCenterortheUniversityofColorado.InstituteofBehavioralScience#6 (303)492.6818TELEFAX:(303)492-6924
ReliabilityofDelayed Self-ReportsinDisaster Research FranH.Norris Georgia State University and Krzysztof Kaniasty Indiana University of Pennsylvania Running Head: Reliability of Delayed Self-Reports This researchwasfundedinpartbya grant from the UniversityofColorado's Natural Hazards Research and Information Center (GrantNo.BP0045468). Aleksandra Kaniasty and Ron Aviram also made important contributionstothis research. Address all correspondence to FranH.Norris, DepartmentofPsychology, Georgia State University, University Plaza, Atlanta, GA, 30303.
ReliabilityofDelayed Self-Reports 2 Abstract In studiesoftraumatic stress, researchers often find themselves asking questions aboutanevent and its aftermath long after the crisis has passed. The purposeofthis study was to assess the reliabilityofthese delayed self-reports. In January, 1990,65residentsofCharleston, SC were interviewedbytelephone about their experiences following Hurricane Hugo, which had devastated the areaonSeptember22,1989. The interview included assessmentsofdisaster-related losses, preparedness, social support received from others, and social support provided to others.InOctober, 1990,53ofthese persons (82%ofthe original sample) were reinterviewed and asked the exact same questions.Forreportsoflosses and preparedness, accuracyofthe later reports was excellent. Both the sample and individuals showed remarkable stability over time.Formeasuresofsocial support, therewasa sample tendency to recall more social supportastime passed,butindividuals generally retained their same rank order. Thus, these reports were also reliable.
ReliabilityofDelayed Self-Reports3ReliabilityofDelayed Self-Reports in Disaster ResearchThemethodological difficulties in disaster research are many. Funding mechanisms are slow, and researchers often find themselves asking questions about the event and its aftermath long after the disaster has passed.Thepurposeofthis studywasto assess the reliabilityofthese delayed self-reports.Ourown interest in this topic stemmed from experiences in proposing a oneand two year follow-up study to NIMH concerning the mental health impactofHurricane Hugo.Oneclear methodological issuewasthe extent to which we could accurately assess the losses and social support exchanges that occurred around the timeofthe event. Very little data appeared to support a claim that disaster victims remember their experiences accurately over time. In fact, some evidence (Hopwood&Guidotti, 1988) suggested that systematic biases couid well be present. These investigators compared symptoms reported at the timeofthe event with symptoms recollected 6 months later among 22 victimsoftoxic exposure. There was a clear tendency for victims to recall more symptoms than they had reported initially. Although these findings raise caution for delayed accountsoftraumatic experiences, we should also not overgeneralize from these respondents' abilities to recollect and distinguish among symptomsassimilarasdizziness, lightheadedness, and eye discomfort. In contrast, thereisindirect evidence in research on autobiographical memory (Rubin, 1986) that delayed self-reportsmaybe quite accurate under certain conditions that disaster studies inadvertently mimic. Disasters conformtothe very typesofevents regardedas"markers"or"landmarks" in a person's temporal frameofreference that help to organize an autobiographical memory search (see Robinson, 1986; Whitten&Leonard, 1981). Studies reportedbyLoftus&Marburger (1983)inthe article entitled, "Since the eruptionofMt. St. Helens, has anyone beaten you up?" constitute an excellent exampleofthe useof"public landmarks" (e.g., eruptionofa
ReliabilityofDelayed Self-Reports4volcano, New Year'sday)to improve the accuracyoftemporal judgments in reporting past experiences. Nonetheless, thereisno direct empirical evidence to support the clpim that disaster victims remember their experiences accurately over time. The present study assessed the test-retest reliabilityofself-reported measures of disasterloss,preparedness, received social support, and provided social using a 9-month interval between tests. Because the support measures tapped experiences over a 3-month interval (boundedbyHugo and New Year's Eve) the first interview took place in January, 1990; the second interview took place in October,1990.The respondents (two-wave Xl of 53)alllived on the Charleston peninsulaatthe time Hugo struck, but their disaster-related losses varied in nature and severity. This variation should have strengthened the study's ability to detect inaccuracies in subsequent, delayed reports. Methods Sampling and Interviewing ProceduresThesamplewasselected to represent a reasonable cross-sectionofresidentsofthe Charleston, SC peninsula.Sixweeks after Hurricane Hugo struck, we toured the peninsula and chose 3 census tracts for further investigation. Criteria for selecting the 3 tracts were that damage should stillbevisible and that they should be occupiedbydemographically different subpopulations. One tract (5)wasprimarily occupiedbywhitesofmiddletoupper-middle socioeconomic status (as judgedbythe qualityofhousing). The second tract (8)washeterogeneous, but predominantly poor. The third tract (17)wasoccupied primarilybyblacksofmiddle class status. Using the Charleston cross-reference directory, each tractwascompletely enumerated. A proportionofnameswasthen selected (e.g., every 16thinTract8)sothat the end result was a sampling frame consisting of approximately equal numbersofresidents from eachtract
ReliabilityofDelayed Self-Reports5Itisdifficult to assess Wave 1 response rates becauseofconsiderable inaccuracyinthe Charleston directory. Contactwasattempted with 160 households.Ofthese numbers,27either turnedoutbenonworkingordid not thenorearlier belong to the name with which itwaslisted. No contact was made within 5 attempts for 36ofthe remaining133persons.Ofthe97households successfully contacted, 4ofthedesignated respondents had died, 28 refused, and65were interviewed. Thus, the response ratewas70%ofthose contacted (and living)butonly41%ofthe original listing. All respondents but 5 had experienced sometypeofdisaster loss.Theresponse rates among those contacted varied across the three tracts from83%(Tract5)to71%(Tract8)to61%(Tract17).Because the response rate was substantially higher in the predominantly middle to upper-middle SES tract (5), our samplemayoverrepresent higher SES persons.Thetotal numberofinterviews conducted was65.Each interview, which averaged17minutes, was conductedbytelephone andbythe same interviewer.Thecharacteristicsofthe sample are presented in TableI. As shown, the samplewasdiverseinrace(37%black), sex (48% female), marital status(45%married), and age (one third each20-34, 35-49,50+).Halfofthesample owned their homes, and half rented.Thesample, overall, was well educated, with a meanof14years. Thisfinding:also suggests that higher SES persons maybeoverrepresented.Ofthese65persons,53(82%)were interviewed again nine months later. As also showninTableI,characteristicsofthe two-wave sample were quite comparable to thoseofthe original sample.Forthis reason, itisnot likely that attrition has influenced our findings. After the first15interviews at Time1,it became apparent thatwehad to revise the social support questions to make it clear that the support receivedorprovided did not have to be
ReliabilityofDelayed Self-Reports6Table1.Sample Characteristics Original Sample Two-Wave Samplen%n%RaceWhite41633464 Black 24 37 19 36 Sex Male 34 522853Female31482547 Marital Status Never Married2335 1834Married 29 45 2343SeparatedlDivorced 8 12 713Widowed 5 8 5 9 Age(M=44,SD=16)(M=45,SD=16) 20-34 20 31 163135-492133 17 33 50-641320112165+10 16 815Education(M=14,SD=3)(M=14,SD=3) Less than121015 81512 Years 8 12 61113-15 Years914 81516+38 59 31 59 HomeownershipRent32 50 24 46Own32 502854
ReliabilityofDelayed Self-Reports7''because''ofHurricane Hugo but could have been received or provided for any reason. The interviewer noticed that many respondents were confused; some were orienting their answers to concern Hugo only, while others were taking a broader perspective. Therefore, foranyanalyses using the social support scales, the sample sizeis50,which excludes the first15persons interviewed.Ofthese50persons,44(88%) were reinterviewed at Time2.Measures Disaster Loss. Hugo-related losses were assessedbya 14-item battery. The first11items concerned specific typesoflosses, subsequently grouped into 5 variables (each coded 1 if that losswaspresent, 0 otherwise).Injury reflected the presenceofan injury to either the respondent or to any other household member. Structural damage referred to damage done to the outsideofthe respondent's dwelling or to any other building on the property. Property damage encompassed damage to trees or gardens, rugs, furniture, or appliances, or a car, truck, or boat. Lossofpersonal belongings encompassed lossesofclothingorbooks or thingsofsentimental value suchasphotographs or keepsakes. Loss of incomewasmeasuredbya single item, "Did you lose any income due to disruptionofemployment or closing a business?" The final item asked whether they suffered some other typeofloss, primarilyasa check on our own completeness. This question elicited18affirmative responses at Time 1 (28%ofthe sample) and13affirmative responses at Time 2 (25%ofthe sample). Most frequently mentioned were losses at placesofwork. Others included rental properties, summer homes, time outofschool, food, and inconveniences. The remaining three itemsinthelossbattery were includedassummary measures. Perceived total impactwasbased on answers to the question, "Whichofthe following statements best describes the total impactofHurricane Hugo on your own property and belongings?"Itwasanswered on a 5-point scale where 4=enormous damage, 3=much damage, 2=some damage, and 1=just a little damage; respondents with no losses received scores ofo.Impact compared
ReliabilityofDelayed Self-Reports8wasbased on answerstothe question, "Compared to other residentsofCharleston, were you hit harder than others(+1),-affected about the sameasothers (0), or affected less than others ( 1)?"Loss in dollarswasthe respondent's best estimate, including insuredaswellasuninsured losses. A fourth summary measure, scope,wasthe numberofaffirmative responses to the eleven items described. previously. Preparedness. There weretwomeasuresofpreparedness. Evacuatedwasa 3-point scale where 0=did not evacuate, 1=left home, and 2=left Charleston. Preparednesswasthe sumof3 items assessing whether the respondents had taped or boarded the windowsoftheir homes, had taken any other steps to prepare, or viewed themselvesasmore prepared than others in Charleston. Each affirmative response contributed one point to the scale score. Received social support. A 12-item measure of received supportwasbased on the InventoryofSocially Supportive Behaviors (ISSB), a scale that attempts to represent the "broad diversityoffunctions that characterize informal supportsystems"(Barrera, Sandler,&Ramsay, 1981).Onthe basis of a previous factor analysis (Barrera&Ainlay, 1983), four items each were selected from threeISSBsubscales. The three subscales were guidance (receiptofadviceorinformation); tangible support (concrete help suchasmoney or transportation); and nondirective support (comfort or expressionofcaring). Questions were worded to emphasize that the help received could have been foranyreason, not just Hurricane Hugo. All items had the same response options, 0=never, 1=once or twice, 2=afewtimes, and 3=many times. Therefore, each 4-item subscale had the same potential range (0-12) which heightens the descriptive valueofthe data. Respondents were asked to think about the support they received in termsoftwotime intervals. The first (A) covered the first4-5weeks following the hurricane, or between Hugo and Halloween. The second (B)was"further down the road," between Halloween and New Year's Eve.
ReliabilityofDelayed Self-Reports9Thus, for each measure, there was an A scale and aB scale referring to the two time intervals. Means for the received support and provided support scales and subscales were consistently higher for theAinterval than for theBinterval, which may indicate that Hugo-related helping behaviors were concentrated in the first measurement interval. Nevertheless, theAscales andBscales were highly correlated; for the total scale, r =.74, R <.001.Provided social support. Measuresofprovided support were created tobedirectly parallel to the measuresofreceived support. Rolesofrecipient and provider were reversed; for example, "Did anyone giveorloan you some money?" became "Did you giveorloan anyone some money?" Thus, for provided support, there were also three subscales, guidance, tangible support, and nondirective support scored for two time intervals, A andB. As for received support,theA scales had higher means than theBscales but were highly correlated with them; for the total scale, r =.75, R <.001.In halfofthe questionnaires, the received support items preceded the providedsupportitems; in half, the reverse was true. Results Loss and Preparedness TableIIpresents sample frequenciesonthemeasuresofloss and preparedness. Time1frequencies are given for the original Time1sample and for the subsetofrespondents who completedbothinte,rviews. Time 2 frequencies,ofcourse, are given only forthetwo-wave sample. Intheoriginal sample, all but8%ofthose interviewed experienced some typeofloss related to Hurricane Hugo (4%ofthe two-wave sample). Structural damagewasthe most common typeofloss, and physical injurywasthe least common (reportedbyonly11-12%).Perceived total impact was normally distnbuted across the five-point scale. Most respondents felt that they experienced fewer losses than others in their community. Withoneexception, losses in dollars ranged from 0 to$75,000.
ReliabilityofDelayed Self-Reports10Table II also shows that the majorityofrespondents had evacuated, at least from their own homes, prior to the hurricane's arrival. Nonetheless, the fact that a third remained in their homes despite the effortsofcity officialsissomewhat striking. There was good dispersiononthe preparedness scale, with mostofthe sample receiving scoresinthe 1-2 range. Methodologically, itisimportant to note the comparabilityofthe losses reportedbythe original and two-wave samples. Means and frequency distributions were virtually identical. Within the two-wave sample, the distributionsofTime 2 responses were strikingly similar to the distributionsofTime 1 responses.Anexception to this general rule was a substantial mean difference for dollars lost.Themedian loss in dollars, however,wasconstant across sample and time. The change in means was primarily due toone"outlier" who reported lossesof$200,000 at Time 1 (4 standard deviations above the mean) but only $50,000 at Time2.This seemed too large a difference to attribute to unreliability (especially since at Time2,this person had stillbeenunable to return to her original residence because it was being rebuilt). Given the unfortunate but very reasonable possibility that her answer was improperly recorded at Time2,we excluded her from the reliability analyses presented in Tableill.With the possible but unlikely exceptionofdollars lost, the consistency in the sample dataISnotable. This finding indicates that no systematic biases were present with regard to remembering fewerorgreater losses over time. Tableillpresents data relevant to assessing the accuracyofthe delayed (Time 2) self reportsofdisaster-related loss and preparedness. Here we are concerned with intra personalasopposed to sample consistency over time. Different measuresofassociation are given. Firstisthe
ReliabilityofDelayed Self-Reports11Table II. Sample Frequencies: Loss and Preparedness Measures Original Sample Time 1(%)Two-Wave Sample Two-Wave Sample Time 1(%)Time 2(%)PercentofSample with: Injury in Household1211 11Structural Damage868993Property Damage 66 68 72 LossofPersonal Belongings43 4542 LossofIncome 32 32 28 Scope M= 4.4 M= 4.5 M=4.3SD = 2.2 SD =2.1SD = 2.0None8 4 2 Low (1-3 "yeses")2328 34Moderate(4-6) 49 4953High (7-10) 201911Perceived Total Impact M= 2.0 M=2.1M= 2.0 SD = 1.2SD=1.1SD =1.1None(0) 8 4 4 Little 32 3433Some 25 2631Much23 2320 Enormous (4)121312Impact Compared M= -0.6 M= -0.6 M= -0.6 SD = 0.6 SD = 0.6 SD = 0.5 Less than others (-1) 696561Sameasothers (0) 25 29 37Morethan others(+1)6 6 2 Loss in Dollars M= $17,209 M =$16,729 M= $13,804 SD = 31,502 SD = 33,228 SD = 18,895 Median = 5,000 Median = 5,000 Median = 5,000 011710100-1,0001920 20 1,100-9,000334028 10,000-25,000 201528 36,000-75,0001615 15200,000 2 2 0 table continues
ReliabilityofDelayed Self-Reports12Table II Continued Evacuated No 34 34 34 Left home 2023 23Left Charleston4543 43Preparedness M=1.7 M=1.7 M=1.8 SD=0.9 SD=0.8SD=0.9None(0) 8 811Low (1)31 3121Moderate (2)43 4547 High (3) 18 1621
ReliabilityofDelayed Self-Reports13simple percentofagreement across Time 1 and Time 2 responses. For example, 89%ofrespondents gave the same answers to questions concerning household injuries each time. Secondisthe proportion of affirmative responses at Time 1 that became negative at Time 2 expressed in actual n's. For example,ofthe 6 persons who reported injuries at Time1,3 did not at Time2.Nextisthe proportion of negative responses at Time 1 that became affirmative at Time2.For this same item, 3 of the47Time 1 "no's" said"yes"at Time2.Finally, two statistical measuresofassociation aregiven.The firstisCohen's kappa (Cohen, 1960). Originally developedasa measureofinterrater agreement, it has since been appliedasa more general measureofassociation when marginals (probabilities of a given response) are unequal (Fleiss,1981;Hopwood&Guidotti, 1988). Values greater than0.4suggest good agreement.Thismeasureismost useful for comparing one item to another that has a different probabilityofaffirmative response. Continuing with the household injury example, it has a lower kappa than other items despite its high percentageofagreement because the probability of this loss occurringwasquitelow.Thus chance alone would predict high agreement across time. The second statistical measureofassociationwasthe product-moment correlation; thiswasphi for dichotomous measures, Cramer's V for categorical measures, and Pearson r for items with fourofmore categories.Thedata presented in Table III provide evidenceofexcellent reliability. Percentagesofagreement ranged from 82% (impact compared)to93% (structural damage). All kappa coefficients were greaterthan.4.Given the restricted range of most variables, the product-moment correlations were also substantiaL The test-retest correlation (.82) for scopewasparticularly notable. This finding indicates that the battery tendedtoelicit a consistent numberof"yes"responses even when different items were endorsed.
ReliabilityofDelayed Self-Reports14Table III. ReliabilityofLoss and Preparedness Measures (Two-Wave Sample). ConcordanceofResponseonNominal Measures%Same12"no"12 "yes" Phiof ofT1-12T1"yes"Tl"no"korV Injury in Household 88.7 3/6 3/47 .44 .44*** Structural Damage 92.5 1/47 3/6 .56 .57*** Property Damage 92.4 1136 3/17 .82 .82*** LossofBelongings 84.9 51243129 .69 .70*** LossofIncome 88.7 4/17 2136 .73 .73*** Impact Compared& 82.0 6134 3/16 .61 .78*** Evacuatedb86.8 51302(23 .80.78*** Time 1Time 2 CorrelationsonContinuous Measures Scope Perceived Total Impact Loss in Dollars PreparednessV.63*** .67*** .84*** .43***r.82*** .60*** .73*** .60***aBecauseofthe low frequency in the "more" category, this variable was recoded into dichotomous form.Thesecond column represents the proportionofthe"less"category changing their responses and the third column represents the proportionofthe "same or more" category changing their response.bThis variable had 3 categories.Thesecond column splits into 2/18 (proportionofthe "stayed home" category changing their response) and 3/12 (proportionofthe "left home" category changing their response).Thethird columnisthe proportionofthe "left Charleston" category changing to another category. Cramer's Visgiven insteadofPhi.
ReliabilityofDelayed Self-Reports is Social Support Measures Table IV presents descriptive statistics and reliability coefficients the social support measures (A scales only). Here, the original sample consists onlyofthe 50 respondents interviewed after the scales were revised (see Sampling and Interviewing Procedures).Thetable presents the Time 1 meansofthe original sample and both the Time 1 and Time 2 meansofthe two-wave sample.Asfor the loss measures, the Time 1 responsesofthe original and two-wave samples were quite comparable, thereby reducing the threat that attrition could have influenced these findings. Within the two-wave sample, however, systematic differences between Time 1 and Time 2 responses were evident.AtTime2,respondents tended to remember having receivedorprovided greater social support after the crisis than they had reported at the earlier interview. For 3ofthese measures (all received support scales), the Time 1Time 2 difference achieved statistical significance: for the total scale, (43)=3.30, Q <.002; for guidance, (38)=4.63, Q <.001; for nondirective support, (41)=2.68, Q <.02. Thus, unlike recollectionsoflosses and preparedness, there was some systematic bias in the delayed reportsofreceived social support. For assessing the test-retest reliabilityofthese measures, the internal consistencyofeach scaleisofparticular importance. Thatis,any unreliability evidenced at a given point in timewillserve to attenuate reliability that can be demonstrated over time (Pedhazur, 1982). For this reason, Table IV also presents the Time 1 alphas for each scale. With the exceptionofreceived tangible support, all scales demonstrated good to excellent internal consistency (alphasof.69 to .80)Thefour items accessing tangible support, adapted from Barreraetal.'s (1981) general scaleofreceived support,maynot have been congruent with the particular needsofdisaster victims, and thus elicited responses that were not internally consistent (alpha=.29).
ReliabilityofDelayed Self-Reports16Table IV. Descriptive Statistics and Reliability Coefficients for Social Support Measures. Original Two-Wave Sample Sample Time 1 Time 1 Time 2 Time 1T-RM SD MSDMSDAlpha r Received Support15.16.314.8 6.317.16.8 .76 .75 Guidance 4.4 3.1 4.2 3.0 5.2 2.9 .69.84 Tangible 18.104.22.168 2.1 3.3 2.5 .29.60Nondirective 6.8 3.8 7.6 3.4 8.5 3.1 .80 .70 Provided Support 17.9 7.5 17.5 7.5 18.6 6.2 .84 .62 Guidance 5.1 3.3 5.2 3.4 5.7 2.6 .82 .37 Tangible 4.4 3.1 4.2 3.1 4.3 2.5 .68 .82 Nondirective 8.4 3.3 8.1 3.3 8.7 2.9 .59 .66 All coefficients significant at 12 <.001.
ReliabilityofDelayed Self-Reports17Given the long interval between tests (9 months), the test-retest (T-R) correlations were quite high. With one exception, the correlations were in the .60 to.85range.TheT-R correlationsofthe received support measures (.60 to .84) were somewhat higher than theT-Rcorrelationsofthe provided support measures (.37 to .75).The.37correlation between Time 1 and Time 2 measuresofprovided guidancewastroubling. Muchofthis unreliability couldbeattnbutedto two respondents: the person who received the highest score at Time 1 received a below average at Time2,whereas the person who received the highest score at Time 2 had a below average score at Time1.Without these two persons, the T-R coefficient increased to.51--better, but still indicativeofsome reliability problems for this particular subscale. Otherwise, however, the correlations wereofadequate strength to evidence reliability over time. Thus, although sample means tended to be higher at Time2,individuals within the sample generally maintained their same rank order at each timepoint. Such biases therefore should have minimal impactonthe correlational analyses characteristicoftraumatic stress research. DiscussionItisnot surprising that people remember an eventasnewsworthyaswasHurricane Hugo. Autobiographical memory research (e.g., Robinson, 1986) would even suggest that,inthe years to come, Hugowillbecome a major "landmark" in these persons'lives.In certain partsofthe United States, for example,itisnot uncommon to hear older people describe their youths in termsofbefore and after the '37 flood. Whatismore notable hereisthe striking accuracy with which people remembered the detailsofthe event; generally, our respondents reported the same losses at Time 2asthey had at Time1.In life-events research, delayed accounts have long been viewedasproblematic because memoriesofeventsmayfade (Cohen, 1988; Funch&Marshall, 1984; Jenkins, Hurst,&Rose, 1979). Our findings do not dispute the validity of these concernsasthey apply to researchon
ReliabilityofDelayed Self-Reports18eventsofa more "ordinary" variety, suchasretirement or relocation. Nonetheless, our findings do imply that such concernsmaybe overstated when the research pertains to more "extraordinary" events, suchasnatural disaster.Thisconclusionisfurther supportedbyFunch and Marshall's (1984) finding that delayed reportsofhighly salient events (e.g., deathofa spouse) were notably more reliable (indeed almost perfectly so) than were delayed reportsofless salient events (e.g., illness in the family). Moreover,asimpliedbytraumatic stress theory, itisthe very vividness or oftraumatic events that distinguishes them from other life events. Clinically, the intrusivenessoftraumatic events has beenofmore concern than the ease with which they are forgotten (Horowitz, 1976). Findings concerning the reliabilityofdelayed accountsofsocial support were more mixed.AtTime2,respondents tended to remember having received greater social support after the crisis than they had reported at the earlier interview. Some research suggests that itmaynot be the later interview that was inaccurate here. Early accountsmayhave been deflated if victims received less help than they thought they would in situations suchasthis. This appears to happen in the contextofcollective trauma where many victims simultaneously need help and resources are sparse (Kaniasty, Norris,&Murrell,1990;Solomon, 1986).Ontheotherhand, research also has documented the omnipresence of positivity bias when individuals are asked to assess their personal and environmental resources (e.g., Taylor&Brown, 1988). From this perspective, it would be the later report that was more vulnerable; the furtheronemoves from the event, the more likely that personal factors contribute to the reports (e.g., Hobfoll&Lerman, 1989).Moreresearchofthis issue would be useful. The threat this shift posesmaybe minimal. Although sample means tended to be higher at Time2,individuals within that sample generally maintained their same rank order at each
ReliabilityofDelayed Self-Reports19timepoint. Reliability coefficients for the social support measures were generally quite substantial With one exception,.theywere no lower than .60 andashighas.84.In closing,weshould mention some weaknessesofthis study.Thiswasa "pilot study" and,assuch, suffered from someoftheflawscharacteristicofthem. We had to make some changesinthe interview schedule shortly after data collection had begun.Onesubscale, received tangible support, did not workwellwith this populationofvictims. Our fundingwasminimal, which imposed limits both on the lengthofthe interview and on the numberofinterviews that could be conducted. Two more serious limitationsofthe study must also be acknowledged. Unfortunately, bothofthese biases could have served to enhance this sample's accuracyinreporting past experiences. The first limitationfollowsfrom thehighmean levelofeducation observed in this sample. These persons may have been better able to remember their experiences accurately over time than would have a less educated sample (Funch&Marshall, 1984). More than a quarterofour respondents, however, did have only a high school education or less; and the samplewasquite diverse in termsofother demographic characteristics, suchassex,race, and age. Unfortunately, our sample sizewasnot sufficient for assessing reliability within demographic subgroups. The second limitation concerns the possibility of testing effects, thatis,the potentialofa given interview to enhance memory at a later time. However, the long interval between tests (9 months) should have minimized this threat. Nonetheless, within the limitsofitsmethodology, this study indicates that disaster victims remember their experiences quite accurately over time. Thisisa simple point, but one that should be reassuring to researchersinthis field.
ReliabilityofDelayed 20References Barrera,M.,Sandler,1.,&Ramsay,T.(1981). Preliminary developmentofa scaleofsocial support: Studies on college students. American JournalofCommunity Psychology, 2, 435447.Barrera, M.,&Ainlay,S.(1983). The structureofsocial support: A conceptual and empirical analysis. JournalofCommunity Psychology, 11,133-143. Cohen, J. (1960). A coefficientofagreement for nominal scales. Educational and Psychological Measurement, 37-46. Cohen,L.(1988). Measurementoflife events.InL.Cohen (Ed.), Life events and psychological functioning: Theoretical and methodological issues. (pp. 11-30). Beverly Hills: Sage. Fleiss, J. (1981). Statistical methods for rates and proportions. New York: Wiley. Funch, D.,&Marshall, J. (1984). Measuring life stress: Factors affecting fall-offinthe reportingoflife events. JournalofHealth and Social Behavior, 453-464. Hobfoll,S.E.,&Lerman, M. (1989). Predicting receiptofsocial support: A longitudinal studyofparents' reactions to their child's illness. Health Psychology, 61-77. Hopwood, D.,&Guidotti, T. (1988). Recall biasinexposed subjects following a toxic exposure incident. ArchivesofEnvironmental Health, 234-237. Horowitz,M.(1976). Stress response syndromes. New York: Jason Aronson. Jenkins,c.,Hurst,M.,&Rose, R. (1979). Life changes:Dopeople really remember? ArchivesofGeneral Psychiatry, 379-384. Kaniasty,K.,Norris,E,&Murrell,S.(1990). Received and perceived social support natural disaster. JournalofApplied Social Psychology, 85-114.
ReliabilityofDelayed Self-Reports21Loftus, E.,&Marburger,W.(1983). Since the eruptionofMt. St. Helens, has anyone beaten you up? Improving the accuracyofretrospective reports with landmark events. Memory and Cognition, 11,114-120. Pedhazur,E.(1982). Multiple Regression in Behavioral Research. New York: Holt, Rinehart,&Winston, Inc. Robinson,J.(1986). Temporal reference systems and autobiographical memory. In D.C. Rubin Autobiographical Memory (pp. 159-188). Cambridge: Cambridge University Press.Rubin,D.C.(1986). Autobiographical Memory. Cambridge: Cambridge University Press. Solomon,S.D. (1986). Mobilizing social support networks in timesofdisaster. InC.R.Figley (Ed.), Trauma anditswake: Vol.2.Traumatic stresstheory, research and intervention (pp. 232-262). New York: BrunerlMazeI. Taylor,S.,&Brown,J.(1988). llIusion and well-being: A social psychological perspective on mental health. Bulletin,103,193-210. Whitten,W.B.,&Leonard,J.M.(1981). Directed search through autobiographical memory. Memory and Cognition, 2, 566-579.