5 z... ....... . ..v. yl. n. a. . . 5...... > "nffl re ... a ‘ n... a.“ . u 5. IX... it... di 1, . 9.35%: ,.. v . 1 .fl... at“ 9.9%.. 2 dw,n...‘vm..w; {hi .5... . A»? I 4%.”... ufififlf stun ..‘ Lu: .2. . :8» a 3%.: hfifiv .... V. .. a .. ‘. 5 “muemmyxfiww . to. .1. i h: €93.31 my. . 72. .. 1! .u ‘1 09:: r . . vn . ‘ I 2: 7:12.. 5.. . uvlt 1.... n! . . i . I: :7! L. Yilrayuflnu , L2. .17.? J9 . A 52 : mun! , . e‘ . :3 1w . . _. LP? 21.: 1:55.] 1;: I. . L . . A $735.”; ‘ «”332 ”III! u I ‘ . . “\.J ‘J LIBRARY Michigan State University This is to certify that the dissertation entitled COMPARING UNCERTAINTY REDUCTION IN FACE-TO-FACE AND COMPUTER-MEDIATED COMMUNICATION: A SOCIAL INFORMATION PROCESSING THEORY PERSPECTIVE presented by DAVID WESTERMAN has been accepted towards fulfillment of the requirements for the PhD. degree in COMMUNICATION Major Professor’s Signature “‘08“07 Date MSU is an affinnative-action, equal-opportunity employer PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 6/07 p:/CIRC/DaIeDue.indd-p.1 COMPARING UNCERTAINTY REDUCTION IN FACE-TO-FACE AND COMPUTER-MEDIATED COMMUNICATION: A SOCIAL INFORMATION PROCESSING THEORY PERSPECTIVE By David Westerman A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Communication 2007 ABSTRACT COMPARING UNCERTAINTY REDUCTION IN FACE-TO-FACE AND COMPUTER-MEDIATED COMMUNICATION: A SOCIAL INFORMATION PROCESSING THEORY PERSPECTIVE By David Westerman Social Information Processing Theory (SIPT; Walther, 1992) posits that the lack of nonverbal cues in computer-mediated communication (CMC) poses challenges for accomplishing interpersonal goals using this channel. However, the theory assumes that individuals can find ways to overcome this limitation and that goals can be accomplished equally well in CMC as in face-to-face (FtF). One interpersonal goal that previous CMC research has examined is uncertainty reduction (Tidwell & Walther, 2002; Westerman & Tamborini, 2005, 2006). Consistent with SIPT, past research has shown that people can reduce uncertainty in both CMC and FtF (Westerman & Tamborini, 2006), and that one way people overcome the lack of nonverbal cues is by asking more questions and disclosing more in CMC (Tidwell & Walther, 2002). Although past studies have examined uncertainty reduction in both CMC and FtF, these studies have not examined SIPT processes over time. The current study was designed to examine the uncertainty reduction process in CMC and FtF over time, by addressing the two main assumptions of SIPT. To this end, patterns of uncertainty before, during, and after fifteen minute interactions in FtF and CMC were examined. Patterns of interactive uncertainty reduction strategy use across condition were examined over time due to their expected potential to reduce uncertainty. Two additional CMC with avatar conditions manipulated the informativeness of an avatar used in interaction to examine the role avatars play in circumventing the lack of nonverbal information in CMC. In order to examine SIPT processes for uncertainty reduction an experiment was designed comparing patterns of uncertainty and IURS use across mode. One hundred and twenty- four participants interacted with a stranger in one of four conditions: FtF, CMC-no avatar, or one of two CMC with avatars conditions in which the informativenss of the avatar differed (CMC-high informativeness or CMC-low informativeness). After a fifteen minute interaction with their partner, participants filled out measures of uncertainty (Parks & Floyd, 1996) about their partner, as well as measures of other judgments related to uncertainty. Then participants recorded their own certainty minute by minute throughout the interaction using a form of protocol analysis (while reviewing their interaction to aid recall). Finally, trained coders reviewed the transcripts and coded the frequency and type of IURS use minute by minute throughout the interaction. Consistent with predictions of SIPT, uncertainty levels began higher in CMC than FtF, although the gap between the two channels appears to narrow as time elapses. Patterns of uncertainty did not differ across communication modes over time as expected. Instead, the patterns of uncertainty for CMC and FtF interaction were very similar. However, as predicted, IURS use in F tF appeared to start high, quickly drop, and then level off, whereas IURS use in CMC started high but dropped more slowly, and was always lower than FtF use as predicted by channel restrictions. Finally, the use of avatars in the study did not have strong effects on CMC interactions. These findings are discussed, as are potential limitations of the study. Last, the study’s implications for fiIture research are discussed, considering questions both old and new for CMC. ACKNOWLEDGEMENTS First and foremost, I want to thank my advisor, Dr. Ron Tamborini. There were many times when I did not think this day would come. I could write pages and pages about all you have done for me. I am generally a man of few words, however, so I will simply say that I would not be at this point if not for you. I know it is cliche’, but it is true. Thank you for all your effort and time. I also would like to give thanks to my committee members, Dr. Sandi Smith, Dr. Tim Levine, and Dr. Mark Levy. Thank you for all of your helpful suggestions on this manuscript. And thank you for all of the helpful advice and comments you have given me throughout my graduate school career. They are lessons I will continue to use throughout my life. An extra special thank you goes out to Dr. Joseph Walther for inspiring this line of research and for offering a lot of advice and suggestions along the way. Thank you for being an unofficial committee member and for all the help you have given me. I also want to thank all of the undergraduate and graduate assistants who have helped me collect and code data for this project. There are too many to name individually, but I appreciate all the help I have received. Finally, I would like to thank my family and fi’iends for your constant love and support. You have helped make me who I am, and not only academically, but as a person. I especially want to thank my wife, Catherine, for all of her help, love, and support. I am looking forward to repaying the favor for you for the rest of my life. iv TABLE OF CONTENTS LIST OF TABLES ............................................................................................................ vii LIST OF FIGURES ......................................................................................................... viii Literature Review ................................................................................................................ l Uncertainty Reduction Theory ........................................................................................ 5 Predicted Outcome Value Theory ................................................................................... 7 Online Interaction ........................................................................................................... 9 Social Information Processing Theory .......................................................................... 12 Westerman and Tamborini's Model of Uncertainty Reduction across Modes ............. 14 Patterns of Uncertainty Reduction in FtF and CMC ..................................................... 18 The Influence of IURS on Uncertainty across Modes .................................................. 24 Avatars and Uncertainty Reduction in CMC ................................................................ 28 Method .............................................................................................................................. 31 Overview ....................................................................................................................... 31 Participants .................................................................................................................... 31 Procedure ...................................................................................................................... 32 Measures ....................................................................................................................... 33 Mood measure. .......................................................................................................... 34 Uncertainty measure. ................................................................................................ 35 Liking measure .......................................................................................................... 36 Predicted outcome value measure ............................................................................. 38 Thought listing procedure. ........................................................................................ 40 Modified protocol analysis. ...................................................................................... 41 Pilot test of measures used for modified protocol analysis ....................................... 41 Transcript coding. ..................................................................................................... 46 Treatments and Stimulus Materials .............................................................................. 47 Pilot test of avatars used in current study. ................................................................ 48 Results ............................................................................................................................... 51 Analysis plan ................................................................................................................. 51 Tests of hypotheses ....................................................................................................... 53 Post hoc analyses .......................................................................................................... 60 Uncertainty over time across all communication mode conditions. ......................... 61 Liking and comprehensive measures of uncertainty and predicted outcome values after 15 minutes ......................................................................................................... 64 Percentage of IURS use over time. ........................................................................... 69 Predicted outcome values over time across all communication mode conditions... 71 Discussion ......................................................................................................................... 75 Observations made prior to interaction ......................................................................... 77 Observations made after interaction ............................................................................. 79 Observations made during interaction .......................................................................... 81 Limitations .................................................................................................................... 84 Directions for Future Research ..................................................................................... 90 What information matters? ....................................................................................... 90 Does CMC lack nonverbal information? .................................................................. 96 Are all limitations of CMC overcome? ..................................................................... 99 Conclusion .................................................................................................................. 101 APPENDICES ................................................................................................................ 105 Experimental Scripts ................................................................................................... 105 Consent Form .............................................................................................................. 116 Mood Measure (Oliver, 1993) .................................................................................... 118 Uncertainty Measure (Parks & Floyd, 1996) .............................................................. 120 Liking Measure (McCroskey & McCain, 1974) ......................................................... 121 Predicted Outcome Value Measure (Sunnafrank, 1988) ............................................ 122 Thought Listing Procedure (Cacioppo & Petty, 1981) ............................................... 123 Protocol Analysis Measure ......................................................................................... 125 Study Assessment Measure ......................................................................................... 131 Debriefmg Sheet ......................................................................................................... 132 IM interaction for modified protocol analysis measure pilot study ............................ 133 Coding Instructions ..................................................................................................... 135 Avatars used in main study ......................................................................................... 137 Correlation Matrix Of Variables ................................................................................. 138 Table of Means and Standard Deviations ................................................................... 139 References ....................................................................................................................... 140 vi LIST OF TABLES Observed Inter—item Correlations and Factor Loadings for Negative Mood .................... 35 Observed Inter-item Correlations and Factor Loadings for Uncertainty .......................... 36 Observed Inter-item Correlations and Factor Loadings for Liking .................................. 37 Observed Inter-item Correlations and Factor Loadings for Liking (without item 5) ....... 38 Observed Inter-item Correlations and Factor Loadings for Predicted Outcome Values .. 39 Observed Inter-item Correlations and Factor Loadings for Predicted Outcome Values (no 3 or 7) ................................................................................................................................ 40 Uncertainty Means Over Time (standard deviations) ....................................................... 55 Number of IURS Means Over Time (standard deviations) .............................................. 58 Uncertainty Means Ovethime (standard deviations) ....................................................... 62 Uncertainty Multi-level Modeling Values ........................................................................ 68 Liking Multi-level Modeling Values ................................................................................ 68 Predicted Outcome Value Multi-level Modeling Values .................................................. 69 IURS Percentage Means Over Time (standard deviations) .............................................. 70 Predicted Outcome Value Means Over Time (standard deviations) ................................ 72 vii LIST OF FIGURES Westerman and Tamborini’s model of uncertainty reduction across modes .................... 14 Uncertainty across mode as a function of time. ................................................................ 24 IURS use (total number) across mode as a function of time. ........................................... 27 Uncertainty (means) over time (one-item measure). ........................................................ 44 Uncertainty (standard deviations) over time (one-item measure) ..................................... 45 Predicted outcome value (means) over time (one-item measure). .................................... 46 Uncertainty means over time. ........................................................................................... 56 Number Of IURS means over time .................................................................................... 59 Uncertainty means over time. ........................................................................................... 63 IURS percentage means over time. ................................................................................... 71 Predicted outcome value means over time ........................................................................ 73 viii Literature Review The arrival of new communication technologies is ofien closely followed by questions and concerns about the potential effects that technology will have on its users (Wartella & Reeves, 1985). In many Of these cases, the concern resulted from misgivings that new media will alter patterns of social interaction in a dysfunctional manner. Each new major communication technology, from the telegraph to television, has been followed by concerns about a breakdown in community ties (Katz, Rice, & Apsden, 2001). For example, early concern about the telephone was centered on the potential for this channel to isolate people and hurt interpersonal relationships. At the time, critics reasoned personal ties would be weakened by the fact that people would no longer have to physically visit each other (Fischer, 1992). When looked at today, the telephone actually seems to have had the opposite effect (Matei & Ball-Rokeach, 2001). One of the most prominent examples of this concern in recent decades is the Internet, and questions about this technology’s influence on human interaction. A great deal of interest in both the positive and negative effects the Internet has had on its users followed its arrival as a communication channel (e.g., Bargh & McKenna, 2004). Although questions about possible effects stemming from use of communication technologies are important and interesting, focusing solely on outcomes leaves other potentially more central topics unanswered. For example, the Internet has the potential to influence not only the outcomes from communication, but the processes that shape these outcomes. Similar to past technologies such as the telegraph and telephone, the Internet has changed the way many people interact by making communication more available and delivery more rapid for individuals remotely located (Bargh & McKenna, 2004). Nevertheless, criticisms of the Internet, similar to those offered against telegraphs and telephones, suggest that communicating in this manner is not as good as interacting with someone in person (Bargh & McKenna, 2004). The potential for intemet technology to bring about fundamental changes to the communication process seems especially likely for interpersonal communication due to the paucity of nonverbal cues offered in current online communication technology (Walther & Parks, 2002). Some of the central questions facing scholars investigating computer-mediated communication (CMC) examine issues related to the availability of information online. How does information available in CMC influence interpersonal interaction? How do people adapt to the lack of nonverbal cues in online communication? There is little question that characteristics present in the contemporary communication technology of any time period will alter information exchange. How current CMC technology affects interaction with and perception of people online is largely unknown. The present investigation adds to existing efforts to explicate this influence. Social Information Processing Theory (SIPT; Walther, 1992) explains some of the different attributes that characterize online interaction and how people are likely to respond to them. Generally speaking, SIPT suggests that CMC is characterized by several limiting features not present in traditional face-to-face (FtF) communication. However, it also suggests that users of CMC will work to find ways to overcome these limitations, and that given enough time they can accomplish their goals as completely as in FtF interaction. Most notable among the limitations highlighted by SIPT is that CMC lacks many of the nonverbal cues that are prevalent in FtF. Because nonverbal cues are heavily utilized in the impression formation process (Berger & Calabrese, 1975), their shortage in CMC poses a problem for impression formation online. However, the theory maintains that people overcome this limitation, in part by asking more questions and disclosing more in CMC (Walther, 1992), and though research has not examined these processes in ongoing CMC interaction, initial evidence consistent with SIPT indicates the increased use of questions and disclosures in CMC (Tidwell & Walther, 2002). Past research by Westerman and Tamborini (2006) has also examined the relationship between the nonverbal information available in a communication mode and the use of questions and disclosures. Westerman and Tamborini (2006) measured questions and disclosures as indicators of interactive uncertainty reduction strategies (IURS). Their research proposed that mode-based differences in the availability of nonverbal cues would influence the use of IURS to reduce uncertainty and subsequent liking, and tested a model of uncertainty reduction across different communication modes. The results of this study provided outcomes consistent with claims based upon SIPT (Walther, 1992) arguing that: 1) uncertainty reduction goals can be accomplished through similar processes both in FtF and CMC, and 2) these processes take longer in CMC due to the lack of available nonverbal cues (Tidwell & Walther, 2002; Walther, Anderson, & Park, 1994). Westerman and Tamborini’s evidence for these claims was found in data comparing how uncertainty changed over time in FtF and CMC. The pattern of results observed shows that upon initial contact with strangers in both FtF and CMC interactions, uncertainty was not only high, but was equivalent across the two modes. Afier five minutes of interaction, uncertainty had decreased in both modes; however, the rate at which uncertainty decreased differed across conditions. Uncertainty was lower in FtF than CMC at this time point. Although these results showed general support for Westerman and Tamborini’s (2006) hypothesized model, several questions remain concerning incomplete or unexpected findings associated with the processes governing uncertainty reduction across modes. The current study is designed to address three major objectives that correspond to the two core assumptions of Social Information Processing Theory (SIPT; Walther, 1992). The first core assumption of SIPT considered here is that, given enough time, impressions formed through CMC and FtF interaction will be equivalent. Related to this assumption, the current study has the following objective: (1) providing more comprehensive longitudinal data to show how accomplishing uncertainty reduction goals (to completion) differs though initial FtF and CMC interaction. The second core assumption of SIPT considered here is that people find ways to overcome the lack of nonverbal information present in CMC in order to form impressions of others. Related to this assumption, the second and third objectives of the current study are: (2) providing more comprehensive longitudinal data to show how the use of IURS differs over time in initial FtF and CMC interaction, and (3) explicating the extent to and conditions under which adding an avatar (and the increased nonverbal information it makes available) to text-only CMC may or may not reduce uncertainty. Before discussing SIPT, the paper begins by discussing two more general theories of human interaction that help provide the logic for the research undertaken in this study: Uncertainty Reduction Theory (URT; Berger & Calabrese, 1975) and Predicted Outcome Value Theory (POVT; Sunnafrank, 1986). URT provides the background for key SIPT processes applied to current research examining online uncertainty reduction (Tidwell & Walther, 2002) and POVT offers a reformulation of URT that challenges previous explanations about why people reduce uncertainty in initial interactions (Sunnafrank, 1986). Uncertainty Reduction Theory Uncertainty Reduction Theory (Berger & Calabrese, 1975) began as a call for interpersonal communication scholars to formulate their own concepts and theories instead Of relying solely on social psychological ones. Although Berger and Calabrese (1975) do not discount the potential usefulness of social psychological approaches, they note that most social psychological theories do not consider interpersonal communication directly. Thus, they also noted that sole reliance on social psychological theories leaves important aspects of interpersonal communication potentially overlooked as areas of study. To address the lack of theory specific to the field, Berger and Calabrese (1975) postulated URT as an axiomatic theory to explain and predict interpersonal communication in initial interactions, although the authors fiIlly intended it to be applied to later interactions, and it has been by other scholars (e.g., Gudykunst, Yang, & Nishida, 1985; Parks & Adelman, 1983). The core assumption of URT is that the primary concern strangers have when they meet is to reduce uncertainty about themselves and each other in a effort to increase predictability about each other’s (and their own) behaviors. The theory’s structure is that of seven axioms (causal statements) and 21 theorems (statements of covariation) derived from those axioms. In general, the theory states that verbal and nonverbal communication cause decreases in uncertainty, and this causes future increases in communication. Under conditions of high uncertainty, people are likely to engage in higher levels of information seeking, and under conditions of low intimacy, people are likely to exhibit greater reciprocity in their communication behaviors. The theory also states that increased similarity between people will reduce uncertainty, and that decreases in uncertainty increase liking, up to a certain level. An eighth axiom about the role of shared social networks in uncertainty was later added by Berger and Gudykunst (1991). Kellerrnan and Reynolds (1990) also suggested another axiom that states, “as the target’s behavior becomes more deviant, level Of uncertainty increases” (p. 67). Berger (1979) extended URT in three ways. First, he identified two different types of uncertainty: cognitive and behavioral. Cognitive uncertainty deals with not knowing one’s own or others’ beliefs and attitudes. Behavioral uncertainty deals with unpredictability of behaviors in different situations. Second, he identified three conditions that are likely to increase the activation Of uncertainty reduction attempts: The first is if others are perceived as likely to provide rewards or costs. Second, if behavior violates expectations, uncertainty reduction attempts are likely to take place. Third, if future interactions are likely, uncertainty reduction attempts will increase. Third, he identified three types of uncertainty reduction strategies: passive, active, and interactive. Passive strategies are unobtrusive attempts to gain information about a relational partner. Active strategies involve gaining information from people other than the target or manipulating a target’s environment to observe how the target will respond. Interactive strategies involve obtaining information directly from the target. Ramirez, Walther, Burgoon, and Sunnafrank (2002) have identified a fourth type of strategy they label extractive. This is a strategy exclusive to CMC environments, and involves searching existing databases to gain information from the target about the target. Although empirical support has been found for parts of URT (e.g., Berger & Calabrese, 1975; Gudykunst & Nishida, 1984) many challenges to the theory have been Offered. Axiom six, dealing with similarity and uncertainty, has been found to be particularly problematic (Clatterbuck, 1979; Gudykunst & Nishida, 1984). Gudykunst and Nishida (1984) offer a potential boundary condition for this axiom. They found that cultural similarities reduced uncertainty, whereas attitudinal similarities did not. Others have argued that dissimilarity can reduce uncertainty as much as, if not more than, similarity (Sunnafrank, 1986). Overall, support for URT is not as solid as might be expected. Sunnafrank (1986) noted that at that point in time, only about half of the individual tests conducted on uncertainty reduction theory resulted in empirical support. This weak support and contradictory evidence led Sunnafiank to propose a reformulation of URT called Predicted Outcome Value Theory (Sunnafrank (1986). Predicted Outcome Value Theory POVT (Sunnafrank, 1986) is a theory of human interaction that was formulated in response to URT (Berger & Calabrese, 1975). At its core, POVT takes issue with the underlying assumption of URT. Instead of stating that the central concern of strangers who meet is uncertainty reduction, POVT starts with the assumption that people have a more general goal in mind. According to POVT, people who meet strangers attempt to form impressions of these strangers, and these impressions are used to predict what outcomes are likely to be obtained by communicating with that person. Thus, if people seek to reduce uncertainty, it is with predicting future outcomes goal in mind, rather than uncertainty reduction for uncertainty reduction’s sake. According to POVT, uncertainty reduction is not the central goal of individuals in beginning relationships. Instead, the primary goal is achieving positive relational outcomes. Uncertainty reduction may be a means to that end, but it is not the end unto itself. With this assumption in mind, Sunnafrank (1986) Offers an axiomatic theory that reforrnulates URT's axioms. The first five axioms of URT are reworked to say that after uncertainty is reduced, people will behave differently based on whether they predict positive or negative outcomes from further interaction. Axiom 6 of URT is reworked to say that both similarities and dissirnilarities will reduce uncertainty, but similarities will do so more when the dissirnilarities are not familiar. Axiom 7 is reworked to state that decreased uncertainty leads to liking only if positive outcome values are predicted (if negative outcome values are predicted, reduced uncertainty will lead to dislike). Based on these seven new propositions, 15 hypotheses are developed. Nine conflict with theorems set forth fi'om URT. Support for this reformulation has come from Sunnafrank (1988), who found that predicted outcome value was positively associated with amount of verbal communication, intimacy level of communication content, nonverbal affrliative expressiveness, liking, perceived attitude similarity and perceived background similarity. Critical tests of the two theories (Grove & Werkman, 1991; Sunnafrank, 1990) have also supported POVT over URT. Although, POVT has fared better than URT in direct empirical tests between the two, it is not necessarily meant as an alternative theory. Berger (1986) argued that not all predictions about individuals’ communication behavior are related to outcome values. His position is that predicted outcome values are no more or less important to relationship development than uncertainty reduction. As such, Berger argues that POVT is an expansion of URT, instead of an alternative theory. Even Sunnafrank would say that URT works early in initial interactions, but POVT expands it to make predictions about what will happen as the relationship develops. Sunnafrank (1990) says “The expectations of POV generally do not diverge from those of URT until interlocutors have had sufficient time to make outcome value assessments” (p. 87). If this is the case, then one might expect each theory to be equally sufficient in predicting communicators’ behaviors in interaction between initially meeting strangers, at least until these strangers have had a chance to form predictions about outcome values. However, neither theory was originally formulated with the plethora of communication channels that have been made available today in mind. Specifically, the increase in communication through the Internet requires that both theories be examined anew. Online Interaction Throughout history, technology has had a large impact on the ways humans communicate with each other. When McLuhan (1964) said “The medium is the message” (p. 7), he focused attention on the role that technological advances play in our communication patterns. The alphabet, books, telegraphs, and televisions are a few of the many innovations that have allowed people to communicate in new ways. Whether they be in a strict deterministic pattern (Ellul, 1964) or more from a social shaping perspective (Williams & Edge, 1999), it is hard to deny that such technologies have played a role in changes in human communication. One recently popularized technology that both has and will likely continue to revolutionize the human communication process is the Internet. As a large and somewhat amorphous channel, the Internet has great potential to impact communication in many ways. Tasks such as information seeking, entertainment, and commerce all can be completed though utilization of the Internet. However, another basic and common use of the Internet is interpersonal communication (Kraut, Mukhopadhyay, Szczypula, Kiesler, & Scherlis, 2000; Stafford, Kline, & Dimmick, 1999). In fact, of all the applications afforded by the Internet, e-mail, a communication channel, has repeatedly been the most used application of the Internet over the last decade (Rainie & Shermak, 2005). The type of communication that is most common is social communication. Although early researchers doubted that relationships could be started and built online (e.g., Rice, 1984), and presented evidence that online communication was less social than FtF communication (e.g., Hiltz, Johnson, & Turoff, 1986), more recent evidence suggests that the Internet is rife with social interaction (e. g., Merkle & Richardson, 2000; Parks & Floyd, 1996; Parks and Roberts, 1998). Cummings and Kraut (2002) found an increase in personal e-mails of 50% from 1995 to 1998, without a similar increase in work-related e- mails. Stafford, Kline, and Dimmick (1999) reported that 61% of all e—mail usage was personal in nature. Even other CMC applications, such as video games, which are not primarily designed for social interaction, have been found to contain a great deal of social communication (Pena & Hancock, 2006). This research demonstrates that the majority of intemet use is for social interpersonal communication. 10 Although critics and scholars seem to agree that the Internet is used heavily for social communication purposes, not all agree on the effects that this large amount of online communication has on users. Several scholars have drawn attention to potential negative effects of communication through the Internet. For example, Putnam (2000) has argued that participation in traditional civic institutions has declined as communication technologies have isolated people from each other. Kraut, Patterson, Lundmark, Kiesler, Mukophadhyay, and Scherlis (1998) found that heavy use of the Internet made people feel socially isolated, lonely and depressed. Other scholars suggest that people do not find online relationships to be as rewarding as their offline counterparts (Parks & Roberts, 1998), especially when these relationships substitute for Offline relationships (Cummings, Butler, & Kraut, 2002). In fact, some scholars (Putnam, in particular) believe that it is impossible for online communities to offer the same benefits as traditional FtF ones (W ellman, Haase, Witte, & Hampton, 2001, p. 439). However, not all critics and scholars share this negative view of online communication. Addressing Putnam's (2000) arguments on social capital, Wellman et al. (2001) argue that the Intemet has the potential to create new communities and increase communication, albeit in a different manner, and also found that heavy intemet use was actually related to increased civic participation. Walther and Boyd (2002) offer reasons why the Internet may be an even better place to find social support than FtF communities. And in contradiction to their earlier findings, Kraut, Kiesler, Boniva, Cummings, Helgeson, and Crawford (2002) found that intemet use had positive effects on social involvement, communication and general well-being. Interestingly, a meta-analysis conducted on 16 studies of intemet use found that the relationship between intemet use 11 and social interaction was moderated by the time-frame considered. Cross-sectional designs show that intemet communication can be negatively related to social interaction with friends, whereas longitudinal designs show the Opposite effects (Shklovski, Kiesler, & Kraut, in press). This research suggests that the effects of intemet communication may not be as negative as some critics and scholars have reported, and in fact, may be positive. Other research has found that online relationships can be just as meaningful and rewarding as offline relationships (Chan & Cheng, 2004; Mesch & Talmud, 2006). Mesch and Talmud found that when duration of relationship was controlled for, there was no effect of channel (online vs. offline) on relational quality among Israeli adolescents. Although the Mesch and Talmud research does not use this approach, their arguments and findings are consistent with and can be explained by SIPT (Walther, 1992). Social Information Processing Theory One approach to explicating the processes at work in online interactions is Social Information Processing Theory (Walther, 1992). This approach assumes that communicators make attempts to achieve communication goals in online settings as much as in oflline settings. When the lack of cues available in an online setting presents Obstacles to accomplishing their goals, users adapt their behaviors to the cues that are available. Given enough time, people can utilize these circumventions to accomplish goals online just as well as FtF. One metaphor posits that the broader bandwidth of F tF interaction with regard to carrying social information for uncertainty reduction is analogous to water flowing through a very wide hose, whereas the relatively narrower bandwidth of social 12 information online is like water flowing through a more narrow hose (Griffin, 2006). In the end, the same amount of water—or social information—can pass through either hose; it simply takes longer through the narrow hose. Over time, social information should accrue online the same as it does offline. Another way of thinking about this phenomenon is that online communication is like “sipping” from a glass of water, whereas offline interaction is like “gulping” from the same glass (Griffith, 2006). Both methods can result in an empty glass, but “sipping” takes a longer time. A growing number of studies use SIPT to study CMC in a variety of settings (Gibbs, Ellison, & Heino, 2006; Hobman, Bordia, Irmer, & Chang, 2002; Pena & Hancock, 2006; Tidwell & Walther, 2002; Utz, 2000; Walther, 1993, 1994, 1996; Walther & Bunz, 2005; Walther & Burgoon, 1992; Walther, Loh, & Granka, 2005; Walther & Tidwell, 1995, Westerman and Tamborini, 2005, 2006). One major area of SIPT research examines how people overcome limitations of CMC to accomplish their goals. These circumventions include the use of emoticons (Walther & D'Addario, 2001) and chronemics (Walther & Tidwell, 1995). Of particular relevance to the present line of research is a study by Tidwell and Walther (2002) that examined URT (Berger & Calabrese, 1975) in CMC from a SIPT perspective. Tidwell and Walther predicted that individuals adapt to the barriers that CMC presents for uncertainty reduction by finding other ways of getting to know someone that compensate for the limitations that CMC technology imposes. Consistent with their predictions, they found that individuals communicating with a stranger over a text—based CMC system asked more direct questions and disclosed more than those interacting FtF. 13 Westerman and T amborini '5 Model of Uncertainty Reduction across Modes Although the study by Tidwell and Walther (2002) helps illuminate the process Of interacting with strangers in text-based CMC systems, a variety of differences between FtF and CMC leave many issues lefi to address. The use of avatars is one aspect of CMC particularly relevant to our understanding of how URT might apply to online interaction. Avatars are a unique feature of CMC that distinguishes some forms of online interaction from others. Moreover, the visual information associated with the use of avatars has direct implications for URT. The present investigation is concerned with general questions about how people communicate with strangers over text-based CMC systems, and the manner in which the use of avatars can influence this communication. Based on principles from SIPT and URT, it follows that increased information provided from the cues that are available in different online settings (such as an avatar) can help reduce uncertainty in a manner that alters the patterns of communication and outcomes of CMC interaction. Westerman and Tamborini (2005, 2006) proposed a model based on this logic. The model can be found in Figure 1. Figure 1. Westerman and Tamborini’s model of uncertainty reduction across modes Mode _, Available Nonverbal ___, Uncertainty _, Liking Information Interactive Uncertainty Reduction Strategies The model describes the manner in which information available over CMC influences uncertainty and liking during stranger interactions. In this model, available nonverbal information represents differences presumed to exist as a function of 14 communication mode - a factor manipulated in past investigations. The model posits that available nonverbal information influences uncertainty both directly and indirectly through its stimulation of IURS. Directly, available nonverbal information reduces uncertainty. Indirectly, a lack of available nonverbal information limits the capacity for passive uncertainty reduction strategies and increases the need for the use of IURS in order to reduce uncertainty. As such, the direct effect of available nonverbal information works to reduce uncertainty and the indirect effect works to decrease accompanying behaviors that reduce uncertainty. The importance of nonverbal information in the model mirrors the central role it plays in uncertainty reduction (Berger & Calabrese, 1975). Research supports the notion that physical features are central in the person categorization process (e. g., Argyle, 1975; Ichheiser, 1970), which is a type of uncertainty reduction itself. When people find that past categorizations based on physical features are useful for perceiving others, they are likely to use those categorizations in the future (e. g., gender, Hamilton & Sherman, 1994). Notably, the use of physical features for categorization appears to be an automatic process (Patterson, 1995) and is utilized in virtual as well as actual environments (N ass & Moon, 2000; Reeves & Nass, 1996). This logic is represented in the model by the negative path from available nonverbal information to uncertainty. The application of the model to different modes of communication is based on a mode’s provision of nonverbal cues. If uncertainty reduction behaviors are truly adapted to the cues available through a medium, then different communication patterns should be expected when the available cues differ. Consistent with SIPT predictions, Tidwell and Walther (2002) demonstrated that individuals communicating with a stranger over a text- 15 based CMC system adapted to the limited information availability by asking more direct questions and disclosing more often. Westerman and Tamborini (2005, 2006) applied the same logic to compare individuals interacting over a text-based system using avatars (avatar interaction) with individuals interacting over a text-based system without avatars (text-only interaction) and individuals interacting FtF. They argued that the presence of an avatar should increase the amount of information available to partners in online interaction, and therefore, individuals involved in text-only interaction should have less information available than those involved in avatar interaction. Likewise, individuals involved in avatar interaction should have less information available than those interacting FtF. As such, IURS use (e. g., more direct questions and self-disclosures) was expected to be greatest among individuals who engaged in text-only interaction, next greatest among those who engaged in avatar interaction, and lowest among those interacting FtF. This logic was the basis for the negative path fi'om available nonverbal information to IURS in their model. Not surprisingly, IURS use is thought to be motivated by and result in the reduction of uncertainty (Berger, 1979), a principle consistent with the negative path from IURS to uncertainty in the model. The model concludes with liking, one of the major outcomes of uncertainty reduction. Drawing on theory supporting the notion that people attempt to make sense of their environment and become anxious when they cannot (Heider, 195 8), a central axiom of URT states that uncertainty level and liking are negatively related (Berger & Calabrese, 1975). According to this logic, people will like others more if there is less uncertainty in the situation, especially if positive outcomes are predicted from future 16 interaction with that person (Sunnafrank, 1986). This inverse relationship is represented in the model by the negative path from uncertainty to liking. The observations of Westerman and Tamborini (2005, 2006) were generally consistent with this model. However, as stated above, important questions remain. First, while these studies suggest that similar goals can be accomplished across different modes of interaction, not enough time was allowed to observe predicted patterns of uncertainty reduction through to their projected completion. This is particularly limiting for the observation of patterns in CMC where uncertainty reduction is expected to extend over longer periods of time (Walther, Park, & Anderson, 1994). Second, the predicted negative path from [U RS to uncertainty in the model did not differ significantly from zero (and was actually positive). This may have resulted from the fact that those most uncertain after five minutes were still using more IURS, and the procedures used in earlier studies were unable to examine patterns of IURS and uncertainty over time. Third, expected differences between the two CMC conditions were not apparent in any analyses. In fact, at both the initial point of contact (Westerman & Tamborini, 2005) and after five minutes (W esterrnan & Tamborini, 2006), uncertainty was actually highest (although not statistically so) in the avatar conditions. This raises questions concerning the nonverbal cues present in the particular avatars chosen for these two studies and their associations with uncertainty. Taken together, these limitations due to the design utilized by the Westerman and Tamborini (2005, 2006) studies highlight the importance of examining uncertainty reduction and IURS use over the course of an interaction. SIPT (Walther, 1992) processes and individual studies examining uncertainty levels at individual points 17 in time can be culled together to make predictions about expected patterns of uncertainty reduction in both FtF and CMC. Patterns of Uncertainty Reduction in F tF and CMC URT argues that the primary concern of strangers when they first meet is to reduce uncertainty about each other (Berger & Calabrese, 1975). Central to the present study is the notion that both verbal and nonverbal information are useful in reducing uncertainty about other people (Berger & Calabrese, 1975; Berger, 1979). Notably, the first two axioms of URT state that verbal and nonverbal communication decrease uncertainty, which in turn increases further communication. The usefulness of both verbal and nonverbal communication in the uncertainty reduction process would seemingly make uncertainty reduction a difficult goal to attain in CMC due to the comparatively low level of nonverbal information available in these types of interaction. Nevertheless, past research shows that uncertainty can be reduced in CMC (Westerman & Tamborini, 2006), and SIPT (Walther, 1992) provides a useful framework to help explain how this may occur. SIPT’s claim that people utilize whatever cue systems are available in their medium and adapt their communication to the limits of the medium is central to this explanation. As stated above, one way in which communicators can circumvent problems presented by limited nonverbal information is through continued effort over extended periods of time. CMC interactions generally take longer (Walther, Anderson & Park, 1994) due in part to the time needed to type, and the lack of capacity to encode messages while decoding others. Thus, an important factor for accomplishing goals in CMC is having the time needed for enacting the processes necessary to be successful (such as 18 increased use of IURS). If it is true that FtF and CMC can be used to accomplish similar goals, but the process simply takes longer in CMC, then similar patterns of uncertainty reduction should be seen across modes. The only difference is that this pattern should be stretched out over a longer period of time in CMC. Although no prior research comparing FtF and CMC interaction tracks the process of uncertainty reduction through to its expected completion, individual studies can help inform us of what the pattern of uncertainty reduction will look like across communication modes. Theory and research informs about how uncertainty in FtF and CMC should compare at three different time intervals with important implications for the present research. These include comparisons of uncertainty measured before initial interaction, comparisons of uncertainty measured after completing interaction, and comparisons of patterns of change in uncertainty over time during ongoing interactions. Prior to talking in stranger interactions, uncertainty should be higher in CMC than FtF communication. This is because of the increased nonverbal cues available in FtF compared to CMC, and the importance that nonverbal cues have in reducing uncertainty. This difference should be especially pronounced at this point in time (i.e., prior to talking) because of the very limited cues available for reducing uncertainty in CMC interactions. Although Westerman and Tamborini (2005) found uncertainty to be equally high for both FtF and CMC interactants prior to conversation, experimental artifacts present in their study curtail the influence of their findings on hypotheses offered in the present investigation. These scholars asked participants to respond to a measure of uncertainty about their “partners” before having any interaction with them. It seems likely that people found this to be a strange thing to do, and thus had heightened feelings l9 of global uncertainty. It also seems likely that this global uncertainty surrounding a procedure that participants found to be strange was what caused people to be equally highly uncertain in both FtF and CMC conditions in the Westerman and Tamborini study especially when these results are weighed against theoretical predictions from the well reasoned logic of SIPT. The logic of SIPT conflicts with Westerman and Tamborini’s (2005) findings and produces the prediction that uncertainty should be higher in CMC than FtF communication prior to conversation. This theory and reason leads to the first hypothesis in the current investigation. H1: Prior to conversation (time 0), people will be more uncertain about a stranger in CMC compared to FtF. After extended conversation in stranger interactions, uncertainty in CMC and FtF communication might differ very little if it differs at all. SIPT (Walther, 1992) suggests that people can accomplish interpersonal goals in both CMC and F tF . Though research shows that accomplishing these goals takes longer in CMC (Walther, Anderson, & Park, 1994), given enough time, people should be able to reduce uncertainty to a satisfactory level through CMC just as they can through FtF interaction. What is less clear is whether or not people interacting for an equal amount of time in CMC and F tF would reach equally low levels of uncertainty. One reason this might be expected to occur is if uncertainty reduction efforts cease at some point in interaction. Uncertainty in an interaction may be reduced to a manageable point, and then attention may turn to accomplishing other goals. If this point is reached rapidly in F tF , then uncertainty levels may remain generally unchanged after that point in time because interactants are focusing on other goals. At the same point in time during CMC interaction, because uncertainty 20 would not have been reduced to a manageable point yet, attention should stay focused on accomplishing uncertainty reduction goals. Ongoing CMC after this point in time would allow for uncertainty levels in CMC to arrive at the same manageable point reached in FtF. As such, when interactions last long enough for people in CMC to reach this point, people interacting in CMC and FtF for an equal amount of time will reach equally low levels of uncertainty. Interestingly, Nowak (2004) found no statistically significant differences in uncertainty between CMC and FtF conditions after 15 minutes of interaction. Thus, it seems likely that 15 minutes is enough time to see the expected convergence of uncertainty levels between F tF and CMC conditions. The research and logic offered here based on SIPT leads to the second hypothesis in the current investigation. H2: After 15 minutes of interaction, the difference between CMC and FtF (in uncertainty about a stranger) will be lower than the difference prior to conversation (time 0). Individual studies examining uncertainty after different lengths of interaction offer clues on the projected patterns of change in uncertainty during ongoing interactions and how these patterns should compare between FtF and CMC. In research on FtF interactions only, studies show that uncertainty reduction occurs within the first two and a half minutes (Afifi & Burgoon, 2000), and that people form judgments of strangers in the first three minutes (Sunnafrank & Ramirez, 2004). In a study comparing uncertainty after five minutes of interaction both in FtF and CMC, Westerman and Tamborini (2006) found that uncertainty was somewhat reduced in each of these modes (as compared to initial levels), but that it declined significantly more during FtF interactions. Their 21 findings are consistent with the SIPT notion suggesting that similar goals can be accomplished through the two different modes. At the same time, they are consistent with the view that it takes longer to accomplish goals in CMC (Walther, Anderson, & Park, 1994). Finally, in a study that compared uncertainty after 15 minutes of interaction, once again in both F tF and CMC, Nowak (2004) found no significant differences existed in uncertainty between FtF and CMC. These individual studies can be combined to inform us about the patterns of change in uncertainty expected from ongoing stranger interactions both in F tF and CMC over time. For purposes here, the patterns under consideration begin following the initial contact point (i.e., the pre-conversation point) and continue up to the point at which initial uncertainty needs have been satisfied. Here are the projected patterns: Following initial contact with strangers (before conversation has begun), high levels of uncertainty are expected across both modes (Westerman & Tamborini, 2005), although SIPT (Walther, 1992) would suggest that uncertainty will be higher in CMC than FtF. This uncertainty should decline somewhat quickly in FtF interactions (Sunnafrank & Ramirez, 2004), but more slowly in CMC (Westerman & Tamborini, 2006) since CMC interactions take longer to achieve similar goals (Walther, Anderson & Park, 1994). In both cases, uncertainty should continue to decline until it reaches the point at which satisfying initial uncertainty needs has transpired and addressing other goals can be attempted. When participants in both modes reach this point, uncertainty levels for FtF and CMC should be similar (Nowak, 2004). Combining the findings and logic from theSe studies to create predictions about patterns of uncertainty reduction compared across modes allows for a stronger direct test 22 of SIPT processes. In order to offer compelling evidence of these processes, there is a need for research that provides continuous measures of change in uncertainty reduction over time from the point of initial interaction until its completion for both F tF and CMC interaction. Past studies have shown that some goals take between four to five times as long to accomplish in CMC than they do in FtF interaction (Dubrovsky, Kiesler, & Sethna, 1991; Weisband, 1992). Although comparisons of the length of time taken to accomplish uncertainty reduction goals has not been examined directly, it is interesting to note that separate investigations of uncertainty reduction Show outcomes consistent with this four to five times rule of thumb. In FtF interactions people reduced uncertainty about strangers enough to make judgments about future relationships within the first three minutes (Sunnafrank & Ramirez, 2004), whereas after fifteen minute of interaction measured uncertainty did not differ significantly across F tF and CMC (Nowak, 2004). Regrettably, efforts to provide moment by moment measures of uncertainty are Often hindered by techniques that interfere with ongoing interaction processes. One way to help overcome this problem is through protocol analysis (Ericson & Simon, 1993), where people vocalize their thoughts while reviewing their past interactions. For example, after engaging in conversation long enough to acquaint themselves fully through either FtF or CMC interaction, participants using a form of protocol analysis (while reviewing their interaction to aid recall) can report their own certainty levels minute by minute during interactions with strangers. This would allow examination of the uncertainty process over time and comparison among different communication modes. Based upon the expectations culled from the literature on interactions, uncertainty is expected to follow patterns of reduction that look something like the curves 23 represented in Figure 2. Prior to talking/chatting with each other, uncertainty will be higher in CMC conditions than FtF conditions. After discussion begins, uncertainty will begin to decrease in both conditions; however, it will decrease more quickly in FtF interactions. After a few minutes uncertainty will level off in FtF, although it will continue to decrease in CMC, until it reaches a similarly low point to FtF. Thus, the patterns of uncertainty over time for FtF and CMC interaction will both be exponential functions. Figure 2. Uncertainty across mode as a fimction of time. 3.5 3 2.5 r 2 1.5 M o .. , o 1 2 3 4 s 6 7 a 9 101112131415 -0- FtF Uncertainty -l- CMC/No Av. Uncertainty The patterns predicted from this reason and logic lead to the third hypothesis in the current investigation. H3: Uncertainty levels in FtF and CMC interactions will follow divergent exponential reduction patterns similar to those seen in Figure 2 over the course of 15 minute interactions. The Influence of IURS on Uncertainty across Modes 24 The ability of people to surmount the limits of reduced visual information over time in CMC is explained in part as a function of change in their use of interactive strategies. When Tidwell and Walther (2002) found that strangers communicating over a text-based CMC system asked more direct questions and disclosed more than those who interacted FtF, they explained this increased use of IU RS according to SIPT as the best way afforded by the medium to reduce uncertainty. Because CMC users do not experience the wealth of nonverbal cues and other visual cues found in F TF interaction, users take advantage of IURS to compensate for these limitations. Westerman and Tamborini (2006) also found that people used a greater percentage of IURS in CMC compared to FtF interaction. Based on beliefs that the increased use of IURS would reduce uncertainty in CMC, their research also predicted that IURS would be negatively correlated with uncertainty. In other words, uncertainty would be lowered when people used more IURS. However, in contrast to expectations, virtually no relationship was found between IURS use and uncertainty in any interaction mode. One explanation for the lack of significant findings in research by Westerman and Tamborini (2006) is that their procedures were insensitive to the dynamic nature of IURS use over time. The cross-sectional analysis employed in their research was unable to account for the moment by moment changes expected to occur. Whereas IURS use may serve to lower uncertainty, heightened levels of uncertainty also motivates greater IURS use (Berger, 1979). The sequential causal nature of the two variables should follow a pattern in which initial uncertainty prompts the use of IURS which should, somewhat instantaneously, serve to lower uncertainty. Immediately following this, the newly attained level of uncertainty should determine the ensuing use of IURS. This process 25 should repeat itself in a recursive manner until the uncertainty is reduced to the point where it is no longer a strong enough force to shape interaction. As such, instead of a study predicting a negative correlation between uncertainty and [U RS at one point in time and using forms of cross-sectional analysis, observing the patterns of IURS use over time across conditions in comparison to patterns of uncertainty over time is a useful step in understanding how IURS use operates in both FtF and CMC interactions. Past studies involving 1U RS (Tidwell & Walther, 2002; Westerman & Tamborini, 2006) conceptually defined and analyzed IURS use as a percentage of total utterances. This was done to address issues associated with the greater number of overall utterances expected in F tF. The current study begins with logic suggesting that the increased capacity for utterances overall in FtF has important implications for the uncertainty reduction process. Along with the increased capacity in FtF for nonverbal cues, the simple fact that there is a greater number of IURS should cause uncertainty to reduce faster during FtF compared to CMC. As such, the total number of IURS is important to consider separately from IURS use as a percentage of total utterances. For example, if someone only has one utterance, and it happens to be a self-disclosure, their percentage would be 100%. If another person has twelve utterances, and six are self-disclosures, their percentage would be 50%. Although person one would have a higher percentage of IURS, person two should reduce more uncertainty. If IURS both predicts and is predicted by uncertainty, then predicted patterns of IURS use over time can be made based upon patterns of uncertainty in combination with expectations based on the influence of channel limitations. If uncertainty is highest in both FtF and CMC before conversation, then the IURS use should be highest in the first 26 minute for each condition. If uncertainty is reduced very quickly in FtF, then the use of IURS should lower quickly as well. If uncertainty levels off in FtF, then the use of IURS should level off as well. If uncertainty reduction is a slower process in CMC, then the use of IURS should decrease more slowly in CMC. From this, it would follow that patterns of IURS use across FtF and CMC should be very similar to predicted patterns of uncertainty. However, patterns of IURS use will not be exactly the same as patterns of uncertainty because fewer utterances are possible in CMC than F tF due to technical limitations. Therefore, fewer IURS will be used in CMC compared to FtF over the course of the entire interaction, and this difference will be especially pronounced in the early stage of the interaction. The logic offered here suggests that IURS use will follow patterns of change across both modes in a pattern seen in Figure 3. Figure 3. IURS use (total number) across mode as a function of time. mar“ FtF IURS *CMC/No Av. IURS 0123456789101112131415 H4: Total number of IURS used in FtF and CMC interactions will follow patterns similar to those seen in Figure 3. 27 Avatars and Uncertainty Reduction in CMC Research positing how IURS use is increased and extended for a longer period of time is understood mainly as an effort to overcome barriers to uncertainty reduction created by the limited availability of nonverbal information in CMC. Yet some forms of CMC offer more nonverbal cues than others. One nonverbal feature of CMC that has received a great deal Of attention in this regard is the use of avatars. The word “avatar” originates from Sanskrit and means “God’s appearance on Earth” (Damer, 1998). The first use of this term in a computer setting appears in the novel Snow Crash (Stephenson, 1993). In such settings, the word “avatar” has been defined as a graphic image that represents a user in a virtual environment (N owak, 2000) and as an icon that users choose to represent themselves (Suler, 1997). The use of avatars is increasing in business applications, especially in customer service applications (Qiu & Benbasatrn, 2005). Avatars are also commonly found in online interactions and are easily created by such CMC tools as Yahoo or AOL instant messenger. The increased uses and potential uses for avatars in such settings underlie the importance of understanding their impact on communication processes. Westerman and Tamborini (2005, 2006) examined the role of avatars in uncertainty reduction, predicting that the increased nonverbal information offered by avatars in CMC would help reduce uncertainty more quickly than text-only CMC interactions. Contrary to their predictions, however, the addition of an avatar to text-only CMC (and the increased nonverbal information it made available) did not reduce uncertainty at the initial point of contact (Westerman & Tamborini, 2005) or after five minutes of interaction (Westerman & Tamborini, 2006). In fact, although not statistically 28 significant, uncertainty in CMC was actually higher at both time points when an avatar was present. Moreover, the presence or absence of an avatar had no influence on other studied variables including liking and IURS use. This finding seems somewhat at odds with the underlying assumption that the amount of nonverbal information available in a communication exchange will influence initial levels of uncertainty and liking for a partner, and raises questions about why the addition of an avatar was unrelated to uncertainty. Even beyond the basic theoretical issues concerning how people adapt to the lack of nonverbal cues in online interaction, growing interest in online avatar use makes the need to explain their role in CMC more pertinent. One potentially important factor overlooked in the studies by Westerman and Tamborini is the informativeness of the avatar. Close inspection of the procedures used in Westerman and Tamborini (2006) reveals several factors that might explain why the addition of avatars did not influence outcomes as expected. For example, the fact that participants did not select their own avatar might have made them question the extent to which the avatar encountered accurately represented their interaction partner. In this case, instead of providing informative nonverbal cues the avatar might have created questions, thus increasing uncertainty instead of decreasing it. To some extent, such simple matters of protocol are easily corrected by providing a context prompting users to believe that their partner chose their own avatar. At the same time, it focuses attention on the conspicuous drawbacks concerning the extent to which an avatar is informative in the CMC context, and how this affects attributions made by users. 29 For example, research by Westerman and Tamborini (2006) presented participants with somewhat ambiguous avatars in the context of a rather ambiguous task. Participants were told only that they were going to perform a task in which people did better if they got to know each other first. Then they were presented with one of two male or two female avatars chosen somewhat arbitrarily from an inventory of available avatars containing head-and-shoulder shots of young adult, Caucasian, cartoon-like characters. Given the lack of any understood context for interaction, it is difficult to imagine that participants had any specific expectations in mind. Indeed, combined with questions about whether or not participants selected their own avatar, it is hard to imagine what information participants might have gained fi'om the avatar to help reduce uncertainty. By contrast, whether it is for social or professional purposes, most CMC interaction is driven by more clearly defined reasons and occurs within a more clearly defined context. As such, the perceived appropriateness of the avatar for the interaction context should be an important factor in how the avatar’s presence influences uncertainty. One way to test this would be to modify the procedures used by Westerman and Tamborini (2006) to present participants with the type of clearly defined interaction context that creates preconceived notions of what to expect, and then provide groups of participants with avatars that differ in their level of informativeness for the context. Based on the logic described above, the patterns of uncertainty predicted in this situation can be expressed in the following hypothesis. H5: A more informative avatar will reduce uncertainty more than a less informative avatar for the same interaction context. 30 Method Overview An experiment was designed to test the hypotheses relating communication mode with patterns of uncertainty and IURS use based on previous findings (Westerman & Tamborini, 2005; Westerman & Tamborini, 2006). Participants interacted with a stranger in one of four conditions: FtF, CMC-no avatar, or one of two CMC with avatars conditions in which the informativenss of the avatar differed (CMC-high informativeness or CMC-low informativeness). Before interacting with a stranger, participants filled out Oliver’s (1993) mood measure. After a fifteen minute interaction with their partner, participants filled out measures of uncertainty (Parks & Floyd, 1996), liking (McCroskey & McCain, 1974) and predicted outcome value (Sunnafiank, 198 8) about their partner, and a thought-listing questionnaire (Cacioppo & Petty, 1981). Then participants recorded their own certainty and predicted outcome value levels minute by minute throughout the interaction using a form of protocol analysis (while reviewing their interaction to aid recall). Finally, trained coders reviewed the transcripts and coded the frequency and type of IURS use minute by minute throughout the interaction. Participants One hundred and twenty-four (62 males, 62 females) undergraduates from a large Midwestern university participated in the main study. These participants were recruited fiom a variety of communication classes. A college student sample is well suited to this study not only for its pragmatic availability advantages but more importantly because college students are one of the largest users of instant messaging (Jones, 2002), which was the channel used for CMC interactions in this study. Participants signed up to take 31 part in two separate sign-up spaces to insure that one male and one female comprised each interaction dyad, and to make it unlikely for a male and female who were familiar with each other to sign up for the same time. Overall, two dyads were excluded from analysis due to prior knowledge of each other. This left 60 dyads (120 individuals) included for hypothesis testing (FtF, N = 15, CMC-no avatar, N = 15, CMC-high informativeness avatar, N = 15, CMC-low informativeness avatar, N = 15). Procedure The procedure differs by condition and required the use of different locations. The FtF condition took place in a research laboratory room furnished simply with a table and chairs where participants sat across from each other. The CMC conditions took place in two separate rooms, each with a computer present. Pairs of participants participated in the separate labs simultaneously (for experimental scripts, see Appendix A). In the F tF condition, when a participant arrived, s/he was instructed to sit at the table in the lab, given a consent form to sign (see Appendix B), and asked to wait for the next person. Once the second participant arrived, s/he was brought into a different room in the lab, and given a consent form. The participants were then told that they were involved in a study on meeting new people. As such, they were told they would interact with two different people, each for fifteen minutes, and after both interactions they would choose who they would prefer to hang outwith. Before the beginning of the interaction, participants filled out a mood measure (Oliver, 1993; see Appendix C). Participants then interacted for fifteen minutes. After the time elapsed, they moved back into the separate rooms to complete several measurement procedures. First, they filled out an uncertainty scale (Parks & Floyd, 1996; see Appendix D), a modified liking scale (McCroskey & 32 McCain, 1974; see Appendix E), and a measure of predicted outcome values (Sunnafrank, 1988; see Appendix F). Then they completed a thought listing procedure (Cacioppo & Petty, 1981) describing what information they used to make their judgments of uncertainty and liking (see Appendix G). After completing these procedures, participants used a modified form of protocol analysis to report their thoughts during interaction (see Appendix H). They watched a video of the interaction they just experienced (in order to aid recall) and then gave a minute by minute account; reporting at each point how certain they felt about their partner and what outcomes they predicted fi'om future interaction. After this, participants were given a study assessment sheet (see Appendix I), debriefed (see Appendix J for debriefing sheet), and asked if they knew each other prior to the experiment. In the CMC conditions, when a participant arrived they were seated in one of two rooms with a computer displaying an AOL IM screen. In the text-only CMC condition, screens had no avatar. In the avatar conditions, participants were told that people were usually allowed to choose their own avatar, but the avatar capture software was inoperable on their computer, so an avatar had been assigned to them. The rest of the procedure was the same as the FtF condition, except for the means used to aid recall in the modified protocol analysis. Instead of watching a video, participants in these conditions were given a printed transcript of their interaction broken down by minute to help guide recall. Measures Several measures were used in the current research. Prior to meeting each other, participants filled out a mood measure (Oliver, 1993). After interacting for 15 minutes, 33 participants responded to an uncertainty measure (Parks & Floyd, 1996), a liking measure (McCroskey & McCain, 1974), a measure of predicted outcome values (Sunnafrank, 1988), and a thought listing questionnaire. Each of these was meant to address their post- interaction judgments. Finally, participants responded to items designed to address their judgments of their partner before talking and during interaction. In order for a scale to be considered acceptable for use in this study, it had to be both unidimensional and reliable. To check for unidimensionality, observed inter-item correlations were compared to expected inter-item correlations (calculated as the product of each item's factor loading). If more observed inter-item correlations fell outside the 95% confidence interval of their respective expected inter-item correlation than would be expected by chance, the measure was not considered unidimensional. In this situation, examination of the problem correlations led to choosing items to be dropped from the measure. Items were chosen if they appeared as part of multiple prOblematic inter-item correlations. Items were dropped until a unidimensional solution was found. Finally, a Cronbach's alpha equal to .70 or higher was required for measures to be considered acceptably reliable. Mood measure. Before interacting with each other, participants responded to Oliver's (1993) mood measure. This measure was included to make sure the lack of avatar choice did not upset participants. The scale consists of fifteen items and uses a seven-point Likert-type scale, with l = Not at all and 7 = Very much. A subset of five items comprising a negative mood scale was used for analysis (angry, upset, disturbed, negative, unhappy). Confirmatory factor analysis was performed on the five items using Hamilton and Hunter’s (1988) CFA program by testing inter-item correlations for 34 internal consistency. Observed correlations are compared to expected correlations, which are the product of the two items’ factor loadings. The observed correlations and factor loadings can be found in Table 1. None of the observed correlations fell outside of the 95% confidence intervals of their respective expected values. Thus, the data are consistent with a unidimensional solution, and all items were retained for further analyses. Coefficient alpha of the scale was a = .82. Table 1 Observed Inter-item Correlations and Factor Loadings for Negative Mood Item 1 Item 2 Item 3 Item 4 Item 5 Factor Loading Item 1 .64 Item 2 .55" .73 Item 3 .34“ .39“ .55 Item 4 .40" .45" .43" .69 Item 5 .52" .62" .46" .65“ .86 Note. * p < .05, ** p < .01 by two-tailed t-test. Uncertainty measure. Parks and Floyd’s (1996) measure of interpersonal predictability/understanding was used to measure uncertainty about each participant’s partner. This scale was used because it has proven reliable (or = .82, Parks & Floyd, 1996) and unidimensional (Westerman & Tamborini, 2006) in the past and was developed in a study of intemet relationships. The scale contains five items (e. g., I do not know this person very well) and uses a five-point Likert scale, with 1 = strongly disagree and 5 = strongly agree. Confirmatory factor analysis was performed on the five items using Hamilton and Hunter’s (1988) CFA program. The observed correlations and factor 35 loadings can be found in Table 2. None of the observed correlations fell outside of the 95% confidence intervals of their respective expected values. Thus, the data are consistent with a unidimensional solution, and all items were retained for further analyses. Coefficient alpha of the scale was a = .80. Table 2 Observed Inter-item Correlations and Factor Loadings for Uncertainty Item 1 Item 2 Item 3 Item 4 Item 5 Factor Loading Item 1 .62 Item 2 .34" .61 Item 3 .43" .45M .71 Item 4 .42” .52" .56" .76 Item 5 .49“ .35" .43" .45" .64 Note. * p < .05, ** p < .01 by two-tailed t-test. Liking measure. McCroskey and McCain’s (1974) measure of interpersonal attraction was used to measure liking for each participant’s partner. Specifically, this scale is designed to measure social attraction (as opposed to physical or task attraction). The scale used in the present study contained six items including the five original items fi'om McCroskey and McCain (e.g., We could never establish a friendship together) and a sixth item (I would like to talk to him/her again) added for this study. Participants responded using a five-point Likert scale, with 1 = strongly disagree and 5 = strongly agree. Confirmatory factor analysis was performed on the six items using Hamilton and Hunter’s (1988) CFA program. Observed inter-item correlations and factor loadings can be found in Table 3. The observed correlation between items 3 and 4 fell above the 95% 36 confidence interval around its corresponding expected correlation, P (.13 S r S .37) = .95. The observed correlation between items 3 and 5 fell just outside the lower edge of the 95% confidence interval around its corresponding expected correlation, P (.16 S r S .40) = .95. The observed correlation between items 5 and 6 fell well above the 95% confidence interval around its corresponding expected correlation, P (.31 S r S .53) = .95. These are more problems than would be expected by chance with 15 total correlations. Table 3 Observed Inter-item Correlations and Factor Loadings for Liking Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Factor Loading Item 1 .63 Item 2 .31" .50 Item 3 .35" .29" .49 Item 4 .34" .30" .42" .52 Item 5 .36" .24" .15* .19" .58 Item 6 .40" .32" .24" .28” .72" .72 Note. * p < .05, ** p < .01 by two-tailed t-test. Because item five was part of two of the problematic correlations (including the largest one) and the two smallest observed inter-item correlations, confirmatory factor analysis was run again without this item. The observed correlations and factor loadings can be found in Table 4. None of the observed correlations fell outside of the 95% confidence intervals of their respective expected values. Thus, the data are consistent with a unidimensional solution, and all items except item 5 were retained for further analyses. Coefficient alpha of the scale was a = .70. 37 Table 4 Observed Inter-item Correlations and Factor Loadings for Liking (without item 5) Item 1 Item 2 Item 3 Item 4 Item 6 Factor Loading Item 1 .63 Item 2 .31" .52 Item 3 .35" .29" .57 Item 4 .34“ .30" .42" .59 Item 6 .40" .32" .24" .28“ .54 Note. * p < .05, *‘I‘ p < .01 by two-tailed t-test. Predicted outcome value measure. Next, participants responded to Sunnafrank’s (1988) predicted outcome value measure. This scale consists of ten items designed to assess how positive people predict a future relationship with a person would be (e.g., What kind of conversations do you think would be likely to occur in a relationship with this person?) People were instructed to judge their partner relative to their expectations about how relationships normally continue from a beginning acquaintance, and to assess whether they think that interaction with this person would be more or less positive than normal using a six-point scale, with l = much less positive and 6 = much more positive. Confirrnatory factor analysis was performed on the ten items using Hamilton and Hunter’s (1988) CFA program. The observed correlations and factor loadings can be found in Table 5. Six observed correlations fell outside the confidence interval around their corresponding expected correlation: between items 1 and 2, P (.38 S r S .58) = .95, between items 1 and 7, P (.33 S r S .55) = .95, between items 3 and 4, P (.46 S r S .64) = .95, between items 3 and 9, P (.43 S r S .63) = .95, between items 6 and 7, P (.38 S r S 38 .58) = .95, between items 7 and 8, P (.33 S r S .55) = .95. These are more problems than would be expected by chance with 45 correlations. Table 5 Observed Inter-item Correlations and Factor Loadings for Predicted Outcome Values Iteml Item2 Item3 Item4 Item5 Item6 Item7 Item8 Item9 Factor Loading Item 1 .66 Item 2 .65“ .73 Item 3 .49" .56“ .72 Item 4 .49" .53" .68" .77 Item 5 .58" .54" .53“ .57" .72 Item 6 .39" .56" .51 ** .52" .48“ .73 Item 7 .32" .38“ .44" .49" .44" .64" .66 Item 8 .36" .43" .45“ .47" .48" .51" .58" .67 Item 9 .48" .53" .42" .55" .50" .52" .49" .56" .73 ItemlO .52" .55” .55“ .66" .54“ .56" .51" .54“ .67" .80 Note. "‘ p < .05, ** p < .01 by two-tailed t-test. Because item seven was part of three of the problematic correlations and item three was part of two different problematic correlations, confirmatory factor analysis was run again without these two items. The observed correlations and factor loadings can be found in Table 6. Only the observed correlation between items 1 and 2 fell above the 95% confidence interval around its corresponding expected correlation, P (.38 S r S .58) = .95. Because we might expect one error with 28 total correlations by chance, these eight items 39 were determined to form a unidimensional solution. Finally, because removing the two items did not decrease reliability greatly (down to .90 from .91 ), items three and seven were not included in later analyses. Table 6 Observed Inter-item Correlations and Factor Loadings for Predicted Outcome Values (no 3 or 7) Item 1 Item 2 Item 4 Item 5 Item 6 Item 8 Item 9 Factor Loading Item 1 .68 Item 2 .65** .75 Item 4 .49" .53" .75 Item 5 .58M .54" .57" .73 Item 6 .39" .56" .52" .48** .69 Item 8 .36" .43M .47" .48M .51" .65 Item 9 .48" .53" .55" .50" .52" .56“ .75 ItemlO .52M .55" .66" .54" .56" .54" .67" .81 Note. * p < .05, ** p < .01 by two-tailed t-test. Thought listing procedure. A thought listing procedure, similar to that used by Petty and Cacioppo (1981) was used to provide insight into the study at large. Participants were asked to write down what thoughts they had about the interaction, the information they used to make judgments of uncertainty, liking and predicted outcome values, and any other thoughts they had before or during the interaction. They were given up to four minutes to complete the task. These responses were examined for insight into potential problems and other thoughts about the study. 40 Modified protocol analysis. The modified protocol analysis used in this study is a method that attempts to understand the thoughts people have while enacting a task (Ericsson & Simon, 1993). In customary form, people talk aloud and try to vocalize their thoughts while working through a task. While this method has been employed successfully in CMC interactions (V angelisti, Corbin, Lucchetti, & Sprague, 1999), such a talk-aloud procedure is not feasible during an ongoing FtF conversation. Instead, an aided recall task method based on protocol analysis was used. Participants were given a transcript of their interaction or watched a video of their interaction, and were asked to reflect on their interaction. For each 1 minute interval during the interaction they reported 1) the levels of uncertainty they felt toward their partner and 2) their prediction regarding the likelihood of positive outcomes from the interaction. Using a five point Likert type scale, participants responded to the following two items in retrospect beginning with the pro-conversation point and continuing at the end of each 1 minute section: “New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point?” (1 = very uncertain, 5= very certain) and “How positive did you think future interaction with your partner would be?” (1 = very negative, 5 = very positive). Scores for the uncertainty item were then reverse coded to give a measure of uncertainty such that higher scores represented greater uncertainty. Pilot test of measures used for modified protocol analysis. One-item measures were utilized to avoid the lengthy time requirement and potential burnout that may accompany responding to a longer measure (for example, the uncertainty and predicted outcome value measures discussed above) sixteen times each throughout the interaction. 41 Although the quality of these one-item measures cannot be evaluated using traditional methods, there is evidence to support that they are good measures. First, both items have strong face validity. They were written to reflect exactly the information desired from each measure. Second, a pilot test using both items was conducted. In this pilot test, the text of a fake 5 minute IM interaction was revealed to thirty- four participants from an upper-level communication course. Fifteen (44.1%) of these participants were male, and nineteen (55.9%) were female. This text was created to emulate an interaction between a male (“spartanfan90”) and a female (“funtymebabe”) (see Appendix K). During the first three minutes of the interaction, the male says he is a senior English major who is a big fan of the local university's football and basketball team. During the fourth minute, the female asks the male if he likes to visit bars, to which the male replies he isn't old enough. During the fifth minute, the female questions that he isn't old enough because he said he was a senior, and then the male reveals that he is a senior in high school who hopes to go to the local university. The revelation was included to create a situation where new information would potentially increase uncertainty, and thus test the measure's sensitivity to such change. This was done to determine if some demand characteristic of the testing procedure or some intuitive participant reaction might inadvertently produce the predicted pattern of change in uncertainty reduction over time. In this case, we might see responses that show an unbroken pattern of decreasing uncertainty as time goes on, regardless of the information provided to participants in the interaction. These findings might be interpreted erroneously as support for the hypothesized pattern of change in uncertainty. Inclusion of the revelation helps eliminate this as a potential rival interpretation. If the 42 measure is sensitive to change in uncertainty, this should be apparent in scores from the pilot test following the point where the revelation occurs. This interaction was presented to participants in the pilot study line by line. At each “minute” point, participants responded to the one-item measure of uncertainty and the one-item measure of predicted outcome value about the male (“spartanfan90”) in the fake IM interaction. The interaction was created and presented in this manner to address two potential concerns with the one-item measure. First, concerns may be raised about respondents’ ability to be cognizant of and the measure’s ability to reveal potentially small changes in uncertainty that result from new information in an ongoing interaction and to make judgments that are sensitive to these changes. Second, concerns may exist about experimenter demand effects and the potential that participants may believe that they should be more certain as time elapses, and thus respond accordingly. The interaction was designed to present people with small amounts of information over time to determine if the one-item measure of uncertainty was sensitive to such changes. However, it was also designed to show that people would not simply respond with greater certainty over time. The information in the fourth and fifth minute that would increase uncertainty was intentionally included for this purpose. If people simply responded with lower uncertainty over time, we would expect means for uncertainty to decrease each minute of the interaction, regardless of the information that was actually provided in that minute. Examination of the pattern of uncertainty means over the five minute interaction shows a pattern inconsistent with the potential demand effects discussed above. Participants did not just respond with lower uncertainty as time elapsed (Figure 4). After 43 an initial drop from the first to the second minute, uncertainty levels remained relatively stable. More notably, the pattern of standard deviations for uncertainty shows an increase as time elapsed (Figure 5). This suggests that although the mean level of uncertainty may not have shown great change in response to the scripted revelation, individual responses during and following the revelation did. People became more extreme in their judgments of uncertainty over time. Thus, although the overall means of uncertainty changed relatively little, some individuals reported increased uncertainty and others reported decreased uncertainty as the new information become available, as evidenced by the changes in standard deviations. Because people's levels of uncertainty are changing as the interaction unfolds, these findings suggest this one-item measure is sensitive to changes in uncertainty over time. Figure 4. Uncertainty (means) over time (one-item measure). Uncertainty as a function of time I? h 2.91 -B-Uncertainty Time 44 Figure 5. Uncertainty (standard deviations) over time (one-item measure). —1 Uncertainty as a function of time 1~45 "*‘r~-- ~~._.., ~-_,.133 : 1.35 , 1,25 1,15 ._ 0.95 0.85 0.75 Uncertainty -a- Uncertainty § The patterns of means and standard deviations over time were also examined for the one-item measure of predicted outcome value. The fake 1M interaction was designed to give participants a reason to change predictions on the positiveness of future interaction with the person as the interaction transpired, and thus shows that the measure is sensitive to these types of change. This analysis shows that the measure is sensitive to change. At the point in the interaction at which one of the interactants provides information that questions their actual age (minute four) the mean for predicted outcome value drops (Figure 6). Thus, as predicted, participants expect less positive outcomes from future interaction after the person offers this information. Because predicted outcome values lower after the interactant provides information that allows one to question previous impressions about him, the one-item measure is deemed acceptable for measuring and sensitive to changes in predicted outcome values over time. 45 Figure 6. Predicted outcome value (means) over time (one-item measure). Predicted Outcome Value ii: . - M - .M:- ' Predicted outcome 0 .J .. “0......W. ...._.. . . . . _ , ,. Maw...” ..., W..- h....._-_........,.~.W va'ue «rm—4. i . 4 w... . ._.A.WM~—"—A Predicted Outcome Value Transcript coding. In order to determine the number of IURS used in each condition, the text of each interaction was coded for IURS by two trained coders. First, using procedures used previously by Tidwell (1997) each utterance was identified by placing vertical lines around it. An utterance was defined as "a single assertion about some subject" (Holsti, 1969, p. 116). Also using procedures used by Tidwell (1997), each utterance was coded as one of three types of utterances: question, self-disclosure or other. Questions were defined as "An expression of inquiry that invites or calls for a reply; an interrogative sentence, phrase, or gesture" (Morris, 1976, p. 1070). Any utterance that was an invitation for the other person to share a piece of information was coded as a question. A self-disclosure was defined as “A verbal response (thought unit) which describes the subject in some way, tells something about the subject, or refers to some affect the subject experiences" (Chelune, 1975, p. 133 cited in Tardy, 1988). The important part of this definition is that the phrase had to disclose something about the person uttering it. For example, saying “I like Jim” would be a self-disclosure because it 46 reveals the speaker's feeling about Jim, whereas “Jim went to the store” was not coded as a self—disclosure because it revealed information about someone other than the speaker. The third category was other. Anything that was not a question or self-disclosure was coded as other. Questions and self-disclosures are both types of ID RS, and were combined into the variable called IURS. In contrast to previous research that used this procedure to determine the total number of IURS used throughout an interaction, IURS use was broken down by time and coded separately for each 1 minute segment throughout the 15 minute interaction. Each coder first coded the same sample of ten interactions taken from a separate pool of interactions not included in the current study to establish intercoder agreement. A total of 1689 utterances were recorded, with both coders agreeing on 90.7% (1532) of them. After utterances were identified, each one was coded for its category. Of the agreed upon utterances, the coders agreed on the category of 147 8 utterances (96.5%). Scott's pi for this code = .94. Because of the high levels of intercoder agreement, each coder was given half of the sample to code on their own (see Appendix L for coding instructions). Treatments and Stimulus Materials The context for the interaction setting was established by first telling participants in all conditions (F tF , CMC-no avatar, and the two CMC with avatars conditions) that they were taking part in a study on meeting new people. Those in the three CMC groups were told that their interactions would be done online. All participants were told the following: You will be taking part in a study on fi'iendship formation. In this study, you will take part in two sessions where you and another person will spend 15 minutes 47 getting to know each other. In the first session you will interact with and get to know another volunteer participant in the study. In the second session you will interact with a different volunteer participant in the study. After both interactions, you will choose which person you would prefer to hang out with. The two CMC with avatar conditions were created by manipulating features of the avatar to be more or less informative about their partner. In each condition, participants were led to think their interaction partner chose the avatar shown. In general, these avatars were static graphical representations of a user. In the CMC-high informativeness avatar condition the avatar shown was a photograph. In order to control for the effects of different sex combinations and provide more information, male participants in this condition were shown a female photograph for their partner, whereas female participants were shown male photograph of their partner. Thus, people were given a same—sex avatar to represent them in the interaction. In the CMC-low informativeness avatar condition, the avatar shown was a non-human icon with no identifiable gender. Pilot test of avatars used in current study. The selection of avatars used in this study was based upon two pilot tests conducted on the avatars. The first pilot test was part of a larger study (Westerman, Tamborini, & Bowman, 2007). A group of fifteen avatars were included in profiles and shown to 206 participants enrolled in introductory communication classes. The participants ranged in age from 17-24, with a mean of 19.95 years (SD = 1.30) and 165 (80.1%) were female and 38 (18.4%) were male, with 3 (1.5%) not responding. Participants were told that the profiles were created by students in another class, and asked to rate these supposed people on uncertainty and attraction in one of three contexts: task, social, or dating. A one-item measure of uncertainty was 48 employed, asking whether or not the profile provided enough information to make judgments about choosing that person. Participants responded on a five point Likert-type scale (1 = not at all, 5 = definitely). The avatars rated as most certain in the social context were two photographs (one male, one female) and the avatar rated as second most uncertain was a nonhuman icon of books, a graduation cap , and the word “achieve” (see appendix M for the avatars used in the main study). This avatar was used for the low informative condition because it can be used for both males and females, as opposed to the avatar rated most uncertain (a female cartoon character). Paired samples t-tests on this one-item measure of uncertainty found a significant difference between the male photo (M = 2.58, SD = 1.45) and the nonhuman icon (M = 2.00, SD = 1.02), t (25) = 3.26, p < .01, 112 = .30, and between the female photo (M = 2.71, SD = 1.36) and the nonhuman icon (M= 1.82, SD = .94), t (27) = 3.95, p < .01, n2 = .37, but not between the male (M= 2.69, SD = 1.31) and female photos (M= 2.53, SD = 1.24), t (31) = 1.41,p > .05, n2 = .06. The second pilot test on the avatars was part of the pilot study discussed above for establishing the validity of the one-item uncertainty and predicted outcome value measures. Though not central to the main purpose of this pilot study, analyses were also conducted on initial responses to the three avatars shown to respondents at the start of the pilot test. On the Parks and Roberts (1996) scale, paired samples t-tests found no significant differences (p < .10) for any combination of two avatars. However, the nonhuman icon was rated as the least certain (M = 2.39, SD = .63), followed by the female photo (M = 2.47, SD = .68) and the male photo (M = 2.61, SD = .81). On the social attraction uncertainty measure, no difference was found between the male photo 49 (M= 3.26, SD = .75) and the female photo (M= 3.26, SD = .70), t (33) = .00, p > .10, n2 = .00. However, the male photo was rated as more certain than the nonhuman icon (M = 3.02, SD = .73), t (33) = 1.70, p < .10, n2 = .08, as was the female photo, I (33) = 1.91, p < .10, n2 = .10. On the one-item measure of uncertainty, patterns were similar to those found for the social attraction uncertainty measure. The nonhuman icon was rated the least certain (M = 2.55, SD = 1.02), followed by the female photo (M = 2.88, SD = 1.17) and the male photo (M = 3.06, SD = 1.07). A significant difference was found between the male photo and the nonhuman icon, t (33) = 2.57, p < .05, n2 = .17. Although strong differences were not found among these avatars on the multiple measures of uncertainty, the patterns of uncertainty are similar to the findings of the first pilot study. The nonhuman icon was rated the least certain across all measures of uncertainty. There were no differences between the male and female photographs across the multiple measures of uncertainty. Although large differences between the icon and the two photos were not found, this is likely due to the fact that this pilot study also lacked the contextual information that the first pilot study provided (because the main focus was on the one-item measure of uncertainty). Therefore, the information provided by the two pilot studies suggests that these three avatars are acceptable ones to manipulate the informativeness of the avatar for this study and are the three that were used. 50 Results Prior to the main analyses, mood before interacting was analyzed using a 4 (condition) x 2 (sex) ANOVA to examine the possibility that people in different conditions began the interaction in different moods due to the different instructions they were given. Because it is unlikely that partners would have influence on each other's mood before initial exposure to each other, each person was treated as an individual data point in this analysis. The ANOVA on mood failed to reveal significant differences for both the main effect ofcondition, F(3, 119) = 1.16,p > .05, n2 = .03, of sex, F(1, 119) = .01,p > .05, n2 = .00, and the interaction between sex and condition, F (3, 119) = .65, p > .05, n2 = .02. Overall, negative mood was low (M = 1.53, SD = .76). Based on this low mean, combined with the fact that a seven point response scale was used, it appears that not only did people not differ in their negative mood as a result of the different instructions given across conditions, but people were generally not in bad moods to start their interactions. Analysis plan Because respondents in this study participated in dyads, issues related to interdependence remained salient throughout data analysis. Decisions on how to treat concerns about interdependence were made on a case-by-case basis using the goals particular to each stage of analysis and assumptions about the potential threat of interdependence at that stage to determine how potential interdependence threats would be addressed. In some instances, tests specifically designed to account for these threats were utilized. In other instances, the potential for such threats were either irrelevant for 51 the analyses being conducted or, though relevant, the threats were downplayed and alternative tests better suited for addressing the immediate questions at hand were used. These instances fall into four different groups. The first group involves only data that were collected prior to interaction using single-item measures. In these cases, it was assumed that the extent to which interdependence between respondents and their partners affected scores was negligible. As such, interdependence was considered moot, and analysis techniques that do not account for interdependence were used. The second group involves data collected after interaction that were completed using the same single-item measures. In a few cases, analyses of these single-item post-interaction responses were conducted in order to compare them with the pre-interaction responses collected using the same measurement technique. Although the potential for interdependence in the post-interaction data is apparent, analysis on these data were conducted with the same tests used on the pre- interaction data in order to provide comparisons at the two points in time. These analysis techniques did not account for interdependence. As such, in these few instances, interdependence threats were downplayed and analyses designed to address the immediate goals at hand were used instead of tests specifically designed to account for interdependence. The third group involves data from the single-item measures collected minute-by- minute across all time points including those prior to, during, and after interaction. In these cases, analyses focused on the visual examination of mean patterns over time without the use of statistical tests. Techniques that account for interdependence are designed to control for the amount of variance in an individual that is caused by the group 52 or dyad of which they are a member. These tests have no effect on overall computed means. Because the goal here is to provide a simple analysis of the patterns of means, techniques that account for interdependence were not used. The fourth group involves data collected using multiple-item measures collected only afier interaction was completed. Not only were these cases considered susceptible to potential interdependence problems, but these problems presented a direct threat to the goals for analysis on these data As such, tests specifically designed to account for these threats were utilized in these instances. Instances of all four different groups appear in the analyses that follow. Tests of hypotheses Hypothesis one predicted that prior to conversation, uncertainty about a stranger would be higher in CMC compared to FtF. Because it is unlikely that interdependence would have influenced a partner’s judgment of uncertainty prior to any discussion, each person was treated as an individual data point in this analysis. To test this hypothesis, an independent samples t-test comparing the means for the one-item measure of pre- conversation uncertainty taken from the modified protocol analysis across the two conditions was conducted. The independent samples t-test shows that uncertainty is statistically significantly higher in the CMC-no avatar condition (M = 4.23, SD = .90) than the FtF condition (M= 3.57, SD = .77), t (58) = -3.08, p < .01, n2 = .14. Thus, the data are consistent with hypothesis one. Hypothesis two predicted that differences in uncertainty between CMC and FtF would be smaller after 15 minutes of interaction than prior to interaction. To test this hypothesis, an independent samples t-test was first conducted to see what differences existed at 15 minutes. Although interdependence of data may be expected for judgments 53 of uncertainty after interaction, people were still treated as individual data points for this analysis. This was done because the main objective was to compare mean differences of uncertainty between FtF and CMC at two separate times (pre-conversation and afier 15 minutes). Mean difference will be the same whether the data are treated individually or dyadically. An independent samples t-test of uncertainty afier 15 minutes of interaction shows that CMC (M = 2.00, SD = .59) is not significantly different from FtF (M = 1.77, SD = .73), t (58) = -1.37, p > .05, n2 = .03. The mean difference between CMC and FtF prior to conversation (.66) is greater than the mean difference between CMC and FtF after 15 minutes of interaction (.23). This suggests that the gap between the two conditions is lessening. To test this possibility in greater detail, a mixed AN OVA was performed, with condition as a between-subjects variable and uncertainty levels (pm-conversation and post-conversation) was the within-subjects variable. If a strong narrowing of the gap between conditions is occurring over time, a statistically significant interaction effect would be found. Although statistically significant main effects were found for condition, F(1, 58) = 11.74,p < .005, n2 = .17, and time, F(1, 58) = 196.12,p< .001, n2 = .76, the interaction effect was not statistically significant, F (1, 58) = 2.26, p > .05, 112 = .01. Thus there is no statistically significant effect for a convergence over time, although the mean differences follow a pattern consistent with hypothesis two. Hypothesis three predicted the following pattern of uncertainty over time for comparisons involving FtF and CMC-no avatar conditions: 1) Uncertainty would start higher in CMC than FtF, 2) uncertainty would decrease rapidly in FtF and then level off, 3) uncertainty would decrease more slowly in CMC than FtF, but 4) uncertainty would 54 finally level off at a level similar to FtF. In order to address this hypothesis, patterns of uncertainty means over time for the FtF and CMC-no avatar conditions were compared. Because the goal of this analysis was to compare patterns of means for two conditions over time, tests involving interdependence were not conducted, as discussed in the summary of the analysis plan above. Means and standard deviations are included in Table 7. A graphic depiction of means for both conditions is shown in Figure 7. Table 7 Uncertainty Means Over Time (standard deviations) Time FtF No Avatar 0 3.57 (.77) 4.23 (.90) 1 3.27 (.69) 3.73 (.98) 2 2.63 (.72) 3.17 (.95) 3 2.40 (.77) 2.93 (.74) 4 2.33 (.66) 2.73 (.78) 5 2.13 (.57) 2.60 (.89) 6 2.07 (.58) 2.57 (.82) 7 2.10 (.66) 2.33 (.71) 8 2.10 (.66) 2.30 (.53) 9 1.93 (.52) 2.30 (.70) 10 1.87 (.57) 2.30 (.60) 11 1.73 (.64) 2.07 (.58) 12 1.77 (.57) 2.10 (.61) 13 1.80 (.76) 2.17 (.59) 14 1.67 (.66) 2.03 (.56) 15 1.77 (.73) 2.00 (.59) 55 Figure 7. Uncertainty means over time. Uncertainty Over Time 3.50 Uncertainty in O O —o- FtF Uncertainty +CMC/No Av. Uncertainty 0 1 2 3 4 5 6 7 8 9 101112131415 Time Examination of these means suggests that uncertainty change over time in FtF and CMC- no avatars conditions occurs through similar patterns. Uncertainty starts out higher in the CMC condition and remains higher over the course of the 15 minute interaction, although there is evidence (fiom analyses conducted for hypothesis 2) that uncertainty levels are converging by the end of the 15 minutes. Interestingly, the patterns of uncertainty reduction are nearly identical over the fifteen minutes, with uncertainty reduction occurring rapidly in the first two minutes in both conditions, continuing at a moderate pace until the sixth minute, and then continuing slowly until the end of 15 minutes. Thus, although uncertainty starts higher in CMC compared to FtF as predicted, and uncertainty levels appear to be converging afier fifteen minutes as predicted, the overall patterns of data are not consistent with hypothesized patterns. Instead of following a more linear pattern as predicted, uncertainty in CMC follows a very similar pattern as uncertainty in FtF. Thus, the data are not consistent with hypothesis three. 56 Hypothesis 4 predicted patterns of decreasing IURS usage over time that would differ in CMC compared to FtF. IU RS use would start relatively high in FtF, decrease rapidly, and level off as the interaction ends. IURS use in CMC would also be at its highest point at the start, but would decrease more slowly as time elapsed. Due to channel limitations, IURS use in CMC would always be lower than FtF. A similar procedure as that used to test hypothesis 3 was used to test hypothesis 4, and patterns of IURS use means over time for the FtF and CMC-no avatar conditions were compared. Again, similar to the procedure for hypothesis 3, no tests of interdependence were conducted because this analysis centered on visual description of patterns of means. Means and standard deviations are included in Table 8. A graphic depiction of means for both conditions is shown in Figure 8. 57 Table 8 Number of I URS Means Over Time (standard deviations) Time F tF No Avatar 12.23 (3.27) 2.47 (1.01) \OWQO‘M-§WNu—s t—it—‘h-‘tl-Il—lt—i (II-tht-‘o 9.30 (3.72) 6.80 (2.76) 6.57 (3.10) 7.40 (2.37) 6.17 (2.38) 6.17 (2.88) 7.13 (3.46) 5.60 (2.82) 5.90 (3.46) 5.57 (2.28) 5.27 (2.53) 4.60 (1.96) 4.93 (2.85) 4.90 (3.38) 2.17 (1.37) 1.87 (1.22) 1.73 (1.26) 1.53 (1.20) 1.67 (1.09) 1.73 (1.05) 1.30 (1.06) 1.47 (1.43) 1.33 (1.03) 1.40 (1.28) 1.60 (1.35) 1.53 (1.17) 1.40 (1.25) 1.30 (1.09) 58 Figure 8. Number of IURS means over time. IURS Over Time 10 , “f . 1 ‘3‘ 2 i £3“:me - fifA , ‘ *5 *4 333-13 J A___A 3 “”1123. AL}. 4 .-,... .4 ~ he ~ . -9-CMCINoAv.IURS 2 “i o _ 1 2 3 4 5 6 7 8 9 101112131415 Time Examination of this pattern of means reveals that IURS usage decreased over time in both conditions. Over the course of the interaction, the number of IURS used decreased from 2.47 to 1.3 per minute in the CMC-no avatar condition and from 12.23 to 4.90 per minute in the FtF condition. In both conditions, most of the total decrease (74% of FtF and 51% of CMC) occurred in the first three minutes of interaction. The decrease in IURS usage occurs more rapidly in FtF as compared to CMC. In the third minute of FtF, people used just over 50 percent of the IURS they used in the first minute. In comparison, in the third minute of CMC, people used about 75 percent of the IURS they used in the first minute. By the fifteenth minute of interaction, people in F tF are down to about 40 percent of their first minute IURS usage, whereas, people in CMC still use over 50 percent of their first minute 1U RS totals. Thus, IURS use appears to follow the predicted patterns for both CMC and FtF, and the data are deemed as consistent with hypothesis four. 59 Hypothesis 5 predicted that the informativeness of an avatar provided to a person would influence levels of uncertainty such that more informative avatars would decrease uncertainty more than less informative avatars. Because it is unlikely that that interdependence would have influenced a partner’s judgment of uncertainty prior to any discussion, each person was treated as an individual data point in this analysis, as discussed in the preview paragraph describing the data analysis plan. To test hypothesis 5, an independent samples t-test was performed on uncertainty prior to discussion for the two CMC with avatar conditions. Results failed to show statistically significant differences, t (58) = -1.01, p > .05, n2 = .02. Although the difference between the two CMC with avatar conditions was not significant, it is notable that the pattern of means was in the hypothesized direction. Participants in the high informativeness avatar condition reported less uncertainty (M = 4.23, SD = 1.04) than participants in the low informativeness avatar condition prior to conversation (M = 4.47, SD = .73). Post hoc analyses Additional analyses were conducted to examine characteristics of the data set not represented in the formal tests of hypotheses. First, replicating the analysis plan used to examine hypotheses one through three, over-time patterns of uncertainty for all four avatar conditions were examined in order to compare the uncertainty reduction (or creation) capability of the two avatar conditions with the F tF and CMC—no avatar conditions. Second, three variables that were measured across modes only after the 15 minute interactions (liking, and the comprehensive measures of uncertainty and predicted outcome values) were analyzed using multilevel modeling procedures as suggested by Kenny, Kashy, and Cook (2006). This was done in order to examine differences in these 60 three variables across sex and condition after 15 minutes of interaction, while attempting to control for possible interdependence issues in these data. Third, the patterns of IURS over time were examined as percentages (as opposed to total numbers) in order to allow comparisons to past studies examining IURS usage. Finally, the single-item measure that assessed predicted outcome values during ongoing interactions was examined in a manner similar to the one used to examine uncertainty in hypothesis three. This was done in order to examine general patterns of predicted outcome values for potential differences across communication mode during interactions between strangers. Uncertainty over time across all communication mode conditions. No formal hypotheses were made for uncertainty regarding all four conditions. This was due to ambiguity concerning the role of avatars in uncertainty reduction. It is possible that any nonverbal information would be better than no nonverbal information in CMC, and thus both CMC conditions would fall between the FtF and the CMC-no avatar condition. It is also possible that a low informative avatar might serve as a distracter and a high informative avatar might be useful, and therefore uncertainty levels for the CMC-no avatar condition would fall in between those of the two avatar conditions. Finally, it is also possible that any avatars serve as distracters (consistent with Westerman & Tamborini, 2005, 2006), and thus both avatar conditions would show higher levels of uncertainty than the CMC-no avatar condition. In order to examine these possibilities in greater depth, similar analyses as those used to examine hypotheses one through three were conducted for all four conditions. First, patterns of uncertainty means over time for these conditions were compared. Because this analysis focused on the description of mean patterns for the four conditions, 61 no tests of interdependence were conducted. Means and standard deviations are included in Table 9. Figure 9 shows a graphic depiction of means for all conditions. Table 9 Uncertainty Means Ovethime (standard deviations) Time FtF No Avatar Hi Info Avatar Lo Info Avatar 0 3.57 (.77) 4.23 (.90) 4.23 (1.04) 4.47 (.73) 1 3.27 (.69) 3.73 (.98) 3.63 (1.03) 3.87 (.94) 2 2.63 (.72) 3.17 (.95) 3.37 (1.07) 3.57 (.97) 3 2.40 (.77) 2.93 (.74) 2.87 (.94) 3.27 (.69) 4 2.33 (.66) 2.73 (.78) 2.77 (.97) 2.90 (.84) 5 2.13 (.57) 2.60 (.89) 2.43 (.94) 2.73 (.69) 6 2.07 (.58) 2.57 (.82) 2.40 (.89) 2.47 (.68) 7 2.10 (.66) 2.33 (.71) 2.33 (.80) 2.47 (.63) 8 2.10 (.66) 2.30 (.53) 2.30 (.84) 2.47 (.73) 9 1.93 (.52) 2.30 (.70) 2.31 (.81) 2.40 (.67) 10 1.87 (.57) 2.30 (.60) 2.13 (.73) 2.43 (.77) 11 1.73 (.64) 2.07 (.58) 2.10 (.80) 2.33 (.80) 12 1.77 (.57) 2.10 (.61) 2.00 (.83) 2.20 (.71) 13 1.80 (.76) 2.17 (.59) 2.03 (.81) 2.07 (.58) 14 1.67 (.66) 2.03 (.56) 1.93 (.74) 2.07 (.64) 15 1.77 (.73) 2.00 (.59) 2.03 (.76) 2.03 (.67) 62 Figure 9. Uncertainty means over time. Uncertainty Over Time —+- FtF Uncertainty +CMC/No Av. Uncertainty Uncertainty +CMC/Hi Info Uncertainty +CMC/Lo Info Uncertainty 0 1 2 3 4 5 6 7 8 9 101112131415 Time Examination of the means shows that the two avatar conditions follow patterns of uncertainty reduction over time that are similar to each other and to both the FtF and CMC- no avatar conditions. Notably, the actual levels of uncertainty at any point in time seem highly dependent upon the starting point of uncertainty. Thus, the effect that condition has on uncertainty seems more indirect; it appears to have an influence on initial uncertainty, which in turn has an influence on subsequent uncertainty, with final differences appearing smaller than initial differences. To address this possibility, a one-way ANOVA was first performed on pre- conversation measures of uncertainty across the four communication mode conditions. Similar to tests of hypothesis one, because interdependence is very unlikely at the pre- conversation stage, no tests of interdependence were run. This analysis shows a significant difference, F (3, 119) = 5.99, p < .01, n2 = .13. Subsequent Neuman-Keuls analysis shows that pre-conversation uncertainty in the FtF condition is significantly 63 lower (M = 3.57, SD = .77) than the CMC-no avatar (M = 4.23, SD = .90), CMC-high informativeness avatar (M = 4.23, SD = 1.04), or CMC-low informativeness avatar conditions (M = 4.47, SD = .73). A one-way AN OVA across the four conditions was also conducted on the one- item measure of uncertainty after 15 minutes of interaction. Despite the potential for interdependence in this variable after 15 minutes of interaction, people were treated as individual data points, similar to tests run for hypothesis 2. This approach was used because the main goal of this analysis was to examine post-interaction differences in uncertainty across the four conditions in comparison with earlier analyses that examined pre-interaction differences across the four conditions. Since these earlier analyses were conducted treating respondents as individual data points, the same approach was used here. These analyses show that after 15 minutes, no significant differences were found for condition, F (3, 119) = 1.04, p > .05, n2 = .03. Uncertainty in FtF (M = 1.77, SD = .73) is still lower than uncertainty in the CMC-no avatar (M = 2.00, SD = .59), CMC-high informativeness avatar (M = 2.03, SD = .7 6), or the CMC-low informativeness avatar conditions (M = 2.03, SD = .67). In addition, the mean differences between FtF and each of the three CMC conditions are smaller after 15 minutes (no avatar = .23, high informativeness avatar = .26, low informativeness avatar = .26) than they were at the pre- conversation point (no avatar = .66, high informativeness avatar = .66, low informativeness avatar = .90). These data are consistent with the notion that uncertainty levels are converging across conditions after fifteen minutes of interaction. Liking and comprehensive measures of uncertainty and predicted outcome values after 15 minutes. Prior to the post-hoe examination of scores on liking and on the 64 comprehensive measures of uncertainty and predicted outcome values, steps were taken to account for the potential influence of interdependence in the data. Because the goal of these analyses was neither the visual inspection of means nor the comparison to pre- conversation levels of the same measures, interdependence was considered a more central concern. Moreover, since these observations occurred after fifieen minutes of dyadic interaction, it can be expected that scores were heavily influenced not only by respondents’ own droughts and actions, but also by what their partners did and said. As such, it seems plausible that these data contain levels of interdependence that necessitate the use of tests designed to account for interdependence. To address possible sex and condition differences for these post interaction measures, and also to address the possible interdependence of these data after 15 minutes of interaction, multilevel modeling procedures were utilized as suggested by Kenny et a1. (2006). These scholars suggest treating individual responses in dyads as interdependent if there is any reason to believe that scores for one person have an influence on scores for the other person and, as previously stated, there is reason to believe the data fiom the three measures taken only after interaction will be interdependent. Multilevel models are conducted to examine data that are nested within multiple levels (Kenny et al., 2006). A basic multilevel model has two levels. A level one variable is one that occurs at the individual level. A level two variable is one that occurs at the group/dyad level. This means that all individuals with a given group or dyad will have the same score for any second level variable. Analyzing data in such a way allows one to examine the effects of both individual and dyadic level variables on outcome measures. For analyses conducted on the three post-interaction outcomes, the outcome was treated 65 as a level one variable because it was measured at the individual level. Respondent sex was treated as a level-1 predictor variable because it is an individual level variable nested within each dyad. Condition was treated as a level-2 predictor variable because it is a dyadic level variable, and as such, each member of a given dyad has the same score for condition. For each outcome variable, an intraclass correlation was also calculated. An intraclass correlation provides a measure of the extent to which interdependence (in this instance between respondents and their partners) affects scores on some variable. Dyadic intraclass correlations are interpreted similarly to Pearson correlations. A positive intraclass correlation means that as one dyad member's score increases, so will the other member's, whereas a negative intraclass correlation means that as one dyad member's score increases, the other member’s score will decrease. A positive intraclass correlation can also be interpreted as a percentage of the variation in scores on a variable that can be accounted for by the dyad to which the individuals belong (Kenny et al., 2006). The first measure that was taken only after the fifteen minute interaction was complete was the Parks and Roberts (1996) measure of uncertainty. Because the Parks and Roberts (1996) measure was obtained at the end of the fifteen minute interaction, it can be used to provide a more stringent test of differences in post-interaction uncertainty than the one-item measure obtained from the protocol analysis. When the one-item protocol measure was used, no differences were found among the three CMC modes (high informativeness, low informativeness, and no avatar conditions) prior to or after fifteen minutes of interaction. Not surprisingly, FtF uncertainty was significantly lower than the three CMC conditions prior to discussion. Notably, however, though F tF 66 appeared lower than all three CMC conditions after the 15 minute interaction, these differences were no longer statistically significant. Since these CMC conditions showed no differences after 15 minutes and the goal of the present analysis was to provide a more stringent test of post-interaction uncertainty, the three CMC conditions were coded the same in analyses using the comprehensive measure of uncertainty. Therefore, condition was effect coded such that FtF = 3, CMC-high informativeness avatar = -1, CMC-no avatar = -1, and CMC-low informativeness avatar = -1. This allowed a comparison of post-interaction uncertainty between FtF and CMC in general using the more established measure. Sex was effect coded such that male = 1 and female = -1. Multilevel modeling analysis for the five-item measure of uncertainty revealed a grand mean of 3.3 1. A significant effect of condition was found, t (59.36) = -3.27, p < .005. With a coding scheme in which FtF is coded as positive (equal to 3 in this case) and all CMC conditions are coded as negative (equal to -l), the negative estimate for condition suggests that the positively coded FtF condition produced decreased outcomes. This suggests that uncertainty after 15 minutes of interaction was lower in the FtF condition compared to each of the CMC conditions when using the Parks and Roberts (1996) measure. No significant effect of sex was found, t (58.15) = -l .50, p > .05. There was no significant effect for the interaction effect between sex and condition, t (59.08) = -.98, p > .05. The intraclass correlation for uncertainty was .37. Thus, 37% of the variance in individual uncertainty scores after 15 minutes of interaction was accounted for by dyad membership. 67 Table 10 Uncertainty Multi-Ievel Modeling Values Estimate D] T Intercept 3 .3 1 58.49 43 .92 Sex -.08 58.15 -1.50 Condition -.14 59.36 -3.27 Sex‘Condition -.03 59.08 -.98 Because liking is predicted to be directly related to uncertainty based on URT principles, the same effect coding used for uncertainty was also used for liking. Multilevel modeling analysis for liking after fifteen minutes revealed a grand mean of 3.67. No significant effects were found for sex, t (57.55) = 1.43, p > .05, or condition, t (57.44) = 1.98, p > .05. There was also no significant effect for the interaction between sex and condition, t (57.23) = 1.63, p > .05. The intraclass correlation for liking was .26. Thus, 26% of the variance in individual liking scores was accounted for by dyad membership. Table 11 Liking Multi-Ievel Modeling Values Estimate Df T Intercept 3.67 57.72 72.23 Sex .06 57.55 1.43 Condition .06 57.44 1.98 Sex*Condition .04 57.23 1 .63 68 In this study, predicted outcome values are conceptually parallel to liking, and items used to assess predicted outcome values measured predictions of positive future interaction. Therefore, the same effect coding used for liking was used for predicted outcome values. Multilevel modeling analysis for the comprehensive measure of predicted outcome value found a grand mean of 4. 1 8. No significant effects were found for sex, t (58.18) = 1.03, p > .05, condition, t (58.07) = .37, p > .05, or the interaction effect between sex and condition, I (57.87) = 1.80, p > .05. The intraclass correlation for predicted outcome value was .15. This means that 15% of the variance in individual predicted outcome value scores was accounted for by dyad membership. Table 12 Predicted Outcome Value Multi-level Modeling Values Estimate Df T Intercept 4.18 58.38 69.60 Sex .05 58.18 1.03 Condition .01 58.07 .37 Sex*Condition .05 57.87 1 .80 Percentage of I URS use over time. The current study argues that considering the total number of IURS is a useful way to measure IURS for the minute by minute analysis employed in the study. However, past studies (Tidwell & Walther, 2002; Westerman & Tamborini, 2006) treated IURS as a percentage of total utterances. In order to compare the current study to past studies, a similar procedure as that used to test hypothesis 4 was conducted on the percentage of IURS used each minute. Patterns of IURS use as a percentage of total utterances over time for the F tF and CMC-no avatar conditions were 69 compared. As with hypothesis 4, because this analysis centered on visual description of patterns of means, no tests of interdependence were conducted. Means and standard deviations are included in Table 13. A graphic depiction of means for both conditions is shown in Figure 10. Table 13 I URS Percentage Means Over Time (standard deviations) Time F tF No Avatar 1 .70 (.11) .73 (.20) 2 .63 (.17) .83 (.24) 3 .52 (.19) .83 (.27) 4 .54 (.21) .67 (.30) 5 .56 (.17) .76 (.28) 6 .50 (.19) .77 (.29) 7 .49 (.16) .83 (.21) 8 .52 (.21) .59 (.31) 9 .47 (.20) .69 (.35) 10 .51 (.25) .65 (.34) 11 .53 (.23) .63 (.37) 12 .46 (.20) .77 (.35) 13 .40 (.18) .75 (.34) 14 .45 (.20) .63 (.41) 15 .44 (.23) .66 (.33) 70 Figure 10. IURS percentage means over time. IURS (Percentage) Over Time +FtF IURS% "—CMC/No Av. IURS% lURSiPercentage) 1 2 3 4 S 6 7 8 9101112131415 Time Examination of these means shows that IURS use as a percentage of total . utterances conforms to the following pattern for FtF interactions: The percentage of IURS drops most rapidly from minute 1 to minute 3, and then continues to slowly decrease over time. However, although the percentage of IURS starts out at similar levels in FtF and CMC, the percentage of IURS use over time in CMC does not follow the same pattern. Instead of a rapid drop in IURS use right from the start, there appears to be an initial increase in the percentage of IURS use, resulting eventually in a slight downward trend overall. It is also important to note that the percentage of IURS is greater for the CMC-no avatar condition than for the FtF condition for each of the fifteen minutes. Predicted outcome values over time across all communication mode conditions. Predicted outcome values were measured retroactively before, during and after the 15 minute interaction using a one-item measure. Patterns of means for predicted outcome values over time for the four experimental conditions were compared. Once again, because this analysis focused on visual inspection of patterns of means, tests for 71 interdependence were not conducted, despite the potential for interdependence to exist. Means and standard deviations are included in Table 14. A graphic depiction of means for all conditions is shown in Figure 11. Table 14 Predicted Outcome Value Means Over Time (standard deviations) Time F tF No Avatar Hi Info Avatar Lo Info Avatar 0 3.57 (.63) 3.30 (.79) 3.47 (.63) 3.23 (.68) 1 3.33 (.61) 3.40 (.72) 3.60 (.67) 3.33 (.84) 2 3.57 (.57) 3.63 (.67) 3.60 (.77) 3.47 (.82) 3 3.70 (.70) 3.67 (.66) 3.73 (.69) 3.57 (.68) 4 3.87 (.68) 3.83 (.59) 3.90 (.76) 3.67 (.61) 5 3.87 (.57) 3.93 (.58) 3.97 (.81) 3.63 (.67) 6 3.83 (.59) 3.87 (.78) 3.93 (.74) 3.90 (.55) 7 3.80 (.81) 3.83 (.59) 3.90 (.71) 4.00 (.37) 8 3.90 (.55) 3.90 (.56) 3.97 (.72) 4.00 (.53) 9 4.07 (.64) 3.90 (.71) 3.83 (.76) 3.93 (.52) 10 4.03 (.76) 3.73 (.87) 4.03 (.67) 3.83 (.59) 11 4.10 (.84) 3.83 (.79) 4.03 (.67) 3.93 (.52) 12 4.13 (.73) 3.83 (.70) 4.20 (.61) 4.00 (.53) 13 4.03 (.76) 3.90 (.71) 3.97 (.81) 3.97 (.56) 14 4.10 (.76) 3.93 (.74) 4.03 (.85) 3.93 (.64) 15 4.07 (.78) 3.87 (.82) 3.97 (.85) 4.03 (.56) 72 Figure 11. Predicted outcome value means over time. POV Over Time --FtF POV +CMC/No Av. POV —0—CMC/Hi info POV -fi-CMC/Lo Info POV 0 1 2 3 4 5 6 7 8 9101112131415 Time Examination of the patterns of means for predicted outcome values over time suggests that change occurs for all four conditions in a similar pattern. Predicted outcome values start at their lowest point and increase over time until the seventh minute. As interactions continue, predicted outcome values generally level off, although they increase slightly until the end of the interactions. To examine starting and ending points of predicted outcome values across conditions, two one-way AN OVAs were performed similar to analyses conducted for the one-item measure of uncertainty at the pro-conversation point and after fifteen minutes of interaction. Although issues of interdependence are unlikely at the pre-conversation point, there is potential for interdependence after 15 minutes of interaction. However, similar to analyses for the one-item uncertainty measure, individuals were treated as separate data points for this analysis because the focus was on comparing differences in 73 uncertainty across the four conditions before and after interaction. First a one-way AN OVA was performed on pre-conversation predicted outcome values across conditions. This analysis showed no significant difference, F (3, 119) = 1.48, p > .05, n2 = .04. A one-way AN OVA was also conducted across condition for predicted outcome values after 15 minutes of interaction. No significant differences were found for condition, F (3, 119) = .40,p > .05, n2 = .01. Although differences did not exist across conditions at either individual point in time, post-conversation predicted outcome values were more positive than pre- conversation predicted outcome values in all four conditions. These values were as follows: for the FtF condition, post-conversation (M = 4.07, SD = .78) and pre- conversation (M = 3.57, SD = .63); for the CMC-no avatar condition, post-conversation (M = 3.87, SD = .82) and pre-conversation (M = 3.30, SD = .79); for the CMC-high informativeness avatar condition, post-conversation (M = 3.97, SD = .85) and pre- conversation (M = 3.47, SD = .63); and for the CMC-low informativeness avatar condition, post-conversation (M = 4.03, SD = .56) and pre-conversation (M = 3.23, SD = .68). 74 Discussion SIPT (Walther, 1992) holds that the lack of nonverbal cues in CMC creates problems for online impression formation not found in FtF communication, but assumes both that individuals find ways to overcome this deficiency and that, given enough time, the impressions they form through CMC will be equivalent to those formed FtF. One way people are thought to do this is through the increased use of IURS in CMC, however prior to this research no attempt has investigated these processes during ongoing CMC interaction. This study examined these assumptions by observing patterns of IURS use and uncertainty reduction during ongoing interactions in FtF and CMC, and explored the role played by one type of nonverbal information used in CMC: avatars. Respondent perceptions and behaviors prior to, after, and during CMC and FtF interaction were observed and compared. The findings have important implications for communication, both computer-mediated and otherwise. Two outcomes stand out from observations made prior to interactions. First, inconsistent with predictions, there were no strong differences in initial uncertainty among avatar conditions designed to vary in informativeness. However, as hypothesized, initial uncertainty was higher overall in CMC than in FtF interaction. This second finding is interesting when considered in connection with observations made after interaction. Consistent with logic on how time would affect perceptual change, initial differences in CMC and FtF uncertainty appeared to converge after 15 minutes of interaction. Whereas observations made prior to and after interaction were replications of previous research, the repeated measures of uncertainty and IURS made during interaction were unique to this study and offered some of the study’s more interesting 75 findings. Perhaps most notable in this regard are findings showing that the over time patterns of uncertainty reduction did not differ across communication modes as expected. Instead, the patterns of uncertainty reduction for CMC and FtF interaction appeared to be surprisingly similar. At the same time, patterns of IURS use were consistent with predictions. As predicted, IURS use in FtF appeared to start high, quickly drop, and then level off, whereas IURS use in CMC started high but dropped more slowly, and was always lower than FtF use as predicted by channel restrictions. Finally, in addition to examining uncertainty reduction and IURS, this study explored related issues coupled with observations of predicted outcome values prior to, during and after interaction. One occurrence of note from these observations was that regardless of whether communication occurred through FtF or in CMC, strangers began to predict similarly more positive outcomes after a few minutes of interaction. This suggests that there might be key common things to learn about a stranger (such as the fact that they are a student just like you) that lead to these feelings. The discussion that follows begins by detailing these and other findings from observations made prior to, after, and during interaction. This is followed by a consideration of the conceptual and methodological limitations in this study and their ramifications for the reported findings. Finally, the section concludes with a discussion of this study’s implications for understanding the processes that distinguish FtF and CMC interaction and how future research might help address old questions that remain and new questions that were raised. 76 Observations made prior to interaction Uncertainty and predicted outcome values were both measured prior to conversation in the current study. Examining uncertainty at the pre-conversation point was important because any differences found between conditions of FtF and CMC prior to interaction are likely due to the utilization of nonverbal cues that are present in FtF settings and lacking in CMC. Notably, people in all three CMC conditions reported higher levels of uncertainty before talking to their partners than those interacting FtF. This was consistent with hypothesized patterns, and it also contradicts the findings of Westerman and Tamborini (2005) who found no significant differences in uncertainty pre-conversation between CMC and FtF interactions. The inconsistent finding between these two studies invites questions as to why they differed. One potential explanation is based on a belief that the procedure used to measure uncertainty by Westerman and Tamborini (2005) may have attenuated pre- conversation uncertainty differences. Westerman and Tamborini asked participants to make judgments of uncertainty about a person before any discussion with that person occurred, and used the same Parks and Roberts (1996) scale to measure uncertainty that was administered in the present study after interacting for 15 minutes. This scale asks for responses to statements such as “I can accurately predict how this person will respond to me in most situations.” It is not hard to imagine that participants in the Westerman and Tamborini study found it strange to respond to these questions before ever conversing with the person, and participant comments during debriefing seem to support that notion. Moreover, we might speculate that a sense of global uncertainty was created for participants by ambiguity about why they were being asked answer such questions at that 77 time. This should have affected all participants and attenuated differences in uncertainty that might have been observed between FtF and CMC interactants. If this occurred, it would explain why no significant differences were found by Westerman and Tamborini. Notably, the pre-conversation uncertainty levels in the current study were consistent with predicted patterns. People interacting in FtF had more nonverbal information for use in forming judgments than those in CMC, and they reported lower levels of uncertainty before talking to partners than did CMC participants. These findings are consistent with previous scholarship highlighting the importance of nonverbal information in both the uncertainty reduction process (Berger & Calabrese, 1975) and the person categorization process (Argyle, 1975; Ichheister, 1970), which is itself a form of uncertainty reduction. The findings are also consistent with prior theorizing (Patterson, 1995), suggesting that the person categorization process is a nearly automatic one, as people reported lower uncertainty in FtF conditions prior to conversation. This suggests that short exposure to the nonverbal information provided in FtF interactions allowed reduction in uncertainty prior to gathering other types of information that could be used for uncertainty reduction. As noted above, predicted outcome value was also examined pre-conversation between participants. No differences across condition were found for predicted outcome values before conversations began. One interpretation of this finding is that when people began this study they were equally willing to communicate with strangers regardless of the channel in which they met. Predicting positive outcomes prior to conversation means that these people were probably engaged in having those conversations. Interestingly, this also suggests that people were equally trusting of strangers, regardless of the channel 78 through which they were about to interact. Notably, this contradicts qualitative evidence found in Westerman and Tamborini (2005) to the contrary. In that study, people generally reported keeping an open mind when meeting strangers F tF (predicted positive outcomes), but said they were more wary when meeting strangers online (predicted negative outcomes). As there is no eminently apparent reason to have greater confidence in the results of the present investigation compared to those of Westerman and Tamborini, answers regarding questions about channel differences in pre-conversation predicted outcome value remain inconclusive. Observations made after interaction Uncertainty, liking, and predicted outcome value were all examined after fifteen minutes of interaction. Uncertainty was examined post-interaction across conditions in two different ways. First, it was measured with the single-item measure used as part of the protocol analysis. No statistically significant post-interaction differences were found across conditions with this measure. Uncertainty after fifteen minutes was also assessed using Parks and Roberts’ (1996) measure of interpersonal predictability. Statistically significant differences were found with this measure, showing that uncertainty was lower in FtF than CMC. A relatively high correlation (r = .44, p < .01) between the Parks and Roberts’ measure and the single-item measure used in the protocol analysis suggests that the measures were examining the same concepts. Combining the one-item measure with the Parks and Roberts’ scale also creates a unidimensional solution, providing more evidence for the suggestion that these two measures are examining the same concept. Thus, there may still be differences in uncertainty after fifteen minutes. 79 Whether or not statistically significant differences exist after fifteen minutes of interaction, it does seem that the gap in uncertainty between FtF and CMC closed considerably compared to pre-conversation levels. Although finding significant differences in uncertainty across FtF and CMC conditions after 15 minutes of interaction may be inconsistent with Nowak (2004), the fact that these differences appear to be narrowing is consistent with SIPT (Walther, 1992). It also suggests that although uncertainty reduction achieved through CMC may eventually equal FtF, fifteen minutes may not be enough time for this to occur. Liking was also addressed after 15 minutes of interaction to search for potential differences based on condition, and none were found. This is consistent with Westerman and Tamborini (2006) who also found no differences in liking between CMC and FtF after five minutes of interaction. Thus, it seems that channel-based differences in liking that occur pre-conversation (Westerman & Tamborini, 2005) are overcome rather quickly, and changes in liking produced through further interaction occur similarly across conditions. Predicted outcome value was also analyzed post-interaction. Both a one-item measure and a longer measure (Sunnafrank, 1988) showed no differences due to channel in predicted outcome value at this point in the interaction. It was interesting to note that predicted outcome value increased from the pre-conversation point to the post- conversation point in all conditions. This suggests that interaction itself has more effect on predicted outcome values than the channel used for interaction. It also indicates that people were still interested and actively involved in their conversations as they neared the end of them. If people still predicted positive outcomes from future interaction (as 80 evidenced by the higher scores in predicted outcome values after interaction), they were also more likely to continue interactions. This suggests that participants in the study were motivated to partake in the interactions, and continue the conversation, even toward the end of the fifteen minutes. It also indicates that people maintained motivation to continue reducing uncertainty (Berger, 1979). Observations made during interaction Perhaps the most important component of SIPT is time. SIPT suggests that judgments can be made in CMC just as well as FtF, as long as there is enough time (Walther, Anderson, & Park, 1994). In other words, it takes longer for people to accomplish the same goals in CMC, but it can be done. In order to address this claim, people were asked to retroactively respond to single-item measures of uncertainty and predicted outcome value at each minute point in their interactions. These measures, as well as the number of IURS during each minute, were used to examine SIPT patterns over time. Predicted outcome values were also examined during the interaction. Patterns of uncertainty suggest that people may be able to accomplish goals in both CMC and F tF if given enough time. This is supported by evidence that uncertainty levels across condition seem to be converging. The mean difference between the most and least uncertain condition before conversation was .90, whereas this mean difference was only .26 after 15 minutes of conversation. Thus, consistent with SIPT theorizing and research, it seems that goals such as uncertainty reduction can be accomplished in CMC as they are in FtF when given sufficient time. Although these patterns of uncertainty suggest that CMC may become more like F tF over time, hypothesized patterns showing how this occurs did not materialize as 81 predicted. Hypothesized patterns were based on past findings showing that uncertainty levels in F tF and CMC were similar at the pre-conversation point (Westerman & Tamborini, 2005), levels in FtF were lower than CMC after five minutes of conversation (Westerman & Tamborini, 2006), and these levels were once again similar after fifteen minutes of conversation (Nowak, 2004). Consistent with Ramirez and Sunnafiank (2004) and Afifi and Burgoon (2000), the findings demonstrate that a great amount of uncertainty reduction in FtF takes place in the first 3 minutes of interaction. However, contrary to predictions of the current study, uncertainty is also reduced rather quickly in CMC. This would suggest that the differences that occur in uncertainty at any given point in time stem not from a slower uncertainty reduction process, but instead from the initial differences in uncertainty observed between FtF and CMC. Though not hypothesized in the present study, this finding is consistent with the logic of SIPT. Unlike prior studies comparing F tF and CMC that have allowed more time for CMC interactions (Tidwell & Walther, 2002), this study kept time consistent across both channels. This was done to determine if there was a point when uncertainty reduction stopped in F tF , allowing CMC to converge, or if uncertainty reduction would persist as long as interaction continued. As stated above, evidence combining the single-item and five-item measures of uncertainty after fifteen minutes suggests that although uncertainty levels may still differ from each other after fifteen minutes, they do appear to be converging. The current study design does not allow an answer to questions about whether differences in uncertainty at any point during an interaction are determined by characteristics of uncertainty reduction inherent to the channel, or from differences 82 between FtF and CMC in the initial level of uncertainty observed. Additional research is needed to provide a convincing answer to this question. In order to provide insight into the issue of differences between CMC and FtF in uncertainty reduction over time, the use of IU RS was observed continuously throughout the interaction. Although past studies (Tidwell & Walther, 2002; Westerman & Tamborini, 2006) have analyzed this variable as a percentage of total utterances to account for the disproportionally greater number of utterances possible in F tF , the goals of the current study made the total number of IU RS used more pertinent than the percentage of IURS used. Therefore, the patterns of IURS use over time were compared for FtF and CMC and were found to be consistent with hypothesized predictions. IURS use dropped more slowly over time in CMC than in FtF, and IURS was always lower in CMC than in FtF. However, this finding coupled with the patterns of uncertainty over time in these two conditions leaves questions about the effectiveness of IURS to reduce uncertainty unanswered. Uncertainty levels followed similar patterns over time in both F tF and CMC, although fewer IURS were used in CMC conditions. Taken together, these findings suggest that the sheer number of IURS alone may not be enough to predict uncertainty across multiple channels of interaction. Predicted outcome values were also observed over time to see if motivation to intemet with strangers varied in different communication channels. Predicted outcome values appeared generally high and no differences were observed across conditions before conversation, meaning that people were predicting generally that future interactions with strangers would be positive and equally so across conditions. Predicted outcome values also showed no differences across conditions after 15 minutes of 83 interaction. Together these findings suggest that channel neither affects predicted outcome values for interacting with strangers before conversation, nor does it affect actual conversations in ways that affect predicted outcome values during interaction. This is also documented by the finding that predicted outcome values became more positive after interactions for all four conditions in the study. In general these data offer support for SIPT (Walther, 1992). Uncertainty reduction can occur in CMC, but it takes longer than F tF . People were able to reduce uncertainty regardless of the channel in which they were interacting. However, questions remain about the length of time needed for CMC and FtF to be equal and the extent to which initial uncertainty differences between FtF and CMC persist, and what mechanisms people use to circumvent the channel restrictions inherent in CMC. Limitations Inevitably, all studies have limitations. Most notable in the present investigation are generalizability issues related to the selected sample, reliability issues associated with measures of uncertainty and predicted outcome value, questions about variance in the uncertainty reduction capacity of different IURS, questions about the uncertainty reducing potential of the avatars chosen for use in the study, and potential artifact resulting fi'om control of online access to non-experimental websites during the study. All of these potential drawbacks are considered here. The first of these potential limitations stems from the composition of the sample. This study used a solely student sample with all male-female dyads. Although there were good reasons for both of these choices, it is plausible that the findings would differ for subjects fi'om other populations or dyads with different gender combinations. A college 84 student sample was selected for this study because college students are among the heaviest users of IM technology (Jones, 2002). In order to observe the processes under consideration, it was important to have participants who where familiar with the technology and heavy users. Yet despite the sound logic for the composition of the sample selected, it would be interesting to see if the findings from this investigation would replicate in a sample of less experienced IM users. It is possible that those not familiar with IM would be more affected by the reduction in nonverbal cues because they would not be used to interacting with a channel that lacks them. If this were the case, then less experienced users might not have possessed the knowledge, skills, or practice needed to effectively reduce uncertainty when communicating through channels with more limited nonverbal cues. Similarly, the exclusive use of male-female dyads made sense in order to control variance associated with the make-up of different dyadic combinations. However, past research found that same-sex stranger dyads ask more questions than opposite-sex stranger dyads in FtF interactions (Douglas, 1987). If questions are important and useful for reducing uncertainty, then dyads that ask more questions should be able to reduce uncertainty quicker. Thus, it would be interesting to see if patterns found in the current study would hold for dyads comprised solely of males or females in both FtF and CMC. A second limitation stems from the techniques used to observe uncertainty and predicted outcome value over time. The use of single-item measures was difficult to avoid given the likelihood that more lengthy devices would have caused problems with participant burnout and impeded the respondents’ uninterrupted recall of their interaction experience. Although there is evidence that the single-item measures are sensitive to 85 actual change in uncertainty and predicted outcome value, traditional indices of measurement quality cannot be performed on these measures. This is particularly problematic because significant differences were found for uncertainty after 15 minutes using the Parks and Roberts (1996) measure that were not found after 15 minutes using the one-item scale. Although a moderately strong correlation (r = .44, p < .01) exists between these two measures, the correlation is not perfect, which raises measurement quality concerns regarding the techniques used for gathering these over-time observations. However, due to the nature of over-time measurement goals for this study, a better solution is not eminently apparent. Perhaps a study could be conducted where length of interaction was a separate variable controlled in the experimental design. This study could be conducted by allowing different dyads to interact for different lengths of time, and having them respond to uncertainty measures after their interactions. This would allow for comparisons of uncertainty after different interaction time lengths without asking people to retroactively recall what they were thinking during their interaction. The trade off with this approach is that it would require a large number of participants to achieve acceptable power in each condition, and a great deal of time to conduct the study. Issues regarding the potential for different IURS to vary in their uncertainty- reduction capacity introduce several concerns related to this study. All questions and disclosures were observed and counted as though they were equal, but it seems likely that this is not the case. The current study predicted that pe0ple would not be able to use as many IURS in CMC compared to FtF, and this would result in a more linear pattern of uncertainty reduction in CMC compared to FtF. Although the data are consistent with the 86 hypothesized patterns of IURS use in both CMC and FtF, the data are inconsistent with the hypothesized patterns of uncertainty in CMC. Instead of following a more linear pattern in CMC, uncertainty followed a similar pattern in both CMC and F tF , despite the use of fewer IURS in CMC. These patterns of IURS use and uncertainty cause a need to reconsider the role of IU RS use as an uncertainty reducing strategy. Because sheer numbers of IURS did not follow similar patterns to uncertainty, it could be that the uncertainty reduction capacity of different IURS varies. Each utterance coded as a question or disclosure was counted as an IURS. This approach not only makes it impossible to determine if questions and disclosures differ in their capacity to influence uncertainty, it also precludes us from discerning variance in the ability of different questions or disclosures to account for uncertainty reduction. It may be that people find questions more or less useful than disclosures in impression formation online. Certainly it is not difficult to imagine that some questions and self-disclosures reduce more uncertainty than others. It is possible that strangers that are interacting through CMC realize (either consciously or not) that they are under time pressure, and thus they focus on the questions and self-disclosures that are going to reduce the most uncertainty fastest. This would be consistent with the pattern of data found. Research that attempts to explicate and examine the attributes of IURS that shape their uncertainty-reduction capacity might help to determine if some questions and disclosures are more capable of facilitating the uncertainty reduction process than others. For example, one attribute of IURS that may have an influence on uncertainty- reduction capacity arises from the manner in which questions and self-disclosures were defined. A self-disclosure was defined as a statement that revealed some information 87 about the speaker. However, a question was defined as something that “invited a reply”, but not necessarily a reply about oneself. Therefore, some questions, coded as interactive uncertainty reduction strategies under current definitions, may not have actually served to reduce uncertainty about the other person. For example, based on the way these two concepts were defined, a statement such as “How is your sister doing?”, would have been coded as a question, despite the fact that this question was not asked in order to acquire information that would reduce uncertainty about the person involved in the interaction, and the response of “She's doing ok” was not coded as a self-disclosure under current definitions even though it might provide important information about the interactant’s state of mind. Past research (Berger & Kellerrnan, 1983; Douglas, 1987) examining FtF stranger interactions have employed a coding scheme intended to draw distinctions along these lines. In these studies, questions were coded into three major categories (about partner's self, about third parties, and about general information), each with a variety of subcategories. Although these past studies did not consider the uncertainty reducing potential of these different categories, it is clear that questions asking about third parties and those asking for general information seem much less likely to reduce uncertainty about one's interaction partner than those directly asking about the partner’s self. This is one example of how more careful consideration of different question and self—disclosure attributes, as opposed to a less discerning scheme that provides only global counts, might offer more detail and insight on the uncertainty reduction process. The next limitation centers on the avatars used in this study. Although uncertainty levels were in the hypothesized patterns for the two avatar conditions, such that uncertainty was lower in the high informativeness avatar condition, the differences were 88 not as strong as predicted. One possible reason for the lack of strong differences is that none of the avatars used in this study were avatars that have high potential for uncertainty - reduction. It is possible that a dynamic, 3-dimensional avatar might be able to relay even more information than the static avatars used in the current study. For example, these types of avatars might be able to convey information such as changes in mood (if an avatar changes from smiling to frowning). These avatars may also allow for an increase in nonverbal cues from the allowance of avatar gesturing and physical movements. In general, the use of avatars that can provide even more information than those used in the crurent study may allow for stronger effects in future studies. The final limitation considered here deals with the potential for artifacts resulting from experimental control. Participants were explicitly asked not to visit other websites during the study. Although this constraint helped provide conditions well suited to test the mechanisms of SIPT, it may have barred respondents fiom access to alternative avenues of uncertainty reduction and limited the generalizability of the study’s findings. For example, restricting access to non-experimental websites prevented participants from utilizing the type of extractive strategies commonly found in CMC interactions (Ramirez, Walther, Burgoon, & Sunnafrank, 2002). When trying to learn about strangers, information gleaned from social networking sites has the potential to reduce uncertainty, and is exactly the kind of information people would likely seek (Berger, Gardner, Parks, Schulman, & Miller, 1976). There are also possible warranting Walther & Parks, 2002) opportunities offered by these sites such as information posted by others that may be telling about newly met acquaintances. Such information may have greater uncertainty reduction capacity than things that new acquaintances have to say about themselves. 89 Although preliminary research has looked at some aspects of social networking sites in the impression formation process (Walther, Kim, Van Der Heide, Westerman, Tong & Langwell, in press) unanswered questions remain concerning how these sites are utilized and their function in this process. Directions for Future Research The findings and limitations of this study highlight several important avenues for examining technology's influence on traditional communication processes, and how people adapt to the lack of nonverbal cues in computer-mediated communication. These avenues focus on addressing three major questions: (1) What information matters?, (2) Does CMC lack nonverbal cues?, and (3) Are all limitations of CMC overcome? What information matters? Questions surrounding the mechanisms people used in this study to circumvent the channel restrictions of CMC highlight the point that not all information is created equal. Some information might be better suited for reducing uncertainty than others. Information theory (Shannon & Weaver, 1949) suggests that uncertainty exists when there are multiple possible outcomes in a given situation, and the uncertainty reduction potential of a piece of information will be determined by how many possible outcomes it eliminates in a person’s mind. Thus, a piece if information that removes two possible outcomes is more informative then a piece that removes only one possible outcome. This reasoning suggests that many things might influence the informativeness of a piece of information. For example, information that activates stereotypes might be particularly useful in reducing uncertainty. The trust (warranting potential) one places in the source of the information might heavily influence that piece of inforrnation’s uncertainty reducing potential. These potential influences on uncertainty 9O reduction potential, coupled with the previously noted potential for questions and self- disclosures to vary in their uncertainty-reduction capacity, highlight the fact that some pieces of information will be better than others at reducing uncertainty. The offshoot of this reasoning is that much may be gained by research designed to more accurately identify the uncertainty reduction potential of utterances in FtF and CMC interactions. The section that follows combines information theory logic with a discussion of several attributes thought useful in identifying an utterance’s uncertainty reduction potential. This is used to suggest an improved method for coding interactions with a standardized measure of an utterance’s uncertainty reduction potential. Based on this discussion, a promising approach for future research is presented that describes a procedure for recoding the interactions in the current study, and an analysis plan that combines these new codes with the existing data on observed uncertainty reduction. The potential for IU RS to vary in their uncertainty-reduction capacity draws attention to the fact that various features of interaction can influence the perception of others in CMC (or even FtF communication). For example, although the present study found evidence consistent with predictions that an avatar with high information value would reduce uncertainty more than an avatar with low information value, the highly informative avatar had no more influence on uncertainty reduction than having no avatar at all. If the mere presence of nonverbal information availed by the inclusion of an avatar was the only thing affecting uncertainty reduction, this outcome would make little sense. But a better understanding is possible here by considering other forces at work in these interactions. An important matter in this regard is attention to how information varies in its uncertainty reduction capacity. Generally speaking, information is thought to reduce 91 uncertainty (Berger & Calabrese, 1975) although some information can increase uncertainty (Planalp & Honeycutt, 1985). One potentially useful approach to understanding the strength and capacity of information to reduce (or even increase) uncertainty is information theory (Shannon & Weaver, 1949). Uncertainty reduction theory (Berger & Calabrese, 1975) in its original formulation was heavily influenced by information theory (Shannon & Weaver, 1949). Under information theory tenets, uncertainty arises when the occurrence of multiple possible events are equally likely (the more possible events, the greater the uncertainty). This is implicit in URT’s definition of uncertainty as the “number of alternative ways in which each interactant might behave” (Berger & Calabrese, 1975, p. 100). In information theory, information is something that removes possible events from likelihood. Thus, by definition, information is a reduction of uncertainty (implicit in URT predictions that verbal and nonverbal communication will reduce uncertainty). Thinking about it in these terms, a one that removes a greater number of possible outcomes from likelihood, and thus reduces more uncertainty, would be more informative than one that removes fewer possible outcomes. Although it would not be defined as information by Shannon and Weaver (1949), a cue that increases the number of possible outcomes would be likely to increase uncertainty. Planalp and Honeycutt's (1985) list of uncertainty increasing events is consistent with this position. Based on this logic it can be reasoned that cues that eliminate the greatest number of possible outcomes will be more informative. For example, asking the question “Are you female?” and receiving the answer “Yes” removes the possibility that the other person is a male (assuming the information is trusted). However, a picture used to answer 92 the question “Are you female” might remove many more possibilities. For example, not only could receivers see if the person in the picture is male or female, but they might also determine whether or not the person seems attractive, fashionable, pleasant, etc. It is also possible that the information responsible for removing possible outcomes is not actually contained in the verbal or nonverbal components of a cue, but comes from the receiver’s perception. In some situations, the influence of schema activated by a cue may fill in blanks and reduce more uncertainty than the cue itself provided. This might explain what happens in the hyperpersonal model (Walther, 1996). For example, when we want to reduce uncertainty about a person, but have insufficient information available to remove possible outcomes, we have to fill in the gaps somehow. If we choose to do this by filling in gaps to imagine people the way we want them to be, we are more likely to think more positively of them, as the hyperpersonal model proposes. If cues that eliminate more possible outcomes have greater informative value, it makes sense to identify features of interaction that increase the informativeness of an utterance or other cue in FtF and CMC. Though many attributes are likely to influence a piece of inforrnation’s informative value, two potentially strong forces made salient by this study’s focus on avatars in online impression formation are stereotype influence and trust. Stereotypes and trust are central concerns when considering inforrnation’s ability to influence uncertainty. Whereas information that conforms to expectations established by an interaction context is likely to help reduce uncertainty, information that violates expectations is just as likely to inflate it. In most situations stereotyped avatars may be particularly useful in helping to reduce uncertainty. Stereotypes are commonly defined as “cognitive structures 93 that contain our knowledge, beliefs, and expectations about a social group” (Kunda, 2001 , p. 315). One reason stereotypes are utilized is to make impression formation easier (Macrae, Milne, & Bodenhausen, 1994). In this sense, stereotyping can be seen as a form of uncertainty reduction heuristic, especially because stereotypes guide expectations about group members and interpretations of members’ behaviors and traits (Kunda & Thagard, 1996). Thus, stereotypes can be used as a tool to reduce uncertainty about a person quickly, and can create expectations about what that a person will be like. If people respond to avatars as they do actual people (Reeves & Nass, 1996), and their response to people is governed by the type of common stereotypes as mentioned above, then the same manner of stereotype influence should shape the way people respond to avatars. Some evidence for this comes from Koda (2004), who found that people rated online others using an avatar with glasses as more intelligent than those using avatars without glasses. Extension of this reasoning leads to predicted patterns of uncertainty reduction consistent with the receiver’s ability to perceive stereotyped attributes present in online avatars. Just as in F tF interaction, when this information conforms to expectations, uncertainty is reduced, whereas information that violates expectations should lead to greater doubt. Indeed, the likelihood of this influence may be strong given that the lack of visual and vocal cues in CMC should allow cues available in avatars to activate automatic judgments (i.e., stereotypes) about other people (Patterson, 1998). Future research should examine more closely how enacting stereotypic nonverbal information influences the uncertainty reduction process in CMC. Though an avatar in CMC may have great potential to activate stereotypes useful in reducing uncertainty, the ability of this feature to help reduce uncertainty is strongly 94 determined by the extent to which it is consistent with expectations set by context. When information received is inconsistent with expectations, uncertainty may be increased. Photographs or comments by others, for instance, might override avatars or textual self- disclosures in terms of perceived veracity. The warranting potential (Walther & Parks, 2002) of information, or the likelihood that a piece of information is true, may cause some pieces of information to be more heavily utilized than others when forming an impression of new online acquaintances. The interplay between text and visual imagery calls attention to other forces that might govern uncertainty reduction capacity. When text and imagery seem incompatible, how is uncertainty affected? For example, if a person says on their facebook page that they do not drink, but the pictures posted all show people smiling and holding drinks, this could increase levels of uncertainty, as it would be unclear which piece of information presents the person's actual feeling toward alcohol consmnption. Several features of information and the circumstances within which information is received should influence the uncertainty reduction weights given to some pieces of information, and even seem capable of altering infonnation’s influence fi'om reducing to increasing uncertainty. Yet such factors that should make some pieces of information better than others at reducing uncertainty have not been examined in research on online impression formation. One strong prospect for future research suggested by this line of thought is a study that would recode the individual utterances in the present study for uncertainty reducing potential. The study could be conducted in the following manner. First, a list consisting of 20 questions/disclosures that are likely to reduce, or not reduce, uncertainty would be created. A sample of undergraduate students would be asked to list 95 10 things they would want to know if they were meeting a stranger for the first time. From these, the 20 things most frequently listed would be used to comprise a master list. A separate sample from the same population would be asked to assign a score to each of these 20 items rating each item for its uncertainty reducing potential. Respondent ratings would be combined to create a standardized score representing the uncertainty reduction potential for each item. The interactions from the current study would then be analyzed in a new study to locate these 20 items and record the minute during which each appeared in conversation. The uncertainty reduction potential scores for each of the 20 items occurring in any one minute would be added together to calculate minute-by-minute uncertainty reduction scores for each subject. Thus, if one minute contained a statement coded as a 10 and another minute contains 3 statements each coded as a 2, the first minute would have a score of 10 on uncertainty reduction potential and the second minute would have a score of 6. These scores would then be correlated with the minute- by-minute single-item measures of actual uncertainty reduction. The study would hypothesize a strong positive correlation between the minute-by-minute scores of subjects on uncertainty reduction potential and actual uncertainty reduction. Does CMC lack nonverbal information? Many research questions center on the assumption that CMC lacks nonverbal cues. This assumption is the foundation for much of the theory and research on CMC, including the four major areas of CMC impression formation theory: SIPT (Walther, 1992), the hyperpersonal model (Walther, 1996), the SIDE model (Spears & Lea, 1992) and cues filtered out (Culnan & Markus, 1987) approaches. Yet clearly CMC is not completely devoid of nonverbal cues as some CMC approaches seem to imply. Though the concentration of early research on the lack of 96 nonverbal cues focused attention on an important distinction of CMC, greater attention to the role of those nonverbal cues that are contained in CMC has started to develop and will undoubtedly intensify as technology increases the availability of these cues in online interaction. Closer examination of the CMC landscape shows that many of the nonverbal cue systems offered by Burgoon and Hoobler (2002) have the potential to exist in CMC. Walther and Tidwell (1995) point to chronemics as one nonverbal cue system that functions in CMC. People make different judgments about people and aspects of their relationship based on the time that messages are sent. Moreover, many other nonverbal cue systems have the potential to operate in CMC. For example, Walther, Slovacek, and Tidwell (2001) found that people form judgments of an online acquaintance based in part on photographs they see of that person. This suggests that people do in fact rely on physical appearance cues when making judgments in CMC. The manipulation of artifacts is also common in CMC. For example, people add objects and items to their myspace and facebook pages in an attempt to say something about themselves, perhaps making themselves appear more interesting to potential online observers. Research on the reaction to people's homepages (Marcus, Machilek, & Schlitz, 2006) has found that people do make different judgments of others based upon these nonverbal visual cues. Massively-multiplayer online role playing games (MMORPG's), such as “Second Life” and “World of Warcraft” also provide a platform for the use of nonverbal cues online. Williams, Caplan and Xiong (2007) examined the effects that inclusion of voice with text had on players of “World of Warcraft” and found that voice capabilities increased players’ trust and liking of each other. Yee, Bailenson, Urbanek, Chang, and Merget (2007) outline a series of nonverbal social norms in Second Life that mirror 97 nonverbal social norms in FtF interactions. These include several surprising influences such an influence on interaction that results from proxemic cues between avatars. Future systems with more advanced avatars and hardware will undoubtedly be developed to allow for greater inclusion of proxerrric and haptic cues. For example, Rheingold (1991) coined the phrase “teledildonics” to refer to systems designed to simulate physical sex though the inclusion of haptic cues. We might reason that CMC does not lack nonverbal cues to the extent suggested by some of the theories upon which much CMC research is based. This should come as no surprise when considering that the capacities of technology often improve faster than theory in any given field of study (O' Sullivan, 11. d.). As recently as 2002, Walther and Parks noted that most online interaction took place in text-only systems. However, even if this is still the case, it may not be true for long. It would seem that nonverbal cues have crept into CMC, and although they may be somewhat limited today, their availability will surely increase. Already their presence challenges the basis of many of the theories in this area, and these challenges will continue to grow. The dearth of research on the use of nonverbal cues in CMC would be greatly augmented by increased attention to the role nonverbal cues play in CMC. It may be that level of cue availability (as opposed to the absence of nonverbal cues) simply operates as a continuum from no cues, to limited cues, to maximal cues in FtF when considering processes such as impression formation. Notably, the reliance on FtF as a gold standard has been criticized in the past (Hollan & Stometta, 1992), and it may be that the presence of limited cues may actually advance impression formation as well as the attainment of other goals by reducing distracting information. 98 Although many goals can be accomplished in CMC, and some may even be better accomplished, SIPT seems to suggest that all goals can be accomplished equally well in CMC as FtF if given enough time. This leads to the next major question to be asked: Are all limitations of CMC overcome? SIPT focuses attention on how people overcome limitations of a channel to accomplish their goals. However, there may be situations when people do not wish to overcome the limitations of a channel, and instead attempt to utilize a channel's limitations to accomplish their goals. For example, the hyperpersonal model (Walther, 1996) explains how a receiver may use the reduced cues of CMC in order to idealize another person, and thus accomplish their goal of finding someone they deem worthy of having interactions and a relationship with. The hyperpersonal model also discusses the use of limited cues to selectively self-present oneself more effectively, thus utilizing the limitations of the channel to accomplish the goal of putting one's best foot forward. Other scattered evidence also discussed the selective utilization of channels fiom goal-oriented perspectives. For example, O'Sullivan (2000) examined how people utilize the limitations of different channels to manage both their impressions and those of their partners. Kayany, Wotring and Forrest (1996) found that people say they use different channels based on the extent to which a channel allows for control over aspects of an interaction that are thought to be contradictory. Thus, although people may have the capacity to overcome the limitations of channels, as SIPT suggests (Walther, 1992), people do not always try to overcome the limitations of a channel. Instead they use these channel limitations strategically to their own advantage. Discussion of utilizing channel limitations in order to maximize goal attainment also highlights the necessity of explicating the different boundary conditions for the 99 applicability of SIPT and other CMC impression formation theories. What moderators help to identify the circumstances under which different theories are applicable? One possible approach to answering this question is to consider the communication goals pursued in each theoretical context and their relationships to channel limitations. That CMC can be used to accomplish different goals has been demonstrated. However, it is important to consider the conditions under which people have certain goals in mind. For example, if one's goal can only be accomplished by getting to know someone on a more intimate interpersonal level, then SIPT processes will likely explain and predict communication patterns. Under these circumstances, people will likely be more motivated to overcome the limitations a medium presents, thus doing things such as asking more questions and self-disclosing more in an effort to reach deeper levels of intimacy. However, if a person's goal can be accomplished without increasing intimacy and by interacting at a superficial level with others in that context commensurate to simple roles that they fill, then SIDE processes may be more likely to apply. For example, if you go to McDonald's and your only goal is to get food, you can successfully accomplish this goal by responding to the person behind the counter solely as a food service provider. Goals of this ilk seem especially likely in the context of initial interactions where the likelihood and desire for future interaction is nil. Alternatively if my goal is to be together with someone, and not necessarily to really get to know that person, then hyperpersonal may be the most applicable theory (indeed, desiring a relationship is predicted to increase hyperpersonal probabilities). Determining when different theories may apply to different situations would help improve our understanding of CMC interaction. 100 Conclusion The current study was designed to examine SIPT (Walther, 1992) by examining the theory’s two major assumptions: that people can accomplish goals in CMC, and that they do so by finding ways to circumvent channel limitations. An examination of the findings related to these two assumptions along with some of the limitations linked to these findings highlight several important avenues for examining big picture issues in the field of CMC. These avenues focus on addressing three major questions concerning what information matters in online impression formation, how much CMC suffers from the lack nonverbal cues, and whether or not all limitations of CMC are overcome. In order to examine the assumption that, if given enough time, people can accomplish goals in CMC as they would in FtF, patterns of uncertainty reduction were observed over time for CMC and FtF interactions. Data were consistent with the assumption of SIPT that goals can be accomplished in both channels. Although levels of uncertainty were never as low in CMC as FtF, differences at the end of interaction were smaller than those at the start, suggesting that not only is uncertainty reduction occurring, but given enough time, this goal may be accomplished to similar levels in CMC and FtF. However, data were not consistent with the hypothesized patterns of uncertainty reduction in CMC. Instead of following a more linear pattern as expected, uncertainty followed a similar pattern over time in CMC as in F tF. In order to examine the assumption that people find ways to circumvent a channel’s limitations and accomplish their goals, the current study observed both the use of IURS over time and the effects of including nonverbal information in the form of avatars in some conditions. With regard to the first set of observations, the patterns of 101 1U RS use over time followed the generally predicted patterns. Due to channel restraints, the number of IURS uses was lower in CMC than F tF for each one-minute segment of the interaction, although IURS as a percentage of total utterances was always higher in CMC than FtF. The patterns showed that total number of IURS used started out at its highest point in the first minute of interaction in both F tF and CMC. The number of IURS declined in each condition over time. The decline was very steep in the first three minutes for FtF, and then the IURS leveled off with a slight downward trend; however, in CMC, the decline was more gradual over the course of the 15 minute interaction. With regard to the second set of observations, circumvention associated with the inclusion of nonverbal information in the form of avatars was examined by varying the information available in avatars included in some CMC conditions. Avatars were selected based on their expected informativeness about the person they represented. They were expected to have the greatest effect at the start of an interaction, because they would provide more cues to use for impression formation when compared to a CMC condition lacking an avatar. Although strong differences did not exist between the two avatars in terms of their observed effect on uncertainty levels, comparison of pre-conversation means suggested that more uncertainty existed for the low-infonnativeness avatar than the high-informativeness avatar. Notably, however, neither avatar condition had lower uncertainty at this initial point than the CMC-no avatar condition (although the high informativeness avatar had the same level as the CMC-no avatar condition), so the effects of these avatars are unclear. Some of the limitations linked to the findings in this study suggest possible avenues for future exploration. First, a college-student sample comprised entirely of 102 male-female dyads was used in the current research. Using samples that are less familiar with [M and are comprised of same-sex dyads may reveal different patterns of findings. Second, potential problems related to using a single-item scale to measure uncertainty over time suggest the need for a design providing a more comprehensive measure of uncertainty while still allowing comparisons of uncertainty after different lengths of interaction. Third, refinements to the definition of IURS may allow for better predictions of uncertainty reduction based upon the types of questions and self-disclosures examined. Focusing attention on the quality of IURS instead of simply using quantity highlights the potential for some questions and selfedisclosures to reduce uncertainty more than others, and future research should attempt to determine the uncertainty reducing potential of different type of questions and disclosures. Finally, experimental artifacts centering on restricting participants’ use of other websites during the study is another limitation of the study that future research should address. Specifically, the usefulness of social networking sites like facebook.com should be examined as they potentially provide the type of information that has been identified as useful for uncertainty reduction (Berger et al., 1 976). The data in this study also suggest three key questions in the field of CMC in general that should be addressed. First, consideration of limitations of the definition of IURS in this study combined with the lack of strong effects found for the avatars used suggest that more work is necessary to identify what information matters most in forming impressions of others in online (and FtF) interactions. Second, central to the major theories of online impression formation is the extent to which CMC lacks nonverbal cues. Although extensive differences in the availability of nonverbal cues may have been true 103 in early eras of CMC, it does not seem to represent the reality of CMC today, or what capabilities future CMC systems are likely to feature. Thus, examination of how nonverbal cues operate in CMC, and what effects they have on interaction and impression formation in CMC, would provide important knowledge for this field. Finally, SIPT (Walther, 1992) suggests that people overcome limitations of a mediated environment to accomplish their interpersonal goals. It does not address cases in which people do not want to overcome limitations, and instead prefer to use and profit from a channel’s limitations. Addressing these questions in future research would serve as a great step forward in sharpening our understanding of how people interact in mediated environments. 104 APPENDIX A Experimental Scripts F tF. Bring first participant into one room in the lab. Have them sit and wait until other participant comes. When second participant comes, sit them in the other room. When both participants have arrived, give them consent forms. GIVE CONSENT FORMS At five past the hour, if other participant has not shown up, administer alternate study. If second person comes after this, allow then the chance to do alternate as well. After participant has filled out consent form, tell them the following: “You will be taking part in a study on friendship formation. In this study, you will take part in two sessions where you and another person will spend 15 minutes getting to know each other. In the first session you will interact with and get to know another volunteer participant in the study. In the second session you will interact with another volunteer participant in the study. After both interactions, you will choose which person you would prefer to hang outwith. Before we start this interaction, we would like to know what kind of mood you are in. Please respond to the following questionnaire about your feelings at this time” GIVE PARTICIPANTS MOOD MEASURE Once participant has finished this mood measure, collect it. Then bring the participants together in one room and say: “Please note that these interactions will be videotaped for later analysis, so please do not say anything you do not want people to see. However, remember that your 105 interactions will be kept confidential. If at anytime you feel uncomfortable during this interaction, please report this to the experimenter.” We will tell the participants to begin interaction and leave the room. We will let them interact for fifteen minutes. When the time is up, we go back into the room and say the following: “Ok. Your fifteen minutes is up.” Bring the participants back into their original rooms and say the following: “Now that this interaction is complete, we would like you to respond to it. Please fill out this questionnaire at this time.” GIVE PARTICIPANTS UNCERTAINTY PACKET When participant has finished this packet, collect it, and say the following: “Thank you. Now we would like to know your thoughts about this interaction. On the following thought listing questionnaire, please read the directions. We are particularly interested in what information you used to answer this past questionairre. Any other thoughts about this interaction that you would like to share would also be valued.” GIVE PARTICIPANTS THOUGHT LISTING While participants are responding to the thought listing Dave will set out playback of the interaction in each room. “Now we would like to ask more questions about the interaction you just had. We are going to ask you to think about and respond to how you were feeling as the interaction went on. We are going to show you the interaction you just had minute by minute. Then respond to the appropriate questions on the sheet, think about 106 how you were feeling at that part of the interaction. Please do this for each minute mark. Please feel free to ask questions if you need to. GIVE PARTICPANTS PROTOCOL PACKET Answer any questions participants may have. When they have finished responding to this, you may begin the debriefing procedures. Start by asking the following question: “Now that you have finished that part, we have a few more questions to ask you. First, had you ever met the person you interacted with before this interaction?” Note their response. Then say the following: “Now we would like you to respond to the following questions.” GIVE PARTICIPANT DEBRIEFING QUESTIONNAIRE Once this is finished, please say the following: “I am now giving you a sheet that tells you more about this study, but I will also tell you more verbally. This study is designed to examine how people interact with strangers. There is no second person to interact with. You were told this to give you a reason and a flame for your interaction. We would like to apologize if anything about this study upset you. Please remember that all of your responses will be kept confidential, but if for any reason you do not want us to use your date, please indicate that now, and it will be destroyed. Wait for participant's response, and then say: “You may keep that debriefing sheet if you wish. If not, please leave it so we can recycle it. Do you have any questions?” Answer them, and then thank them for coming in. 107 IM without avatars. Bring first participant to one room (Oyer 210). Have them sit and wait until other participant comes. When second participant comes, sit them in the other room (Oyer 210). When both participants have arrived, give them consent forms. GIVE CONSENT FORMS At five past the hour, if other participant has not shown up, administer alternate study. If second person comes after this, allow then the chance to do alternate as well. After participant has filled out consent form, tell them the following: “You will be taking part in a study on friendship formation. In this study, you will take part in two sessions where you and another person will spend 15 minutes getting to know each other. In the first session you will interact with and get to know another volunteer participant in the study. In the second session you will interact with another volunteer participant in the study. After both interactions, you will choose which person you would prefer to hang out with.” Once this has been done, tell participants that we would like them to fill out a simple measure before we start. Say: “Before we start this interaction, we would like to know what kind of mood you are in. Please respond to the following questionnaire about your feelings at this time” GIVE PARTICIPANTS MOOD MEASURE Once participant has finished this mood measure, collect it and radio the other experimenter. Then tell the participant: “Please note that these interactions will be printed for later analysis, so please do not say anything you do not want people to read. However, remember that your interactions will be kept confidential. If at anytime you feel uncomfortable during 108 this interaction, please report this to the experimenter. We also ask you to refrain from adjusting the IM system, and please do not go on other websites during this time.” When both are done, we will tell the participants to begin interaction and leave their rooms. We will let them interact for fifteen minutes. When the time is up, we go back into the rooms and say the following: “Ok. Your fifteen minutes is up. Please stop typing. Now that this interaction is complete, we would like you to respond to it. Please fill out this questionnaire at this time.” GIVE PARTICIPANTS UNCERTAINTY PACKET When participant has finished this packet, collect it, and say the following: “Thank you. Now we would like to know your thoughts about this interaction. On the following thought listing questionnaire, please read the directions. We are particularly interested in what information you used to answer this past questionairre. Any other thoughts about this interaction that you would like to share would also be valued.” GIVE PARTICIPANTS THOUGHT LISTING While participants are responding to the thought listing Dave (or Brandon) will print out two copies of the interaction. Each participant will get one. Each minute break will be noted. Tell the participants the following: “Now we would like to ask more questions about the interaction you just had. We are going to ask you to think about and respond to how you were feeling as the interaction went on. We are giving you a copy of the interaction to aid in this task. 109 Please read the interaction up to the point that says 1 minute. Then respond to the appropriate questions on the sheet, think about how you were feeling at that part of the interaction. Please do this for each minute mark. Please feel free to ask questions if you need to.” GIVE PARTICPANTS PROTOCOL PACKET Answer any questions participants may have. When they have finished responding to this, you may begin the debriefing procedures. Start by asking the following question: “Now that you have finished that part, we have a few more questions to ask you. First, had you ever met the person you interacted with before this interaction?” Note their response. Then say the following: “Now we would like you to respond to the following questions.” GIVE PARTICIPANT DEBRIEFING QUESTIONNAIRE Once this is finished, please say the following: GIVE PARTICIPANT DEBRIEFING SHEET “1 am now giving you a sheet that tells you more about this study, but I will also tell you more verbally. This study is designed to examine how people intemet with strangers. There is no second person to interact with. You were told this to give you a reason and a frame for your interaction. We would like to apologize if anything about this study upset you. Please remember that all of your responses will be kept confidential, but if for any reason you do not want us to use your date, please indicate that now, and it will be destroyed. Wait for participant's response, and then say: 110 “You may keep that debriefing sheet if you wish. If not, please leave it so we can recycle it. Do you have any questions?” Answer them, and then thank them for coming in. 111 [M with avatar. Bring first participant to one room (Oyer 210). Have them sit and wait until other participant comes. When second participant comes, sit them in the other room (Oyer 210). When both participants have arrived, give them consent forms. GIVE CONSENT FORMS At five past the hour, if other participant has not shown up, administer alternate study. If second person comes after this, allow then the chance to do alternate as well. After participant has filled out consent form, tell them the following: “You will be taking part in a study on fiiendship formation. In this study, you will take part in two sessions where you and another person will spend 15 minutes getting to know each other. In the first session you will interact with and get to know another volunteer participant in the study. In the second session you will interact with another volunteer participant in the study. After both interactions, you will choose which person you would prefer to hang outwith.” Next, tell participants the following: “As part of this interaction, we have been asking people to choose an avatar to use. Unfortunately, the capturing software on this computer seems to be malfunctioning today....so I went ahead an assigned you the default avatar (High informativeness avatar condition) ..... so the avatar you see will be the one your partner has chosen.” (Low informativeness avatar condition) Once this has been done, tell participants that we would like them to fill out a simple measure before we start. Say: 112 “Before we start this interaction, we would like to know what kind of mood you are in. Please respond to the following questionnaire about your feelings at this time” GIVE PARTICIPANTS MOOD MEASURE Once participant has finished this mood measure, collect it and radio the other experimenter. Then tell the participant: “Please note that these interactions will be printed for later analysis, so please do not say anything you do not want people to read. However, remember that your interactions will be kept confidential. If at anytime you feel uncomfortable during this interaction, please report this to the experimenter. We also ask you to refrain from adjusting the IM system, and please do not go on other websites during this time.” When both are done, we will tell the participants to begin interaction and leave their rooms. We will let them interact for fifteen minutes. When the time is up, we go back into the rooms and say the following: “Ok. Your fifteen minutes is up. Please stop typing. Now that this interaction is complete, we would like you to respond to it. Please fill out this questionnaire at this time.” GIVE PARTICIPANTS UNCERTAINTY PACKET When participant has finished this packet, collect it, and say the following: “Thank you. Now we would like to know your thoughts about this interaction. On the following thought listing questionnaire, please read the directions. We are particularly interested in what information you used to answer this past 113 questionnaire. Any other thoughts about this interaction that you would like to share would also be valued.” GIVE PARTICIPANTS THOUGHT LISTING While participants are responding to the thought listing Dave (or Brandon) will print out two copies of the interaction. Each participant will get one. Each minute break will be noted. Tell the participants the following: “Now we would like to ask more questions about the interaction you just had. We are going to ask you to think about and respond to how you were feeling as the interaction went on. We are giving you a copy of the interaction to aid in this task. Please read the interaction up to the point that says 1 minute. Then respond to the appropriate questions on the sheet, think about how you were feeling at that part of the interaction. Please do this for each minute mark. Please feel free to ask questions if you need to.” GIVE PARTICPANTS PROTOCOL PACKET Answer any questions participants may have. When they have finished responding to this, you may begin the debriefing procedures. Start by asking the following question: “Now that you have finished that part, we have a few more questions to ask you. First, had you ever met the person you interacted with before this interaction?” Note their response. Then say the following: “Now we would like you to respond to the following questions. In the section that says list here for the second item, please write in and respond to the capturing software malfunction.” 114 NOTE: When delivering the second sentence, please try to sound as if you are not reading that from the sheet, so as to avoid sounding like that was planned. GIVE PARTICIPANT DEBRIEFING QUESTIONNAIRE Once this is finished, please say the following: GIVE PARTICIPANT DEBRIEFING SHEET “1 am now giving you a sheet that tells you more about this study, but I will also tell you more verbally. This study is designed to examine how people interact with strangers. There is no second person to interact with. You were told this to give you a reason and a frame for your interaction. We would like to apologize if anything about this study upset you. Please remember that all of your responses will be kept confidential, but if for any reason you do not want us to use your date, please indicate that now, and it will be destroyed. Wait for participant's response, and then say: “You may keep that debriefing sheet if you wish. If not, please leave it so we can recycle it. Do you have any questions?” Answer them, and then thank them for coming in. 115 APPENDIX B Consent F orm Consent Form Communication with Partners The present research study is being conducted by faculty in the Department of Communication at Michigan State University. In this study we are examining how people interact over different modes of communication. You will interact with another person for fifteen rrrinutes, and then be asked several things about this interaction. If you choose to participate, the study will take about 60 minutes and you will receive one hour of course credit for your participation. All of your responses will be confidential. The data will be stored on a computerized disk, and along with your paper responses and interactions, will be kept in a locked office for the next two years. Matter of fact, your privacy will be protected to the maximum extent allowable by law. If you agree to participate in the study but then change your mind, you are free to stop at any time. You may also skip any part of the study you do not wish to complete. Participation in this study is voluntary, and your refusal to participate or decision to discontinue your participation at any time will involve no penalty or loss of benefits to you. The foreseeable risks of this study are minimal. You will be interacting with someone else, so the interaction will be unscripted. If at any time you feel uncomfortable with the interaction, let the researcher know, and the interaction will be stopped. There are no foreseeable benefits to you. If you have any questions, contact the investigators: David Westerman, 556 Communication Arts and Sciences, (517) 432-1286, westerm4@msu.edu, or Dr. Ron Tamborini, 570 Communication Arts and Sciences, (517) 355-0178, tamborin@msu.edu. If you have any questions or concerns regarding your rights as a study participant or are at any time dissatisfied with any aspect of this study, please feel free to contact, anonymously if you wish, Peter Vasilenko, Ph.D., Director of the Human Subject Protection Programs at Michigan State University by phone: (517) 355-2180, fax: (517) 432-4503, e-mail: irb@msu.edu, or regular mail: 202 Olds Hall, East Lansing, M148824. Your signature below indicates your voluntary agreement to participate in this study. Class Title (i.e., COM 200) Instructor and TA 116 Printed Name Signature Date 117 APPENDIX C Mood Measure (Oliver, 1 993) Please respond to the next set of words based upon how you feel at this moment. Sad Not at All 2 3 4 5 Very Much Glad Not atAll 2 3 4 5 Very Much Angry Not atAll 2 3 4 5 Very Much HaPPY Not at All 2 3 4 5 Very Much Moumful Not at All 2 3 4 5 Very Much Positive Not atAll 2 3 4 5 Very Much Upset Not at All 2 3 4 5 Very Much Melancholy Not at All 2 3 4 5 Very Much Joyful Not at All 2 3 4 5 Very Much Sorrowful Not atAll 2 3 4 5 Very Much Cheerful Not at All 2 3 4 5 Very Much Blue Not at All 2 3 4 5 Very Much Disturbed Not at All 2 3 4 5 Very Much 118 Negative Not atAII 1 2 3 4 5 6 7 Very Much Unhappy Not atAll 1 2 3 4 5 6 7 Very Much 119 APPENDIX D Uncertainty Measure (Parks & F loyd, 1996) l-strongly disagree 2-disagree 3-neither agree nor disagree 4—agree 5-strongly agree This group of questions asks about your confidence in your partner. Please use the scale above to respond. 1. I am very uncertain about what this person is really like. 2. I can accurately predict how this person will respond to me in most situations. 3. I can usually tell what this person is feeling inside. 4. I can accurately predict what this person’s attitudes are. 5. I do not know this person very well. 120 APPENDIX E Liking Measure (McCroskey & McCain, 1974) The next group of questions asks about some perceptions of your partner. Please use the scale above to respond. 1. I think he (she) could be a fiiend of mine. 2. It would be difficult to meet and talk with him (her). 3. He (she) just wouldn’t fit into my circle of fiiends. 4. We could never establish a personal friendship together. 5. I would like to have a fiiendly chat with him (her). 6. I would like to talk to him (her) again. 121 APPENDIX F Predicted Outcome Value Measure (Sunnafiank, 1988) 1 = much less positive 2 = less positive 3 = slightly less positive 4 = slightly more positive 5 = more positive 6 = much more positive This set of items asks you to make judgments about future interaction with the person you interacted with. Make estimates relative to your normal expectations of relationships continuing from initial meeting. Please use the above scale to respond to the following items. 1. How positive do you think future relationships with this person would be? 2. How do you think a future relationship with this person would be? 3. Based upon your partner’s likely future behaviors toward you, how positive do you think a future relationship would be? 4. Based upon your partner’s likely communicative responses to you, how positive do you think a future relationship would be? 5. Based upon your feelings about your partner, how positive do you think a future relationship would be? 6. Based upon your partner’s likes and dislikes, how positive do you think a future relationship would be? 7. Based upon your partner’s interests, how positive do you think a future relationship would be? 8. Based upon your partner’s attitudes/values, how positive do you think a future relationship would be? 9. Based upon your partner’s general behavior patterns, how positive do you think a future relationship would be? 10. Based upon likely conversations that would occur in the relationship, how positive do you think a future relationship would be? 122 APPENDIX G Thought Listing Procedure (Cacioppo & Petty, 1981) THOUGHT LISTING QUESTIONNAIRE We are interested in what you were thinking about during the interaction. More specifically, we would like to know what information you considered while making the judgments you just did. We would also like to know what was your mind while interacting with the other person. All of this is important to us. The next page contains the form we have prepared for you to use to record your thoughts and ideas. Simply write down the first idea you had in the first box. Please state your thoughts and ideas as briefly as possible... a phrase is sufficient. IGNORE SPELLING GRAMMfi AND PUNCTUATION. You will have 4 minutes to write your thoughts. We have deliberately provided more space than we think most people will need to insure that everyone would have plenty of room to write their ideas. So don’t worry if you don’t fill every space. Just write down whatever your thoughts were during and after the interaction. Please be completely honest and list all of the thoughts that you had. 123 List your thoughts in the spaces provided below. Use a different line for each thought. Ignore spelling, grammar and punctuation. Use the back of this form if you run out of room. 124 APPENDIX H Protocol Analysis Measure At the initial point of interaction (prior to any talk) New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 125 After 1 minute New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 2 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 3 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 126 After 4 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think firture interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 5 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 6 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 127 After 7 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 8 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 9 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 128 After 10 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 11 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 12 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 129 After 13 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 14 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think future interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative After 15 minutes New information can both increase and decrease uncertainty about another person. Keeping that in mind, how certain were you in your impression of your partner at this point? Very uncertain Uncertain Neither certain Certain Very certain nor uncertain How positive did you think firture interaction with your partner would be? Very negative Negative Neither positive Positive Very positive nor negative 130 APPENDIX I Study Assessment Measure 1. Were you bothered by anything about this study in general? NotatAll 1 2 3 4 5 6 7VeryMuch 2. Were you bothered by anything specific about this study (list here ) NotatAll l 2 3 4 5 6 7 VeryMuch 131 APPENDIX J Debriefing Sheet Debriefing Sheet This study was designed to investigate how people communicate in initial interactions with strangers. More specifically, it is an attempt to identify differences between different communication channels. In this study, you were asked to interact with another person either face-to-face, over a text-only computer system, or over one of two text systems with an avatar. Past studies have found that strangers interacting over a text-only system used more direct questions and were more self-disclosive than those interacting face-to-face. It is has been found that people rated their partners as more effective in these text-only systems. The current study is an attempt to replicate and extend these findings. Specifically, the addition of a condition with a text with avatar system is an attempt to examine how the information provided by an avatar impacts these uncertainty reduction strategies and subsequent uncertainty levels. The avatars used were designed to either be informative or uninformative about the person with which you interacted. Although it may seem counterintuitive, there is a large and growing body of evidence to support the notion that people treat computers like people and that they judge people differently based solely on avatars. If this is the case, then we should expect that people in the text with informative avatar condition should behave more closely to people who interact face-to- face and that the avatar used also impacts these behaviors. Results of this study should be available in Fall 2006. If you have any questions about this study, please contact the Responsible Project Investigator. Dr. Ron Tamborini at tamborin@msu.edu or 517-355-0178, or David Westerman at westerm4@msu.edu or 517-432-1286. Thanks again for participating in this study. We ask that you do not discuss this study with anyone other than the investigators listed above or UCRIHS. Have a good day. 132 APPENDIX K [M interaction for modified protocol analysis measure pilot study fiintymebabe: hi spartanfan90: how are u? funtymebabe: good...where are you? spartanfan90: a/s/l? spartanfan90: I'm in East Lansing One minute mark funtymebabe: 21/f/east lansing spartanfan90: cool spartanfan90: do you go to school funtymebabe: so u like the spartans? spartanfan90: lol spartanfan90: i love themllllllll spartanfan90: basketball and football are my faves funtymebabe: are you a student spartanfan90: yes Two minute mark funtymebabe: what year? spartanfan90: senior funtymebabe: cool, me too funtymebabe: what is u major? spartanfan90: english funtymebabe: cool spartanfan90: yeah, its ok spartanfan90: a lot of reading funtymebabe: no doubt Three minute mark spartanfan90: you like to read funtymebabe: not really funtymebabe: i can, but i don't like too spartanfan90: lol spartanfan90: what do you like to do funtymebabe: i like to go out with my girls spartanfan90: cool funtymebabe: we had a good time at Rick's last night funtymebabe: you like to go out? spartanfan90: maybe when I get to college Four minute mark funtymebabe: what?????? funtymebabe: you said you were a senior! l!!! spartanfan90: i am 133 spartanfan90: in high school spartanfan90: i want to go to msu funtymebabe: how old are you? spartanfan90: 17 funtymebabe: well, i gotta go....look me up when you get to msu Five minute mark 134 APPENDIX L Coding Instructions Instructions for Coding Utterances and IURS Designate utterances by placing vertical lines around each utterance. Use Holsti’s (1966) definition of an utterance as "a single assertion about some subjec " (p. 116). Utterances typically contain a subject and verb, although the subject could be understood rather than stated. For example, the participant may have said, "[I] know what you mean, exactly." Further, dependent clauses were included with the main clause to which they were subordinate. Additionally, all compound sentences were divided into two or more utterances. For instance, “I could see and hear my partner" should be coded as one utterance, whereas "I was face-to-face with my partner and / I was also able to listen" should be coded as two utterances. Unintelligible or incomplete remarks such as "I think [inaudible or illegible] or merely [inaudible or inaudible] will not count as complete utterances. After all utterances have been identified, they need to be coded into one of three content categories: Question, Self-disclosure, and Other. Questions: Will be defined as "An expression of inquiry that invites or calls for a reply; an interrogative sentence, phrase, or gesture" (Morris, 1976, p. 1070). Self-disclosures: Messages that reveal personal information about the sender. “A verbal response (thought unit) which describes the subject in some way, tells something about the subject, or refers to some affect the subject experiences" (Chelune, 1975, p. 133 cited in Tardy, 1988). Basically, to be coded as a self-disclosure, the utterance has to reveal something about the person or how they feel or think about themselves or others. 135 Other: As it says, anything that is not a question or self-disclosure. Both exclamations and imperatives are among the “other” category. Other expressions include conversation elements such as statements of fact that are nonpersonal in nature, statements about third parties, exclamations, imperatives, preview and summary statements, greetings, backchanneling elements, and other filler items that are not clearly questions or self-disclosures. 136 APPENDIX M Avatars used in main study High informativeness (male) High informativeness(female) Low informativeness (male and female) 137 APPENDIX N Correlation Matrix of Variables UNC POV IURS UNC POV IURS UNC POV (T0) (T0) (T1) (T15) (T15) (T15) COMP COMP UNC (T0) POV n (TO) -.29 IURS *4- (T1) -.36 .16 UNC (T15) .1 l -.O9 -.10 POV , u _ u (T15) .15 .27 .05 .50 IURS _ an: n - (T1 5) .25 .09 .68 .07 .03 UNC it. _ a1- _ u an _ an: ._ a COMP .21 .21 .28 .44 .28 .22 POV _ u _ an: an _ _ u COMP .16 .32 .05 .34 .63 .03 .28 UK _ :1: a1: , u an: _ u u COMP .12 .18 .22 .34 .56 .07 .40 .69 Note. * p < .05, *"‘ p < .01 by two-tailed t-test. 138 APPENDIX 0 Table of Means and Standard Deviations Mean SD UNC 4.13 .92 (T0) POV 3.39 .69 (T0) IURS 4.91 4.62 (T1) UNC 1.96 .69 (T 15) POV 3.98 .75 (T15) IURS 2.18 2.51 (T15) UNC 3.31 .76 COMP POV 4.18 .61 COMP LIK 3.67 .51 COMP 139 References Afifi, W. A., & Burgoon, J. K. (2000). The impact of violations on uncertainty and the consequences for attractiveness. Human Communication Research, 26, 203-233. Argyle, M. (1975). The syntaxes of bodily communication. In J. Benthal & T. Polhemus (Eds.), The boafy as a medium of expression (pp. 143-161). New York: E.P. Dutton & Co., Inc. Bargh, J. A., & McKenna, K. Y. A. (2004). The intemet and social life. Annual Review of Psychology, 55, 573-590. Berger, C. (1979). Beyond initial interaction. In H. Giles & R. St. Clair (Eds.), Language and social psychology (pp. 122-144). Oxford: Basil Blackwell. Berger, C. R. (1986). Uncertain outcome values in predicted relationships: Uncertainty reduction theory then and now. Human Communication Research, 13, 34-38. Berger, C. R., & Calabrese, R. J. (1975). Some explorations in initial interaction and beyond: Toward a developmental theory of interpersonal communication. Human Communication Research, 1, 99-112. Berger, C. R., Gardner, R. R., Parks, M. R., Schulman, L., & Miller, G. R. (1976). Interpersonal epistemology and interpersonal communication. In G. R. Miller (Ed.), Explorations in interpersonal communication (pp. 149-171). Beverly Hills, CA: Sage. Berger, C. R., & Gudykunst, W. B. (1991). Uncertainty and communication. In B. Dervin & M. J. Voight (Eds.), Progress in communication sciences (V 01. 10, pp. 21-66). Norwood, NJ: Ablex. Berger, C. R., & Kellerrnan, K. A. (1983). To ask or not to ask: Is that a question? In R.N. Bostrom (Ed.), Communication yearbook 7 (pp. 342-368). Beverly Hills, CA: Sage. Burgoon, J. K., & Hoobler, G. D. (2002). Nonverbal signals. In M. L. Knapp & J. A. Daly (Eds.), Handbook of interpersonal communication (3"d ed., pp. 240-299). Thousand Oaks, CA: Sage. Cacioppo, J ., & Petty, R. (1981). Social psychological procedures for cognitive response assessment: The thought listing technique. In T. Merluzzi, C. Glass, & M. Genest (Eds.), Cognitive assessment. New York: Guilford. Chan, D. K. S., & Cheng, G. H. L. (2004). A comparison of offline and online friendship qualities at different stages of relationship development. Journal of Social and Personal Relationships, 21, 305-320. 140 Clatterbuck, G. (1979). Attributional confidence and uncertainty in initial interaction. Human Communication Research, 5, 147-157. Culnan, M. J ., & Markus, M. L. (1987). Information technologies. In F. M. Jablin, L. L. Putnam, K. H. Roberts, & L. W. Porter (Eds.), Handbook of organizational communication: An interdisciplinary perspective (pp. 420-443). Newbury Park, CA: Sage. Cummings, J. N., Butler, B., & Kraut, R. (2002). The quality of online social relationships. Communication of the ACM, 45, 103-108. Cummings, J. & Kraut, R. (2002). Domesticating computers and the intemet. The Information Society, 18, 221-232. Darner, B. (1998). Avatarsl: Exploring and building virtual worlds on the internet. Berkley, CA: Peachpit Press. Douglas, W. (1987). Question-asking in sarne- and opposite-sex initial interactions: The effects of anticipated future interaction. Human Communication Research, I 4, 230-245. Dubrovsky, V. J ., Kiesler, S., & Sethna, B. N. (1991). The equalization phenomenon: Status effects in computer-mediated and face-to-face decision making groups. Human-Computer Interaction, 6, 119-146. Ellul, J. (1964). The technological society. New York: Alfred A. Knopf. Ericson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports as data. Cambridge, MA: MIT Press. Fischer, C. (1992). America calling: A social history of the telephone to 1940. Berkeley, CA: University of California Press. Gibbs, J. L., Ellison, N. B., & Heino, R. D. (2006). Self-presentation in online personals: The role of anticipated future interaction, self-disclosure, and perceived success in intemet dating. Communication Research, 33, 153-177. Griffrn, E. (2006). A first look at communication theory. Boston: McGraw-Hill. Grove, T. G., & Werkman, D. L. (1991). Conversations with able-bodied and visibly disabled strangers: An adversarial test of predicted outcome value and uncertainty reduction theories. Human Communication Research, I 7, 507-534. Gudykunst, W. B., & Nishida, T. (1984). Individual and cultural influences on uncertainty reduction. Communication Monographs, 51, 23-3 6. 141 Gudykunst, W. B., Yang, S. M., & Nishida, T. (1985). A cross-cultural test of uncertainty reduction theory: Comparisons of acquaintance, friends, and dating relationships in Japan, Korea and the United States. Human Communication Research, 11, 407- 455. Hamilton, D., & Sherman, J. (1994). Stereotypes. In J. Robert, S. Wyer, & T. K. Srull (Eds.), Handbook of social cognition, 2"“! Edition. (3-58). Hillsdale, NJ: LEA. Hamilton, M. A., & Hunter, J. E. (1988). Confirmatory Factor Analysis [Computer software]. East Lansing, MI: Michigan State University. Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley. Hiltz, S. R., Johnson, K, Turoff, M. (1986). Experiments in group decision making: Communication process and outcome in face-to-face versus computerized conferences. Human Communication Research, 13, 225-252. Hobman, E. V., Bordia, R., Irner, B., & Chang, A. (2002). The expression of conflict in computer-mediated and face-to-face groups. Small Group Research, 33, 43 9-465. Hollan, J ., & Stometta, S. (1992). Beyond being there. In P. Bauersfeld, J. Bennett, & G. Lynch (Eds.), CHI '92 conference proceedings: ACM conference on human factors in computing systems (pp. 119-125). New York: Association for Computing Machinery. Holsti, O. (1969). Content analysis for the social sciences and humanities. Reading, MA: Adddison-Wesley. Ichheiser, G. (1970). Appearances and realities: Misunderstanding in human relations. Jossey-Bass, Inc. Jones, S. (2002). The intemet goes to college: How students are living in the future with today’s technology. Retrieved August 2nd, 2005 from http://wwwmewintemetorg/pdfs/PIP College Reportpdf. Katz, J. E., Rice, R. E., & Aspden, P. (2001), The Internet, 1995-2000. American Behavioral Scientist, 45, 405-419. Kayany, J. M., Wotring, C. E., & Forrest, E. J. (1996). Relational control and interactive media choice in technology-mediated communication situations. Human Communication Research, 22, 399-421. Kellerrnan, K., & Reynolds, R. (1990). When ignorance is bliss: The role of motivation to reduce uncertainty in uncertainty reduction theory. Human Communication Research, I 7, 5-75. 142 Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006). Dyadic data analysis. New York: The Guilford Press. Koda, T. (2004). Impression on different faces. Retrieved September 30, 2004 from ht_tp://xenia.media.mit.edu/~tomoko/survev.html. Kraut, R., Kiesler,S., Boneva, B., Cummings, J ., Helgeson, V. & Crawford, A. (2002). Internet paradox revisited. Journal of Social Issues, 58, 49-74. Kraut, R., Mukhopadhyay, T., Szczypula, J ., Kiesler, S., & Scherlis, B. (2000). Information and communication: Alternative uses of the Internet in households. Information Systems Research, 10, 287-303. Kraut, R., Patterson, M., Lundmark, V., Kiesler, S., Mukophadhyay, T. & Scherlis, W. (1998). Internet paradox: A social technology that reduces social involvement and psychological well-being? American Psychologist, 53, 1017-1031. Kunda, Z. (2001). Social cognition: Making sense of people. Cambridge, MA: MIT Press. Kunda, Z., & Thagard, P. (1996). Forming impressions fiom stereotypes, traits, and behaviors: A parallel-constraint-satisfaction theory. Psychological Review, 103, 284-308. Macrae, C. N., Milne, A. B., & Bodenhausen, G. V. (1994). Stereotypes as energy-saving devices: A peek inside the cognitive toolbox. Journal of Personality and Social Psychology, 66, 37-47. Marcus, 3., Machilek, F ., & Schiitz, A. (2006). Personality in cyberspace: Personal web sites as media for personality expressions and impressions. Journal of Personality and Social Psychology, 90, 1014-1031. Matei, S., & Ball-Rokeach, S. J. (2001). Real and virtual social ties. American Behavioral Scientist, 45, 550-564. McCroskey, J. C., & McCain, T. A. (1974). The measurement of interpersonal attraction. Speech Monographs, 41, 261-266. McLuhan, M. (1964). Understanding media: The extensions of man. McGraw-Hill: New York. Merkle, E. R., & Richardson, R. A. (2000). Digital dating and virtual relating: Conceptualizing computer mediated romantic relationships. Family Relations, 49, 187-192. 143 Mesch, G., & Talmud, I. (2006). The quality of online and offline relationships: The roles of multiplexity and duration of social relationships. The Information Society, 22, 137-148. Morris, W. (Ed.). (1976). The American heritage dictionary of the English language. Boston: Houghton Mifflin. Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56, 81-103. Nowak, K. (2000). The influence of anthropomorphism on mental models of agents and avatars in social virtual environments. Unpublished doctoral dissertation, Michigan State University, East Lansing, MI. Nowak, K. L. (2004). The influence of anthropomorphism and agency on social judgment in virtual environments. Journal of Computer-Mediated Communication, 9. Retrieved June 22, 2004 from hEp://www.ascusc.org[jcmc/vol9/issue2/nowak.html. Oliver, M. B. (1993). Exploring the paradox of the enjoyment of sad films. Human Communication Research, I 9, 315-342. O'Sullivan, P. B. (2000). What you don't know won't hurt me: Impression management functions of communication channels in relationships. Human Communication Research, 26, 403 -43 1. O'Sullivan, P. B. (n.d.). Masspersonal communication: Rethinking the mass- interpersonal divide. Unpublished manuscript, Illinois State University, Normal, I_L_._ Parks, M. R., & Adelman, M. B. (1983). Communication networks and the development of romantic relationships: An expansion of uncertainty reduction theory. Human Communication Research, 10, 55-79. Parks, M. R., & Floyd, K. (1996). Making friends in cyberspace. Journal of Commuication, 46, 80-97. Parks, M. R., & Roberts, L. D. (1998). Making MOOsic: The development of personal relationships on-line and a comparison to their off-line counterparts. Journal of Social and Personal Relationships, 15, 517-537. Patterson, M. (1995). A parallel process model of nonverbal communication. Journal of Nonverbal Behavior, 19, 3-29. 144 Patterson, M. L. (1998). Parallel processes in nonverbal communication. In M. T. Palmer & G. A. Barnett (Eds.), Progress in communication sciences: Volume XIV (pp. 1- 18). Ablex Publishing Corporation: Stamford, CT. Pena, J ., & Hancock, J. T. (2006). An analysis of socioemotional and task communication in online multiplayer video games. Communication Research, 33, 92-109. Planalp, S., & Honeycutt, J. (1985). Events that increase uncertainty in personal relationships. Human Communication Research, I I, 593-604. Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. New York: Simon & Schuester. Qiu, L., & Benbasat, I. (2005). Online consumer trust and live help interfaces: The effects of text-to-speech voice and three-dimensional avatars. International Journal of Human-Computer Interaction, 19, 75-94. Rainie, L., & Shermak, J. (2005, November). Search engine use November 2005. Pew Internet and American Life Project. Retrieved June 6, 2007 fi'om http://wwwpewinternet.org/pdfs/PlP_Sea_;chDga_l 105.pdf Ramirez, Jr., A., Walther, J. B., Burgoon, J. K., & Sunnafrank, M. (2002). Information seeking strategies, uncertainty, and computer-mediated communication: Towards a conceptual model. Human Communication Research, 28, 213-228. Reeves, B. & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Stanford, CA: CSLI Publications. Rheingold, H. (1991). Virtual reality. New York: Simon & Schuster. Rice, RE. (1984). Mediated group communication. In R.E. Rice & Associates (Eds.), The new media: Communication, research, and technology (pp. 129-156). Beverly Hills, CA: Sage. Shannon, C. E., & Weaver, W. (1949). The mathematical theory of communication. Urbana, IL: University of Illinois Press. Shklovski, 1., Kiesler, S. & Kraut, R. E. (In press). The Internet and social interaction: A meta-analysis and critique of studies, 1995-2003. In R. Kraut, M. Brynin, and S. Kiesler (Eds). Domesticating Information Technology. Oxford University Press. Retrieved July 9, 2007, from http://www.cs.cmu.edu/~kraut/RKraut.site.filesiarticles/Shklov$i04~ IntemetUseSociaLRleationships-mem—analysis.pdf 145 Spears, R., & Lea, M. (1992). Social influence and the influence of the "social" in computer-mediated communication. In M. Lea (Ed.), Contexts of computer- mediated communication. (pp. 30-65). London: Harvester-Wheatsheaf. Stafford, L., Kline, S. L., & Dimmick, J. (1999). Home e-mail: Relational maintenance and gratification opportunities. Journal of Broadcasting & Electronic Media, 43, 659-669. Stephenson, N. (1993). Snow Crash. New York: Bantam Books. Suler, J. (1997). The psychology of avatars and graphical space in multimedia chat communities. Retrieved September 29, 2004, from http://www.rider.edu/~suler/psvcvber/psyav.htrnl. Sunnafrank, M. (1986). Predicted outcome value during initial interactions: A reformulation of uncertainty reduction theory. Human Communication Research, I 39 3'33. Sunnafrank, M. (1988). Predicted outcome values in initial interactions. Communication Research Reports, 5, 169-172. Sunnafrank, M. (1990). Predicted outcome value and uncertainty reduction theories: A test of competing perspectives. Human Communication Research, I 7, 76-103. Sunnafi'ank, M., & Ramirez, Jr., A. (2004). At first sight: Persistent relational effects of get-acquainted conversations. Journal of Personal and Social Relationships, 21 , 361-379. Tardy, C. H. (Rd). (1988). A handbook for the study of communication: Methods and instruments for observing, measuring, and assessing communication processes. Norwood, NJ: Ablex. Tidwell, L. C. (1997). A comparison of anticipated interaction, channel, and communication goal an impression formation, self-disclosure, and relational orientation: Getting to know one another a bit at a time. Unpublished doctoral dissertation, Northwestern University, Evanston, IL. Tidwell, L. C., & Walther, J. B. (2002). Computer-mediated communication effects on disclosure, impressions, and interpersonal evaluations: Getting to know one another a bit at a time. Human Communication Research, 28, 317-348. Utz, S. (2000). Social information processing in MUD’s: The development of friendships in virtual worlds. Journal of Online Behavior, 1 . Retrieved June 6, 2004 from http://www.behavior.net/JOB/v1nl/utz.html. 146 Vangelisti, A. L., Corbin, S. D., Lucchetti, A. E., & Sprague, R. J. (1999). Couples' concurrent cognitions: The influence of relational satisfaction on the thoughts couples have as they converse. Human Communication Research, 25(3), 370-398. Walther, J. B. (1992). Interpersonal effects in computer-mediated interaction: A relational perspective. Communication Research, 19, 52-90. Walther, J. B. (1993). Impression development in computer-mediated interaction. Western Journal of Communication, 5 7, 381-398. Walther, J. B. (1994). Anticipated ongoing interaction versus channel effects on relational communication in computer-mediated interaction. Human Communication Research, 20, 473-501. Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23, 1-43. Walther, J. B., Anderson, J. F., & Park, D. W. (1994). Relational tone in computer- mediated communication: A meta-analysis of social and anti-social communication. Communication Research, 21, 460-487. Walther, J. B., & Boyd, S. (2002). Attraction to computer-mediated social support. In C. A. Lin & D. Atkin (Eds.), Communication technology and society: Audience adoption and uses (pp. 153-188). Cresskill, NJ: Hampton Press. Walther, J. B., & Bunz, U. (2005). The rules of virtual groups: Trust, liking and performance in computer-mediated communication. Journal of Communication, 55, 828-846. Walther, J. B., & Burgoon, J. K. (1992). Relational communication in computer-mediated interaction. Human Communication Research, 19, 50-88. Walther, J. B., & D'Addario, K. P. (2001). The impacts of emoticons on message interpretation in computer-mediated communication. Social Science Computer Review, 19, 323-345. Walther, J. B., Kim, S.-Y., Van Der Heide, B. Westerman, D., Tong, S. T., & Langwell, L. (in press). The role of fiiends’ appearance and behavior on evaluations of individuals via Facebook: Are we known by the company we keep? Human Communication Research. Walther, J. B., Loh, T., & Granka, L. (2005). Let me count the ways: The interchange of verbal and nonverbal cues in computer-mediated communication and face-to-face affinity. Journal of Language and Social Psychology, 24, 36-65. 147 Walther, J. B., & Parks, M. R (2002). Cues filtered out, cues filtered in: Computer- mediated communication and relationships. In M. L. Knapp & J. A. Daly (Eds.), Handbook of interpersonal communication (3rd ed., pp. 529-563). Thousand Oaks, CA: Sage. Walther, J. B., & Slovacek, C., & Tidwell, L. C. (2001). Is a picture worth a thousand words? Photographic images in long term and short term virtual teams. Communication Research, 28, 105-134. Walther, J. B., & Tidwell, L. C. (1995). Nonverbal cues in computer-mediated communication, and the effect of chronemics on relational communication. Journal of Organizational Computing, 5, 355-378. Wartella, E., & Reeves, B. (1985). Historical Trends in Research on Children and the Media: 1900-1960. Journal of Communication, 35, 118-133. Weisband, S. (1992). Group decision and first advocacy effects in computer-mediated and face-to-face decision making groups. Organizational Behavior and Human Decision Processes, 53, 352-380. ' Wellman, B., Haase, A. Q., Witte, J ., & Hampton, K. (2001). Does the intemet increase, decrease, or supplement social capital?: Social networks, participation, and community commitment. American Behavioral Scientist, 45, 436-455. Westerman, D., & Tamborini, R. (2005, November). “I don ’t have to ask, I can tell you ’re a dog The eflects of avatars on initial interactional judgments. Paper presented to the National Communication Association Conference, Boston, MA. Westerman, D., & Tamborini, R. (2006, June). The efi’ects of avatars on the use of interactive uncertainty reduction strategies. Paper presented at the International Communication Association Conference, Dresden, Germany. Westerman, D., Tamborini, R., & Bowman, N. D. (2007, November). The eflects of avatars an impression formation in difi'erent contexts. Paper to be presented at the National Communication Association Conference, Chicago, IL. Williams, D., Caplan, S., & Xiong, L. (2007). Can you hear me now?: The impact of voice in an online gaming community. Human Communication Research, 33, 427-433. Williams, R. and Edge, D. (1999). The social shaping of technology, In W. Dutton (Ed.), Information and communication technologies: Visions and realities (pp. 53-68). Oxford University Press, Oxford, pp. 53-68 148 Yee, N., Bailenson, J. N., Urbanek, M., Chang, F., Merget, D. (2007). The unbearable likeness of being digital: The persistence of nonverbal social norms in online virtual environments. CyberPsychology and Behavior, 10, 115-121. 149 lliliilgililliiiiljijii