Normative and Informational Influences in Online Political Discussions

How do the statements made by people in online political discussions affect other people's willingness to express their own opinions, or argue for them? And how does group interaction ultimately shape individual opinions? We examine carefully whether and how patterns of group discussion shape (a) individuals' expressive behavior within those discussions and (b) changes in personal opinions. This research proposes that the argumentative "climate" of group opinion indeed affects postdiscussion opinions, and that a primary mechanism responsible for this effect is an intermediate influence on individual participants' own expressions during the online discussions. We find support for these propositions in data from a series of 60 online group discussions, involving ordinary citizens, about the tax plans offered by rival U.S. presidential candidates George W. Bush and Al Gore in 2000. This journal article is available at ScholarlyCommons: http://repository.upenn.edu/asc_papers/99 Normative and Informational Influences in Online Political Discussions Vincent Price, Lilach Nir, & Joseph N. Cappella 1 Annenberg School for Communication, University of Pennsylvania, Philadelphia, PA 19104 6220 2 Department of Communication and the Department of Political Science, Hebrew University of Jerusalem, Jerusalem, Israel, 91905 How do the statements made by people in online political discussions affect other peo ple’s willingness to express their own opinions, or argue for them? And how does group interaction ultimately shape individual opinions? We examine carefully whether and how patterns of group discussion shape (a) individuals’ expressive behavior within those discussions and (b) changes in personal opinions. This research proposes that the argu mentative ‘‘climate’’ of group opinion indeed affects postdiscussion opinions, and that a primary mechanism responsible for this effect is an intermediate influence on individ ual participants’ own expressions during the online discussions. We find support for these propositions in data from a series of 60 online group discussions, involving ordi nary citizens, about the tax plans offered by rival U.S. presidential candidates George W. Bush and Al Gore in 2000. Investigations of social influence and public opinion go hand in hand. Opinions may exist as psychological phenomena in individual minds, but the processes that shape these opinions—at least, public opinions—are inherently social–psychological. The notion that group interaction can influence individual opinions is widely accepted. Indeed, according to many participatory theories of democracy, lively exchanges among citizens are deemed central to the formation of sound or ‘‘true’’ public opinion, which is forged in the fire of group discussion. This truly public opinion is commonly contrasted with mass or ‘‘pseudo’’-opinion developed in isolation by disconnected media consumers responding individually to the news (e.g., Blumer, 1946; Fishkin, 1991, 1995; Graber, 1982). Although discussion is celebrated in democratic theory as a critical element of proper opinion formation, it also brings with it a variety of potential downsides. These include a possible tyranny of the majority (e.g., de Tocqueville, 1835/1945), distorted expression of opinions resulting from fear of social isolation (Noelle-Neumann, 1984), or shifts of opinion to more extreme positions than most individuals might actually prefer (see, e.g., Janis, 1972, on dangerous forms of ‘‘group think,’’ or more recently Sunstein, 2001, on the polarizing effects of ‘‘enclave’’ communication on the Web). The problem of how to foster productive social interaction while avoiding potential dysfunctions of group influence has occupied a large place in normative writings on public opinion and democracy. Modern democracies guarantee freedom of association and public expression; they also employ systems and procedures aimed at protecting collective decision making from untoward social pressure, including not only the use of secret ballots in elections but also more generally republican legislatures and executive and judicial offices that by design are insulated from too much democracy, that is, from direct popular control (e.g., Madison, 1788/1966). However, steady advances in popular education and growth of communication media have enlarged expectations of the ordinary citizen and brought calls for more direct, popular participation in government. In particular, dramatic technological changes over the past several decades—and especially the rise of interactive forms of electronic communication enabled by the Internet and World Wide Web—have fueled hopes for new, expansive, and energized forms of ‘‘teledemocracy’’ (e.g., Arterton, 1987). Online political discussion is thus of considerable interest to students of public opinion and political communication. It has been credited with creating vital spaces for public conversation, opening in a new ‘‘public sphere’’ of the sort envisioned by Habermas (1962/1989), (see, e.g., Papacharissi, 2004; Poor, 2005; Poster, 1997). Though still not a routine experience for citizens, it has been steadily growing in prevalence and likely import for popular opinion formation. Recent surveys indicate that close to a third of Internet users regularly engage with groups online, with nearly 10% reporting that they joined online discussions about the 2004 presidential election (Pew Research Center, 2005). Online political discussion offers new and potentially quite powerful modes of scientific observation as well. Despite continuous methodological improvements, the mainstay of public opinion research, the general-population survey, has always consisted of randomly sampled, one-on-one, respondent-to-interviewer ‘‘conversations’’ aimed at extracting precoded responses or short verbal answers to structured questionnaires. Web-based technologies, however, may now permit randomly constituted respondent-withrespondent group conversations. The conceptual fit between such conversations and the phenomenon of public opinion, itself grounded in popular discussion, renders it quite appealing. Developments in electronic data storage and retrieval, and telecommunication networks of increasing channel capacity, now make possible an integration of general-population survey techniques and more qualitative research approaches, such as focus group methods, that have become popular in large part owing to the sense that they offer a more refined understanding of popular thought than might be gained from structured surveys (e.g., Morgan, 1997). Perhaps most important, the study of online discussion opens new theoretical avenues for public opinion research. Understanding online citizen interactions calls for bringing together several strands of theory in social psychology, smallgroup decision making, and political communication that have heretofore been disconnected (Price, 1992). Social influence in opinion formation Certainly, the most prominent theory of social influence in public opinion research has been Noelle-Neumann’s (1984) spiral of silence. Citing early research on group conformity processes, such as that of Asch (1956), Noelle-Neumann argued that media depictions of the normative ‘‘climate of opinion’’ have a silencing effect on those who hold minority viewpoints. The reticence of minorities to express their views contributes to the appearance of a solid majority opinion, which, in turn, produces a spiral of silence that successively emboldens the majority and enervates the minority. Meta-analytic evaluations of research on the hypothetical silencing effect of the mediated climate of opinion suggest that such effects, if they indeed exist, appear to be fairly small (Glynn, Hayes, & Shanahan, 1997); nevertheless, the theory has garnered considerable empirical attention and remains influential. In experimental social psychology, group influence has been the object of systematic study for over half a century. Although no single theoretical framework is available for explaining how social influence operates, some important organizing principles and concepts have emerged over time (Price & Oshagan, 1995). One of the most useful heuristics, proposed by Deutsch and Gerard (1955), distinguishes two broad forms of social influence (see also Kaplan & Miller, 1987). Normative social influence occurs when someone is motivated by a desire to conform to the positive expectations of other people. Motivations for meeting these normative expectations lie in the various rewards that might accrue (self-esteem or feelings of social approval) or possible negative sanctions that might result from deviant behavior (alienation, excommunication, or social isolation). Normative social influence is clearly the basis of Noelle-Neumann’s (1984) theorizing about minorities silencing themselves in the face of majority pressure. Informational social influence, in contrast, occurs when people accept the words, opinions, and deeds of others as valid evidence about reality. People learn about the world, in part, from discovering that they disagree (e.g., Burnstein & Vinokur, 1977; Vinokur & Burnstein, 1974). They are influenced by groups not only because of group norms, but also because of arguments that arise in groups, through a comparison of their views to those expressed by others (see also the distinction between normative and comparative functions of reference groups in sociology, e.g., Hyman & Singer, 1968; Kelley, 1952). Although the distinction between informational and normative influence has proven useful and historically important in small-group research, it can become cloudy in many instances. This is so because normative pressure and persuasive information operate in similar ways within groups, and often with similar effects. For example, the tendency of groups to polarize—that is, to move following discussion to extreme positions in the direction that group members were initially inclined—can result either from adjustments to a perceived normative position of the group or from limited or biased pool of group arguments. It can also be difficult in practice to separate the informational content of persuasive messages from the normative pressure they may bring to bear, in other words, to separate the normative function of comments made within a group (i.e., conveyance of a group’s overall preference) from the informative functions served by those same comments (Postmes, Haslam, & Swaab, 2005, pp. 29–30; Turner, 1991, pp. 144–147). Turner (1982) argued for a third approach to group influence—one he labeled referent informational influence—which he saw as something of an amalgam of Deutsch and Gerard’s (1955) informational influence and notions of reference-group power to shape behavior (e.g., French & Raven, 1959; Kelman, 1961). Turner’s explanation centered on basic cognitive, self-categorization processes, following Tajfel (1978). He argued that, when people categorize themselves and others as belonging to distinctive groups (even broad social categories), they will tend to impute the perceived stereotypical characteristics of their own group to themselves. Unlike normative or informational influence as conventionally described, this form of influence stems not only (or not even) from normative or argumentative pressures brought by other group members but rather from one’s own beliefs about the appropriate behavior of people belonging to whatever social category is used at the moment to define oneself (Turner, 1982, 1991). This form of influence is in essence a ‘‘self-stereotyping’’ process, involving internalized norms rather than external pressures. It is fully consistent with long-standing evidence that merely increasing the situational salience of group membership leads people to express more normative group opinions and attitudes, even in the absence of any direct social influence from others (e.g., Charters & Newcomb, 1952; Doise, 1969; Kelley, 1955; also reconfirmed by recent evidence that anonymous encounters online can exert normative group influence, e.g., Postmes, Spears, & Lea, 1998). The Turner and Tajfel cognitive redefinition of social group influence, under the rubric of social identity theory, has met with substantial empirical support and informed large bodies of scholarship in social psychology, particularly so in the study of intergroup behavior, stereotyping, prejudice, and minority group relations (see, e.g., reviews in Hewstone, Rubin, & Willis, 2002; Turner & Reynolds, 2001). Given that the theory explains how people can be socially influenced even without direct interpersonal contact, it has been studied primarily in connection with ‘‘minimal,’’ noninteracting groups or in relation to large-scale intergroup processes (where group categories are constructed experimentally or where preexisting group differences are made salient). More recently, however, research has begun to apply social identity theory in the analysis of small-group interaction as well, particularly with respect to identity formation in computer-mediated communication (e.g., Postmes et al., 1998). Noting that the bulk of research to date has focused on the effects of making salient some existing or externally imposed group identity (what Turner, 1982 called the deductive aspect of referent informational influence), Postmes and colleagues (Postmes et al., 2005; Postmes, Spears, Lee, & Novak, in press) point out that relatively few studies have explored what Turner termed the inductive aspect: the ‘‘means by which the criterial attributes of some [group] category are inferred from one or more individual members’’ (1982, p. 28). Harkening back to Sherif’s (1935) early work on norm formation through group interaction, Postmes and colleagues have begun to explore not only the ‘‘top-down’’ means by which group norms influence members, but also the ‘‘bottom-up’’ means by which members actively construct their own group norms. Postmes, Spears, and Lea (2000), for example, monitored e-mail exchanged by 87 students enrolled in an online statistics course over 4 months. They identified 11 groups of students who regularly exchanged messages to each other, and examined patterns in messaging over time. Evidence of norm formation was found in terms of distinctive within-group message profiles based on features such as message categories, requests, complaints, humor, expression of emotion, flaming, and message length. Moreover, the data suggested that group members tended overtime to produce messages that were increasingly prototypical of the group norms, which the authors interpreted as evidence of normative influence. Tracking the routes of influence The mechanisms by which normative or informational influence might flow among members of a group discussing politics online, then, are myriad. Some theorists (e.g., Noelle-Neumann, 1984) heavily emphasize normative pressures to conform; others emphasize the role of argumentation (e.g., Sunstein, 2001, in his examination of ‘‘informational cascades’’ as a source of group polarization). Many outcomes, for example apparent conformity to a group majority, may stem from either normative or informational influences, or some amalgam (Asch, 1956; Deutsch & Gerard, 1955; Turner, 1991). As predicted by conformity theories, the dominant sense of feeling of a group (the opinion climate) might cause its separate members to misrepresent their private views, either by refraining from expressing them or by adapting them to the majority opinion. This process may occur and still not produce individual changes of opinion. Indeed, studies of small-group influence indicate that conformity processes often produce mere compliance rather than internal conversions of opinion (Kelman, 1961; Moscovici, 1985). The particular arguments expressed, on the other hand, might produce internal conversions of opinion even without extracting much behavioral conformity. That is, a group might generally agree on a particular issue and yet, in their discussions of it, uncover a range of both proand con-arguments. These may, over the course of time, lead to movements of group members to adopt new viewpoints, particularly if their initial opinions were not very well formed. Finally, as predicted by social identity theory, group members might conform their behavior to their perception of a prototypical group member, not out of any majority pressure or because they are persuaded by the substance of the arguments, but rather because they are enacting their understood roles as group members. The perceived group norms to which members conform could be preexisting, and brought into the group from the outside, or built up inductively through behavioral exchanges within the group. Examining these sorts of interactive processes requires more than just input/ output measures of how group members feel about an issue before and after discussion. Also needed are detailed assessments of what each person actually says in discussion, including both expressions of preference and arguments that are raised. Data that closely track group interaction were usually not gathered in early social– psychological experiments. Rather, such experiments measured inputs to and outputs from group interaction, without close examination of actual processes of group discussion. And, oddly enough, many studies of group influence did not actually examine groups communicating in any ordinary sense; instead, they involved rounds of decision-making tasks, for example, perceptual judgments of line lengths or slides that are shared within the group (often constituted of experimental confederates in order to create opinion majorities or minorities, see Moscovici, 1985, for a review). As noted earlier, recent research has begun to track patterns of group interaction. However, the focus of such interaction analysis has not been of the kind that illuminates the opinion formation and change processes of interest to political communication or public opinion researchers. The Postmes et al. (2000) study of online communication discussed above, for instance, coded features such as self-reference, requests, complaints, flaming, use of humor or slang, and the like. Research in formal decision-making groups has studied patterns of argumentation in some detail, both in face-to-face settings (e.g., Meyers, 1989) and in computer-mediated groups (e.g., Brashers, Adkins, & Meyers, 1994). These studies, however, have focused on group outcomes rather than on individual choices, finding evidence that group decisions are influenced both by the total proportion of arguments made supporting or opposing the proposition and by the number of group members offering support or opposition (Gouran, 1994; Hoffman, 1979; Lemus, Siebold, Flanagin, & Metzger, 2004; McPhee, Poole, & Siebold, 1982; Meyers & Brashers, 1998). The burgeoning array of studies into Group Decision Support Systems has thus made significant advances in tracing patterns of argumentation in groups; but the applicability of this work to online political discussion, where people are under no expectation of arriving at a group decision or shared judgment, remains open to question. Moreover, these studies have not typically examined the role of group arguments or normative pressure in shaping individual participants’ expression or concealment of their own private opinions, or traced this behavior to subsequent individual opinion formation—processes of primary interest to public opinion researchers (Glynn et al., 1997). Research questions and hypotheses The present study focused on the influence of groups on individuals in settings where they engaged in political discussion without any explicit expectation of coming to a shared judgment. We took advantage of a unique data set to explore the processes— both normative and informational—through which group influences might operate. Specifically, we examined a series of 60 online group discussions that occurred in 2000, involving ordinary citizens, about the tax plans offered by rival presidential candidates George W. Bush and Al Gore. These discussions, described below, were broadly deliberative in nature but did not lead to any common group judgment. Neither votes were taken nor were participants urged to arrive at a consensus. We recorded all of the group interactions and classified all statements made by participants into either (a) mere expressions of preference or opinion, and (b) reasons or arguments in support of particular points of view. Mere expressions of opinion, while they can clearly create a normative climate of group opinion, should not contribute to opinion formation through any rational, argumentative mechanism. Arguments, on the other hand, should constitute a fund of potential informational value to participants. Both traditional conformity approaches to social influence and more recent social identity approaches would predict that the people’s arguments and expressions of opinion during online discussions should be affected by the tenor of group argumentation, such that: H1a: People are likely to express views favorable to candidate Bush when in groups where others express support for Bush; conversely, people will tend to express support for Gore when in groups where others support Gore. H1b: People are likely to make arguments for Bush’s tax plan (and against Gore’s plan) when in groups where others argue for Bush; conversely, people will tend to make arguments for Gore’s plan (and against Bush’s plan) when in groups where others argue for Gore. The exchange of viewpoints is also expected to shape individual participants’ opinions: H2: People are likely to report opinions, postdiscussion, supporting Bush’s tax plan when others in their group supported Bush; conversely, people will tend to support Gore’s plan, postdiscussion, when others in their group supported Gore. Research has suggested that computer-meditated groups, relative to face-to-face interactions, produce less individual dominance (Walther, 1995), greater equality of member participation (Siegal, Dubrovsky, Kiesler, & McGuire, 1986), and at least in formal decision-making settings, a greater number of unique ideas (Dennis, 1996; Gallupe, DeSantis, & Dickson, 1998). Research findings on the comparative impact of computer-mediated communication on argument generation are somewhat sparse and mixed to date. However, we have cause to expect that computer mediation should facilitate disagreement in a political context (Stromer-Galley, 2003) and, in part because the exchanges focus entirely on textual messages (e.g., Rice, 1993), we expect that in the present context, arguments should prove especially influential: H3a: Individual’s arguments and expressions of opinion will be more greatly influenced by arguments made by others in the group than by mere expressions of opinion among members of the group. H3b: People’s postdiscussion opinions will be more greatly influenced by others’ arguments than by others’ mere expressions of opinion. Finally, we expect that the particular arguments and expressed viewpoints proving most influential on postdiscussion opinions would be each participant’s own expressions. This is predicted in part on account of the higher salience of one’s own statements, but also due to self-persuasion and self-attribution processes (Bem, 1970). Expressive behavior is a step in the process of self-definition and opinion change, a form of ‘‘test’’ behavior that helps a person along toward a well-formed judgment (Kelman, 1974; Price, 1992). This prediction is also consistent with social identity theory, in that it posits people induce the prototypical features of a group member from others’ behavior, and then conform their own behavior to perceived group norms. We thus hypothesize H4: The influence of others’ arguments and viewpoints on an individual’s postdiscussion opinions concerning the candidates’ tax plans will be mediated by the arguments and views that individual personally expresses during the discussion. We carefully examined these hypotheses, inquiring as to whether and how mere opinion statements and arguments shaped both (a) individuals’ expressive behavior in the discussions and (b) changes in personal opinions. Our results suggested that the argumentative ‘‘climate’’ of group opinion indeed affected postdiscussion opinions. Importantly, a primary mechanism responsible for this effect appears to have been an intermediate influence on individual participants’ own expressions during the online deliberations.