|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
The Impact of Computer Mediated Communication Systems Monitoring on Organizational Communications Content by Carolyn F. Holton A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Information Sy stems and Decision Sciences College of Business University of South Florida Co-Major Professor: Rosann Webb Collins, Ph.D. Co-Major Professor: Richard P. Will, Ph.D. Robert M. Fuller, Ph.D. Stanley J. Birkin, Ph.D. Date of Approval: March 28, 2008 Keywords: communications monitoring, surveillance, self-aware ness theory, CMC, hazard communications, figure-ground contrasts, instant messaging, IM Copyright 2008, Carolyn F. Holton
Dedication I dedicate this dissertation to my ch ildren, Liam and Calvin, without whom it would have been completed in a third less ti me but with more than double the stress. Thanks to you, playgroups, playgrounds, pre gnancy, painting, pre-school, and puns have been staples of my years in the Ph.D. program. It is hard to imagine that any student could have smiled more, been hugged more, or enjoyed motherhood more than I have during your infancies and early ch ildhoods. Your love, silliness, and need for lots of time with your Mama have meant myopic focus on sc hoolwork was never a risk for me. I love you both so fully. I hope that you each may enjoy the satisfaction that education has given your overeducated parents.
Acknowledgements Dr. Robert M. Fuller has taught me the lion s share of what I have learned in the Ph.D. program. His mentorshi p, insights, and example leave me deeply in his debt. When I commented in a seminar that no dissertation is the work of one person, I heard a resounding, Thank you! from its leader, Dr. Rosann Webb Collins. Dr. Rick Will and Dr. Stanley J. Birkin, along with Rosann and Robert, have generously guided and strengthened this work. For their service on my committee, I am grateful. Dr. Monica Tremblay recruited me to join her in the ISDS doctoral program and turned me back from tempting, green pastures. Her encouragement has been invaluable. My sister, Peggy Frisbie, has a willing ear and a practical creativity that have helped me around several research dilemm as. Her Aunt Peggy Day visits are rejuvenating. My brother and sister-in-law, Loyal and Julie Frisbie-Knudsen, encouraged me by their examples, also returning to sc hool while planning a family together. My parents, S.L. and Mary Frisbie, have de monstrated through their careers that doing worthy, satisfying work provides far reaching be nefits. I am grateful for the foundation they laid and the model they have provided. My husband, Jim, supported my pursuit of a doctoral degree even after seeing how the sausage is made when he completed his own. This act of deep love and generosity, and many others that followed, enabled this work. Finally, I thank God, the ultimate docto r of philosophy, whose persistent and sometimes humorous intervention made it difficult not to conclude that he intended for me to pursue this path. Trusting h im is the best decision of my life.
i Table of Contents List of Tables......................................................................................................................v List of Figures................................................................................................................... vii Abstract....................................................................................................................... ..... viii Chapter 1. Introduction and Motivation.............................................................................. 1 Introduction............................................................................................................. 1 Motivation...............................................................................................................3 Electronic communications monitori ng is prevalent and increasing .......... 3 Monitoring is likely to change communications behavior.......................... 6 Chapter 2. Literature Review and Research Model Development................................... 10 Performance monitoring....................................................................................... 11 Computer mediated communications................................................................... 12 Surveillance...........................................................................................................13 Self awareness theory...........................................................................................14 CMC monitoring encourages organi zational standard selection ..............19 Organizational CMC monitoring changes communications behavior ...... 22 CMC monitoring changes the incide nce of hazard communications and related denials ...........................................................................................28 CMC monitoring reduces communi cations volum e beyond hazard communications effects............................................................................ 30 Chapter 3. Research Design..............................................................................................34
ii Subjects.................................................................................................................34 Experimental manipulations................................................................................. 35 Monitoring manipulation.......................................................................... 35 Organizational standard manipulation...................................................... 37 Measurement......................................................................................................... 43 Study variables..........................................................................................44 Communications variables............................................................44 Standards applied.......................................................................... 46 Control variable........................................................................................ 46 Manipulation checks................................................................................. 47 Experimental procedure........................................................................................ 48 Instant messaging discussion.................................................................... 49 Pilot test findings.................................................................................................. 51 Summary of changes between pilot and main studies.............................. 59 Chapter 4. Results............................................................................................................. 61 Assessing variable adequacy.................................................................................61 Demographics...........................................................................................61 Communications variables........................................................................63 Manipulation checks................................................................................. 73 Standards applied...................................................................................... 76 Control variable........................................................................................ 79 Evaluation of the hypotheses................................................................................80
iii Post hoc analyses.................................................................................................. 92 Total hazard communications incidence................................................... 92 Total observed incidents incidence........................................................... 92 Organizational standard recency............................................................... 93 Effects of familiarity with the technology................................................95 Chapter 5. Discussion...................................................................................................... 97 Key findings..........................................................................................................97 Contributions to research....................................................................................100 Theory development...............................................................................100 Research framework...............................................................................101 Hazard communications taxonomy and coding scheme......................... 102 Relative standards influence measure and methodology........................ 103 Areas for future work.......................................................................................... 105 Contributions to practice.....................................................................................108 Implications for CMC monitoring in organizations................................ 108 Implications for organizational st andard training and presentation ........ 110 Concluding thoughts........................................................................................... 112 References.......................................................................................................................114 Appendices......................................................................................................................127 Appendix 1: Coding scheme...............................................................................128 Appendix 2: Standards Applied Measure........................................................... 134 Appendix 3: Self-Focus Scale Items, as Adapted............................................... 135
iv Appendix 4: Organizational St andard Manipulation Check ...............................136 About the Author...................................................................................................End Page
v List of Tables Table 1. Pilot Study Descriptive Statistics Overall and by Treatment Condition............ 56 Table 2. Pilot Test Results................................................................................................ 59 Table 3. Demographics.....................................................................................................62 Table 4. Interrater Reliabili ty Correlation Assessm ent.................................................... 67 Table 5. Pearson Correlations Betwee n Ratings Arrays by Interview.............................. 68 Table 6. Communications Statement, Volume a nd Incident Descriptive Statistics....... 71 Table 7. Standard Recency Descriptive Statistics............................................................. 75 Table 8. Number of Subjects by Treatment Condition.................................................... 76 Table 9. Organizational and Personal Standa rds Influence Descri ptive Statistics ...........76 Table 10. Standards Applied Factor Loadings................................................................. 77 Table 11. Organization:Personal Standards Influence Ratio Descriptive Statistics ........ 78 Table 12. Self-Focus Scale Descriptive Statistics.............................................................79 Table 13. Factors in Relative Importance As cribed to Personal and Organizational Communications Standards ..............................................................................................81 Table 14. Impact of Monitoring and St andard P resentation Recency on Hazard Communications Statement Content................................................................................. 83 Table 15. Significance of Monitoring on Statem ent Dependent Variables...................... 84 Table 16. Impact of Monitoring on Hazar d Communicatio ns Incident Content.............. 84 Table 17. Significance of Monitoring on Incident Dependent Variables ......................... 85
vi Table 18. Descriptive Statistics for Content Dependent Variables with S ignificant Monitoring Differences, Grouped by Hazard In tensity, Neutral Beliefs and Denials...... 90 Table 19. Impact of Monitoring on Communications Volume.........................................91 Table 20. Total Hazard Statements a nd Inciden ts by Monitoring Condition................... 92 Table 21. Impact of Organizational Standa rd Presentation and Training Recency .......... 93
vii List of Figures Figure 1. Figure-ground contrasts under m onitoring that m ay induce self focus............. 19 Figure 2. Relative intensity of hazard communications.................................................... 25 Figure 3. Illustration of hypothetical hazard topic dom ain effects................................... 27 Figure 4. Research model................................................................................................. 33 Figure 5. Monitored condition screen............................................................................... 37 Figure 6. Instant messaging interview prompts................................................................ 51 Figure 7. Relative influence of or ganizational and personal standards ........................... 82 Figure 8. Average denial st atem ents per incident............................................................. 89 Figure 9. Sample general negative self-d isclosure statem ents from non-monitored subjects receiving the organizational standard treatment................................................. 94
viii Abstract The Impact of Computer Mediated Communication Systems Monitoring on Organizational Communications Content Carolyn F. Holton ABSTRACT Employer monitoring of communications is prevalent and on the rise due in part to the Sarbanes-Oxley Act, the Heal th Insurance Privacy Protection Act, and other legislation in the U.S. and other countries. However, the critical effect of this new activity on what is communicated in companie s has not been assessed. This dissertation examines the impacts of computer medi ated communication systems monitoring on neutral, incriminating and exculpatory c ontent, as well as the overall volume of communications issued on monitored a nd non-monitored computer mediated communication systems. Incriminating communication is cataloged in a hazard communications taxonomy for this investigatio n. A controlled laboratory experiment has subjects participate in an in stant messaging discussion on a topic for which they are likely to be aware of information that is incriminating to their or ganization, or its members, or both. Consistent with self awareness theory, monitored subjects engage in significantly less overall and neutral comm unication. They volunteer fewer high intensity hazard communications, but are less likely to curtail low intensity hazard communications. They
ix issue denials about more incriminating topics. Contributions to rese arch include theory development, especially in the area of standa rd selection; applicat ion of self-awareness theory to the new domain of computer medi ated communications monitoring; a research framework; a taxonomy and coding scheme for the new hazard communications constructs; and a relative st andards influence instrument and methodology for use in studying competing standards. Implic ations for corporate monitoring and communications policies are discussed, and a research agenda is outlined.
1 Chapter 1 Introduction and Motivation Introduction Employer monitoring of communications is prevalent and on the rise. Electronic monitoring has long been used for custom er service quality assurance, gathering employee feedback data, increasing securi ty, and for the prom ise of productivity enhancements and management efficiencies. Do zens of statutes in many countries also now encourage or require that electronic communications be archived and sometimes reported on or produced on demand, strengthen ing monitoring motivations. Among the better known are the U.S. Sarbanes-Oxley Act ("Sarbanes-Oxley Act" 2002), Canadas C-SOX ("Bill 198"); United Kingdom Regulati on of Investigatory Powers Act Chapter 23 ("Investigatory Powers Act"), Securi ties and Exchange Commission Rule 17A-4 ("Title 17 of the Code of Federal Regulations (CFR), Rule 17a-4" 2001), and U.S. Department of Defense Rule 5015.2-STD ("El ectronic Records Management Software Applications Design Criteria Standard" 2007). This legislative trend has strengthened monitoring motivations. Given the increasing prevalence of recording of com puter mediated communications (CMC) and the peril these records may create, pr oactive companies are monitoring their communications to take appropriate liability -limiting actions. For instance, many subject to the U.S. Health Insurance Portability and Accountability Act ("HIPAA") are choosing
2 to monitor e-mail to prevent disclosure of protected information and to identify employees who require additional education in this regard, while organizations subject to National Association of Securities D ealers (NASD) Rules 3010 and 3110 ("NASD Conduct Rules 3010 & 3110" 1999), are explici tly required to monitor the content of electronic communications. Thes e monitoring activities may s eek both to capture current communications behaviors and to influence them We have reason to suspect that they also cause unexpected and unintended consequences. Several theories suggest computer medi ation encourages communication that is less inhibited or more extreme in some respects than face-to-face communication. For instance, anonymity effects and reduced social cues may lead to anti-social behaviors like flaming (Sproull et al. 1986). Use of co mputer mediated communications may even encourage an increase in hazard communicati ons, which are messages that might tend to incriminate an organization or its members (H olton et al. 2006). Alternatively CMC may facilitate greater emotional content than face-to-face communications, amplifying social influence (Postmes et al. 1998), encouragi ng social disclosure (McGrath 1991), and leading to extreme, hyperpersonal communicat ions with richer, more highly social content (Walther 1996). Anothe r possibility is th at an individuals submersion in and isolation from a virtual group may limit co mmunication, discouraging full contribution over CMC (social loafing) (Chidambaram et al 2005) While there is ample evidence that computer mediation influences what peopl e communicate, the nature of computer mediated communications is difficult to pred ict. Organizational m onitoring of CMC adds complexity to this consideration. Might it attenuate extreme communications behavior? Or could it promote different extreme communi cations? These issues lead us to the
3 following research question: What impact does computer mediated communication systems monitoring have on the conten t of organizationa l communications? Motivation The prevalence of communications monitori ng, and legal changes that are making it even more commonplace, along with the suspected, but ill understood impact of monitoring on communications content ma ke this an area ripe for study. Electronic communications monitoring is prevalent and increasing The incentives for communications monito ring are extensive and increasing. The growth in monitoring of CMC represents a substantial environmental shift with potentially profound impacts on what is comm unicated in organizations, and ultimately on the work that gets done within them. We define this monitoring activity as reviewing computer mediated communications inte nded for other parties by organizational members whose formal role it is to perform this function. For instance, an internal auditor might read instant messages to determine whether restrictions on passing information between representatives of co mpeting customers are being followed, or a security staff member might r ead email related to a suspici ous activity to determine its legitimacy. In each case, monitoring is part of the monitoring individuals role-consistent organizational duties. Intercepting comm unications for any purpose other than the fulfillment of an organizational role, or intercepting communications outside of the monitors organization lies outside of this definition. Electronic monitoring has been employed for the promise of various productivity enhancements (Douthitt et al. 2001), for cu stomer service quality assurance (Sherry 1998), for objective gathering of data for em ployee feedback (Urbaczewski 2000), and to
4 limit corporate risks, including legal liab ility (Varon 2003), and security breaches (Nadeem et al. 2006), among other applicatio ns. Computer mediated communications monitoring is a less studied practice. Recently a great deal of legislation has strengthened communications monitoring incentives (e.g. "Bill 198" 2003; "Electronic Records Management Software Applications Design Criteria Standard" 2007; "HIPAA" 1996; "NASD Conduct Rules 3010 & 3110" 1999; "Sarbanes-Oxley Act" 200 2; "Title 17 of the Code of Federal Regulations (CFR), Rule 17a-4" 2001; "Inves tigatory Powers Act" 2000). For instance, the Sarbanes-Oxley Act was described by Pres ident Bush as landmark legislation that adopts tough new provisions to deter and punish corporate and accounting fraud and corruption, ensure justice for wrongdoers, a nd protect the interests of workers and shareholders (Bush 2002). It has had a dramatic effect not just on American companies, but also on their subsidiaries in other countries, and on companies which are based in other countries but listed on U.S. exchanges. The rules issued by the SEC to enforce the Act are being interpreted to require every public company to store every document that influences the audit process for seven year s (Varon 2003). A clear intention of these rules is to maintain records of communications originating inside a company that suggest, reveal or otherwise pertain to illicit or unethical activity occurring within it. The legislation requires that computer mediated communications be recorded, stored, and produced on demand. The Acts real-time disclosure rule also requires companies to self-monitor their records, both structured like those in accounting systems, and unstructured like those in communications systems, for events that must be disclosed under the provisions of the Act in real time (Worthen 2005).
5 While Sarbanes-Oxley is perhaps the be st-known driver of the communications monitoring trend, other legislation carries provisions that more directly require or promote CMC monitoring. For instance, HIP AA has proliferated products to prevent email, blog and other communications leakage of protected information ("HIPAA email compliance policy" 2008; "HIPAA email encryption and security compliance for healthcare" 2008; "HIPAA secure e-mail" 2008). Organizations subject to National Association of Securities Dealers (NASD) Rules 3010 and 3110 ("NASD Conduct Rules 3010 & 3110" 1999) have no discretion in the ma tter. They are exp licitly required to monitor the content of elect ronic communications, and instant messaging is known to have been monitored as early as the year 2000 for some firms subject to its provisions.("IM Shop?" 2004). Given the increasing prevalence of reco rding CMC and the peril these records may create, proactive companies are monitoring their communications to take appropriate liability-limiting actions or to meet the e xplicit requirements of laws and regulations (Tam et al. 2005). With this backdrop, it is not surprising that monitoring workplace communications is now the norm for U.S. companies, and various monitoring practices are still increasing in preval ence ("Electronic Monitoring Survey" 2005). For instance, between 2004 and 2005 (the last years for whic h comparable numbers were reported), email monitoring increased 57 per cent in U.S. companies (AMA "E-Mail and IM Survey" 2004; AMA "Electronic Monitoring Survey" 2005) Monitoring is now so prevalent that over half of employers have now fired workers for e-mail or Internet misuse, with approximately equal incidence of each ( 28 percent and 30 percent respectively)
6 ("Electronic Monitoring and Surveillance Survey" 2007). Despite the myriad and growing uses, little is known about the im pacts of employee communications monitoring on the content of communications. Monitoring is likely to change communications behavior While monitoring may seek both to capture current behaviors and to influence them, we have reason to suspect that it also causes unexpected and unintended consequences. A great deal of anecdot al evidence suggests CMC changes when monitoring ceases. For instance, meeting hosts turn off recording equipment when they want people to speak more freely. Reporters perk up when a source offers information off the record. Entire academic conferences have been held off the record to encourage the freer flow of ideas ("Speedbump Conf erence" 2004). Large industry conferences have also had an off the record restric tion, like this one from a Google customer innovation conference, All speeches and discussi ons at Zeitgeist are off the record. To ensure that our presenters and attendees can speak openly, no press coverage or blogging is permitted (Google 2005). In this light it is surprising that academic literature is largely silent on the question of the imp act of monitoring on computer mediated communications content. Several theories also suggest that CMC encourages communication that is less inhibited in some respects than face-to-f ace communication. For instance, when users remain anonymous, they may contribute mo re candid views (El-Shinnawy et al. 1997) without fear of suffering nonconformity consequences (Nunamaker et al. 1991) or of embarrassing themselves. A meta-analysis c oncludes that anonymity does impact what people communicate over CMC (Baltes et al. 2002).
7 Other depersonalizing aspects of the natu re of CMC media may encourage similar behavioral changes. CMC have been describe d as lacking in social presence given the limited range of both nonverbal and verbal cu es and the communications context (Rice 1993; Short et al. 1976). The number of cues transmitted, channels used, and the degree of personalization are all also said to place limits on the information richness of a medium, with computer mediated communicatio ns channels ranking low in this respect as well (Daft et al. 1986). Cues filtered out theory (Sproull et al. 1986) explains that distance, posture, facial expressions, gaze, voi ce variations, and other social cues may not be conveyed at all (or only partially conveyed with more fully-featured CMC tools) and thus fail to influence or cont rol the behavior of others (H iltz et al. 1989). Both anonymity effects and reduced persona lization and social cues may result in decreased awareness of those with whom one communicates or even of ones personal identity. They may also decrease the power of social norms to regulate behavior. Other paths through which CMC may reduce soci al influence include removing physical barriers so communicating parties may expect that they are more diverse from one another, and eliminating knowledge of similariti es between parties. Each of these factors can lead to communications that are impulsive, extreme, and even anti-social, as in the case of flaming (Sproull et al.) or which flout authority (Hiltz et al. 1989). These same factors may be responsible for an increased willingness to communicate negative information (Sproull et al. 1986). Just as CMC may break down certain soci al boundaries, it is also possible that CMC may draw focus away from self and towards the group, providing the group with greater social influence over individual beha vior (Postmes et al. 1998). According to the
8 social identity model of deindividuation e ffects (Lea et al. 1991), as deindividuation occurs (Festinger et al. 1952), group member s become more sensitive to situational norms and responsive to environmental cues for appropriate behavior for the context. In absence of individuating information, CMC us ers may subrogate their own opinions and values to the groups (Postmes et al. 1998). In this way, groups and dyads within which CMC occur serve as a source of social iden tity, and the medium is socially engaging. The social influence of the group may even lead to hyperpersonal communications with richer, more highly social content (Walther 1996). Another possibility is that drawing focus away from the individual and towards the group may make individuals feel isolat ed and submerged within a virtual team, factors which, according to social impact th eory, can lead to suppression of effort (Chidambaram et al. 2005). This may be expressed as fewer or lower quality communications than in face-to-face groups. On the other hand, given its ability to provide work structure, CMC may help to maintain focus on task deliverables, discouraging social loafi ng (Shepherd et al. 1995). While there are ample anecdotes, theoretical bases, and empirical evidence that CMC influences what people communicate, the nature of computer mediated communications in a particular circumstance re mains difficult to predict. Organizational monitoring of CMC complicates the predicti on task. Under monitoring, will people be less candid and less willing to communicate negativ e information? Or will they be more deliberative about their communications a nd more likely to give balanced views including negative information? Will they seek to avoid notice, or be likely to contribute greater effort in hopes of being noticed? Will th ey be more tentative, or more expressive
9 of their opinions? Will certain topics be c onsidered off limits, or will sensitive topics be discussed differently? Will people avoid extr eme communications, or engage in different extreme communications? We begin to addres s these uncertainties and others with the following research question: What impact does comput er mediated communication systems monitoring have on the cont ent of organizational communications? Consistent with self awareness theory, th is study finds that monitored subjects engage in significantly less overall and ne utral communication. They volunteer fewer high intensity hazard communications, but are le ss likely to curtail low intensity hazard communications. They issue denials about more incriminating topics. Contributions to research include theory development, partic ularly in the area of standard selection, application of self-awareness theory to the new domain of computer mediated communications monitoring, a research framework, a taxonomy and coding scheme for the new hazard communications constructs, a nd a relative standards influence instrument and methodology for use in studying competing standards. Implications for corporate monitoring and communications po licies are discussed, and a rese arch agenda is outlined. This document proceeds with the following sections: review of the relevant literature, with the research model developed in the course of that review in chapter 2; research design, which includes a discussion of p ilot test results in chapter 3, findings of the full study in chapter 4, and discussion, c ontributions to research and practice, and limitations in chapter 5.
10 Chapter 2 Literature Review and Research Model Development In a pilot study (Holton et al. 2006; Holton et al. 2008) researchers found a significant relationship between monitoring CM C and changes in hazard communications content. While high intensity confessional communications decreased in frequency, no changes in moderate intensity whistle-blowing type comm unications were observed, and modest intensity hazard reports about which individuals had no first hand knowledge increased. Denials of knowledge of hazardous topics also increased under monitoring. In this section, we seek to elucidate the psychological processes involved in these changes. First, we consider how the pe rformance monitoring, CMC and surveillance literatures jointly predict changes in communications beha vior under monitoring. Next, self awareness theory is applied to build the research model, which also draws on CMC and surveillance literature streams. Ultim ately, we posit that monitoring of computer mediated communications induces self focus, which increases the regulatory role of perceived organizational communications standards relative to personal standards, and precipitates changes in communicat ions behavior. When negative or positive misalignment with the standards applied is detected, changes in communications are precipitated. The specific nature of these changes is predicted for sensitive hazard communications topics.
11 Performance monitoring Performance monitoring is probably the best known and most studied use of computers to monitor activity within orga nizations. The impacts of performance monitoring via computer on performance, satis faction, health, stress and other outcomes, along with attitudes towards monitoring and perc eptions of supervision, have been widely examined. In the psychology literature, studies typically fi nd higher quantity of performance (sometimes with lower quality), along with lower satisfa ction, poorer health, higher stress, and other negative impacts on those monitored (Douthitt et al. 2001). Work within the information systems domain has found mixed results, sometimes suggesting that the nature of the monitoring and other factors su rrounding it can lead to different outcomes (Georg e; Irving et al. 1986). While much of the performance monitori ng literature is at heoretical, social facilitation theory, which predicts that in the presence of others, people will perform better on easy or well-learned tasks but worse on difficult tasks or t hose not well-learned is one explanation sometimes given for performance impacts in the presence of others. Social facilitation is said to take place due to the arousal provided by the presence of others. The classic interpretation is that this arousal encourages habitual responses, appropriate for simple tasks, but which may impair performance of tasks calling for nonroutine approaches (Zajonc 1965). While computer mediated communications tasks range from simple and routine to novel and co mplex, it is the latter with which we are concerned, those for which monitoring seem s most likely to impair performance. Both this theory and the performance m onitoring studies in general have focused on highly structured tasks like clerical work. Typical use of e-mail, instant messaging,
12 and other CMCs by knowledge workers is less st ructured and more complex (Rice et al. 1988). Further, we are not expl icitly addressing performance variables. Even so, this literature indicates that the extent to which one is focused on others, as contrasted with experiencing oneself as an isolated individua l, will impact ones approach to a task. Computer mediated communications Computer mediated communications (CMC) encompass all forms of communication transmitted between two or mo re people via computer networks. CMC applications are many and ever expand ing with significant consequences for organizations (Cameron et al. 2005). In a ve ry broad sense, CMC systems include those providing e-mail, text chat and instant messa ging, group decision support/group support, bulletin boards, listservs, virtual workspaces, online conferencing, massively multi-player online games (MMOs), weblogs (blogs), wikis, and the exchange of RSS (web feeds). CMC systems may support communications that are synchronous or asynchronous; sequential or parallel; anonymous or identified ; ephemeral (not reco rded) or persistent (recorded); rehearsable (allowing review and editing of a draft message before sending) or instant; dyadic, one-to-many, or many-to-m any. They can reach around the world and be used by those in the same room. They diffe r in the degree to which they convey social presence. The simplest merely convey type d messages, but most have an array of features to support communication. Anonymity, or lack thereof, is a well-st udied characteristic of CMC. While full anonymity is not a typical feature of organizational CMC use, visual anonymity, or communication without visual c ontact with other communicating parties, often is. That
13 visual anonymity has been found to have si milar impacts to full anonymity causes us to consider it here (Spears et al. 1990). Anonymity is an important driver of de individuation, a sense of losing ones own identity to a group (Festinger et al. 1952). Anonymous indi viduals experience a reduced state of self awareness in which they feel unidentified and unaccountable. Deindividuation is said to reduce the restrain ts one normally places on ones behavior to inhibit unsanctioned behavior. Consequentl y, unsanctioned behavior increases (Sproull et al. 1991). When this line of reasoning is applied to visual anonymity in a CMC system, it predicts that unsan ctioned communication increases as compared with less anonymous media or face-to-face communications. To determine the effect of monitoring on unsanctioned communication, we next turn to the surveill ance literature. Surveillance Researchers have not converged on a single definition of surveillance, but much academic literature on the topic seems to im plicitly define surveillance as close observation of a person or group by law enforcement. While the same communications monitored by organizations could conceiva bly be surreptitiously monitored by law enforcement, it is intra-organizational role consistent interception of communication intended for other parties with which we are concerned in this study. Despite the differences in definitions, monitoring can have the aspect of close observation by an authority, thus we consult the surveillance literature for insigh ts relevant to the study of communications monitoring. Surveillance has been described as a means of exercising social control (Deci et al. 1985), and the social implications of closed circuit television ( CCTV) have received
14 some academic attention. Displacement of mo nitored activities is one of the commonly observed outcomes of CCTV surveillance (Shor t et al. 1998). Beha vioral changes under surveillance are often explained with refere nce to Foucaults panopticon (1977): belief that one might currently be monitored is sufficient to induce changes just as if the monitors presence were certain. The possibility that deviant or otherwise undesirable behavior may draw a monitors unwanted attention suppresses it. If the attention is oppressive, when possibl e, individuals escape the panoptic gaze, resulting in displacement of undesirable behaviors to an unmonitored setting. Although visual anonymity in a CMC system may lead to less desirable communications behavior, surveillance literature predicts that when m onitored, such behavior may be displaced to an unmonitored channel or suppressed. It is also possible that organizational CMC monitoring is considered far less oppressive than the panoptic prison, limiting its ability to induce effects through th is mechanism (D'Urso 2004). Surveillance research has been criticized as being under-theorized (Vorvoreanu et al. 2000). To better understand the impact of monitoring CMC on communications content, we turn from the broad, non-speci fic panoptic approach to a well-developed literature stream, self awareness theory, which better addresses the aims of this research. Self awareness theory Reduced self awareness has been called th e key psychological constituent of the manipulation of deindividuation effects, (Postmes et al. 1998), the same effects which are believed to underlie mu ch computer mediated communication s research. In contrast, as we will see, monitoring CMC may increase self focus, the construct of primary interest in self awareness theory.
15 Self focus has been a topic of academic inquiry for over a ce ntury, gaining early prominence in the work of sy mbolic interactionists (Cooley 1902). It has been studied empirically since 1932 when a research methodology was deve loped (Wolff 1932 ). Self focus is also referred to as self-directed attent ion and is a state construct similar to the trait construct self-consciousness. Self awar eness, which has sometimes been used interchangeably with self focus, is more prop erly defined as the abili ty to recognize ones existence (Joinson 20 01), and as such it is a pr e-cursor to self focus. At its heart, self awareness theory predicts that when self-focused, people evaluate themselves according to a standard relevant to the area of self focus. When they find themselves falling short of or exceeding the standard, they be come motivated to alter their behavior to achieve alignment between self and standard. When they fi nd alignment, they are motivated to maintain it by inhibiting standard-inconsisten t actions. We now examine the relevance of self awareness theory to CMC monitoring. Human attention is a li mited resource. It is selective, and of limited ca pacity (Posner 1982). While we may rapidly switch between object s of attention, the num ber of objects to which we may simultaneously at tend is very limited. According to self awareness theory (Duval et al. 1972), attent ion is bi-direction al. It is oriented either towards oneself or towards ones environment. Self focus is the act of directing attention towards oneself rather than towards ones environment. This same term is applied to the st ate of being enga ged in the act of self focus. When an individual is self-focused, his or her attention is drawn to whatever aspects of self are then salient, typically to situatio nal factors such as the event that drew focus to self, the current or up coming performance of a task, or an attitude or belief relevant to these.
16 Researchers disagree whether public and private self focu s are distinct constructs. Public self focus is the concer n for self as a social object (Buss 1980) and relates to selfpresentation (Scheier et al. 1983). With private self focus, an individual is attuned to his or her own thoughts and em otions (Scheier et al. 1978). Wicklund and Gollwitzer (1987) present several persuasive theo retical arguments that public and private self consciousness are not distinct constructs. They reason that since thinking about on eself per private self focus is said to preclude thinking about anot hers impression of onesel f per public self focus (Buss 1980), measurements of priv ate and public self focus should be inversely correlated. However, across many studies, th ese measures have be en found to be disturbingly high[ly] correlated (Wicklund et al. 1987, p. 502). Fu rther, these authors discuss the tight entanglement of means of induc tion of self focu s and whether publ ic or private self focus is induced, noting inconsistency between the oper ationalization and defini tion of private self focus. Private self focus is said to focus on aspects of se lf which are no t observable. Nevertheless, subjects are presen ted with observable aspects of themselves to induce this state. For additional arguments, the reader is referred to (Wicklund et al. 1987). We find their review of empirical results conc lusive in supporting their reason ing and thus make no further distinction between the two. Engaging in CMC has been fo und to reduce self awareness of onese lf as engaged in social interaction with an audi ence (Matheson et al 1988). We posit th at monitoring CMC restores it. According to self awar eness theory, there are two causes of self focus, namely that it is generated by reminding one of oneself or by placing one in a figural-ground contrast (Silvia et al. 2001). Early researchers experimented with many means of inducing the self-
17 focus state, primarily by providing self-remin ders. An individuals own voice (Wolff 1932), video image (Duval 1976), and reflection in a mirror (Hormuth 1982), the presence of a video camera (Wicklund et al. 19 71), physically present observers (Innes et al. 1975), social disruptions (Shibuta ni), experimenter instruction (Taylor et al. 1975), presen tation of symbols of self (Duval et al. 1972), writing about oneself (Fenigstein et al. 1984), and experiencing oneself as a minority (Duv al 1976) have all been found to be effectiv e means of self focus induction. Several of th e methods of induci ng self focus have the bear ing of monitoring. In particular, the physical presence of others an d video cameras afford means for observing an individuals behavior contemporaneously or after the fact through review of a created record. One study has used vi deo cameras to induce self-focus during a CMC task (Yao et al. 2006). The study uses traditional video camer as and webcams, without explaining to subjects how the video images would be used beyond for a separate research project. Peers were said to be using the images, not one or more authority figures within the organization within which communications occu rred. In one condition, the cameras were used to show subjects their own image. Thus the investigation illustrated that traditional video camera self-focus manipulations are e ffective in a CMC context, but it did not provide a test of the effect of organizational monitoring. While the reminders of self have been the primary form of expe rimental self focus induction, CMC monitoring affo rds a number of poss ibilities for increa sing self focus through Gestalt figu re-ground effects on attention (Koffka 1935). According to this Gestalt principle, a figure attracts atte ntion because it is smaller than the surrounding ground. When studying attention, th e principle was initially ap plied to visual stimuli, but it has also been successfully applied to other stimulus-atte ntion effects (Duval et al. 2001).
18 In the realm of CMC monito ring, simply drawing anothe rs focus by any means is a relatively rare event (Handel 2005 ), and thus the figure agains t the ground of times at which one is not the focus of others attention. Th is figure-ground relati onship brings ones attention to the fact of the others focus. Mo nitoring is an even smaller figure among the class of situations in which one draws anothers focus, cr eating a greater figure-ground contrast that may magni fy this effect. The figure-ground principle al so applies to the properties of people, with the less abundant property servi ng as figure agai nst the more common ground (Duval et al 2001). When monitoring CMC for compliance with comp any policies or other regulations and laws, an observer, particularly one with the power to initiate sanctions, may be thought by the observed to have greater status. In this circum stance, the lower status level of the monitored individual serves as figure, drawing attention to the self. Finally, the CMC monitoring role may be believed to be held by mo re than one person. Multiple monitors serv e as ground against the observed individual who becomes the figure of his or her own focus. If CMC monitoring beco mes conspicuously prevalent in some context, the act of monitoring may be insufficient to cause figu ral attention. Howeve r, accountability is expected to increase in this circumstance, prov iding its own figure-ground contrast. Those to whom one is acco untable are likely to outnumber oneself and to be perceived as more powerful than oneself, creating a ground that brings focus to self. These figure-ground contrasts are presented in Figure 1.
19 Figure 1. Figure-ground contrasts under mo nitoring that may induce self focus CMC monitoring encourages organizational standard selection Once individuals are self aware, self schema ar e brought to mind creating the likelihood of self -evaluation (Gibbons 1990). The outcome of this evaluation depends on the standard applied for assessment (Duval et al. 1972). Standard s are images of correct ways to think, feel, act, and be (Duval et al 2001, p.31). They are nt simply behavior tendencies they are comparison points that on ly influence action when participating in selfregulation (Silvia 2002, p.5). Standards percei ved as relevant to the self-aspect made salient
20 are selected for this comparison Retrieving standards from memory for consideration for the self-standard evaluation thus increases their influence on beha vior regulation. Prior work on self awareness has avoided th e standard selection issue by presenting a single standard or selecting subjects for experimental study based on pre-measured personal standards (Silvia et al. 2001). While convenient for the purpose of acad emic study, neither of these single standard conditions is realistic for organizational environments. In the current study, we consider the st andard selection is sue in a realistic context where standards may be somewhat ambiguous (Chociey 199 7), potentially in conflict with each other, and not explicitly presented at the time of self focus. In the communications monito ring context, two candidate standards are the preexisting communication s standards of the in dividual and the perceived communications standards of the organizational monitor. These stan dards may be congruen t or overlapping in some cases. However, in other cases, orga nizational members pe rsonal communications standards will be misa ligned with a perceive d organizational standa rd for communication. For instance, personal communi cations standards may favor openness while organizational standards support discretion. The attention brought to self by CMC monitoring is used not only to retrieve ca ndidate standards and sele ct an appropriate standa rd, but also to ignore interfering information (Chun et al. 2001; Norman et al. 1986). Thus while more than one standard may be considered, increasing the be havior regulatory influence of each, when standards are in c onflict, the influence of th e most relevant standard will increase while the relative influence of competing standards on behavior regulati on will fall. Issues of perspective-taking a nd salience aid in selecting between competing standards.
21 Self focus has been conceptualized as taki ng the perspective of some other persons viewpoint of the self (Mead 1934). This proper ty encourages adoption of an organizations unchangeable standard (unchangea ble by the monitored at the tim e of monitoring, at least) even when a personal standard is in conflict with it Experimental work finds this to be a typical case, with dozens of st udies showing choice of a just-provided third party standard over the personal sta ndard that operates in its absence (Wicklund et al. 1971). Another possibility is that individual s will resolve any perceived standard discrepancy between candidate standards by bringing the personal standards within their control in line with perceived organizational st andards that they cannot control. Research suggests that rather than aligning standards, choice of whichever standard is salient is more likely. A review by Silv ia et al. (2001) elucidates the primar y role of salience in the selection of standards. In labo ratory studies, when an organizational standard was presented in advance of a task, that standard was ap plied. When it was presented after, subjects adhered to their just applied personal standards. In orga nizational contexts, the organizational standard may not be presented immediately in advance of behavior selection. Howe ver, in the case of CMC monitoring, the act of indu cing self focus via knowledge of an organizati onal monitors activity also encourages re call of or speculati on on the organization s standards for communications, increasing th e likelihood that th ey will be salient. Th is too favors choice of an organizational standard over a persona l standard. When a negative aspect of self is made salient, it can inhibit the transition to some other persons perspective (Stephenson et al. 1983), lessening the likelihood of adoption of anothers standard. In the case of monitored CMC, it is on es communications that are the focus of the monitors atte ntion, and it is this aspe ct of self which theory predicts will become
22 salient. Communications are not typically perceived as an inherently negative aspect of self (McCroskey 1977), so the transiti on to the monitors point of view is not expected to be impeded in this regard. Organizational perspective-taking and organiza tional standard salience, along with it being unlikely that personal and organizational stan dards will be brought into alignment or that a negative evaluation of ones communications ability will impair sh ifting to an organizational standard, all lead to the same conclusion: When self-focus induced by monitoring makes organizational st andards that compete with pers onal standards more salient, the regulatory influence of organizational co mmunications standards will be increased at the expense of the influence of personal standards. H1: Organizational monitoring increases th e regulatory influence of perceived organizational communicati ons standards on computer mediated communications content relative to personal standards. Organizational CMC monitoring changes communications behavior According to self awareness theory, once a standard is selected for the selfstandard comparison, individuals assess their conformity with it. In its original form, the theory predicted that individuals would always find self falling short of the standard (Duval et al. 1972). It was reasoned that sin ce few aspects of self are likely to be assessed to be perfect, holding oneself to a perfect standard would alwa ys lead to the discovery of inadequacies to some degree. Later work by both of the theorys original authors and others has determined this finding of pers onal inadequacy is not essential to its function .(Silvia et al. 2001; Wicklund 1975). However, even when self is assessed to exceed the standard applied, a behavior-motiva ting discrepancy exists (Duval et al. 2001).
23 When self and standard are in alignment, pos itive affect can result and no behavioral change is anticipated (Greenbe rg et al. 1981), but individuals are still motivated to inhibit future standard-inconsistent behavior (Gibbons 1990). When behavior is misaligned with the st andard applied, two pa ths are possible for resolving this aversive affective state: One may either try to avoid the self-focus state, or to more closely align oneself with the sta ndard applied on relevant dimensions. In the circumstance of monitored CMC, avoiding se lf focus means reducing or avoiding use of monitored communications channe ls that induce it. Users se eking to more closely align to a perceived organizational standard can easily do so by changing their communications behavior to be congruent with it. Thus to the extent that a monitored channel cannot be fully avoided, self awareness theory predicts that use will be reduced and that the content of the communications delivered changes from those communications delivered via nonmonitored channels. Since communications can vary so wide ly, understanding the degree to which communications behavior changes and the na ture of those change s depends on a number of factors, among them the characterist ics of the communications topic, the organizational context, and the organizati onal environment. We now examine each, choosing for primary focus a category of commun ications that is suffi ciently broad to be relevant to a variety of organizational communication topics: hazard communications. Hazard communications are messages that might tend to incriminate an organization or its members (Holton et al 2008). These suspicious messages do not necessarily indicate illicit activity. They may be byproducts of productive problem solving processes (brainstorming) as options are generated, refined, and culled. They
24 may reflect misunderstandings or incorrect su spicions. They may pertain to preventing or correcting mistakes. But whatever their motivation, they have in common that they have the potential to draw unwanted atten tion when computer me diated communications are monitored. Hazard communications are organized into a taxonomy of high, moderate, and modest intensity (Holton et al. 2008). High intensity hazard communications are personally incriminating communications by an organizational member. We label this category negative self-disclosure Because any organization is represented by its members, these communications may also te nd to incriminate organizations. Moderate intensity hazard communications offer personal knowledge of incriminating behavior without implicating the communicat or. We label this category observed Observed hazards still tend to incriminate organizations but entail a lesser degree of personal risk. Finally, modest intensity hazard communications are relayed reports of information not personally known to the communicat or. We label this category hearsay. Because they are unsubstantiated, hearsay hazard communicati ons are less incriminating than those in the other two categories. Within these three categories, hazard intens ity is determined by the specificity of the communication. Specific reports are likely to incriminate more individual organizational members and to be more ac tionable than general information. For instance, if petty cash is to be used only for business operating e xpenses, reporting, Ive witnessed that people often use petty cash for office parties, is a lower intensity statement than the more detailed, For Mary Byrnes birthday party in June, her manager told the department secretary to buy a cake a nd candles with money from the petty cash
25 box she keeps. While the former could lead to reform of petty cash reporting and auditing processes, the latter could do both th at and result in punitive action for parties who knowingly violated company policies. This intensity taxonomy is depicted in Figure 2. Figure 2. Relative intensity of hazard communications The specific domain of the communication within an organizational context may also provide an indication of the level of intensity. Where the intensity of communications topics can be ranked within a given organizational context, the taxonomy presented in Figure 2 applies with in a topics ranked position or level. Overlaying a domain-specific taxonomy results in a number of such grids, each within a domain of communications t opics. The example provided above is in the domain of
26 fiduciary responsibility and accountability. In this domain, the amount of money in question is one determinant of the degree of organizational and pe rsonal incrimination. Within the category of infractions un der $100, hazard communications might be considered to be of similar baseline inte nsity so they can be ranked per Figure 2. However, we would not rank the specific repo rt about using petty cash for Marys party above a more general one a bout corporate fraud amounting to hundreds of thousands of dollars. The latter is at a higher level of intensity. A hypothetical distribution of hazard intensity across selected organizational domains is provided in Figure 3. Thus determining hazard communications intensity also requires either considering the importance of the topic within the organizational context, or limiting oneself to topics of a single topic area at approximately th e same baseline intensity level.
27 Figure 3. Illustration of hypothetic al hazard topic domain effects
28 Just as the organizational context is im portant in determining hazard intensity level, so is the operating environment of th e organization. In the U.S., Sarbanes-Oxley and other regulatory incentives for monitoring may influence the assignment of a topic to a hazard level. For instance, with Sarb anes-Oxley receiving so much attention, embezzling may be seen as more hazardous than Occupational Health and Safety Administration regulation violations. Based on this understanding of the nature of hazard communications, we can now predict the nature of behavioral responses to computer mediated co mmunications monitoring. CMC monitoring changes the incidence of hazard communications and related denials Hazard communications are defined by thei r potential for incrimination. While some organizations and organizational m onitors may wish to hear about some incriminating behaviors in some circumstances potentially incriminating topics are often taboo within organizations (Syret t 2001), which is to say in violation of an organizational standard for appropriate co mmunication. Per self awaren ess theory, individuals are motivated to bring their behavi or into alignment with these standards. When monitored channels must be used, this requires suppr essing communications th at are most clearly misaligned with organizational stan dards. Thus we provide H2: H2: Organizational members make fewer higher intensity hazard communications under organizational CMC monitoring than on unmonitored channels. Organizational communications standards are often incomplete and ambiguous with written policies failing to cover many contingencies (Gilsdorf 1992). (Examples are examined in the following chapter.) W ithout hard and fast rules for acceptable
29 organizational communications, where to draw th e line is a judgment call. When there is only a minor potential incrimination of the organization or its members, it is more likely that a communication will be perceived to be aligned with the standards of an organization. For instance, even if borrowi ng petty cash for personal uses is strictly forbidden, borrowing 75 cents for a soft drink is not very incriminating to the individual, and does not represent a level of financial impact that would cause shareholders to charge the organization with fiduciary negligence. Even so, the action is in violation of an organizational policy for petty cash management With such minor policy violations, it is less clear whether communicating about the in cident is misaligned with organizational communications standards. When low intens ity hazard communications are perceived to be in alignment with organizational communi cations standards, no behavioral motivation exists. The smaller or absent motivationa l effects of perceived self-organizational standard misalignment in these cases leads to H3. H3: The decrease in the frequency of high intensity hazard communications between monitored and non-monitored conditions is greater than the change in frequency for low intensity hazard communications. A related class of communications whic h may be influenced by the attempted alignment of ones communications to an orga nizational standard is denials of knowledge of hazards. For example, given an unde rstanding that use of petty cash for nonoperational expenses is prohibited, when the topi c turns to questionable uses of petty cash, one way to avoid engaging in hazard communications, and possibly even guide discussion away from the hazardous topic, is to deny any knowledge of inappropriate uses of cash box monies.
30 Denials corresponding to each level of h azard intensity are possible, as are general denials not specific to any one level. People can deny knowing about incriminating acts, engaging in them, witnessing them, and/or hearing about them. They can also deny knowledge indirectly by ma king statements qualified as suppositions, implying no personal knowledge. For instance, in the petty cash example, one might say outright Ive never he ard about anybody misusing the funds in the cash box, which is a hard denial of relayed repor ts, or alternatively, I woul dnt think anyone here would touch the cash box except for approved uses, a soft general denial not tied to any one of the hazard intensity categories. Because denials are both non-incriminating and potentially exculpatory, they are an alternative means to suppressing hazard communications for changing communications behavior to be congruent w ith a perceived organizational standard. This leads to H4: H4: Organizational members engage in more denial s of knowledge of hazards under organizational CMC monitoring than on unmonitored channels. CMC monitoring reduces communications volume beyond hazard communications effects Self awareness theory tells us not only that individuals will seek to align their communications behaviors to an organizational standard under monitori ng, it also tells us they will seek to avoid the mo tivational affective state induced by monitoring when possible. In most organizational setti ngs, it is not possible to completely avoid monitored communications. As noted herein, most empl oyers routinely monito r communications. However, the extent of use is somewhat within employee control. Mess ages can be terse or more descriptive, including or leaving out details, opinions and beliefs. Live
31 communications sessions like those occurring over instant messaging and virtual conferencing systems can be shorter or longer as participan ts choose. While hypotheses 2-4 predict changes in haza rd communications and as sociated denials, the incentive to limit communication over monitored media suggests that neutral communica tions unrelated to incriminating behavior will also be affected. H5 predicts this effect: H5: Organizational members engage in a lower volume of neutral communications under organizational CMC monitoring. With H2 and H5 predicting decreased communications (of hazards and neutral communications), and H4 predicting increased communication (of de nials), the question arises as to whether communications vol ume increases or fa lls overall. While communications vary across busine ss contexts and cultures, an d we are aware of no studies that try to catalog business communications in a way that is broadl y generalizable, the development of the hazard comm unications construct allows us to explore the volume of projected changes. The hazard communications co nstruct is derived from ne gative self-disclosure, whistleblowing, and gossip literatu re streams (Holton et al. 2008). Negative self-disclosures, the highest intensity hazard communications, are considered to be ris ky in organizations, causing employers to take specific steps to promote them when they are desired, for instance in hospital mortality an d morbidity conferences, military af ter action report s, and corporate exit interviews (Feldm an 1999; Garvin 2000; Orlander et al. 2003). Whistle-blowing is an activity typically perceived by employees to carry heavy risks for the whistleblower, including job loss, discouragi ng this type of communicat ion (Dozier et al. 1985), but legislative incentives have required whistleblower protection in recent years ("Sarbanes-
32 Oxley Act" 2002). While th e incidence of high intensity negative se lf-disclosures and moderate intensity wh istleblowing may be somewhat modest, gossip is a mainstay of organizational communica tion (Noon et al. 1993 ). Gossip may include received reports of incriminating information, whic h are modest intens ity hazard communications. However, the theory presented herein foresees that modest intens ity communications are least susceptible to the effects of m onitoring (H3). Neutral communi cations containing no hazard content are believed to be more common than ha zard communications. That this category is also expected to fall (H5), al lows us to predict that the ex pected reductions in hazard and neutral communications will exceed those of the expected increa se in denials, providing H6: H6: Organizational members engage in a lo wer volume of co mmunications under organizational CMC monitoring. The research model is summarized in Figure 4.
33 Figure 4. Research model
34 Chapter 3 Research Design This impact of CMC systems monitori ng on communications content was studied with a laboratory experiment using a 2 x 2 design consisting of organizational standard presentation and training, or neither presentation nor tr aining; and monitoring vs. no monitoring conditions for an instant messa ging discussion. Instant messaging is of interest since it has b een projected to surpass more fr equently studied communications channels as the primary real time comm unications channel for business by 2013 (Kerner 2007). The subjects, measures and research method are described in this chapter. Subjects Undergraduate subjects were recruited by circulation of a sign-up sheet in upper level information systems courses for the pilot study (n=40), and in upper level courses in each of five disciplines for the main study (n=86): marketing, psychology, public relations, nursing, and history. Subjects in each of the two recruitments were treated as a single pool, and were randomly assigned to tr eatment condition without regard to their source. In pilot testing, age was found to be a signi ficant factor that subjects volunteered to explain their lack of exposure to hazard communications topics. Subjects participating in the pilot also suggested differences in the hazard domain according to GPA. Country of origin is of potential interest since acad emic honesty ethics and practices vary around the world, providing a potentially extraneous source of variation in results. Year of first
35 IM use was collected and converted to a year s of experience measure to allow control for familiarity with the technol ogy if needed. Home campus and year in school were included in the demographics to control for dissimilar experiences that could cause changes in knowledge in the h azard domain if necessary. Experimental manipulations Two manipulations were employed: mon itoring by an organizational authority and presentation of and training on an organizational standard. Monitoring manipulation A basic, text-based instant messaging system that provides rehearsability and which is used non-anonymously and synchronously for dyadic communications, possibly including parallel communicati ons, was used for the experiment. Instant messaging is of rapidly increasing importance for organizations, and the system chosen for the investigation mimics instant messaging syst ems features typical in organizational environments. A benefit of this system is its support for the monitoring manipulation employed. Half of subjects, those in the non-monitoring c ondition, were told that no one would have access to their records be yond the students coll ecting data, and that while the professor assigning participation in the activity would be given their names as a record for course credit, the data the subjects pr ovided would not be shar ed with the professor. The other half of subjects, those in the treatmen t condition, were told that while the data collected was for a student group project, the universitys Office of Academic Honesty ha d asked for copies of the session transcripts. This is a fictitious offi ce. Subjects in this co ndition were then asked
36 for their permission to have their transcripts in cluded in the gr oup that office receives. Upon agreeing to monitoring, subjects were informed that a recording device was being turned on, and a bright red reel-to-reel recorder a ppeared below the chat window as a visual reminder that the session was being recorded for the monitors use. In the main study, the subjects attention was called to the recorder image, an addition to the protocol since the pilot test to increase the manipulation effect A screen shot is provided in Figure 5.
37 Figure 5. Monitored condition screen (Adapted from (Holton et al. 2008)) Organizational standard manipulation In consideration of corporate standard s for IM content, dozens of electronic communications policies were collected from public sources. Normally, one or more communications sections were found within doc uments or on websites with labels like
38 these: electronic communication polic y, acceptable standards of communication, appropriate use of electronic communicati on and technology, human resources policies, business conduct and ethics, network use polic y, computer use policy, appropriate use of IT. Policies that appeared not to have been updated in more than two years; that were issued by Internet service providers, libraries or public use facilities ; that addressed only communications between organizational member s and the public; that limited themselves to wireless communications; and that were samples or templates rather than issued policies were excluded from consideration. Ten policies passing thes e exclusion criteria were randomly selected from the original convenience sample for detailed analysis ("Acceptable computer use policy" 2007; "A cceptable use practice" 2006; "Business Conduct and Ethics" 2007; "Computer use po licy for CDI Corporation and its related companies" 2006; "Computer use policy: Po licy restricting personal use of employer's computer"; "Computer, email and Intern et use policy"; "E-mail and electronic communication policy" 2006; "Electronic co mmunication policy for all Volt entities, affiliates, subsidiaries and divisions"; "Staff acceptable computer use policy" 2006; "Statewide policy: Appropriate use of elec tronic communication and technology" 2006). While the communications dire ctives of the policies vari ed widely, all ten stated that communications systems are subject to monitoring. Four provided no restrictions on the purposes of monitoring, while two others ex plicitly stated that communications could be monitored for any reason. Of the other four, the following monitoring motivations were given: maintaining system integrity a nd ensuring that users are using the system responsibly, confirming policy compliance, confirming law and policy compliance, and for security and network maintenance.
39 All but one of the policie s specifically mentioned e-mail communications among those governed by the electronic communications policy, and eight of the ten specifically listed non-email communications systems as we ll. However, only two mentioned instant messaging, and one those prohibited use of IM (Twenty percent of sampled policies addressing IM is similar to the 31 percent of companies reporting having an IM policy in place in the 2006 AMA survey on this topic ("AMA Workplace CMC Survey" 2006)) Prohibiting IM may be a tempting option for companies given the security concerns it raises. When it is not prohibited, explicitly addressing it seems warranted given how easily unrestricted use can lead to harmful disc losure of data and other security lapses (LeClaire 2006; Macavinta 2007; Muse 2005). All ten policies provided lists of prohibited communicati ons, with violations of law (copyright, trademark, fraud, threats, ha rassment), disclosure of proprietary information, concealing or misleading a bout the senders identity, commercial communications unrelated to the employer, obs cene content, and avoiding spam being the most common restrictions. That anonymous CMC was typically prohibited by the policies suggests that additional research in to non-anonymous instantiations, such as the IM investigation undertaken herein, is merited. Two polic ies prohibited communications contrary to company interests separate from these general categories, including communications contrary to [the company] , and communications on public forums that are in any way disruptive or harmful to the reputation or business. Three noted that communications could be subject to legal disc losure requirements. Despite often lengthy lists of prohibited communications behaviors, one policy noted that no policy can lay down rules to cover every possible situation.
40 While emerging technologies offer myri ad communications benefits, little was included in the ten policies on desirable comm unications, with much shorter lists of appropriate communications qua lities provided (e.g. of high ethical standards or in accordance with the law). Each of the policies allowed lim ited personal use of electronic communications channels. In addition to thes e general guidelines, one policy noted that electronic communications should be used for collaboration, while two others urged communications that use tone and words [t hat] would not cause embarrassment to themselves or the Company if the message were made public, and those that, will reflect favorably on the Company and on the employee. In the case of hazard communications, it is often unclear exactly which communications might be considered contrary to company interests. For instance, one of the more specific of the policies reviewed states both that electronic communications must adhere to high ethical standards and th at they should not cause embarrassment. The no embarrassment directive is similar to the potential incrimination criterion in the hazard communications definition. If unethical behavior were known to occur, how the company would prefer for an employee to co mmunicate about it over CMC is unclear. On the one hand, promoting ethical be havior should not cause the company embarrassment, suggesting CMC aimed at cu rtailing the offending behavior might be welcomed. On the other, the fact that the unethical behavior took place might itself be considered to be potentially embarrassing. Since organizational communi cations policies are often ambiguous with regard to hazard communications, but do sometimes provide guidance, an organizational standard was crafted for presentation to half of the subjects in each the monitored and non-
41 monitored treatment conditions. To encourag e high power while retaining realism, the organizational communications policy presented in the experiment was very similar to the most restrictive of the policies sampled. It was distilled down to just those terms most relevant to hazard communications. The experimental CMC policy did not mention monitoring as that aspect is communicated separately in the m onitoring manipulation. The policy first generally presents the need to avoid certain inappropriate communications, and then specifically require s that communications reflect positively on the organization. At the time of the pretest, subjects in the organizational standard condition were told that the following policy governs electronic communications, including those over instant messaging: Electronic communications like email and instant messaging are considered to be documents subject to legal discovery (subpoena) and information requests under Floridas government in the sunshine law in the same way as contracts and memos. Care should be taken to avoid unprofession al, unethical, and unlawful electronic communications. The privacy of electronic communications cannot be guaranteed. All electronic communications should reflect positi vely on the University of South Florida. Subjects randomly selected to receive this organizational communications policy were presented with three tr aining items to demonstrate a nd test their understanding of the intended restrictive application of the pol icy. For each, they were asked to consider whether the instant messaging c onversation shown is in violation of the just provided policy. The policy remained on the screen at al l times for reference. A pilot test with 17 subjects not included in the e xperiment confirmed the need for feedback on the intended restrictive applicati on of the policy as they tended not to score each example as a
42 violation. Feedback on the appropriateness of each sample communication under the given policy was presented only after subjects had labeled a scenario as consistent with or in violation of the policy. The three examples are as follows: Example 1 : An IM conversation between student advisors. Lorelei Bell: I heard that we got some t-shirts delivered with the logos printed backwards. Know where I can get one? Genevieve Loki: They were supposed to be destroyed to protect [the organizations] image, but Calvins got some listed on eBay. Genevieve Loki: His userid is ugeek so you can search on that. Lorelei Bell: K. Thanks. Feedback: This conversation is in violat ion of the organizational communications policy provided. Genevieve has revealed that an employee used the employers property (misprinted t-shirts) for personal gain. This does not reflect positively on the company. Furthermore, the exchange reveals that a company employee risked harming the organizations image. This too fails to reflect positively on the organization per the policy, instead presenting a negative aspect of the organization. Example 2 : An IM conversation between help desk software support staff members. Mason Scott: Hey, can you help me with a caller? Mason Scott: Hes saying hes got a copy of Ca mtasia under our site license, but we dont have a site license. Delaney Peters: Whats his name? Mason Scott: Greg Bates.
43 Delaney Peters: Oh, thats Dr. Bates son. Hes got it because his dad has a copy for work use. Delaney Peters: Technically, he shouldnt ha ve a copy anywhere but on his desktop computer in his office. Just write the ticket up sayi ng its his office computer so it looks OK. Feedback: Like the last one, this conv ersation is also in violation of the organizational communications policy as it does not reflect positively on the organization. Instead it reveals wrongdoing redistribution of licensed software without a license. Delaneys advice to Mason on getting around th e company rules also reflects negatively, rather than positively, on the organization. Example 3: An IM conversation between students working at a concert venue. Julian Knudsen: Hey did you see Beyonce is coming? Lucy Bailey: Yeah sweet. Julian Knudsen: I know were not supposed to, but Im thinking I might grab one of the staff shirts and put myself on the schedule to work the concert. Lucy Bailey: Oh, I did that when J-Lo was here. Feedback: This conversation is also in violation of the organizational communications policy. Lucy is confessing to unethical behavior, and Julian is considering breaking the rules himself. Both actions reflect negatively, not positively, on the organization. Measurement Measures consist of hazard communi cations, denials, and communications volume, each gleaned from interviews; standa rd applied items; and a short self-focus
44 scale. In addition, items to confirm the succe ss of each manipulation are collected as is a control measure. Study variables Study variables include communications dependent variables measured from instant messaging session transcripts and an original scale providing a subjective estimation of the standards applied. Communications variables The communications dependent variables wer e obtained in the chat session interviews, which were coded into six h azard communications categories, a neutral beliefs category, and eight denials categories. The unit of analysis is the text entry delineated by hitting enter, or a single sentence as noted by concluding punctuation, whichever is shorter. This statement definiti on rule was used to enable more meaningful comparisons between rapid fire instant messaging statements, which are typically conversational and may not be composed into complete sentences, and longer compositions, sometimes consisting of complete paragraphs. Where samples are given in this document, a forward slash (/) indicat es that a communication was broken into a separate text entry by the s ubject at the marked point. The highest intensity hazard communi cations are negative self-disclosure communications that incriminate the pers on conveying them. These may also be considered confessional communications. M oderate intensity hazard communications are reports of observed incriminating behaviors. Language indicating the subject witnessed an incriminating act characterizes communications in this category. Modest intensity hazard communications are relayed reports of information that incriminates others, and
45 are also known as hearsay. These communi cations refer to incriminating behaviors overheard or otherwise reported by others for which the subject has no first hand knowledge. At each level of intensity, general or specific labels were applied to indicate whether a particular incriminating incident was being described or less incriminating broad statements were offered. Related to hazard communications are de nials of knowledge of incriminating information. Hard denials are subdivided in to four categories: denials that one has participated in incriminating activity, witn essed incriminating activity, or heard about incriminating activity occurring; or general denial statements that dont fit into the other categories. Based on pilot st udy observations, soft denials were added to the taxonomy. These are speculative statements that impl y the writer has no personal knowledge of hazards. They may be prefaced with, I would guess that, or Maybe they. The personal impropriety, seeing ot hers experience, heard abou t others experience, and general denial categories also apply to soft denials. In addition to hazard communication statemen ts and associated denials, a neutral beliefs category was also coded. Beliefs are informative, non-procedural statements that do not fall into the incriminating hazard comm unications or denial categories. This category specifically excludes conversationa l elements that do not convey information about the topic of the conversation. Examples of each of these categories within the study context are provided in chapter 4. Appendix 1 provides the coding scheme.
46 Standards applied Because there are no existing scales for th e stand ards applied construct, original items for this construct were created, ba sed on prior research on standard selection (reviewed in Chapter 2). Immediately fo llowing instant messaging communications data collection, subjects were asked to assess to what extent their own standards and organizational standards governed their beha vior during the instant messaging session with six original items. They provide three alternative word ings describing application of personal or organizational standards to the communications domain, such as, How much did complying with organizati onal rules for electronic commun ications affect what you said in the instant message interview? and How important was it to stick to your own views on what its OK to talk about when being interviewed over IM? The items were ordered so that odd items appl ied to organizational standard s and even numbered items applied to personal standards. For each of the items, a 7 poi nt scale with end points Not at all and A great extent was used. The items are provided in appendix 2. Control variable A review of decades of lite rature and dozens of studies has concluded that selffocused attention generally has a positive effect on the validity of self-reports, increasing correlations with direct measures by up to 0.3 or 0.4 points (Gibbons 1983). This verdicality hypothesis applies to self-reports of attitudes, cogn itions and affective states. In the pilot study, subjects in the m onitoring and non-monitoring conditions expressed similar motivations for their behavi or, similar levels of attention towards who might see transcripts of their instant message s, and similar beliefs that a monitor would
47 have no effect on their communications behavior. However, knowledge of monitoring clearly did have an effect on their behavior. Given these observations and a desire to understand the standards applied, which are thought to motivate communicat ions behavior, we use a self -focus scale to control for this source of variation (Matheson et al. 1988). Its four self-focus items are scored on seven point scales anchored by the end points extremely unc haracteristic of me, and extremely characteristic of me. The items were adapted slightly to the context, replacing the word experiment with IM discussion, and by refe rring to a potential monitor in the fourth item. (The fourth item is worded gene rically so that it applies to both monitored and unmonitored treatment c onditions.) The items are provided in appendix 3. The first two scale items are said to m easure private self-focus and have been reported to have a nearly si gnificant correlation of r=.17 (p<.11). The second item had been reverse worded, but was changed to a ffirmative wording in case confusing wording was responsible for the weak correlation. The second pair of items, said to measure public self focus, had a significant correlation of r=0.41 (p<0.01) in a previous study. As discussed in chapter 2, the di stinction between public and private self-focus is not universally accepted, and is disbelieved by th e researcher. Only the scale total was employed. Manipulation checks An open ended item assessed the success of the monitoring manipulation: To the best of your recollection, who will receive transcripts from the interview you just
48 completed? Responses to this open-ended question were compared to the treatment condition for scoring. Scoring details and an assessment are provided in chapter 4. The organizational standard training manipul ation is tested with three yes or no items. The items are written to seem rath er innocuous but they refer to incidents described during training which were said to violate the organi zational communication policy. The prompt asks, From what you re member, to discourage personal views on what the policy should be by people who were e xposed to the standard. Answering no to all three prompts provides an indication that the standard was learned. The items are provided in appendix 4 and assessed in chapter 4. Experimental procedure Upper level undergraduate st udents were recruited by their professors, who circulated a sign-up sheet in class and offere d a small extra credit incentive. The sign-up sheet also provided details of a drawing for iP ods that was open to study participants. To express interest in participating, subjects chos e an available interview time slot from the schedule and provided their names and em ail addresses for study information and reminder notices. They also co mpleted an informed consent fo rm. In the signature block, they were asked to provide a seven digit ID code, memorable to them, but likely to be unique, for survey login. Subjects login ID s were enabled and they were asked by email to complete an online survey ten days prior to their sche duled interview. The survey URL in the e-mail varied according to organizational standard condition, with treatment conditions assigne d based on a random number column in a spreadsheet. Demographics and measures for future investigations were collected on this initial survey, and for subject s in the organizational standa rd condition, the organizational
49 standard was presented and trai ning provided at the end of the survey. This activity took place on average nine days prior to the experiment. On the day of the interview, subjects scheduled for that day were randomly assigned within organizational standard condition to monitoring or non-monitoring conditions to maintain a roughly balanced 2 x 2 design. The assignment could not practically be done prior to the day of interview given continuing recruitment and rescheduling of existing appointments. Instant messaging discussion The IM system selected for the experiment is one component of the larger Elluminate Live! Academic Edition 7.0 system. In a ddition to mimicking organizational instant messaging systems, the Elluminate IM system enables the monitoring manipulation described previous ly and provides the ability to push survey web pages to subjects. Subjects logged onto the system at their appointment time using their first and last names. Using full names was ostensibly to ensure they were credited with participation. It also ensured consiste ncy with typical organizational use of instant messaging which identifies users by their full names, and provided a constant reminder that the sessions were not anonymous as in typical organizational settings. In fact, subjects full names remained on screen at all times during the session. To begin the interaction, subjects were told that the data collect ed was for a student group project. Next a survey website asking subjects to assess the seriousness of various types of cheating was pushed to the subjects, caus ing the survey to open in a separate browser window on thei r computers. The survey was created by the Center for Academic
50 Integrity ("Center for Academic Integrity As sessment" 2005) and is used for university academic honesty audits. The survey served to prime the topics that would be discussed in the interview. Next subjects were interviewed th rough instant messagi ng by one of two interviewers. The researcher served as one interviewer. The sec ond interviewer is an academic who is experienced in large scale interview-driven research projects and has completed coursework in interviewing as a research method. Training was a four step process consisting of 1) a desc ription of the process, and pr ovision of an interview guide developed during the pilot study with software procedures and the interview script; 2) observing the researcher conduct several in terviews and asking questions about the procedures, 3) conducting several interviews as the researcher looked on and provided guidance, and finally 4) conduc ting interviews without observa tion but with live access to the researcher to ask questions Most questions related to appropriate use of follow-up prompts. The researcher reviewed several transcripts from the second interviewer over the course of several weeks to ensure the tw o interviewers remained synchronized in their interviewing technique. No problems were detected. Aiding maintenance of synchronization be tween interviewers was the use of scripted discussion prompts that raised t opics related to academic dishonesty on the campus in question, but did not overtly ask for examples of incidents of cheating. The list of interview prompts is provided in Figure 6.
51 Figure 6. Instant messagi ng interview prompts Following completion of the IM interv iew, another website was pushed to subjects to collect self-focus and perceived organizational standards influence measures, to conduct manipulation checks for each of the two manipulations, and to collect measures for a future study. Pilot test findings The pilot study differed somewhat from the larger study. The main methodological difference was that while the mo nitoring impacts of primary interest were 1. Do you think cheating is a major problem on this campus? a. If yes: Why? What problems do you think cheating causes? b. If no: Why not? 2. How much cheating goes on in your classes? a. How much cheating have you seen? b. How much cheating have you heard about? c. Is there anything youve done that someone might cons ider cheating? 3. How do people cheat? What do they do? a. How do you know about it? 4. Are some methods of cheating more effective than others? a. If yes: Which ones? Why? 5. Are some methods of cheating less successful than others? a. If yes: Which ones? Why? 6. Are certain classes more likely to having cheating activity? a. If yes: Which ones? b. If yes: What makes a class more likely to have cheating going on? 7. Is cheating a group activity or something people do alone? a. Why do you think so? 8. Why do people cheat? 9. Are certain types of peopl e more likely to cheat? a. If yes: Why? b. If yes: In what ways? 10. In a course, what types of activitie s are more likely to be cheated on? a. Why those?
52 hypothesized and measured, no organizational standard manipulation was employed. Improvements based on pilot test results are su mmarized at the conclusion of this section. The pilot test included 40 subjects. Three subjects who were substantially dissimilar from the remainder of the sample were removed. One was based at a different campus than all of the others, and two were in their mid-thirties, as compared with the median age of 24. Of the older subjects, one was a first semester tr ansfer student. All three of these subjects were noteworthy for thei r stated inability to answer most questions, providing frequent responses of I dont know, Not really sure, and No idea. Two of these were in the control group, and one in the treatment group, leaving 18 in the control group and 19 in the treatment group. For the pilot test, two raters (neither of whom served as coders for the main experiment) coded the interviews into specifi c high intensity, moderate intensity, modest intensity, and denials cate gories, as follows. The highest intensity hazard communi cations are negative self-disclosure communications that incriminate the pers on conveying them. These may also be considered confessional communications. In th is context, a negative self-disclosure is subjects description of his or her own cheating behavior, su ch as, In my organizations and systems class two years ago, my friend and I didn't understand the work and some guys thought we were cute ladies and helped us with every assignment. Moderate intensity hazard communications are reports of observed incriminating behaviors. Language indicating the subjec t observed the cheating behavior described characterizes communications in this category. Samples include, i used to be in a frat they had old test and the convience of chea ting was there and available and i saw that
53 some probably 7-15 percent would, and last w eek two friends of mine got together to complete the tests. Modest intensity hazard comm unications are relayed repor ts of information that incriminates others, and are also known as hearsay. In a cheating context, hearsay statements are characterized by language that the subject heard about the behavior in question occurring. Samples include, my roommates were talking about it earlier this semester. / they are all taking internet cla sses and have used th e discussion boards to cheat, and ive heard of frie nds paying others to write them for them if they just dont write very well or dont have the time or basically just dont want to. Denials are statements that the subject did not or has not participated in, seen or heard about academic dishonesty, for example, In my 4 years here, I have never seen any cheating going on in my classes. Within each of these categories, where mu ltiple statements referred to the same incident, common incident numbers were applied. A spreadsheet-based tool was developed for this pilot study coding task (Holton 2006), and improved to support the increased granularity in the main experiment. Several features were included to encourag e coding precision and accuracy. Construct definitions were constantly at the top of the sc reen for reference. Statements were tallied automatically from the codes assigned, with an area provided to keep running tallies of manually applied incident numbers. Known or suspected coding errors like incident counts that exceeded statement counts for a part icular statement type were highlighted by logic checking formulas. These did not cons train the coders disc retion, but encouraged
54 review of codes to reveal errors. They were particularly helpful in identifying incorrect assignment of incident numbers within a given statement type. Initially, four IM transcripts were coded by a single rater, and the ratings were reviewed with a second rater. The two coders were in complete agreement on the codes applied to this small sample. Next, the coders independently coded the remaining transcripts, consulting each other to further refine th eir understanding of construct definitions when close calls were encountere d. Finally, 25 percent of the data for which agreement was not previously determined was coded by each rater. Within three hazard communications categor ies (negative self-disclosure incident statements, specific, observed incident statements, hearsay of a specific incident statements) and the specific denials category, acceptable interrater reliability was achieved (Spearman correlation= 0.996, p=0.000), with perfect agreement in three of the four classifications. Additional general hazard co mmunications and soft denial s were assessed in the larger study, along with neutral beliefs stat ements. These are described in chapter 4. The normality of each variable was consid ered was considered with histograms and the Kolmogorov-Smirnov test. Both revealed that while denials were approximately normally distributed, each category of hazard communications has left skew and a long right tail. Levenes test of equality of va riances was also assessed, with approximately equal variances found for most dependent va riables except negative self-disclosures. When only two groups are compared, in th is case monitored and non-monitored, the MANOVA analysis to be undertaken is robust with respect to each of these characteristics.
55 Descriptive statistics for each variab le, both overall and within treatment condition, are provided in table 1.
56 Table 1. Pilot Study Descriptive Statisti cs Overall and by Treatment Condition 7 Statement dependent variables NMeanSD MinMax Negative self-disclosure incident statements Non-monitored 181.172.09 06 Monitored 190.160.69 03 Total 370.651.60 06 Specific, observed incident statements Non-monitored 181.562.55 08 Monitored 191.794.09 016 Total 371.683.38 016 Heresay of a specific incident statements Non-monitored 180.000.00 00 Monitored 190.891.70 06 Total 370.461.28 06 Hard specific denial statements Non-monitored 182.331.71 06 Monitored 193.681.45 17 Total 373.031.71 07 Incident dependent variables NMeanSD MinMax Negative self-disclosure incidents Non-monitored 180.440.86 03 Monitored 190.050.23 00 Total 370.240.64 03 Specific, observed incidents Non-monitored 180.500.79 02 Monitored 190.370.68 02 Total 370.430.73 02 Heresay of a specific incidents Non-monitored 180.000.00 00 Monitored 190.370.60 02 Total 370.190.46 02 Hard specific denial incidents Non-monitored 182.221.63 06 Monitored 193.263.26 15 Total 372.761.48 06
57 Each hypothesis was assessed in two ways : through statistical comparison of treatment and control group differences in the numbers of hazard communication incidents of each type, and the numbers of statements of each type volunteered by subjects. Separate MANOVAs are appropria te given the high correlations between incidents and statements (r = 0.74-0.91), making it statistically redundant to include both in the same model. We note, however, that results were sufficiently strong to produce a significant combined MANOVA as well as individual MANOVAs on incidents (F=4.181, p=0.008 for each of four MANOVA statistics tested) and statements (F=3.767, p=0.013). H2 proposes that organizational member s make fewer higher intensity hazard communications under organizational CMC mon itoring than on unmonitored channels. This hypothesis is supported by a finding fo r both negative self-disclosure incidents (F=3.71, p=0.06) and statements (F=3.97, p=0.05). On average, one out of every two non-monitored subjects (mean of 0.44 incide nts per non-monitored subject) volunteered several statements with specific, personally incriminating information about their past, current or planned academic dishonesty. In contrast, there was a single case of a monitored subject making this type of hazard communication. H3 posits that the decrease in the frequency of high intensity hazard communications between monitored and non-mo nitored conditions is greater than the change in frequency for low intensity hazard communications. While there will be no one universally accepted standard for the degree of hazard of a particular communication, we implemented a simple scoring algorithm to determine an overall hazard intensity score for the communications of each subject to assess H3. For the initial assessment, confessional incidents, which could potentia lly bring not just shame and disapproval but
58 immediate penalty without furt her corroboration or investig ation, received a score of twenty. Given their post hoc, uncorroborated nature, which the organization and the individual can at least partially combat with equal and opposite statements, reported observation hazard communications received a score of five. On this scale, relayed incident hazard communications received a score of one. Using this initial scoring scheme, differences in hazard communications intensity were found to be significant for hazard communications incidents (F=4.37, p=0.04), but not for statements (F=2.06, p=0.16). As the relative weights are adjusted to give additional gravity to the most severe confessional hazard communications, findings be come more significant. For instance, with a weighting scheme of 50, 5 and 1 fo r confessional, observed, and relayed hazard communications, respectively, th e hazard intensity difference appears more significant (F=4.89, p=0.03 for incidents, and F=3.18, p=0.08 for statements). However, as the scale is adjusted the other way, giving more equal subjective weightings of 3, 2 and 1 to the three intensities of hazard communications findings become non-significant for both incidents (F=2.11, p=0.16) and statements (F =0.32, p=0.58). Although this analysis is necessarily subjective, we conclude that wh en monitored by an or ganizational authority, individuals do reduce the intensity of hazard communications in which they engage. The final hypothesis assessed with pilo t study data, H4, suggests that when monitored by an organizational authority, individuals will make more denials of knowledge of incriminating behaviors. Th is hypothesis was tested with separate ANOVAs for incidents and statements. Both were significant (F=5.10, p=0.03 for incidents, and F=6.70, p=0.01 for statements), providing support for this hypothesis. Pilot test results are reported in table 2.
59 Table 2. Pilot Test Results Findings on all hypotheses tested indicate sufficient power with a sample size of n=37. As the main study would introduce a new organizational training standard contrast, the target sample si ze was increased to 80. Summary of changes between pilot and main studies The following changes were made in the main study as compared with the pilot study. The monitoring manipulatio n was strengthened with noti ce called to the recording indicator. Additional data was collected and analyzed including general hazard communications and denials that did not meet the specificity threshold applied for the data recorded in this chapter; soft denial s, which provide an additional way to avoid engaging in hazard communications; neutral belief statements; and volume measures including time spent in the in terview portion of the online se ssion, total text entries, and total word count. Based on these new categories, additional coding guidelines were Statement dependent variables Type III Sum Sq Mean SqF Sig.R2Negative self-disclosure incident statements 9.4069.4063.965 0.054.102 Specific, observed incident statements 0.5060.5060.043 0.837.001 Hearsay of a specific incident statements 7.4007.4005.001 0.032.125 Hard specific denial statements 16.86816.8686.701 0.014.161 Incident dependent variables Type III Sum Sq Mean SqF Sig.R2Negative self-disclosure incidents 1.4191.4193.709 0.062.096 Specific, observed incidents 0.1600.1600.296 0.590.008 Hearsay incidents 1.2551.2556.839 0.013.163 Hard specific denial incidents 10.01510.0155.095 0.030.127
60 developed. Finally, spreadsheet intelligen ce was improved to support coding accuracy. Each of these changes is described in chapter 4.
61 Chapter 4 Results The pilot study was repeated on a larg er scale, producing support for all hypotheses. This chapter provides a report on th e suitability of the collected measures for hypothesis testing, followed by hypothesis tests of the relatio nships between them and post hoc analysis. Assessing variable adequacy Demographics Subjects in the main study were on averag e 24 years old and reported an average of 7 years of work experience (with a range of 0-35 years). 93 percent of subjects reported that they were from the United States, with one each listing Albania, Canada, Colombia, Guatemala, Haiti, and Panama as their home country. They reported an average GPA of 3.1 on a four point scale. Twelve subjects reported that the study provided their first experience using instant messaging, while they had an average of 6.5 years of experience with the technology. Other demographics are reported in table 3.
62 Table 3. Demographics One hundred three subjects completed the pre-test, with 86 of those completing the interview (83.5 percent). Of the 17 subjects lost to attr ition, 59 percent were from the United States with one each from Belize, Br azil, Canada, the Dominican Republic, Haiti, India, and Jamaica, making this group less hom ogeneous in country of origin than the experimental group, but still largely from the U.S., with all but one other subject from other parts of North and South America like those completing the study. In other respects, this group was extremely similar to the group completing the experiment with an average self-reported GPA of 3.1, an av erage age of 24, and an averag e of 6.7 years of experience using instant messaging. All were from the main campus, and 82 percent were seniors. (Years work experience was collected in a post test and therefore is not available for the attrition group.) A MANOVA predicting the qu antitative demographics from whether or not a subject completed the study was non-significant (F= 0.235, p=0.946). Thus the sample is from the same geographic region as the attrition group but more homogeneous NMinMaxMean SD Age 85185524.1 6.3 Years work experience 790357.4 6.4 Years IM experience 860156.6 4.1 Self-reported GPA 842.003.953.13 0.44 N% N % Sophomore 34 Junior 1214 Main campus 82 96 Senior 6981Regional 3 4 Grad 11 85 100 Total 85100
63 in country of origin, and in other respects, the groups appear extremely similar. Based on the available data, no reason is found to suspect attrition biased the sample. Communications variables The same three hazard communications categ ories applied in the pilot study were used in the main study, but an additional leve l of granularity was c onsidered, consistent with the taxonomy presented in chapter 2: At each level of in tensity, coders were directed to label communications as general or specific (whereas only specific statements were considered in the pilot study). A ge neral statement is sweeping with no hooks to isolate a particular incident of cheating, such as people just copy and paste from spark notes, and People use other st udents papers for classes. / and simply re-submit them as their own. A specific statement is one that, if we had full details, could be tied back to a place, time, and person or people. Specific codes were applied when the behavior was described as a discrete incide nt or incidents (people programmed formulas into their calculators, Was a witness to it last week ). The existence of discrete incidents was often made clear by indicators of quantity (one time, a couple of times), details about the subject or course (e.g. large math cl asses, managerial accounting), or by naming particular perpetrators (my friend, my roommate). To provide additional understanding of the effect predicted in H4, hard denials of knowledge of incriminating information were subdivided into four categories: general (e.g. As far as I know, cheating isnt a probl em here), personal impropriety (e.g. I have never been involved with any situation that cheating has been an issue on campus), observed (ive never seen someone cheat on a test), and heard about (I have never heard about anyone cheating in a ny of my classes so far).
64 Based on pilot study observations of an additional type of hazard communications avoidance, a soft denials category was added to the taxonomy. Soft denials are speculative statements that imply the write r has no personal knowledge of hazards but avoid saying so directly. They may be pr efaced with, I would guess that, or Maybe they. Raters were asked to code soft de nials into the same categories used for hard denials: general, personal impropriety, observed, and heard about. Special coding guidance was given for a potential close call Statements similar to, I dont know besides what I already told you, were not coded as denials as long as some other comment referred to in a statem ent like this one did address the question asked. However, if a subject deflected a question with a comm ent of this nature and no other comment did address the question, such a statem ent was coded as a hard denial. In addition to hazard communication statemen ts and associated denials, a neutral beliefs category was coded to test H5. Belie fs are statements informative about the topic at hand but which do not provide incrimina ting information. This category excludes procedural statements as non-informative. Examples include, its easy to obtain the answers from other students who have prev iously taken the course, and looking at someones test paer is easy and less likl eyto be cought than forging a paper. To prepare the data for coders, the instant messaging transcripts were anonymized, stripped of initial pr ocedural discussion including indi cators of treatment condition, and parsed before being pasted into spreadsheet s, one conversation per worksheet, and one line per IM statement (as delineated by the send command). Separate columns recorded the date, time, message author, and message cont ent. Up to six code pairs, one item for the statement type and one for the hazard inci dent or denial inci dent number could be
65 applied by default, and coders were asked to carry any additional code s into free cells in other rows, annotating this practi ce in the comments field if used. The coding worksheet developed for the pilot study was augmented with the additional categories. Prototypical statemen ts and other coding guida nce were also added in rollover pop-up boxes to reinforce the defi nitions and encourage consistency between raters. Two experienced behavioral research coders unfamiliar with study hypotheses and manipulation procedures, and blind to treatment condition were hired. They were not involved in any other portion of the study. In recognition that this is a difficult coding task, extensive training was undertaken. Initia l training of coders consisted of walking through a coding guidance document explaining construct definitions and examples. Next use of the coding worksheet, including features to promote coding accuracy, was explained. Subsequently each coder rated part of an interview in discussion with the trainer. Finally, each of the first ten inte rviews coded by each rater was reviewed by the trainer with feedback provided. At the conclusi on of this process, both raters and trainer agreed that application of the coding cat egories was well understood by the raters. Following training, periodic checks were made of selected code d interviews, with coaching provided where deficiencies or in consistencies between coders were found. Coders were not asked to change ratings, but rather to consider suspect ratings in light of apparent discrepancies with the construc t definitions or disagreements between themselves. They were asked to explai n their reasoning where they felt questioned ratings were accurate.
66 A number of interrater reliability m easures were considered. Since the determination of what constitutes a statement is itself an issue for coder judgment, the interview was the unit of analysis for determination of interrater reliability, the same unit of analysis used for hypothesis testing. Cohens Kappa, perhaps the most common measure of interrater reliability, is insufficien t for assessing this activity since it considers that the data are assigned to mutually exclusiv e categories. In this case, statements may be assigned to multiple categories (for instance because they contain a description of cheating and a belief that justifies the pr actice), and the codes for an interview are compared across the categories. Pearson correlations are sometimes chosen to assess interrater reliability in this case as they are easily understood, but they suffer from the limitation that they rely on normally distributed data. From the sample of 86 interviews, we find zero skew in some categories as well as long high end ta ils. Intraclass correlations, another popular choice, also assume normality. Spearman correlations are free from this assumption. For sake of comp arison, all three of these measures, Pearson, Spearman, and intraclass correlations, are reporte d by ratings category in table 4. Results by interview are given in table 5. High levels of agreement among the coders in all but one category give confidence both in the ratings a nd that the constructs are di stinct and well defined. The final category is one for which one coder f ound three examples across the 86 interviews, while none were identified by the other coder. As there were no findings associated with this variable, soft denials of seeing others experience, it is not further considered.
67 Table 4. Interrater Reliability Correlation Assessment Pearson*Spearman*Intraclass* Negative self-disclosure incident statements 0.94210.9176 0.9701 Neg. self-disclosures unrelated to single incident 0.84980.7662 0.9161 Specific, observed incident statements 0.88250.7855 0.9326 Reported obs. not related to single incident 0.96200.8711 0.9806 Heresay of a specific incident 0.91990.7986 0.9580 Heresay not related to a specific incident 0.90810.7674 0.9498 Beliefs expressed 0.85300.8218 0.9192 Hard denials in general 0.76130.6422 0.8644 Soft denials in general 0.79590.7573 0.8846 Hard denials of personal impropriety 0.90320.8791 0.9029 Soft denials of persona l impropriety 0.80820.7889 0.8082 Hard denials of seeing others' experience 0.85080.8508 0.8507 Soft denials of seeing others' experience -0.0852-0.1085 ** Hard denials of hearing about others' experience 0.79860.7641 0.8848 Soft denials of hearing about others' experience 1.00001.0000 1.0000 *Except for soft denials of seeing others' experience p values for each statistic are 0 out to at least 10 significant digits. ** Incalculable due to zero va riance in the sample from the rater not applying this code.
68 Table 5. Pearson Correlations Between Ratings Arrays by Interview 1 0.995 30 0.97059 0.952 2 0.984 31 0.99360 0.998 3 0.996 32 0.99761 0.995 4 0.995 33 0.99862 0.997 5 0.996 34 0.99563 0.992 6 0.993 35 0.99764 0.989 7 0.981 36 0.98665 0.995 8 0.982 37 0.91766 0.995 9 0.992 38 0.99667 0.998 10 0.997 39 0.99768 0.998 11 0.980 40 0.98569 0.997 12 0.993 41 0.99570 0.991 13 0.995 42 0.97071 0.996 14 0.998 43 0.99872 0.992 15 0.989 44 0.98073 0.981 16 0.989 45 0.99374 0.957 17 0.995 46 0.98875 0.974 18 0.996 47 0.99176 0.983 19 0.994 48 0.98977 0.999 20 0.995 49 0.99478 0.975 21 0.984 50 0.99479 0.996 22 0.993 51 0.92280 0.990 23 0.998 52 0.97581 0.998 24 0.992 53 0.97882 0.997 25 0.995 54 0.93883 0.999 26 0.990 55 0.98884 0.974 27 0.992 56 0.88385 0.977 28 0.991 57 0.96886 0.990 29 0.995 58 0.992 Twenty percent of the data was reviewed with the coders with alignment sought for this subset. One of the two coders was found to be better than the other at making fine construct distinctions from context, while the second relied more legalistically on the use of certain terms to determine categoriza tions. For instance, an opinion prefaced with
69 I guess Id say, can indicate l ack of knowledge (which should be coded as a soft denial) or a judgment based on facts (which should be coded as a belief). This determination is best made after considering other statemen ts by the subject in the chat session. The weaker coder typically coded su ch statements as soft denials even after coaching, while the stronger coder annotated determinati ons with evidence from elsewhere in the transcript to support a belief interpretation where appropriate. The weaker coder was able to support belief interpretations when aske d to consider that code, but rarely spotted them without prompting. After completion of the coding review for 17 interviews (20 percent), the stronger coders ratings were judged by the trainer to be consistently correct, or in the case of close calls, well-defended. Relatively few discrepancies between coders were found. When discrepancies were returned to code rs for review, the stronger coder sometimes more fully explained coding decisions but was justifiably reluctant to change them, while the weaker coder typically changed ratings without discussion. For that reason, and given limitations on budget and time, once it wa s determined that interrater reliability was satisfactory, the coding review was suspe nded and determinations of the coder who more fully considered statements with their context in each discussion were selected for additional analysis. In addition to the interview content vari ables, the dependent variables collected from the instant messaging sessions included le ngth in minutes of the interview portion of each online session (which was the only data provided to the coders) as captured by the chat software and calculated on the interv iew spreadsheet; the total number of text entries (delineated by returns), tallied auto matically on the interview spreadsheet; and
70 total word count of each interview, which wa s also tallied automatically in the coding workbook. Lengths of two interviews for wh ich time stamps were not captured by the chat system (once due to a system problem and once due to interviewer error) were estimated from the timestamps on surveys the subjects took immediatel y before and after those interviews along with average completion times for interview preliminaries and the second survey. Descriptive statistics for all variables resulting from the interviews follow in table 6.
71 Table 6. Communications Statement, Volu me, and Incident Descriptive Statistics The communications belief statements, word count, text entry count and minutes variables are normally distributed as meas ured by non-significant Kolmorgorov-Smirnoff tests (which indicate we are unable to reje ct the normality hypothesis, with values of Statements NMinMax Mean SD Negative self-disclosure incident statements 86011 1.05 2.20 Neg. self-disclosures unrelated to single incident 8606 0.48 1.04 Specific, observed incident statements 86018 0.84 2.44 Reported obs. not related to single incident 86043 4.27 7.65 Heresay of a specific incident 86036 1.42 5.16 Heresay not related to a sp ecific incident 86060 4.56 7.80 Beliefs expressed 861097 33.66 12.49 Hard denials in general 8608 1.80 1.67 Soft denials in general 8608 1.99 1.87 Hard denials of personal impropriety 8608 1.14 1.52 Soft denials of personal impropriety 86011 0.64 1.47 Hard denials of seeing others' experience 86010 1.37 1.66 Soft denials of seeing others' experience 8601 0.01 0.11 Hard denials of hearing about others' experience 8608 0.97 1.20 Volume Words 865832937 1218.91 355.02 Text entries 8645167 71.31 20.89 Minutes 8616.189.4 35.7 12.1 Incidents Specific negative self-disclosure incidents 8604 0.407 0.803 General self-disclosures 8602 0.419 0.563 Specific, observed incidents 8604 0.384 0.738 General reported observations 86020 1.942 3.019 Specific hearsay incidents 8605 0.291 0.795 General reported hearsay 86019 2.616 3.167 Beliefs expressed distinct topics 86854 18.616 6.149 Hard denials general 8606 1.128 1.370 Soft denials general 8606 2.105 1.624 Hard denials of personal impropriety 8605 0.872 0.943 Soft denials of personal impropriety 8602 0.233 0.452 Hard denials of seeing others' experience 8601 0.012 0.108 Soft denials of seeing others' experience 8604 0.779 0.817 Hard denials of hearing others' experience 8601 0.035 0.185
72 p=0.310, p=0.073, p=0.100, and p=0.196, respectively). General negative self-disclosure statements, specific observation statements, sp ecific hearsay statements, belief statements, hard general denials, hard and soft denial s of personal impropriety, hard denials of observation, and hard and soft hearsay denials have approximately equal variances across monitoring conditions as measured by Levenes test of equality of er ror variances (with p values ranging from 0.096-0.818). Normality a nd equality of variance across monitoring conditions are not characterist ics of the other statements dependent variables. Communications incident count s are decidedly non-normal. With few exceptions, zero is the most frequent value. Kolm ogorov-Smirnoff test values range from p=0.0000.018. Many of the incident counts show approximately equal variances across monitoring conditions, including general and specific hearsay communications, beliefs, specific general denials, specific and genera l personal impropriety denials, specific and general observation denials, and general h earsay denials (with p values ranging from 0.068-0.905). While both normality and equality of equality of error variance are assumed by the MANCOVA analysis technique with just two groups and large n, conforming to these assumptions is no t a requirement for reliable results. That many of the communications variable s are skewed towards zero with a long right end tail is to be ex pected and does not reflect a measurement weakness. Many subjects will have no information to offer pe rtaining to some of the hazard categories, particularly during a short duration task; and under monitoring, suppression is expected. Since incidents group related statements, there will necessarily be an even more severely curtailed distribution for incident variable s. Fortunately, we can rely on MANCOVA robustness for the case at hand. Even with the zero-anchored distributions, we note that
73 the means and ranges observed indicate that practical and predictable differences between subjects on these variables typically exist, wi th statistical significance to be assessed according to the hypotheses. The data was also examined for the pres ence of outliers, a consideration which should be made with some knowledge of the expected population distribution. In organizational settings, whistleblowing is rare. According to the taxonomy and experimental results, negative self-disclosure is rarer still. In one university setting, only 3 percent of surveyed students said they had ever reported cheating to an official (Burton et al. 1995). In this stud y, the period for reporting wa s quite limited, thus many zero values are anticipated. Further, the tendenc y to self-disclose has both state and trait components (Stritzke et al. 2004), with the former suggesting high variance between treatment conditions and the latter suggesting high variance within subjects. Once they have decided to whistle-blow, whistleblowers are likely to be repeat offenders (Sawyer et al. 2006). Together, these facts support an expectation of wide variance. The high incidence of zero values and wide variance observed are congruent with the expected variable distributions, thus no outliers are identified. Manipulation checks Two manipulation checks are the next basi s on which eliminating data from the sample was considered. To test the mon itoring manipulation, a post test item asked, To the best of your recollection, who will receiv e transcripts from the interview you just completed? For the pilot test, a very na rrow interpretation of re sponses to this item was applied and 73 percent of subjects passe d the manipulation chec k. In the larger experiment, the monitoring manipulation was strengthened by asking monitored subjects
74 to confirm that they saw the reel to reel recorder that indica tes a transcript is being made for the Office of Academic Honesty when monitoring is turned on. To score the manipulation check, subjects in the non-mon itored condition not indicating a university department or office were considered passes, as were subjects in the monitored condition indicating a university official would receive copies of transcripts. Three subjects not responding to the manipulation check and three subjects giving incorrect responses were excluded on this basis, leaving 80 subjects in the study, for a manipulation success rate of 93 percent. The organizational standard manipulati on had a simpler interpretation: Every subject in the organizational standard conditi on correctly answered th at all three of the behaviors described were prohi bited under the policy that had been presented to them, indicating a successful manipul ation. (The group not receiving the organizational standard included subjects who rated all three behaviors as allowed under the policy, those who judged that one or two behaviors were allowed, and those who said that all were prohibited.) Some subjects took several days to complete the sta ndard training after being given online access to it, shortening the standa rd-communications capture interval, while others rescheduled their interviews, lengtheni ng this interval. This variation provides a means of testing not only whether standard training impacts communications with and without monitoring, but also the extent to which the recency of presentation of an organizational communications standard imp acts the standards effects. Since the standard-communications capture interval is expected to be a more powerful predictor of communications when it is low, trending towa rds zero when it is hi gh, the inverse of the
75 interval in days was used in initial mode ls. This measure approached significance without crossing the threshold. A second transformation allowed for the possibility that the recency effect is underre presented with the inverse of days measure. Additional investigations considered an invers e square transformation, 1/(interval2), a variable called standard recency. Higher values of this variable indicate more recent presentation of the organizational standard. Using this measure, an interval of one week would have four times the weight of an interval of two week s. The investigations using this variable found statistically signif icant results and are reported in the hypothesis tests. Descriptive statistics for standard recency appear in table 7. (This table and all subsequent tables exclude data for subjects who failed the monitoring manipulation check.) Table 7. Standard Recency Descriptive Statistics N 43 Min 0.0015 Max 1.0000 Mean0.0431 SD 0.1504 (N is equal to approximately half of the da ta sample since the statistic is irrelevant for those not in the organi zational standard group.) This variable is to be applied as a cova riate, so outliers may be influential. A boxplot revealed an apparently legitimate but unusual data point. The subject completed the organizational standard training one day pr ior to the experiment. The subject number was noted for further consideration at the time of hypothesis testing. The cell sizes for the two conditions tested are nearly, but im perfectly equal due to attrition and exclusion on the basis of th e monitoring manipulation check (table 8).
76 Table 8. Number of Subjects by Treatment Condition Standards applied The six items assessing the extent to which personal and organizational standards applied (described in chapter 3 and provi ded in the appendix) were combined to determine perceived relative influence of th ese standards. Items 1, 3 and 5 of the standards items, which assess personal standa rds influence, were summated. Separately items 2, 4 and 6, which assess organizational standards influence, were summated. Descriptive statistics for these summated scal es appear in table 9. Both minimum and maximum scores were realized for each scale when subjects rated each of the three items in a scale 1 or 7. That the means of each sc ale were more than one standard deviation from the scale end points provides an indication that range restriction did not impair scale validity. Table 9. Organizational and Personal Sta ndards Influence Descriptive Statistics NMinMaxMean SD Org Standards Influence 8032112.5 4.4 Personal Standards Influence 8032113.2 4.6 The personal influence scale had a Cronb achs alpha of 0.662. The organizational standard influence scale had a Cronbachs al pha of 0.657. Both are thus considered to have sufficient internal consiste ncy reliability. These scales ar e used in this form only in Trained on organizational standard Not trained organizational standardTotal Monitored 21 1839 Unmonitored 22 1941 Total 43 3780
77 confirmatory factor analysis which has no distributional assumptions. Confirmatory factor analysis finds that each item load s as predicted and without crossloadings. Together these indicators provide evidence of acceptable convergent and discriminant validity. Factor loadings ar e provided in table 10. Table 10. Standards Applied Factor Loadings Personal factor Org factor How important was it to stick to your own views on what its OK to talk about when being interviewed over IM? 0.828 -0.095 How much did your own thoughts about how to use IM to discuss various topics affect what you said during the interview? 0.808 0.275 To what extent did your own personal standards for IM communications impact the IM discussion you just had? 0.652 -0.088 To what extent did USF's organizational standard impact the IM discussion you just had? -0.098 0.887 How much did complying with or ganizational rule s for electronic communications affect what you said in the instant message interview? -0.109 0.825 How great a role did the organi zations acceptable use policy for its communications networks pl ay in how you handled the IM discussion? 0.308 0.557 Eigenvalue 1.793 1.957 Percent of variance explained 29.879 32.611 To examine the relative influence of personal and organizational standards on communications, a ratio of the two summated scales was taken, with organizational standards in the numerator. This measure, which approximates a continuous variable, is different from the typical ways in which self-awareness theory studies have been operationalized. It is most co mmon for subjects to be presen ted with a single standard, or
78 to be placed into groups based on scores on a values measure. However, standards have been described as unique, personal, and conf licting, (Silvia et al 2001) characteristics that are inconsistent with these operationalizations. In this case, rather than measuring subjects on a single standard that fails to capture uni que, personal and conflicting motivations for behavioral change, subjects ar e asked to self-assess th e roles of personal and organizational standards as they define them The ratio of organizational to personal standards is a measure that captures relative influence of the standards A ratio of more than one indicates that organization st andards were reported to be governing communications behavior to a greater extent than personal standards. Descriptive statistics are provi ded in table 11. Table 11. Organization:Personal Standards Influence Ratio Descriptive Statistics N 80 Mean 1.12 S.D. 0.78 Minimum 0.17 Maximum 5.00 Levenes test confirms reasonably eq ual variance across monitoring groups for this variable (Levenes te st statistic=0.256, p=0.615). While a Kolmogorov-Smirnov test rejects a normality hypothesis, the variable produced in this way has a mound-shaped distribution approximating normalit y. As might be expected, or ganizational standards are, on average, attributed to have greater influence, which is reflected in the mean of 1.12 and right tail on the distribution.
79 Control variable The self-focus scale employed (Matheson et al. 1988) has acceptable internal consistency reliability as assessed by a Cronbachs alpha of 0.736. Descriptive statistics are provided in table 12. Subjects did use bot h extremes of the scale, with a minimum summated score of 4 produced when all items were scored as extremely uncharacteristic of me and a maximum score of 28 produced wh en all items were scored as extremely characteristic of me. That the endpoints are more than a full standard deviation from the mean score of 21 suggests range restricti on did not impair the scales operation. The scales author proposed that two subscales exist, but as discussed previously, the researcher believes theory supports the existenc e of just one. Explor atory factor analysis finds only one principle component with an eigenvalue greater than one, with a sharp drop in the scree plot after that point, suppor ting the interpretation that the four items represent a unitary scale. Table 12. Self-Focus Scale Descriptive Statistics N 80 Mean 20.88 S.D. 5.00 Minimum 4.00 Maximum 28.00 A Kolmolvorov-Smirnoff test is unable to reject the hypothesis that the distribution of this scale is normal (p=0.131). As this variable will be used in a model with the monitoring variable, we also check for equality of variances. Levenes test indicates marginal rejection of the hypothesi s that variances are equal (p=0.084). Since this statistic will be applied as a covari ate in a MANCOVA model with the monitoring
80 factor, we check for outliers, to which this test is sensitive. A boxplot identified one potential outlier. The observ ation was determined to contai n some suspicious values, as if the subject were choosing values for conve nience rather than afte r consideration. The subject number was noted for considerat ion at time of hypothesis evaluation. Evaluation of the hypotheses H1 posits that organizational monitoring increases the regulatory influence of perceived organizational communications standards on computer mediated communications content relative to persona l standards. The initial ANOVA predicting this governing standards ratio from monitori ng condition was non-significant. As noted in chapter 3, subjects in the pilot test ha d difficulty accurately reporting their motivations for communication changes. Self-focused attention generally has a strong, positive effect on the validity of self-reports such as this one (Gibbons 1983). Furthe r, self-focus is the mechanism through which monitoring is theorized to cause downstream responses, including the change in the relative influence of organizational standard s. For this reason, a new model including a four item state self-focus summated scale (Matheson et al. 1988) to control for this extraneous source of variation, and an in teraction term with monitoring was also tested given the theoretical relations hip between the constructs. This model was significant at p=0.013, with significant effect s for the monitoring manipulation and its interaction with the self-foc us measure (table 13).
81 Table 13. Factors in Relative Importance As cribed to Personal and Organizational Communications Standards Type III Sum SqMean SqFSig. Corrected Model 6.27852.09283.85650.0126 Intercept 10.997510.997520.26540.0000 Monitoring 3.79613.79616.99520.0099 Self-focus 1.12321.12322.06970.1544 Mon x SF Interaction 4.32514.32517.97000.0061 Error 41.24330.5427 Total 147.6374 Corrected Total 47.5217 The estimated marginal means of this m odel, holding self-focus constant at its mean value, are 1.08 for the non-monitored condition, and 1.15 for the monitored condition, indicating higher perceived influen ce of organizational st andards relative to personal standards on communica tions for the monitored gro up (figure 7). Consistent with theory, the model supports an interpretation that as self focus increases, the effect of monitoring on the relationship between orga nizational standards and communications outcomes increases, and reports about this relationship become more accurate.
82 Figure 7. Relative influence of orga nizational and personal standards The model was run again without the susp ected outlier identified by the box plot upon initial consideration of the self-focus scale. Results were slightly stronger with a higher p value for Levenes test, a lower model p value of 0.008 vs. 0.013, more variance explained with an R2 of 14.4 vs. 13.2 with this observation, and an additional two thousandths of one point separa ting the estimated marginal means. Results remained in the projected directions. H1 thus receives support, but only when self-focus effects are considered. H2, H3 and H4 all consider changes in communications variables in the presence of monitoring. All are assessed first with MANCOVA using the monitoring condition factor and standard training recency covariate to predict statement dependent variables. The model is presented first, after which the findings by hypothesi s are examined.
83 All four omnibus statistics are signi ficant for monitoring (F=2.921, p=0.002) and standard recency (F=2.230, p=0.014) as shown in table 14. Table 14. Impact of Monitoring and Standard Presentation Recency on Hazard Communications Statement Content ValueFSig. Intercept Pillai's Trace0.93459.4060.000 Wilks' Lambda0.06659.4060.000 Hotelling's Trace14.14459.4060.000 Roy's Largest Root14.14459.4060.000 Monitoring Pillai's Trace0.4102.9210.002 Wilks' Lambda0.5902.9210.002 Hotelling's Trace0.6952.9210.002 Roy's Largest Root0.6952.9210.002 Pillai's Trace0.3472.2300.014 Standard Recency Wilks' Lambda0.6532.2300.014 Hotelling's Trace0.5312.2300.014 Roy's Largest Root0.5312.2300.014 Significant or marginally significant di fferences were found on the basis of monitoring condition for specific negative self -disclosures, general observed statements, general hearsay, beliefs expressed, hard denial s of personal impropriety, and hard denials of hearing about others experience (table 15).
84 Table 15. Significance of Monitoring on Statement Dependent Variables Type III Sum SqMean Sq FSig. Specific negative self-disclosure statements 39.96139.961 8.6070.004 General negative self-disclosure statements 1.1081.108 1.6250.206 Specific, observed statements 0.1270.127 0.0200.889 General observed statements 328.466328.466 5.5270.021 Specific hearsay statements 2.0612.061 0.0730.788 General hearsay statements 277.998277.998 4.4100.039 Beliefs expressed statements 1417.2251417.225 9.6840.003 Hard denial statements in general 0.2830.283 0.0970.756 Soft denial statements in general 7.4677.467 2.0510.156 Hard denial statements of personal impropriety 7.9857.985 3.3400.072 Soft denial statements of personal impropriety 1.3211.321 0.5640.455 Hard denial statements of seeing others' exp 1.9371.937 0.6650.417 Soft denial statements of seeing others' exp 0.0130.013 1.0140.317 Hard denial statements hearing others' exp 5.6265.626 4.2000.044 Soft denial statements of hearing others' exp 0.0140.014 0.3710.544 Next the analysis was repeated for incident level variables. This time, standard presentation recency was not significant a nd so was dropped from the model, which is significant for monitoring (table 15) (F=1.857, p=0.045). Table 16. Impact of Monitoring on Ha zard Communications Incident Content ValueF Sig. Intercept Pillai's Trace 0.95081.917 0.000 Wilks' Lambda 0.05081.917 0.000 Hotelling's Trace 19.19981.917 0.000 Roy's Largest Root 19.19981.917 0.000 Monitoring Pillai's Trace 0.3031.857 0.045 Wilks' Lambda 0.6971.857 0.045 Hotelling's Trace 0.4351.857 0.045 Roy's Largest Root 0.4351.857 0.045
85 The incidents model finds si gnificant or marginally si gnificant differences on the basis of monitoring for specific and general negative self-disclosures, general observed incidents, beliefs expressed, soft general denial s, hard personal impropriety denials, soft denials of seeing others experi ence, and hard denials of he aring about others experience (table 17). Table 17. Significance of Monitoring on Incident Dependent Variables Type III Sum SqMean Sq FSig. Specific negative self-disclosure incidents 3.2723.272 5.3050.024 General negative self-disclosure incidents 1.2951.295 4.1920.044 Specific, observed incidents 1.2951.295 2.2910.134 General observed incidents 51.23251.232 5.5750.021 Specific hearsay incidents 0.2590.259 0.4060.526 General hearsay incidents 18.73318.733 1.8140.182 Beliefs expressed incidents 219.425219.425 5.9540.017 Hard denial incidents in general 3.7543.754 1.9380.168 Soft denial incidents in general 10.05310.053 3.8910.052 Hard denial incidents of personal impropriety 2.7122.712 2.9890.088 Soft denial incidents of personal impropriety 0.0280.028 0.1290.720 Hard denial incidents of seeing others' experience 0.0000.000 0.0000.992 Soft denial incidents of seeing others' experience 2.3132.313 3.5740.062 Hard denial incidents of hearing about others' exp 5.2445.244 7.2720.009 Soft denial incidents of hearing about others' exp 0.0140.014 0.3920.533 H2, which posits that organizational memb ers make fewer higher intensity hazard communications under organizati onal CMC monitoring than on unmonitored channels, is supported by a significant finding for specific negative self-disclosures statements (F=8.607, p=0.004) and incidents (F=5.035, p=0.024), as well as general negative selfdisclosure incidents (F =4.192, 0=0.044), which are the highest intensity hazard communications. Non-monitored subjects ma de on average 1.76 specific negative self-
86 disclosures statements, while non-monitored s ubjects made an average of just 0.36. Nonmonitored subjects delivered these in an aver age of 0.61 incidents for non-monitored subjects, and 0.21 for those in the monitored condition. Non-monitored subjects also engaged in more general negative self-disclosur e incidents, at a rate of one in every two interviews on average (mean = 0.54) vs. a ra te of about one in four interviews for monitored subjects (mean = 0.28). Since there was no significant finding on specific negative self-disclosure incidents, another model was run grouping all statements an d incidents by hazard intensity category and all denials. This model was significant with a fewer negative self-disclosures made by monitored subjects (F= 11.507, p=0.001, with 0.487 incident s in the monitore d condition and 1.146 in the non-monitored condition). H3, which posits that the decrease in the frequency of high intensity hazard communications between monitored and non-mo nitored conditions is greater than the change in frequency for low intensity hazard communications, is also supported. Significant differences on the ba sis of monitoring are found in low intensity general hearsay statements (F=4.410, p=0.039), which were 57 pe rcent lower in the monitored condition than in the non-monitored condition. Mode rate and high intensity communications saw greater reductions. M oderate intensity general r eported observation statements (F=5.527, p=0.021), fell 63 percent. Monitored communications contained 80 percent fewer high intensity specific negative self-disclosure statements than non-monitored communications statements. The magnitude of the differences positively correlates with the degree of hazard intensity.
87 Since there were no significant changes in one high intensity statements category, additional tests were run to a ssess changes in all high intensity statements and all low intensity statements by treatment condition. Ea ch of the two ANOVAs were significant. High intensity statements were 73 percent lower in the monitored condition (2.414 vs. 0.641 statements, F=13.740, p=0.000), while low in tensity statements were 49 percent lower (7.951 vs. 4.026 statements, F=3.605, p=0.061), providing additional support for this hypothesis. We observed no statistically significa nt drops in low intensity hazard communication incidents and cannot conclude these are different from zero. We do find lower monitored incidence of general moderate observed incidents (F=5.575, p=0.021, a change of 57 percent), general high (F =4.192, p=0.044, 47 percent) and specific high (F=5.305, p=0.024, a change of 66 percent) hazard communication incidents, which compared with a flat rate of low intensity incidents provides furt her support for this hypothesis. H4 posits that organizati onal members engage in mo re denials of knowledge of hazards under organizational CMC monitoring than on unmonitore d channels. This hypothesis is also supp orted with significant differences in five of the coded denials categories, all in the predicted direction. Non-monitored subjects o ffered an average of 0.7 hard denials of personal impropriety as compared with 1.1 for monito red subjects (F=3.340, p=0.072). Likewise, non-monitored subjects 0 .7 hard denials of hearing about others cheating experiences as compared with 1.2 fo r monitored subjects (F=4.200, p=0.044).
88 This hypothesis is also su pported for incidents with significant or marginally significant findings on soft general denials (F=3 .891, p=0.052, with m eans of 1.8 for nonmonitored and 2.5 for monitored communications), ha rd personal impropriety denials (F=2.989, p=0.088, with m eans of 0.7 for non-monit ored, and 1.1 for moni tored), soft denials of seeing incidents (F =3.574, p=0.062, wi th means of 0.6 for non-monitored communications, and 1.0 for monitored communications), and ha rd denials of hearin g incidents (F=7.272, p=0.009, with means of 0.5 for non-monitore d, and 1.0 for monito red communications). Since there were no findings on some denials statements categories, the analysis was run again on consolidated, less granular data, considering all denials in a single group. Denial incidents experienced a significant chan ge under monitoring w ith 7.256 incidents per subject vs. 4.878 in the non-monitored condition (F=7.630, p=0.007). Howeve r, the change in statements was not significant (F=1.535, p=0.219). While monitored subjects denied more distinct hazards, they made fewer statements about ea ch of them on average than nonmonitored subjects (figure 8).
89 Figure 8. Average denial statements per incident Table 17 summarizes these findings with descriptive statistics for the significant variables reported at the lowest level of data granularity.
90 Table 18. Descriptive Statistics for Conten t Dependent Variables with Significant Monitoring Differences, Grouped by Hazard Intensity, Neutral Beliefs and Denials NMeanSD Min Max Non-monitored411.7562.835 0 11 Specific negative selfdisclosure statements Monitored 390.3590.986 0 4 Total 801.0752.243 0 11 Non-monitored 410.6100.919 0 4 Specific negative selfdisclosure incidents Monitored 390.2050.615 0 3 Total 800.4130.807 0 4 Non-monitored 410.5370.636 0 2 General negative selfdisclosure incidents Monitored 390.2820.456 0 1 Highest Intensity Total 800.4130.567 0 2 Non-monitored416.46310.048 0 43 Monitored 392.3593.766 0 14 General observation of others statements Total 804.4637.887 0 43 Non-monitored 412.7803.991 0 20 General observation of others incidents Monitored 391.1791.449 0 5 Moderate Total 802.0003.118 0 20 Non-monitored 416.39010.281 0 60 General hearsay statements Monitored 392.7694.145 0 23 Low Total 804.6258.068 0 60 Non-monitored 4138.22013.947 17 97 Monitored 3929.6159.613 10 47 Belief statements Total 8034.02512.715 10 97 Belief topics (incidents) Non-monitored 4120.3906.815 14 54 Monitored 3917.0775.173 8 30 Neutral Total 8018.7756.258 8 54 Non-monitored 411.8291.283 0 5 Monitored 392.5381.890 0 6 Soft general denial incidents Total 802.1751.636 0 6 Non-monitored 410.6830.756 0 2 Hard denials of personal impropriety incidents Monitored 391.0511.123 0 5 Total 800.8630.964 0 5 Non-monitored 410.6340.662 0 2 Soft denials of seeing others incidents Monitored 390.9740.932 0 4 Total 800.8000.818 0 4 Non-monitored 410.7320.775 0 2 Hard denials of hearing others statements Monitored 391.2311.459 0 8 Total 800.9751.180 0 8 Non-monitored 410.4880.597 0 2 Monitored 391.0001.051 0 4 Denials Hard denials of hearing others incidents Non-monitored 800.7380.882 0 4
91 H5 posits that organizational members engage in a lower volume of neutral communications under organizational CMC m onitoring. Non-monitored subjects offered an average of 38 belief statements (which contained no hazard communications or denial content), while monitored subjects offere d an average of 30, a significant difference (F=10.221, p=0.002), providing support for this hypothesis. Non-monitored subjects used these to describe 3.4 fewe r belief topics (F=5.954, p=0.017) H6 posits that organizational members engage in a lower total volume of communications under organizational CMC monitoring. A MANOVA model predicting interview time in minutes, word count and te xt entry count from the monitoring condition factor was significant (F=4.781, p=0.004), as we re tests on time in minutes (F=9.978, p=0.002), text entry count (F=8.889, p=0.004), and word count (F=9.760, p=0.003). In each case, non-monitored communications vo lume was greater. Means for these variables by monitoring conditi on are presented in table 19. Table 19. Impact of Monitoring on Communications Volume NMeanSDMin Max Minutes Non-monitored4140.412.821.2 89.4 Monitored3932.210.416.1 61.5 Total8036.412.316.1 89.4 Words Non-monitored411359.8365.5805 2937 Monitored391123.2307.5583 2352 Total801244.5356.8583 2937 Text entries Non-monitored4179.024.350 167 Monitored3965.415.045 107 Total8072.421.345 167
92 Post hoc analyses Several additional analyses were undertaken. We examined total changes in hazard communications statements and incidents by monitoring condition, further examined the organizational standard r ecency finding, and considered effects of familiarity with the communications technology employed. Total hazard communications incidence We hypothesized changes in hazard communications by hazard intensity. We now examine overall changes in hazard communications incidents and statements. Separate ANOVA models pred icting total hazard comm unications incidents and statements from monitoring condition were significant (F=6.827, p=0.011 for incidents; F=9.450, p=0.003 for statements). More incidents and statements were offered by the non-monitored group. Means for the two groups appear in table 20. Table 20. Total Hazard Statements and Incidents by Monitoring Condition NMeanSD Min Max Statements Non-monitored 4117.75618.463 0 102 Monitored 397.8977.947 0 36 Total 8012.95015.085 0 102 Incidents Non-monitored 417.8056.889 1 40 Monitored 394.4364.272 0 21 Total 806.1635.973 0 40 Total observed incidents incidence We did not find a significant effect for specific statements of observed incriminating behavior. An examination of th e data suggests that subjects made similar numbers of specific statements in the two c onditions, with non-monitored subjects more likely to then elaborate in general terms. To further understand this phenomenon, we test
93 the combined counts of specific and genera l statements of observed incriminating behavior. Since the organization standard re cency effect was found not to apply to these communications, monitoring treatment condition is used as the sole i ndependent variable. The model is significant (F=5.305, p=0.024) wi th a mean in the non-monitored condition of 7.390 and a mean in the monitored condition of 3.231, indicating a rate more than twice as high for non-monitored subjects. Organizational standard recency The only significant finding for organizat ional standard recency was for general negative self-disclosures (table 21, F=41.988, p=0.000), but it is a powerful predictor, giving a model R2 of 0.336. Table 21. Impact of Organizational Standa rd Presentation and Training Recency Type III Sum Sq Mean Sq F Sig. Specific negative self-disclosure statements 1.0261.026 0.221 0.640 General negative self-disclosure statements 28.62428.624 41.988 0.000 Specific, observed statements 0.9320.932 0.143 0.706 General observed statements 0.7890.789 0.013 0.909 Specific hearsay statements 0.5170.517 0.018 0.893 General hearsay statements 26.28326.283 0.417 0.520 Beliefs expressed statements 23.57123.571 0.161 0.689 Hard denial statements in general 3.4533.453 1.190 0.279 Soft denial statements in general 0.2290.229 0.063 0.803 Hard denial statements of personal impropriety0.7670.767 0.321 0.573 Soft denial statements of personal impropriety 0.4130.413 0.176 0.676 Hard denial statements of seeing others' exp 1.7471.747 0.600 0.441 Soft denial statements of seeing others' exp 0.0000.000 0.002 0.965 Hard denial statements h earing others' exp 1.8121.812 1.352 0.248 Soft denial statements of hearing others' exp 0.0000.000 0.006 0.939 To further understand the relationship, co rrelations between the recency measure and general negative self-disclosures were considered separately in monitoring and non-
94 monitored conditions for subjects also in the organizational standard condition. Under monitoring, the relationship is non-significant (p=0.219). However, when not monitored, more recent presentation of the standard is associated with a higher level of general negative self-disclosures (r=0.714, p=0.000). Examples of these communications are now examined (figure 9). probably not listed where I got all of my information from on a works cited page. i may not have re arranged a sentence enough for Safe Assignment to accept... or enough that it was much different from the original. maybe paraphrased something I saw on the in ternet, like on wikipedia, and used it in a paper without citing I have had people tell me what questions to expect during an exam based on them taking the class in the past. I have listened without sticking my fingers in my ears. However, I think every ti me the questions have changed. Figure 9. Sample general negative self-dis closure statements from non-monitored subjects receiving the organizational standard treatment While not inherent in the construct defin ition or typical of general negative selfdisclosures overall, those f ound in the non-monitored, organizational standard treatment condition tended to be stated tentatively (probably, may not have, maybe), even though subjects have first hand knowledge of th eir own behavior. Subjects have also minimized incrimination by reporting on incide nts which were scored as among the least serious, on average, in the cheating seriousness survey used to prime the discussion task (ranked 13-16 out of 19), or in the case of the final example, by indicating that the information received was not so licited and probably not useful. To further understand this phenomenon, th e correlation matrix of dependent variables measured from subjects in both the non-monitored and organizational standard conditions was consulted. While none are significant, one a pproaches significance, the
95 correlation with specific nega tive self-disclosures (r=0.349, p=0.113). This negative correlation suggests that genera l negative self-disclosures ma y have been substituted for more incriminating specific negative self-d isclosures by those who had recently been presented with the organizational standard. These subjects appear to have chosen examples they judged to be minimally incr iminating, and stated wrongdoing in tentative, qualified terms to better comply with the sta ndard than conveying sp ecific negative selfdisclosures would have done. The findings discussed in th is section were reassessed without one extreme but apparently legitimate observation for a subject who completed the organizational standard training the day prior to the experi ment. With this single subject removed, the recency affect is no longer statistically significant. Although the data point appears to be legitimate, with so little data available to model the effect, the findings for hypotheses 2, 3 and 4, which were assessed with this construct in the model, were reassessed without it. There were no changes in incident pvalues out to two significant digits. For statements, one marginal finding for hard denials tipped past the threshold traditionally applie d for marginal significance (from p=0.072 to p=0.106), but other statistical decisions were unchanged. Effects of familiarity with the technology Since subjects not familiar with the communications technology employed might have responses that are different from t hose for whom novelty would have no impact, years of IM experience was added to monitori ng and organizational standard recency as a predictor of communications statement impact s, and as a covariate for monitoring alone (since recency was found not to be significant) in a model predicting incident impacts.
96 Experience with the technology was non-signi ficant (F=1.110 p=0.367 for statements; F=0.709, p=0.766 for incidents.).
97 Chapter 5 Discussion This dissertation offers an array of si gnificant findings, theory development, a research framework, a coding scheme, implicatio ns for organizations in the areas of both monitoring and organizational standards, and a research agenda for others interested in these topics. Each of these contribu tions is addressed in this chapter. Key findings Workplace monitoring of computer mediated communications changes both how much and what people communicate. Not only is discussion that may inculpate the speaker or writer, organizational colleagues, or the organization itself less prevalent, so are neutral statements. In the controlled laboratory experiment, non-monitored subjects engaged in discussions that were 26 percent longer than those of their monitored counterparts. They made 21 percent more text entries and used 21 percent more words. Unencumbered by the effects of monitoring, th ey issued 29 percent more neutral belief statements about 19 percent more topics (cod ed as incidents). On every neutral and overall volume measure collected, monito red subjects chose a lower level of communications. This is activity that makes up a large portion of a typical workday, with e-mail alone accounting for 28 percent of work er computer use (Taylor 2007). While recent figures are difficult to obtain, several years ago over 60 percent of employees were already spending more than 90 minutes per day on e-mail and instant messaging alone,
98 with many spending more than four hours ("E-Mail and IM Survey" 2004), it is clear that the measured communications represent a substantial portion of what goes on in a typical organization. Potentially incriminating hazard comm unications distinguished monitored and non-monitored subjects to an even greater extent, with those free of monitoring volunteering more than five times as many sp ecific, personally incriminating statements about more than three times as many incident s, with differences of 90-174 percent in five other categories of incriminating statements and incidents. On every category in which significant differences were found, monitore d subjects engaged in lower levels of communication. Unlike with neutral measures, we did not find statistically signi ficant differences in every category into which hazard communi cations were coded. In particular, numbers of hearsay incidents, both general and sp ecific, and numbers of specific hearsay statements were not statistically distinguish able between treatment conditions. This is consistent with theory: t he smaller or absent motivati onal effects of perceived selforganizational standard misalignment for lower intensity hazard communications topics are less likely to lead to communications suppression. The other type of coded hazard communi cation not found to show statistically significant differences between monitored and non-monitored groups was specific, observed incriminating behavior. However, when moderate intensity communications were grouped, monitored subjects were seen to offer far fewer at this middle level of hazard intensity. While the findings by hazard in tensity category were as predicted, this finding within a category suggests people may not, or may not always, make the finer
99 hazard intensity distinction between general and specific communications. This is a question for study in other domains to further explore the effect. We also observed increases in denials of involvement in and knowledge of hazard communications topics under monitoring. Ho wever, even though the number of topics (incidents) denied increased, the number of denial statements made fell. Under monitoring, subjects were quick to issue deni als, but not to discuss them. Although we did not separately hypothesize statement outcome s, this observation is consistent with using denials to guide discussion away from a hazardous topic to comply with a perceived organizational standard. While modest absolute changes experienced in some categories were small, the percentage changes are dramatic. The expe riment manipulated monitoring in a CMC exchange that lasted an average of 36 minutes In organizations, workers are projected to spend 41 percent of their time on e-mail by 2009 ("Addressing Information Overload in Corporate Email: The Economics of User Atte ntion" 2007), and businesses are expected to adopt instant messaging almost univers ally by 2010 (Kerner 2007). For employees who blog at work, and for thos e who use social networking sites, the time commitment is typically much greater (Boyd et al. 2007; Mattson et al. 2007). In this environment, it is tempting to conclude that monitoring is a helpful tool for reducing information overload or the distraction of unimportant communi cations. Unfortunately, the experiment reported herein demonstrates that the changes are not limited to unimportant communications, but affect those focused on important organizational topics. Although a detailed qualitative analysis is beyond the scope of this project, the researcher notes that she could typically co rrectly assign anonymized transcripts devoid
100 of treatment condition information to the co rrect treatment condition after first exposure to a short portion of th e record. The monitored transcript s generally have a different tone that does not invite discussion. For the study, th e interviewers were instructed to follow a pre-determined interview script In a corporate environment, the nature of the monitored comments makes it likely that many conversa tion partners would drop unwelcome paths of inquiry, or that under monito rings influence, they woul d avoid making inquiries over monitored communications channels in the firs t place. Thus monitoring effects measured herein are likely to be u nderstated from those naturally occurring in organizational environments. Studies in more natural setti ngs, while less controlled, may provide insight here. Contributions to research This work advances research with both conceptual and methodological contributions. Conceptual contributions include development in th e area of standards operation in self awareness th eory, an examination of fi gure ground contrasts to apply this theory to a new research domain, CM C monitoring, and a research framework for studying self awareness in this domain. Methodological contributi ons consist of the hazard communications taxonomy and coding sche me, and a relative standards influence instrument and methodology for use in studying competing standards. Theory development While it is well founded in a hundred-plus year history of research on the self, self awareness theory is a relative newcomer to information systems research. In this study, self awareness theory is synthesized with CMC and surveillan ce literature and applied to a new domain, monitoring computer medi ated communications. CMC monitoring has
101 some of the aspects of well test ed manipulations of self-focus and the theorys self-focus predictions prove effective for unders tanding changes to communications in organizations under monitoring. Understanding the operation of self awar eness theory in the face of multiple standards has been described as a longstandi ng open research area (S ilvia et al. 2001). Typically research subjects have been provide d with a standard, or alternatively have had their personal standards measured for a ssignment to treatment groups based on preexisting standards (Silvia et al. 2001). This dissertation undertakes an area in which organizational standards are typically inferre d, and in which persona l standards vary and are seldom made explicit. It has been suppos ed that theory selec tion is an either-or choice, based on what standard is most salient, but typical tests of the theory do not allow assessment of this supposition (Silvia et al. 2001). Herein, the self awareness induced by monitoring is theorized to increase the regul atory influence of organizational standards on organizational communications at the expe nse of personal communications standards rather than having one standard supplant an other. In fact, the experiment provided evidence that multiple standards are in opera tion, but that their relative influence is impacted by monitoring, the means of self-focus inducation employed. Research framework The communications impacts of monitoring model developed and successfully tested is both novel and parsimonious: the presence or absence of monitoring, and the recency of training on an organizational sta ndard, along with self-f ocus as a control variable, predict dramatic changes in co mputer mediated communications content towards the organizational standard or to minimize monitored behavior.
102 The monitoring models value is extended by the fact that it predicts lots of different changes in communications beha vior: to three categories of hazard communications, denials, neutral statements, and communications volume. Monitoring is measured to account for 14.4 percent of the vari ation in the perceived relative influence of organizational standards on communicati ons, 10.2-11.3 percent of the change in communications volume, and 7.1 and 11.6 percent of the change in neutral incidents and statements respectively. Recency of presentati on explained 33.6 percent of the change in general negative self-disclosur e statements. Monitoring also explained 10.8 percent of the observed difference in hazard statements, and 8.0 percent of the difference in hazard incidents. Finally, monitoring accounted fo r 8.9 percent of the difference in denial incidents. Hazard communications taxonomy and coding scheme To study a domain in which multiple communications impacts were anticipated based on a perceived organizational standard, this work defines hazard communications and related denials constructs and places th em within two taxonomies. The hazards taxonomy has three levels of hazard intensity initially developed from negative selfdisclosure, whistleblowing and gossip literatu re streams, with construct definitions adjusted from these domains to produce non-ove rlapping constructs (Holton et al. 2008). Within each, general and specific categories we re distinguished (fi gure 2). Further, hazard effects across domains were explored (f igure 3). Denials, related to hazards by their exculpatory nature, were also organize d into a taxonomy congruent with the hazard communications taxonomy. These were also subdivided into general and specific categories based on observations during the pilot study.
103 Next, a coding scheme was developed that defines these three levels of hazard intensity, including both general and specific types of each; four categories of denials, with both general and specific ca tegories of each; and a beliefs category. Each of these is further refined into incidents and statements groupings. Given the novelty of this research stream, the granularity of codi ng offers two benefits. Firs t, a very detailed level of analysis is possible, building our understandi ng of the components of the new constructs. Second, it allows, new groupings to be computed and tested as new research questions are posed, and as our understand ing of the domain progresses. For instance, future work might explore the possibility that general a nd specific hazard communications are more appropriately examined separately rather than paired within the three hazard intensity categories as proposed in this dissertation. The detailed coding schema allows post hoc exploration and testi ng of new hypotheses. Relative standards influence measure and methodology One of the goals of this research was to progress our understanding of the influence of competing behavioral standards to make a contribution to self awareness theory. Based on behaviors observed during th e 40 interview pilot study and a review of relevant literature, a nine item measure of standards governing instant messaging behavior was developed. In a pilot assessmen t, the measure performed rather well, with high internal consistency reliability and w ithout evidence of other problems. However, when used with the subjects in the main st udy, internal consistency reliability was poor, and remained weak after dropping several ite ms. Further, different items performed poorly in the personal and organi zational versions of this scal e. Results across subjects varied dramatically. The m easures were discarded.
104 The differences in performance across s ubjects provides a lik ely explanation of the problem. While there has been no agreem ent on this matter, it has been previously suggested that standards are high ly personal (Silvia et al. 2001). In this case, it would be very difficult to represent the relevant dom ain across subjects, and particular sets of behaviors would not be expect ed to covary in predictabl e ways across subjects. Strong performance of the discarded scale in the pilot test, where subjects were not engaging in the measured behavior at the time of standards assessment, is similar to the relatively common practice of using standards pre-tests to assign subjects to treatment condition in self awareness studies. The obser vation of dramatically different pilot test and study performance of the tested instru ment suggests this pr e-test methodology may give misleading results. Fortunately, the potential problem with st andards measurement was anticipated. An alternative measure and methodology al low standards influence assessment even without clear and consistent definition of th e behaviors belonging to the standard domain for individual subjects. The new items which measured perceived influence of organizational standards on communications behavior, and perceived influence of personal standards on communi cations behavior, performe d better. The methodology employed both allows for and demonstrates that standard selection is not a matter of choosing one standard to the ex clusion of all others. A ratio of the organizational and personal standards scales provided a measure of relative standards in fluence that did not require all subjects to define standards in the same terms but respected their implicit standards definitions which may include persona l and unique elements. Further, the ratio
105 measure allows for the assessment of compe ting standards which were shown to operate in this case. Areas for future work Computer-mediated communications monito ring is a new research domain. The laboratory experiment tested a realistic inst ant messaging system with recent monitoring notification that took place im mediately prior to the communications session in which dependent variables were measured. It also provided an unobtrusive but constant visual reminder of monitoring. Enterprise IM syst ems may address on screen reminders with periodic broadcast messages, obtrusive randomly timed inline reminders, reminder notices which must be actively dismissed at log-in, or as in this experiment, with a constant visual reminder. Within e-ma il systems, periodic e-mail reminders and statements in e-mail footers have been frequently observed. Despite these myriad communications systems features for m onitoring notices and reminders, not all organizations that monitor regularly provide frequent monitoring reminders. We are aware of no catalog of use of various CM C systems features for monitoring and reminders to allow us to assess which applic ations of which features are most common. Exploration of the human-computer interf ace issues could allow stronger systems performance tailored to an organizations needs. Future work should test different parameters on both of notification and reminde r features, alone and in combination. The experiment included a single communi cations session. Future work should study effects over longer time horizons so attenuation of the effects can be assessed. Separate assessment of monitoring impacts ove r time with frequent reminders, and over time with infrequent or absent reminders is a rich area for future studies.
106 Hazard communications will be difficult to study in field settings as companies are in no hurry to divulge incriminating info rmation. This study used a non-corporate organizational setting, studying students in their role as members of a university organization. We note that this was not a role simulation. Student subjects were not asked to respond as if they were members of so me other organization or in roles that were foreign to them. The task was realistic and appropriate to subjects role in the organization studied. However, this is still but one role in one organization. Several disciplines were represented: informa tion systems, history, marketing, nursing, psychology, and mass communications. While th e sample size is insufficient to study responses separately by discipline, taken together they tell a cohesive story. Interestingly, we found results regardless of whether instant messaging was a new technology for the subjects. Monitoring effects are predicted not to be learned norms but driven by psychological processe s that are expected to transcend the particular CMC medium chosen. That experience with instant messaging had no impact on results supports this interpretation. Nevertheless, monitoring other existing and emerging CMC media merits separa te consideration. The study predicts reduced communications volume overall and of neutral statements in particular based on the motiv ation of people to escape the self-focusing stimulus provided by monitoring, with pred ictions supported by empirical results. The strength of the finding on these outcome variables suggests that channel switching behaviors to non-monitored channels may also be substantial in organizational settings. The necessity to limit the scope of this study dictates that an i nvestigation of these impacts be slated for future work. A number of research questions are possible: When
107 other channels are cultivated for delivery of hazard communications, such as the teaching hospital mortality and morbidity reports, Army after action reports, and exit interviews described previously are employed, can co mmunications volume be maintained or expanded? When such communications are not cultivated, do face to face communications increase? How does total anonymity (as opposed to just visual anonymity tested herein) affect the impact of monitoring on communications outcomes? The study focused on the effects of or ganizational CMC monitoring, along with understanding drivers of the changes. Post hoc analysis also determ ined that organization standard presentation recency may be asso ciated with a reduc tion in non-specific personally incriminating communications. Ho wever, no impacts on overall volume were found, and no beliefs, denials or hazard cate gories other than ge neral negative selfdisclosures, were identified. Future work should study th e impact of organizational standard presentation and training, including a more comprehensive test of the recency effect, evaluation of training and reminder methods. Considering these effects both with and without monitoring is desirable. This first application of self-awareness theory to CMC monitoring also suggests a number of areas for future work with the th eory whether in this domain or others. The figure-ground contrast means of self-focus i nduction are not often studied and should be explored with different app lications of the theory. Th e relative standards influence methodology and measure proved valuable in this initial test and hold promise for future investigations of the impacts of self-awareness whether induced by CMC monitoring or other mechanisms. Other means of examini ng realistic, personal, unique and potentially
108 competing standards, for instance using the repertory grid techni que (Kelly 1955), could further elucidate the operation of standards. Contributions to practice The two independent variable constr ucts define the two major areas for application of this work to organizati ons: monitoring features and practices, and organizational standard practices. Implications for CMC monitoring in organizations With the environmental shift towards vastly increased organizational CMC monitoring, understanding behavioral consequences is critical for or ganizations. In the sensitive hazard communications context exam ined herein, awareness of communications content impacts should inform monitoring fe ature design and use to continue to satisfy legislative, security, or other motivations wh ile enabling new benefits such as increased disclosure of ha zards on which an organization wo uld choose to act, increased communication of non-hazardous relevant information, a nd limiting drawbacks like reduction of in cidence of other desi rable communications. Although the changes both predicted and meas ured are substantia l, the researcher is not aware of organizations having made atte mpts to compensate for them to the extent that they are considered negative, nor to en courage them to the extent that they are assessed to be desirable. To what extent various changes are positive or negative and what specific responses are appropriate is a matter for consideration by organizations employing monitoring. A review of sample considerations follows. Cutting down the overall level of communicatio ns, to the extent that those cut are excessive rather than valuable, could probably benefit most organizations barreling
109 towards the prediction of 41 percent of work time spent on email and a greater total on all electronic communications. While we in no way mean to demean the value of communications, we note that they are not appropriate for every ta sk, and inappropriate communications are estimated to cost co mpanies $198 billion annually ("Anti-Spam Trends" 2003). There are also individual differences in communications preferences with effectiveness implica tions (Spitzberg 2006). Unfo rtunately, the communications lost include valuable information. Reporting of misdeeds is likely to be desira ble for organizations seeking to curtail them. Since monitoring of CMC reduces th e incidence of reporting on potentially incriminating behaviors, organizations wishi ng to promote reports may want to provide whistleblowing hotlines that provide anonymity, are outsource d so reports are not made directly to an organizational authority, and th ey may even wish to reward reports which are corroborated in i nvestigations. Increasing denials of incriminating behaviors may be desirable for an organization subject to external auditing of communications, especially since there is some evidence that individuals align their behavior with stated sta ndards and opinions (Gibbons 1990). Such a company may wish to provide freque nt inline reminders that communications are monitored and subject to audit. In many circumstances, sharing of input and ideas leads to stronger decisions (Jonathan et al. 1998). To facilitate such sharing when monitoring might curtail it, organizations may wish to provide for f ace to face communications free from the perception of monitoring, or alternatively, provide anonym ous communications channels like certain GDSS systems (Jessup et al. 1990).
110 Obviously, whistleblowing hotlines, inline reminders, face to face communications alternatives, and anonymous communications systems do not constitute a comprehensive list of possibilities. The features of CMC systems and alternatives to them provide myriad options that go well beyond the few discussed here. We further note that organizational goal s are not unidimensional. Companies have multiple and competing motives when it comes to the content of their electronic communications. Multiple responses are appr opriate for addressing multiple, disparate organizational goals. Systems provide means of satisfying many of these, but to date, the effects of monitoring have been unknown and so have not been given appropriate heed in decision-making about the design, use, and communication about communications systems monitoring features. Implications for organizational standard training and presentation The experiment provides some evidence that a recently presented organizational standard can impact communications for organizational members who believe their communications not to be monitored. In prac tice, we know that almost all medium and large organizations have some form of an el ectronic communications policy in place, but also that virtually all empl oy some forms of electronic monitoring. While estimates vary, it appears that a substantial minority of organizations still do not monitor e-mail, and more still leave instant messaging unmonito red ("2001 Electronic Poli cies and Practices Survey" 2001; "2003 E-Mail Rules, Policies and Practices Survey" 2003; "Electronic Monitoring Survey" 2005; "E lectronic Monitoring and Surveillance Survey" 2007). Organizations choosing not to monitor a pa rticular communications medium may see
111 some regulatory benefit to providing an orga nizational policy for the medium. For that policy to be effective, however, it may need to be frequently presented. In this study, behavioral intention to comply with the policy after first presentation was high. On a pilot test of the organizationa l standard training manipulation, just 24 per cent said they would not adjust their communications behavior under this policy, each of them stating that th ey believed they were already in compliance (n=17). During the experiment, the policy was presented on average 9 days before communication impacts were measured, and the variation in lag allowed us to observe that the recency effect fit an inverse square rate of decline. There was admittedly limited data available to fit this curve as the study aimed for homogeneity on this point. While frequent, obtrusive policy reminders are possible, such as inline instant message comments and use of broadcast IM features to provide policy notices, the organizational standard effects observed in the non-monitored condition were limited to one the category of general pe rsonally incriminating statem ents. There are far more direct ways to promote this type of communication such as the mortality and morbidity conferences used by teaching hospitals and af ter action reports incl uding poor decision making and outcomes which have been previous ly discussed. Organizations have many other options, including leading by example to demonstrate appropriate behavior and responding appropriately, which may ofte n exclude punitive outcomes for honest mistakes. Full consideration of the op tions is beyond the scope of this work. The observed policy presentation and training effect held only when monitoring was not in place. Regulations like HIPAA and Sarbanes-Oxley require evidence of policy enforcement that goes beyond infrequent presentation of and training on a policy.
112 Monitoring is one of those enforcement mechanisms, and it was shown in this study to have much farther reaching impacts th an policy presentation and training. Since organizations avoiding monitoring ma y do so based on privacy concerns or cost, we note here that CMC compliance optio ns that require mimimal involvement from human readers are available. Monitored instant messaging systems have been around nearly a decade, and email monitoring featur es for more than two ("IM Shop?" 2004; Zuboff 1988). In this time our ability to find very targeted types of content using automated means has grown quite sophisticat ed (Holton In Press; Overly Undated), allowing targeted compliance actions to be undertaken often w ithout direct human intervention. Communications systems that automatically scan content and notify users when content is blocked or that require au thorization before sending messages suspected of being in violation of policy may provide the most potent standard reminders, and they entail using human reviewers only when n ecessary. They also go a great distance towards demonstrating effectiv e policies are in place for en suring regulatory compliance in this regard. Concluding thoughts This dissertation explores an area believed to be of considerable organizational impact: the enactment of monitoring through computer mediated comm unications systems. A quarter century of work on the language/ac tion perspective (Gold kuhl et al. 1982) has provided support for the notion that communicati on is a form of organizational action, an idea explored extensively in the realm of computer-supported cooperative work. Communication is not just a way to exchange information. It determines wh at gets done in organizations, and in particular determines the processes that underlie information systems.
113 This research begins to investigate th e complex interaction of CMC systems and organizational monitoring, specifically exam ining how monitoring computer mediated communications impacts its content. The prim ary contribution of this work is simply theoretical and empirical support for the dramatic influen ce of monitoring on what is communicated over CMC channels. This work extends prior research by br inging together our current understanding of CMC with surveillance and performance monitoring research and self awareness theory. It introduces the influence of or ganizational CMC monitoring on self-focus to explain changes in communications. In addi tion, this research seeks to develop our understanding of standard selection when there are competing candidate standa rds. Standard selection has been described as an open self awareness theory issue. We have both theorized that when monitoring organiza tional communications standards take on increased influence over communications relativ e to personal standards, and measured that self-focused subjects are able to percei ve this change in th eir own motivations. Through its exploration of hazard communica tions and further development of the hazard communications taxonomy (Holton et al. 2008), the dissertation contributes to our knowledge of contextual influences on CMC content. Due to the nature of the research context selected, it also provi des and tests a research framework for classifying hazard communications for analysis, which ma y be applied in other contexts.
114 References "2001 Electronic Policies and Practices Survey," in: US News & World Report American Management Association, The ePolicy Institute, US News & World Report, 2001. "2003 E-Mail Rules, Policies and Practices Su rvey," American Management Association, The ePolicy Institute, and Clearswift, 2003. "2004 Workplace E-Mail and Instant Me ssaging Survey Summary," American Management Association, New York, 2004, p. 8. "2005 Electronic Monitoring and Survei llance Survey Summary," American Management Association, New York, 2005, p. 13. "2007 Electronic Monitoring & Surveillance Survey," American Management Association. "Acceptable computer use policy," Envision Intret, 2007, pp. 1-5. "Acceptable use practice," General Motors, 2006, pp. 1-3. "Addressing Information Overload in Corporate Email: The Economics of User Attention," The Radicati Group, Palo Alto, CA, 2007. "Anti-spam Market Tren ds, 2003-2007," The Radicati Group, Palo Alto, CA. Baltes, B.B., Dickson, M.W., Sherman, M.P., Bauer, C.C., and LaGanke, J.S. "Computer-Mediated Communication a nd Group Decision Making: A Meta-
115 Analysis," Organizational Behavior and Human Decision Processes (87:1) 2002, pp 156-179. Boyd, D.M., and Ellison, N.B. "Social ne twork sites: Defini tion, history, and scholarship," Journal of Computer-Me diated Communication (13:1) 2007. Burton, B.K., and Near, J.P. "Estimati ng the incidence of wrongdoing and whistleblowing: results of a study using randomized response technique," Journal of Business Ethics (14) 1995, pp 17-30. Bush, G.W. "Statement by the President," The White House, Washington, D.C. "Business Conduct and Ethics," Chevron, 2007, pp. 1-16. Buss, A.H. Self-consciousness and social anxiety W. H. Freeman, San Francisco, 1980. Cameron, A.F., and Webster, J. "Uninte nded consequence of emerging communication technologies: Instant messaging in the workplace," Computers in Human Behavior (21) 2005, pp 85-103. "Canada Keeping the Promise for a Strong Economy Act," in: Bill 198 2003. "Center for Academic Integrity Assessment," D uke University Kenan Institute for Ethics, 2005. Chidambaram, L., and Tung, L.L. Is out of sight, out of mi nd? An empirical study of social loafing in technology-supported groups Institute for Operations Research and the Management Sciences, 2005. Chociey, P.A. "Who's reading my e-mail?": A study of professionals' e-mail usage and privacy perceptions in the workplace," IEEE Transactions on Professional Communication (40:1) 1997, pp 34-40.
116 Chun, M.M., and Wolfe, J. "Visual attention," in: Blackwells handbook of perception, E.B. Goldstein (ed.), Blackwell, Oxford, UK, 2001, pp. 272-310. "Computer use policy for CDI Corporation a nd its related companies," CDI Corporation, 2006, pp. 1-5. "Computer use policy: Policy restricting pe rsonal use of employer's computer," WGI International, Inc., pp. 1-5. "Computer, email and Internet use policy," City of Merriam, pp. 1-5. Cooley, C.H. Human nature and the social order Scribner, New York, 1902. D'Urso, S.C. "Electronic m onitoring and surveillance in the workplace: Modeling the panoptic effect potential of communicati on technology, organizat ional factors and policies," 2004. Daft, R.L., and Lengel, R.H. "Organizationa l information requirements, media richness and structural design," Management Science (32:5) 1986, pp 554-571. Deci, E.L., and Ryan, R.M. Intrinsic motivation and se lf-determination in human behavior Plenum Press, New York, 1985. Douthitt, E.A., and Aiello, J.R. "The role of participation and control in the effects of computer monitoring on fair ness perceptions, task satisfaction, and performance," Journal of Applied Psychology (86:5) 2001, pp 867-874. Dozier, J.B., and Miceli, M.P. "Potential Pr edictors of Whistle-Blowing: A Prosocial Behavior Perspective," Academy of Management. The Academy of Management Review (10:4) 1985, pp 823-836.
117 Duval, S. "Conformity on a visual task as a function of personal novelty on attitudinal dimensions and being reminded of the object status of self," Journal of Experimental Social Psychology (12:1) 1976, pp 87-98. Duval, S., and Wicklund, R.A. A theory of objective self-awareness Academic Press, San Diego, CA, 1972. Duval, T.S., Silvia, P.J., and Lalwani, N. Self-awareness and causal attribution: A dual systems theory Kluwer Academic Publishers, Boston, 2001. "E-mail and electronic communication policy," World Publishing Company, 2006. El-Shinnawy, M., and Vinze, A.S. "Technology, culture and persuasiveness: a study of choice-shifts in group settings," International Journal of Human-Computer Studies (47:3) 1997, pp 473-496. "Electronic communication policy for all Volt entities, affiliates, subsidiaries and divisions," Volt Informati on Services, Inc., pp. 1-4. "Electronic Records Management Software Applications Design Cr iteria Standard," Department of Defense, 2007. Feldman, D.C. "The impact of exit questi onnaire procedures on departing employees' self-disclosure," Journal of Managerial Issues (11:1) 1999, pp 13-25. Fenigstein, A., and Levine, M. P. "Self-attention, concept ac tivation, and the causal self," Journal of Experime ntal Social Psychology (20:3) 1984, pp 231-245. Festinger, L., Pepitone, A., and Newcomb, T. "Some consequences of deindividuation in a group," Journal of Abnormal Psychology (47), April 1952, pp 382-389. Foucault, M. Discipline and Punish Vintage, New York, 1977.
118 Garvin, D.A. "The U.S. Army's After Action Reviews: Seizing the Chance to Learn," in: Learning In Action, A Guide to Putt ing the Learning Organization to Work Harvard Business School Press, Boston, 2000, pp. 106-116. George, J.F. "Computer-based monitoring: Common perceptions and empirical results," MIS Quarterly (20:4) 1996, pp 459-480. Gibbons, F.X. "Self-attention and self-report: The 'verdicality' hypothesis," Journal of Personality (51:3) 1983, p 517. Gibbons, F.X. "Self-attention and behavi or: A review and theoretical update," Advances in Experimental Social Psychology (23) 1990, pp 249-302. Gilsdorf, J.W. "Written corporate policy on communicating: A Delphi survey," Management Communication Quarterly (5:3) 1992, pp 316-347. Goldkuhl, G., and Lyytinen, K. "A Language Action View of Information Systems," Third International Confer ence on Information Systems, Ann Arbor, Michigan, 1982, pp. 13-31. Google "FAQ for Google Zeitgeist 05: The Google Partner Forum," 2005. Handel, M.J. "Trends in Perceived Job Quality, 1989 to 1998," Work and Occupations (32:1) 2005, pp 66-94. "Health coverage portability : Health Insurance Portability and Acc ountability Act of 1996 (HIPAA) ", 1996. Hiltz, S.R., Turoff, M., and Johnson, K. "Experiments in group decision making, 3: disinhibition, deindividua tion, and group process in pen name and real name computer conferences," Decision Support Systems (5:2) 1989, pp 217-232. "HIPAA email compliance policy," Sendmail, 2008.
119 "HIPAA email encryption and security co mpliance for healthcare," Authora, 2008. "HIPAA secure e-mail," Orchestria, 2008. Holton, C. "Text Mining E-mail to Iden tify Disgruntled Employee Fraud Risk," Decision Support Systems ) In Press. Holton, C., and Fuller, R. "The Impact of Electronic Monitoring on Hazard Communications," Hawai'i Internationa l Conference on System Sciences 39, Kauai, HI, 2006. Holton, C.F. "Sophisticated Spreadsheet Featur es for Qualitative Data Analysis," ICIS 2006 International Symposium on Resear ch Methods, Milwaukee, Wisconsin, 2006. Holton, C.F., and Fuller, R.M. "Unintended Consequences of Electronic Monitoring of Instant Messaging," IEEE Transactions of Professional Communication (In Press) 2008. Hormuth, S.E. "Self-awareness and drive th eory: Comparing internal standards and dominant responses," European Journal of Social Psychology (12:1) 1982, pp 3145. Innes, J.M., and Young, R.F. "The effect of presence of an audience, evaluation apprehension and objective self-awareness on learning," Journal of Experimental Social Psychology (11:1) 1975, pp 35-42. Irving, R.H., Higgins, C.A., and Safayeni, F. R. "Computerized Performance Monitoring Systems: Use and Abuse," Communications of The ACM (29:8) 1986, pp 794-801.
120 Jessup, L.M., Connolly, T., and Galegher, J. "The effects of anonymity on GDSS group process with an idea-generating task," group decision support system (14) 1990, pp 312-321. Jonathan, K.K., Robert, C.W., and Mary Beth, R. "Coordinating Computer-Supported Cooperative Work: A Review of Res earch Issues and Strategies," Journal of the American Society for Informati on Science (1986-1998); New York (49:9) 1998, p 776. Kelly, G.A. The Psychology of Personal Constructs Norton, New York, 1955. Kerner, S.M. "Gartner: Instant Messaging Reigns Supreme," in: InternetNews.com 2007. Koffka, K. Principles of Gestalt Psychology Harcourt Brace, New York, 1935. Lea, M., and Spears, R. "Computer-mediate d communication, de-ind ividuation and group decision-making," International Journal of Man-Machine Studies (34:2) 1991, pp 283-301. LeClaire, J. "Firms develop policies, technologies to curb IM security dangers," in: TechNewsWorld, 2006. Macavinta, C. "Do you need a workplace IM policy?," in: Inc. Technology, 2007. Matheson, K., and Zanna, M.P. "The impact of computer-mediated communication on self-awareness," Computers in Human Behavior (4:3) 1988, pp 221-233. Mattson, E., and Barnes, N.G. "The Hype is Real: Social Media Invades the Inc. 500," University of Massachusetts Dartmout hs Center for Marketing Research, Dartmouth, MA. McCroskey, J.C. "Oral communication apprehension: A summary of recent theory and research," Human Communication Research (4) 1977, pp 78-96.
121 McGrath, J.E. "Time, intera ction and performance (TIP ): A theory of groups," Small Group Research (22:2), May 1991, pp 147-174. Mead, G.H. Mind, self and society University of Chicago Press, Chicago, 1934. Muse, D. "Enterprises come up half-empty on IM policies," in: InstantMessagingPlanet.com 2005. Nadeem, M.F., Ramin, T., and Jitka, S. "E -Mails in the Workplace: The Electronic Equivalent of 'DNA' Evidence," Journal of American Academy of Business, Cambridge (8:2) 2006, p 71. "NASD Conduct Rules 3010 & 3110," National A ssociation of Securi ties Dealers, 1999. Noon, M., and Delbridge, R. "News from be hind my hand: Gossip in organizations," Organization Studies (14:1) 1993, pp 23-36. Norman, D.A., and Shallice, T. "Attention to action: Willed and automatic control of behavior," in: Consciousness and self-regulation: Advances in research and theory, R.J. Davidson, G.E. Schwartz and D. Shapiro (eds.), Plenum, New York, 1986, pp. 1-18. Nunamaker, J.F., Dennis, A.R., Valacich, J.S., Vogel, D.R., and George, J.F. "Electronic Meeting Systems to Support Group Work," Association for Com puting Machinery. Communications of the ACM (34:7) 1991, pp 40-61. Orlander, J.D., and Fincke, B.G. "Morbidity and Mortality Conference: A Survey of Academic Internal Medicine Departments," Journal of General Internal Medicine (18:8) 2003, p 3. Overly, M.R. "Email Compliance Quick Refe rence Guide: Strate gies for Regulatory Compliance and Legal Risk Management," Undated.
122 Posner, M. "Cumulative development of attention theory," American Psychologist (37) 1982, pp 168-179. Postmes, T., Spears, R., and Lea, M. "Br eaching or Building Social Boundaries? SideEffects of Computer-Mediated Communication," Communication Research (25:6), Dec 1998, pp 689-715. Rice, R.E. "Media Appropriateness. Usi ng Social Presence Theory to Compare Traditional and New Or ganizational Media," Human Communication Research (19:4) 1993, pp 451-484. Rice, R.E., and Shook, D. "Access to, usage of and outcomes from an electronic message system," ACM Transactions on Office Information Systems (6:3) 1988, pp 255276. "Sarbanes-Oxley Act," in: H.R. 2002. Sawyer, K.R., Johnson, J., and Holub, M. The Necessary Illegitimacy of the Whistleblower SSRN, 2006. Scheier, M.F., Buss, A.H., and Buss, D.M. "Self-consciousness, self-report of aggressiveness, and aggression," Journal of Research in Personality (12:2) 1978, pp 133-140. Scheier, M.F., and Carver, C.S. "Self-directed attention and the comparison of self with standards," Journal of Experime ntal Social Psychology (19:3) 1983, pp 205-222. Shepherd, M.M., Briggs, R.O., Reinig, B.A., Yen, J., and Nunamaker, J.F., Jr. "Invoking social comparison to improve electr onic brainstorming: Beyond anonymity," Journal of Management Information Systems (12:3) 1995, pp 155-170.
123 Sherry, G. "Beyond call recording: the ROI for monitoring technology," Call center solutions (17) 1998, pp 90-96. Shibutani, T. Society and Personality: An Interactio nist Approach to Social Psychology Prentice-Hall, Inc., Englewood Cliffs, NJ, 1961. Short, J., WIlliams, E., and Christie, B. The Social Psychology of Telecommuncations Wiley International, London, 1976. Silvia, P.J. "Self-awareness and the regulation of emotional intensity," Self and Identity (1) 2002, pp 3-10. Silvia, P.J., and Duval, T.S. "Objective self-awareness theory: Recent progress and enduring problems," Personality and Social Psychology Review (5:3) 2001, pp 230-241. "Slowing the Stream: Development of a Speed-Bump Approach for Protecting New Releases in the Digital Media Environm ent," Berkman Center for Internet & Society at Harvard Law School, Boston, MA, 2004. Spears, R., Lea, M., and Lee, S. "De-individuation and group polarization in computermediated communication," British Journal of Social Psychology (29) 1990, pp 121-134. Spitzberg, B.H. "Preliminary development of a model and m easure of computer-mediated communication (CMC) competence," Journal of Computer-Mediated Communcation (11:2) 2006. Sproull, L., and Kiesler, S. Connections: New Ways of Working in the Networks Organization MIT Press, Cambridge, MA, 1991, p. 212.
124 Sproull, L.S., and Kiesler, S. "Reduci ng Social Context Cues: Electronic Mail in Organizational Communication," Management Science (32:11) 1986, pp 14921513. "Staff acceptable computer use policy," Gananda School District, 2006, pp. 1-2. "Statewide policy: Appropriate use of el ectronic communication and technology," State of Minnesota, 2006, pp. 1-9. Stritzke, W.G.K., Nguyen, A., and Durk in, K. "Shyness and Computer-Mediated Communication: A Self-Presentational Theory Perspective," Media Psychology (6:1) 2004, pp 1-22. Syrett, M. "Ineffable behaviour," People Management (7:17), August 30 2001, p 49. Tam, P.-W., White, E., Wingfield, N., and Maher, K. "Snooping E-Mail by Software Is Now a Workplace Norm," The Wall Street Journal), March 9, 2005. Taylor, K. "An Analysis of Computer Use Across 95 Organisations in Europe, North America and Australasia," Wellnom ics, Christchurch, New Zealand. Taylor, S.E., and Fiske, S. T. "Point of view and pe rceptions of causality," Journal of Personality and Social Psychology (32:3), September 1975, pp 439-445. "Title 17 of the Code of Federal Regulations (CFR), Rule 17a-4," Securities and Exchange Commission, 2001. "To Be or Not to Be an IM Shop?," Wall Street Technology :February 12, ) 2004. "United Kingdom Regulation of Investig atory Powers Act Chapter 23 ", 2000. Urbaczewski, A. "An examination of the effects of electronic monitoring of employee Internet usage," Indiana Universi ty, United States -Indiana, 2000.
125 Varon, E. "Mandate from SE C Regulators: Save Your Electronic Documents," in: CIO Magazine, 2003, p. 30. Vorvoreanu, M., and Botan, C.H. "Examining el ectronic surveillance in the workplace: A review of theoretical perspectives an d research findings," Conference of the International Communication Asso ciation, Acapulco, Mexico, 2000. Walther, J.B. "Computer-Mediated Communicat ion: Impersonal, Interpersonal, and Hyperpersonal Interaction," Communication Research (23:1), Feb 1996, pp 3-43. Wicklund, R.A. "Objective self-awareness," in: Advances in experimental social psychology, L. Berkowitz (ed.), Acad emic Press, New York, 1975. Wicklund, R.A., and Duval, S. "Opinion change and performance faci litation as a result of objective self-awareness," Journal of Experimental Social Psychology (7:3) 1971, pp 319-342. Wicklund, R.A., and Gollwitzer, P.M. "The fa llacy of the privat e-public self-focus distinction," Journal of Personality (55) 1987, pp 491-523. Wolff, W. "Selbstbeurteilung und Fr emdbeurteilung lm wissentlichen und unwissenthchen Versuch," Psychologische Forschung (16) 1932, pp 251-328. "Workplace E-Mail, Instant Messaging & Blog Survey: Bosses Battle Risk by Firing EMail, IM & Blog Violators," Ameri can Management Association, 2006. Worthen, B. "Message therapy," in: CIO Magazine 2005. Yao, M.Z., and Flanagin, A.J. "A self-awa reness approach to computer-mediated communication," Computers in Human Behavior (22:3), May 2006, pp 518-544. Zajonc, R.B. "Social facilitation," Science (149:3681) 1965, pp 269-274.
126 Zuboff, S. In the age of the smart machine: The future of work and power Basic Books, New York, 1988.
128 Appendix 1: Coding scheme Hazard communications Hazard communications are statements that may incriminate the speaker or the organization to which he or she belongs, in this case, the universi ty. The potentially incriminating topic discussed in the transcripts is cheating activity. For each type of cheating discussed, hazard communications constructs distinguish between whether it was something the subject participated in, saw happen, or heard about happening. Within each of these cate gories, statements are coded as specific or general. A specific incident is one that, if we had full de tails, could be tied back to a place, time and person or people. General incidents are sweeping statements with no hooks to isolate a part icular incident. Specific incidents may be indicated by the following Number of incidents is clear, e.g. (one time, a couple of times) Tied to a particular subject (in history) Written in past tense (people programmed formulas into their calculators, students would write out their notes) Particular people are mentioned (my frie nd, my roommate, a guy, I, people in my class, all of my friends) Statements that do not fit the specific categor ies (e.g. students bring ch eat sheets to class) are coded as general. Confessional (done): Negative self-disclosures
129 Characterized by a first person description of cheating, or other indicators that the subject engaged in cheati ng behaviors described. Specific example : In my organizations and systems class two years ago, my friend and I didn't understand the work and some guys thought we were cute ladies and helped us with every assignment. This is also a W incident since the su bject indicates othe rs were involved. General example : For people like myself who are computer illiterate, if I happen to have a question on how to perform a partic uliar task, you can bet th at I am not going to sit around waiting for a response, when I coul d call a friend and they could tell me the same thing. Witnessed (seen): Reports of ob served incriminating behaviors Characterized by language indicating that the subject obse rved the cheating behavior in question. Specific example : like in the computer class where the gu ys helped us...there were three of them that helped us. (This is also S since the subj ect indicates own involvement.) General example : i used to be in a frat they had old test and the convience of cheating was there and available and i saw that some probably 7-15% would Heresay (heard about): Relayed reports of received information that incriminates others Characterized by language indicating that the subject heard talk or otherwise received a report about the ch eating behavior in question.
130 Specific example : my roommates were talking about it earlier this semester. they are all taking internet classes and have used the discussion boards to cheat. General example : ive heard of friends paying others to write them for them if they just dont write very well or dont have the time or basically just dont want to Should a subject explain that a group or its members engage in a cheating behavior and that he or she is a member of the group (e.g., a fraternity), the incident should be coded as first hand knowledge, witn essed or confessional as appropriate, not heresay. The cheating penalty guideline Sometimes subjects may describe somethi ng they did that might not be cheating, as if they want to be responsive but have not hing to say or arent willing to say anything truly incriminating. Samples like these ma y be somewhat ambiguous: I asked my mom to read my paper to see if it made sense, A fter I wrote it out my essay, I got my friend to type it up because I type so slowly LOL, or Ive been asked for answers, but I didnt give them. If the behaviors described do not appear to be ones subjects feel could have subjected the perpetrators to cheating penalties, they should not be assigned hazard codes. Denials Denials are characterized by exculpat ory language. They are coded into categories congruent with hazard communications. Denials of personal impropriety incl ude language denying participating in cheating activity.
131 Examples : I have never been involved with any situation that cheating has been an issue on campus, I don't know because I don't cheat. Denials of witnessing others cheatin g include language denying firsthand knowledge of others cheating activity. Example : ive never seen someone cheat on a test Denials of relayed knowledge of othe rs cheating include language denying secondhand knowledge of ot hers cheating activity. Example : people dont really talk about cheating General denials do not fall in to the above categories. Example : I don't think cheating goes on here. The denial examples shown here are all hard denials. A hard denial disclaims knowledge of a hazardous topic, e.g., I don t cheat, or when asked about how cheating occurs, I dont know. In some cases, a de nial of knowledge may be suggested rather than stated. A soft denial implies a lack of firstha nd or secondhand knowledge without stating it outright, and often accompanies speculation. A soft denial may include statements like these: I would guess that, Maybe they, This is just a guess, most typically followed by a description of be haviors that would be coded as hazard communications if they were said to have happened rather than being framed as speculative. One category of close calls is wo rth nothing: statements of the type, I dont know besides what I already told you. Th is is not a denial, as long as some other comment refers to did addre ss the question asked. If a subject is deflecting with a comment like this no other comment did ad dress the question it is a hard denial. Beliefs
132 Beliefs are meaningful, non-procedural statements that do not fall into the above categories. They may appear in the same st atement as denials to justify the denial and can appear independently. A single belie f is one contained thought about cheating behavior. Examples : It can be solo [B1] and it can include another by the consent of another to allow it to happen [B2], anyone who is unde r a great deal of st ress of getting a good grade would cheat, i thing. Grouping Related Statements In addition to coding each statement with a hazard code, statements about the same incident of cheating get a common incident number. They are applied consecutively. The first confessional statement gets the incident number one, for instance. If a second statement talks about the same instance of cheating committed by the subject, that also receives an incident number of one A statement describing a different way or time the subject cheated would receiv e a different incident number. Quantifying incidents Incidents are coded conservative ly. For specific incidents, if the statement only indicates that something happened and quantity is not clear, one incident per category is recorded (e.g. per course if multiple courses are mentione d). If something is said to have happened more than once / was ongoing / was typical a nd quantity is not clear, two incidents are coded. For general statements, each statem ent is counted separately. There is no common incident number as th ere is no specific incident.
133 Incident numbers may be revised base d on new information provided during an interview. For instance, a subject may say initially, Ive seen cheating in math and science, which indicates at le ast two incidents. Later it may become clear that both Calculus and Statistics were included in the math description (3 incidents total so far), and that cheating by programming formulas into calculators happe ned in both calculus and statistics while looking on someones paper was seen only in statistics (the original two categories are now broken into at least 4 incidents). The specifi c counting is subject to the guidance listed above (whi ch could result in a count of more than four incidents). Incidents are coded at the grea test level discovered (at least two incidents in statistics, one in calculus, and at least one in science) propagating these back to the original statement. Ive seen cheating in math and science would therefore get at least four incident codes in this example to this statement to the subsequent statements to which it pertains. Denial statements on the same topic are also grouped by incident number.
134 Appendix 2: Standards Applied Measure 1. How much did complying with organizati onal rules for electronic communications affect what you said in the instant message interview? 2. How important was it to stick to your own views on what its OK to talk about when being interviewed over IM? 3. How great a role did the organizations acc eptable use policy for its communications networks play in how you handled the IM discussion? 4. How much did your own thoughts about how to use IM to discuss various topics affect what you said during the interview? 5. To what extent did this organizational st andard impact the IM discussion you just had? 6. To what extent did your own personal st andards for IM communications impact the IM discussion you just had? Each item was rated on a scale of 1 to 7 where 1 is Not at all, and 7 is A great extent.
135 Appendix 3: Self-Focus Scale Items, as Adapted from Matheson et al. 1988 1. During this IM discussion, I've generally been very aware of myself, my own perspectives, and attitudes. 2. Rather than being distracted by my task and what is going on around me, I have been thinking about myself in this IM discussion. 3. In this IM discussion, I have wondered a bout the way I've responded and presented myself in comparison to others who are answering these questions. 4. In this IM discussion, I have been thought ful about what people who later read the transcript of this chat may think of my responses. Each item was rated on a scale of 1 to 7 wher e 1 is extremely uncharacteristic of me, and 7 is extremely characteristic of me.
136 Appendix 4: Organizational Standard Manipulation Check From what you remember, does the USFs electronic communication standard permit IM messages like these? 1. A student saying how she got into a concert for free but against USFs rules 2. Entering a USF help desk call into the system as a desktop issue when the software was installed on a laptop 3. Telling somebody where to find t-shirts with a messed up USF logos on eBay when the shirts were supposed to have been destroyed. Each item is answered yes or no.
About the Author Carolyn F. Holton conducts research on computer mediated communications systems, including the impacts of system monito ring and user culture, the development of group norms, the spread of rumors and organi zational responses to them, their use for detecting and deterring fraud, and pedagogi cal applications. She also examines information systems leadership issues. Her computer mediated communications monitoring work is soon to appear in both IEEE Transactions on Professional Communication and Decision Support System s. She earned an MBA from Duke University, where she was designated a F uqua Scholar, and a BBA from The George Washington University, summa cum laude She was also selected to attend the 2007 ICIS doctoral consortium and has been nominated for the 2008 University of South Florida College of Business Doctoral Research Award.
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 002007631
007 cr mnu|||uuuuu
008 090622s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002440
Holton, Carolyn F.
The impact of computer mediated communication systems monitoring on organizational communications content
h [electronic resource] /
by Carolyn F. Holton.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 136 pages.
Dissertation (Ph.D.)--University of South Florida, 2008.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
ABSTRACT: Employer monitoring of communications is prevalent and on the rise due in part to the Sarbanes-Oxley Act, the Health Insurance Privacy Protection Act, and other legislation in the U.S. and other countries. However, the critical effect of this new activity on what is communicated in companies has not been assessed. This dissertation examines the impacts of computer mediated communication systems monitoring on neutral, incriminating and exculpatory content, as well as the overall volume of communications issued on monitored and non-monitored computer mediated communication systems. Incriminating communication is cataloged in a hazard communications taxonomy for this investigation. A controlled laboratory experiment has subjects participate in an instant messaging discussion on a topic for which they are likely to be aware of information that is incriminating to their organization, or its members, or both. Consistent with self awareness theory, monitored subjects engage in significantly less overall and neutral communication. They volunteer fewer high intensity hazard communications, but are less likely to curtail low intensity hazard communications. They issue denials about more incriminating topics. Contributions to research include theory development, especially in the area of standard selection; application of self-awareness theory to the new domain of computer mediated communications monitoring; a research framework; a taxonomy and coding scheme for the new hazard communications constructs; and a relative standards influence instrument and methodology for use in studying competing standards. Implications for corporate monitoring and communications policies are discussed, and a research agenda is outlined.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Co-advisor: Rosann W. Collins, Ph.D.
Co-advisor: Richard P. Will, Ph.D.
x Information Systems and Decision Sciences
t USF Electronic Theses and Dissertations.