Anonymity in Computer-Mediated Communication

Anonymity in Computer-Mediated Communication: More Contrarian Ideas with Less Influence Russell Haines, Jill Hough, Lan Cao & Douglas Haines

Group Decision and Negotiation Published in cooperation with the Institute for Operations Research and the Management Sciences and its Section on Group Decision and Negotiation ISSN 0926-2644 Volume 23 Number 4 Group Decis Negot (2014) 23:765-786 DOI 10.1007/s10726-012-9318-2

1 23

Your article is protected by copyright and all rights are held exclusively by Springer Science +Business Media Dordrecht. This e-offprint is for personal use only and shall not be selfarchived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.

1 23

Author's personal copy Group Decis Negot (2014) 23:765–786 DOI 10.1007/s10726-012-9318-2

Anonymity in Computer-Mediated Communication: More Contrarian Ideas with Less Influence Russell Haines · Jill Hough · Lan Cao · Douglas Haines

Published online: 30 September 2012 © Springer Science+Business Media Dordrecht 2012

Abstract Anonymity is thought to be an important means for ensuring a free exchange of ideas by encouraging the expression of minority viewpoints. However, we suggest that anonymity’s reduction in awareness of others potentially affects the expression and interpretation of comments that are made during a discussion. In particular, anonymity will increase the likelihood that comments will be made that are contrary to the majority opinion while at the same time decreasing the effect that those contrary arguments have on other group member’s opinions. This paper reports experimental results showing that anonymity led to more overall participation in discussions of ethical scenarios. However, equality of member participation did not differ between anonymous and member-identified groups, and anonymous groups had significantly higher awareness-related comments. This leads to the conclusion that additional participation in anonymous groups accommodates reduced awareness rather than reflecting the increased participation of normally reticent group members. In addition, anonymity

R. Haines (B) · L. Cao Department of Information Technology and Decision Sciences, College of Business and Public Administration, Old Dominion University, Norfolk, VA 23529 USA e-mail: [email protected] L. Cao e-mail: [email protected] J. Hough Department of Management and Marketing, Collins College of Business, University of Tulsa, Tulsa, OK, USA e-mail: [email protected] D. Haines Department of Business, College of Business and Economics, University of Idaho, P.O. Box 307, Garden Valley, ID 83622, USA e-mail: [email protected]

123

Author's personal copy 766

R. Haines et al.

led to more arguments in support of questionable behavior, suggesting that the freeing effects of anonymity apply to the social desirability of arguments. Finally, there was less change in opinion under conditions of anonymity than when comments were identified, suggesting that anonymous arguments have less influence on opinions than identified comments. Keywords Anonymity · Awareness · Computer-mediated communication and collaboration · Laboratory experiments

1 Introduction For many, the protection of free speech offered by anonymity is “sacrosanct” (O’Brien 2002, p. 2475), and some go as far to say “anonymity is an essential tool in protecting free speech and action on the Internet” (Helms 2001, pp. 304–305). Concealing the source of speech by allowing anonymous contributions is thought to allow participation without fear of rejection or harm while permitting information to emerge that might otherwise remain concealed (Helms 2001). This suggests that anonymity will encourage the contribution of ideas from those “whose views were not being heard in the decision-making process” (Stanley and Weare 2004, p. 521), and thereby “increase the number of comments received, expand the range of constituencies that were heard, and broaden the topics on which respondents commented” (Stanley and Weare 2004, p. 520) and, in addition, “encourage greater and more truthful disclosures” (Whitlock et al. 2006, p. 415). Anonymity is frequently employed in electronic meeting systems (EMS), with the idea that it will encourage otherwise reticent group members to participate in discussions and thereby reduce the potential for a discussion and decision dominated by a few members (Kahai et al. 1998). By freeing participants from the possibly negative evaluations of others, anonymity is thought to encourage the contribution of more diverse ideas (El Shinnawy and Vinze 1998). However, while theories of group decision-making via computer-mediated communication (CMC) suggest the potential for anonymity to encourage the presentation of minority viewpoints (DeSanctis and Gallupe 1987; Gopal and Prasad 2000; Dennis and Garfield 2003), the influence of minority opinions may be less via CMC than when opinions are expressed face-to-face (McLeod et al. 1997; Dennis 1996). Furthermore, increased anonymous participation may not necessarily have an impact on a final decision because the views of identified stakeholders may be given more weight (Stanley and Weare 2004). EMS researchers generally believed that anonymity strips away status cues thereby leaving opinions to shift only as a result of the persuasiveness of arguments (Dubrovsky et al. 1991; Weisband 1994; El Shinnawy and Vinze 1998). However, the Social Identity model of Deindividuation Effects (SIDE) proposes that the relative anonymity of CMC gives group members a common identity thereby leaving members more susceptible to group influence (Spears and Lea 1992). For example, members of groups interacting via anonymous CMC were shown to be more susceptible to group influence than face-to-face groups (e.g., Postmes et al. 1998, 2001; Spears et al. 1990; Spears and Lea 1992). In another study, anonymous groups produced riskier decisions than those

123

Author's personal copy Anonymity in Computer-Mediated Communication

767

made by face-to-face groups (Valacich et al. 2002). Anonymity also appears to affect how comments made by others are received, with anonymously shared information seeming less credible (Dennis 1996), or anonymously made comments having less influence on personal opinions (McLeod 2000; Sassenberg and Postmes 2002). This suggests the need for a more complex view of anonymity’s effect on the expression of alternative viewpoints and on group opinion shift. Our motivation for this study comes from unanticipated findings in three studies that compared anonymous and identified groups: one in which subjects were found to shift their opinions less when their comments were anonymous (McLeod 2000), a second in which subjects were found to shift their opinions less when others’ comments were anonymous (Sassenberg and Postmes 2002), and another in which anonymous groups were found to make riskier decisions than face-to-face groups (Valacich et al. 2002). Anonymity alone cannot explain the differences; thus, we propose that the content of comments interacts with the anonymity of group members. We wish to illuminate group processes via CMC by examining the number and the type of comments generated and the associated effect on opinion shift between groups where comments are anonymous and those where comments are identified. We approach anonymity from a new perspective, focusing on anonymity’s effects on awareness of others. We examine this is in an experimental context where group members discuss brief ethical scenarios via synchronous CMC. We suggest that anonymity’s reduction in awareness of others potentially affects the expression and interpretation of comments that are made during a discussion in two ways. First, we suggest that anonymity will shield group members from the negative evaluations of others, enabling them to hold and express arguments that are counter to the norms of the group. Second, we suggest that a lack of awareness of others’ presence and/or identity in an anonymous setting will decrease the influence that members have in shifting group opinion. Thus, anonymity will increase the likelihood that comments will be made that are contrary to the majority opinion while at the same time decreasing the effect that those contrary arguments have on other group member’s opinions. We begin with our theoretical foundation, which is contrasted with the Social Identity model of Deindividuation Effects (SIDE), and develop a research model and hypotheses for examining the effect of anonymity on participant comments and anonymity’s ultimate effect on the intention to engage in a particular behavior (i.e. the opinion shift of the group). Next we describe our experimental design and report our results. Following a discussion of the findings, we identify limitations of this study along with implications for practice.

2 Theoretical Foundation SIDE proposes that members of a group will conform to the group’s norms when identifying with the group becomes more important than their own individual identity (Spears and Lea 1992). This is derived from the group polarization phenomenon (Spears et al. 1990), which has been studied extensively (Isenberg 1986; ML76). Group polarization describes the phenomenon where “an initial tendency of individual group members toward a given direction is enhanced following group discussion” (Isenberg

123


R. Haines et al.

1986, p. 1141). There is a rich history of group polarization theory centered on two probable drivers for the effect (Isenberg 1986; Myers and Lamm 1976). The first driver of group polarization, social comparison, suggests that the opinion shift occurs because individuals adjust their own preferences as they compare their feelings with those expressed by others. During the interaction, an individual will see that others are strongly for or against the topic and then change their opinion in that direction because they feel that such a change would place their opinion in a more socially desirable position. For example, as a group discusses nuclear power, members will see that others feel strongly one way or the other about phasing out nuclear power stations. If the initial tendency of the group was that power stations should be phased out, those that felt less strongly (or were opposed) may change their opinions in order to appear to be in line with other members of the group. Thus, after the interaction, the overall opinion of the group will have shifted because members revised their opinion after seeing that other members favored the initial tendency. The second driver of group polarization, persuasive arguments, suggests that when a group is formed, the members already possess a set of arguments that are predominantly either for or against the decision. As the group discusses the issue, the members are exposed to the arguments and shift their opinions in the direction favored by the arguments. For example, in a discussion of nuclear power, each group member has a set of arguments either for or against phasing out nuclear power plants. If the initial tendency of the group was that power stations should be phased out, the group members would be more likely to possess a set of arguments that is more strongly in favor of phasing out nuclear power plants. As other members that were less strongly against nuclear power plants are exposed to these arguments against nuclear power plants, they are likely to change their opinions as a result. Thus, after the interaction, the overall opinion will have shifted because members responded to arguments favoring the initial tendency. This paper focuses on judgment problems such as these. Polarization tendencies may be different when groups discuss discrete problems with known alternatives where logic rather than persuasion drives the decision process. 2.1 Group Polarization and Anonymous CMC Anonymous CMC filters cues to individual identity that would ordinarily be present in face-to-face communication (Walther 1992). Thus, CMC researchers suggest that the drivers of group polarization are different when a group’s members are anonymous versus settings where members are identified. First, anonymity removes the group’s mechanism for tying comments to individual members, potentially freeing normally reticent group members to express opinions when they ordinarily would not because of evaluation apprehension (Rao and Jarvenpaa 1991). From this viewpoint, anonymity should mitigate the polarization effect by leading to more arguments “for the opposition” and, therefore, a less extreme shift in opinion from the initial tendency. Second, anonymity is thought to filter out many status cues such as age, appearance, and gender (Dubrovsky et al. 1991; Weisband 1994; El Shinnawy and Vinze 1998). According to this reasoning, less social comparison should occur, leaving opinions to shift based primarily on the persuasiveness of the arguments. In effect, anonymity

123


769

should mitigate the polarization effect because group members only respond to the persuasiveness of arguments and don’t feel as strongly that they need to conform to the views of other group members. Third, as noted earlier, SIDE proposes that the “inherent anonymity” of CMC provides a common identity to participants and makes them more susceptible to group pressure (Postmes et al. 1998). Having a common identity, group members feel more pressure to conform to group norms when they are communicating via CMC than when communicating face-to-face (Diener 1980). As a result, polarization would be more pronounced. However, El Shinnawy and Vinze (1998) found that anonymous CMC groups showed no signs of polarization while face-to-face groups polarized. This has led some to assert that anonymous arguments might be less persuasive than identified arguments (Dennis 1996; McLeod 1997, 2000). Two explanations for this have been offered. One explanation focuses on the degree to which social cues remain intact via CMC according to the perceived anonymity (Pinsonneault and Heppel 1997; Dennis and Garfield 2003), or the degree to which group members feel that they are unable to identify the authors of comments (Pinsonneault et al. 1999). As a result, anonymous settings may not mitigate the polarization effect if the participants don’t actually feel anonymous. A second explanation focuses on the quantity of information that can be expressed verbally and non-verbally in a face-to-face situation over a given time period versus how much can be conveyed by a group typing at keyboards. In essence, anonymous groups, given enough message exchanges, should be experienced in much the same way as face-to-face groups (McLeod 1997, 2000; Walther 1992). Over the course of message exchanges anonymous individuals will feel and become more identifiable (cf., Pinsonneault and Heppel 1997; Hayne et al. 2003). We suggest another alternative, that anonymous CMC reduces the awareness of others. In addition to filtering out accountability cues (Prentice-Dunn and Rogers 1982), anonymity in CMC also filters out cues that communicate presence and identity, which are necessary for one to be able to characterize the comments of others (Cooper and Haines 2008). We suggest that awareness of others is a set of fundamental cues that are lost when interacting via anonymous CMC. The loss of awareness cues in anonymous CMC mitigates group members’ ability to coordinate their interaction (Gutwin and Greenberg 2002), leads to disinhibited behavior (Prentice-Dunn and Rogers 1982), and reduces the influence of the group (Short et al. 1976; Diener 1980). Group members interacting via anonymous CMC have been observed accommodating for the loss of awareness cues through practices such as adding identifying labels to comments (McLeod 2000). We suggest that such user practices are one mechanism that enables awareness of others to build to the point where anonymous CMC can be experienced in a way that more closely resembles face-to-face communication (cf., (Walther 1992; Frößler 2006; Riemer et al. 2007). 2.2 Research Model Overall, we examine whether polarization is reduced via anonymous CMC when compared with identified CMC. This reduction should be the result of differences with

123


R. Haines et al.

Fig. 1 The effect of anonymity on comments and their influence

respect to the types of comments made during a discussion via CMC, yielding differences in opinion shift afterward. Figure 1 summarizes our research model. Individual differences such as gender and ego strength have been shown to affect the ethical decision making process (Haines and Leonard 2007), and therefore might affect the number of comments made and/or the effect of comments on ethical decision making, but are outside the scope of this paper. Anonymity gives individual group members the feeling that others cannot evaluate their individual comments (Pinsonneault and Heppel 1997; Valacich et al. 1992). This in turn could affect the quantity of group interaction in two ways. First, when rewards are given at the group level rather than the individual level, anonymity gives rise to free riding and less participation (Barua et al. 1995). Second, when members are ordinarily reticent to participate, anonymity frees them from evaluation apprehension and leads to more participation (Rao and Jarvenpaa 1991; Pinsonneault et al. 1999). Based on prior research on anonymity in CMC, the freeing of reticent members from evaluation apprehension would seem to have a more pronounced effect on participation than free riding. For example, anonymity increased actual participation over identified groups in CMC settings (Kahai et al. 1998). Furthermore, groups using anonymous CMC had higher perceptions of participation than face-to-face groups (Valacich et al. 1992). An awareness perspective provides an alternative explanation for increasing participation when comments are anonymous: that members of anonymous groups will participate more because they feel it necessary to accommodate for the cues that are not present. In anonymous CMC groups, the strongest cue is most likely the lack of presence, and should lead to more group development oriented comments under conditions of anonymity [e.g. “Is anybody out there?” (cf., Jarvenpaa et al. 1998)]. Thus, while there may be more participation in anonymous groups, the increased participation may not be a result of participation by normally reticent members; rather, it may be a reflection of the need to communicate more in order to accommodate for the lack of cues (cf. Walther 1992; Frößler 2006; Riemer et al. 2007; Haines and Riemer 2011). We suggest that the overall participation in anonymous groups will be higher than for identified groups, but the equality of participation among group members will not differ between anonymous and identified groups. Hypothesis 1 Although overall participation by group members will be higher when anonymous than when identified, the equality of participation will not differ. Anonymity may insulate members from evaluation apprehension (Valacich et al. 1992; Pinsonneault and Heppel 1997), leading to more overall comments because of the participation of normally reticent members (Rao and Jarvenpaa 1991;

123


771

McLeod 1997) or because of an increase in awareness cues as we have suggested. However, the content of participation may be more important than the quantity. In prior studies, anonymous groups generated more original solutions than identified groups (Connolly et al. 1990) and produced more novel ideas than face-to-face groups (El Shinnawy and Vinze 1998). However, we suggest that the kinds of comments will vary because anonymity removes accountability cues. In effect, anonymous participants feel free to present socially undesirable arguments (Prentice-Dunn and Rogers 1982). Thus, in a discussion of ethical issues, we expect more arguments in support of an ethically questionable action when comments are anonymous. Hypothesis 2 The number of arguments that are supportive of an ethically questionable action will be higher in anonymous groups than in identified groups SIDE’s propositions have been substantiated by many studies (Lea et al. 2001; Postmes et al. 1998, 2001; Postmes and Lea 2000; Spears et al. 1990; Spears and Lea 1992; Hiltz et al. 1989). However, these studies used a label to operationalize anonymity, which provides a considerable opportunity to gain its own social identity even over a relatively short period of time (McLeod 1997; Walther 1992; Lee and Nass 2002). More recent research suggests that individual’s opinions may not shift to the same extent when other group members’ comments are completely anonymous versus when they are identified (Sassenberg and Postmes 2002; McLeod 2000), and that argument quality does not affect opinion shift when comments are completely anonymous (Lee 2007). We suggest that a common social identity is generally not communicated under complete anonymity and is unlikely to be formed until later in an interaction. Lacking a social identity, completely anonymous comments may be perceived as being less salient than identified or even nominally labeled comments. This is consistent with Dennis (1996), who speculated that his group decision support system groups did not perform better than identified groups because “anonymity… made the information more suspect because it was difficult to verify the source’s credibility” (p. 450), and McLeod (2000), who speculated that members of her anonymous groups were not shifting their opinions as much as the identified groups because “just as people are unattached to their own statements when they communicate anonymously, they are analogously unaffected by the anonymous statements of others” (p. 197). Thus, anonymous comments will be less convincing than identified comments due to the effects of social comparison. The social comparison effect was partially acknowledged earlier in our suggestion that anonymity will free group members to express socially undesirable comments. We extend Dennis’ and McLeod’s arguments by offering the notion that anonymity further reduces social comparison by hindering group members’ ability to be aware of the presence and identity of other members of their group (cf., Cooper and Haines 2008). Consider that in an anonymous setting, a person may see arguments expressed in a discussion, but cannot necessarily determine (1) who made a particular argument, (2) how many different people have expressed a similar argument, (3) the degree to which the person that expressed a particular argument is similar to him/herself, and (4) whether a series of arguments are all coming from the same person (cf., Hayne et al. 2003; Gutwin and Greenberg 2002; McLeod 2000; Dennis 1996; Lee 2007).

123


R. Haines et al.

The effect of arguments on opinion shift in a group setting should depend on whether arguments seem to come from one person or from multiple people and the extent to which those people are considered to be similar to oneself (Lee 2007). Thus, a series of similar arguments can be attributed to the continuous ranting of a single person. This means that anonymity’s removal of the mechanism for social distinction would also mitigate the persuasiveness of arguments and have less effect on opinion shift. We suggest that anonymity will affect the impact of arguments on opinion shift, such that if the same number of arguments is expressed, anonymity will lead to less opinion shift than when comments are identified because of decreased social comparison. Hypothesis 3a As more arguments supportive of an ethically questionable action are expressed, group opinion shift toward intending to engage in questionable behavior will be more under identified than anonymous conditions. Hypothesis 3b As more arguments critical of an ethically questionable action are expressed, group opinion shift away from intending to engage in questionable behavior will be more under identified than anonymous conditions. 3 Research Design and Methods A laboratory experiment was conducted where participants were randomly assigned to either anonymous or identified CMC groups. Groups were given judgment tasks requiring an online discussion of whether the course of action taken in five different scenarios was ethical or unethical. Individual participants first read an informed consent document, entered their first name in a web form, and completed basic demographic information such as gender, age, major, and grade point average. Before the group discussion, participants read the five scenarios and completed a questionnaire assessing the probability that the participant would engage in the same questionable action as the actor in the scenario. The scenarios were presented in a different, random order to each participant. The five scenarios were taken from Robin et al. (1996) (Scenario A) and Reidenbach (1991) (Scenarios B–E). The only changes to the scenarios were to remove all gender references to mitigate the possibility that participants might make gender based judgments. The complete text of the scenarios is given in Appendix 1. Ethical scenarios were chosen for several reasons. First, subjects would be more likely to have something to contribute to the discussion—none would feel that their lack of expertise or limited background precluded their contribution (cf. Kahai et al. 1998, 2003). Second, individuals are more motivated when generating ideas about sensitive issues than more neutral decisions such as “should X make business decision Y” (Pinsonneault et al. 1999). Third, judgmental tasks such as these have been found to generate a larger pool of persuasive arguments than intellective tasks (El Shinnawy and Vinze 1998). Fourth, evaluation apprehension between anonymous versus identified comments would be more pronounced for socially sensitive issues than more neutral issues (Pinsonneault et al. 1999). Multiple scenarios were chosen to make this study less vulnerable to idiosyncrasies of particular topics or particular discussions. Previous studies of group polarization

123


773

via CMC have involved one or two scenarios as a rule; so their results may reflect idiosyncrasies of the topic(s) and limit generalizability (Spears et al. 1990). Once all of the participants completed the scenario questionnaire, they were randomly assigned to a group and groups were randomly assigned to a treatment condition. Each group discussed the five scenarios in a different, random order in a chat room. After his or her group had discussed a scenario, each participant once again answered the questions assessing the probability that they would engage in the questionable action. Chat rooms for the groups were identical, except that in identified chat rooms, each comment was listed on the message window with the person’s first name next to it, while in anonymous chat rooms comments were preceded by a bullet. All comments appeared on the display in the order they were entered, which allowed for normal discussion flow. This condition could be considered a web-based version of the anonymous condition used in prior studies that used a group decision support system to facilitate discussions. In those studies, anonymous comments were displayed with a sequential comment number but without an author-identifying alias or name (cf., Hayne et al. 2003; McLeod 2000; Kahai et al. 1998; Dennis 1996). The first discussion lasted four minutes to allow time for the groups to get used to having a discussion in the chat room, while the remaining discussions lasted three minutes. Consistent with Leonard and Haines (2007), we chose three minutes to allow enough time to present relevant arguments, but not so much time that they began to have off topic conversations. Statistical tests included a control variable to indicate whether or not the scenario under consideration was the first scenario discussed by the group.1 Experimental sessions were conducted during the normal class period of a juniorlevel course required of all business majors at a public western U.S. university, each with between 26 and 47 participants, totaling 219 participants in 42 groups. 33 groups had five members and nine groups had six members. Seventy-four percent of the participants were male and the average age of all participants was 22 years. The discussion was a required part of a learning unit on ethical decision-making. There were no additional incentives for participating. Because of the potential for group processes to vary by group size (Chidambaram and Tung 2005; Pinsonneault et al. 1999), we analyzed data that included only the five-member groups alongside an analysis using all of the groups. To control for differences in perceived anonymity between treatments, all participants in a session were present physically in the same room but were randomly assigned to groups and treatments. Anonymous and identified groups were intermingled and received the same instructions (cf., McLeod 2000). The intermingled groups and treatments controlled for the possibility that participants in anonymous groups would systematically vary in their perceived anonymity by using visual, auditory, and/or textual cues to assign social characteristics to the other group members (cf.,Hayne et al. 2003; McLeod 1997, 2000; Pinsonneault and Heppel 1997; Valacich et al. 1992; Weisband 1994). To our knowledge, this is the first study of anonymity in which group assignments and treatments were random and intermingled within sessions. Class members 1 Employing dummy variables to control for all five discussions did not change the results of the hypotheses,

and showed that the later discussions did not vary significantly from each other.

123


R. Haines et al.

were at least minimally acquainted with each other, meaning that members of identified groups understood that the members of their group were in the same class and present in the room, but may not have been able to locate and/or personally identify all members of their group. Sessions were conducted in a large classroom in which each participant used his or her personal laptop to access the experimental system. The design of the room provided spacing between participants such that, given the relatively small size of the laptop screens, they could not easily observe the screen of the person(s) sitting next to or in front of them. Again, treatments were randomly assigned within sessions, and all participants and sessions received the same on-screen and scripted verbal instructions. Pre- and post-discussion questions regarding the probability that the participant would engage in the questionable behavior were used to measure opinion shift. The number and type of arguments generated during the discussions were obtained through content analysis of the interaction among group members. The effects of comment identification on (a) total participation, (b) type of argument and (c) the opinion shift of group members were examined using Hierarchical ANOVA (Walczuch and Watson 2001).

3.1 Variables Participation was measured as the total number of keystrokes that were entered by group members during a chat period. We chose keystrokes as our measure of participation because it is an objective measure of the total amount of information that was displayed in a discussion. Measuring participation as the number of utterances during a chat period (the number of times text was entered and submitted to the chat) and as the number of complete thoughts (coding unit) yielded the same results. Equality of participation was measured using the Gini coefficient, a continuous variable in which a group would score zero if their contributions were completely equal (i.e., each group member had equal participation), and one if their contributions were completely unequal (i.e., only one group member was responsible for all participation). Complete transcripts for all group discussions were content analyzed. During the coding process, a computer program presented a transcript from a randomly chosen group about a randomly chosen scenario to the coder. The coder was blind as to whether a particular group was identified or anonymous. A total of 5,955 thoughts were analyzed. To rate the reliability of the coding schema and coder, a validating coder followed the same procedure and coded 250 thoughts using the same random, blind coding technique. The overall inter-coder reliability using the Kappa coefficient was 0.71, which is considered “substantial” (Landis and Koch 1977), with an overall percent agreement of 89 %. Arguments were thoughts identified by the content analysis that contained logic or evidence either in support of (supportive arguments) or critical of (critical arguments) the questionable behavior in the scenario. Intention to Engage in Questionable Behavior was measured as the average preinteraction responses minus the average post-interaction responses within each group for moral intent. This measures the probability that a person would engage in the same questionable action as the actor in the scenario. The moral intent items were: (1) if

123


775

you were the [actor], what is the probability you would have done the same thing? (anchored Highly Probable–Highly Improbable), (2) if I had the opportunity, I would do the same thing that the [actor] did. (anchored Likely–Unlikely), and (3) I would never do what the [actor] did. (anchored True–False, reverse coded). All items were on seven point Likert scales, with the highest value (seven) indicating the respondent was least likely to do the same action. Thus, positive numbers for the shift from pre- to post-interaction indicate a shift toward being more likely to engage in the questionable behavior as a result of the discussion. Reliability of the scale was .966 (Cronbach’s alpha). The first item is from Robin et al. (1996); the other two are from Beck and Ajzen (1991). 4 Results As noted earlier, separate analyses were conducted using all of the data (42 groups) and using only the five member groups (33 groups). The conclusions were the same for both sets of data, but we report results for the more conservative, 33 group data set, which included 16 anonymous and 17 identified groups. The means and standard deviations of all of the dependent variables by treatment and scenario are shown in Table 1, and correlations are shown in Tables 2 and 3. The data were analyzed using a nested factorial design (HANOVA) where group within treatment was a nested factor, and treatment and scenario were crossed factors. The error term for all of the statistical tests is the group effect (Walczuch and Watson 2001). The results are summarized in Table 4. Hypothesis 1 Hypothesis one predicted that anonymous groups would have more overall participation, but would not differ from identified groups with respect to equality of participation. As we predicted, and consistent with prior research, anonymous groups had significantly more participation (anonymous average = 878, identified average = 722, F = 6.09, df = 1,31, p = .02) than identified groups. However, also as predicted, anonymous and identified groups did not significantly differ with respect to equality of participation (anonymous average = .249, identified average = .272, F = 1.22, df = 1,31, p = .28). Hypothesis 2 Hypotheses two predicted that anonymity would increase the number of arguments that were supportive of the ethically questionable behavior during the discussions. As predicted, anonymity increased the number of supportive arguments generated in the groups (anonymous average = 4.63, identified average = 2.93, F = 5.73, df = 1,31, p = .02). Hypothesis 3 Hypotheses three (a) and (b) predicted that the impact of supportive and critical arguments on intention to engage in questionable behavior would be less when comments were anonymous. HANOVA indicated a significant interaction between supporting arguments and anonymity (F = 4.32, df = 1,31, p = .046), but did not support the interaction between critical arguments and anonymity (F = 0.74, df = 1,31, p = .40). A simple slopes analysis (Aiken and West 1991) of the interaction between anonymity and arguments on the shift in intention to engage in questionable behavior is summarized graphically in Fig. 2.

123


R. Haines et al.

Table 1 Means and SD of dependent variables by treatment and by Scenario Overall

Scenario A

B

C

D

E

Participation Anonymous Mean

878.0

884.1

919.1

864.5

964.2

757.8

SD

252.3

214.9

304.4

273.8

235.5

201.6

Identified Mean

721.9

767.1

723.5

713.9

742.8

662.1

SD

219.2

223.3

219.4

157.2

241.5

254.3

Equality of participation Anonymous Mean

.249

.260

.254

.233

.254

.245

SD

.094

.086

.113

.089

.099

.092

Identified Mean

.272

.267

.269

.256

.295

.270

SD

.098

.077

.114

.095

.090

.116

Supportive arguments Anonymous Mean

4.63

6.94

2.13

4.19

6.88

3.00

SD

4.11

4.15

2.31

3.73

4.00

3.98

Mean

2.93

5.53

1.53

1.53

4.71

1.35

SD

2.96

2.43

2.70

1.77

3.00

1.84

Identified

Critical arguments Anonymous Mean

3.69

2.25

4.13

5.38

2.94

3.75

SD

3.40

2.35

3.63

4.63

2.98

2.41

Mean

3.68

2.29

4.18

5.59

1.88

4.47

SD

2.76

1.99

2.21

2.98

1.76

2.98

Identified

Intention to engage Anonymous Mean

.228

.611

−.023

.146

.413

−.008

SD

.670

.573

.517

.706

.655

.716

Identified Mean

.193

.865

−.217

−.470

SD

.895

.780

.692

.534

.753 1.00

.034 .586

The graphs show that an increase in supporting arguments generally tended to shift opinions toward being more likely to engage in questionable behavior. On the other hand, critical arguments tend to shift opinions toward being less likely to engage in questionable behavior only under the anonymous condition. Reflecting the

123


777

Table 2 Correlation table of continuous variables

1. Participation

Overall mean

SD

1

2

3

797.55

247.79

2. Equality of participation 3. Supportive arguments

0.26

0.10

3.75

3.65

0.52**

−0.13

4. Critical arguments

3.68

3.08

0.39**

−0.25**

0.79

0.13†

−0.09

5. Intention to engage

0.21

4

−0.30**

−0.13† −0.22**

0.49**

n = 165 † p