Abstract

In this essay, we draw on broader psychological theories of the attitude-behavior relationship to postulate specific reciprocal patterns of causality between the civic attitudes and forms of political and civic engagement featured in contemporary political communication research. We then examine the extent of these reciprocal relationships with a 2-wave panel survey of 2,872 Pacific Northwest residents. Spanning the 2004 elections, structural equation modeling of the panel data shows complex reciprocal causal paths between political/civic attitudes (internal and external efficacy and civic pride and faith) and a range of political and civic behaviors (voting, political action, media use, political/community talk, and group involvement). The conclusion suggests how to conceptualize these variables and model their relationships in future research. doi:10.1111/j. 1460-2466.2010.01484.x Virtually all political communication research rests on implicit assumptions about how attitudes and behaviors relate to one another. Research seeking to explain variation in political attitudes often presumes that they drive key behaviors. Naturally, explanation of attitude variation is often sought in communication exposure and attention variables, as in research on media framing, or the overall tone and content of political communication Both propositions (that attitudes drive behavior, and conversely that behaviors spawn attitudes) have received ample support in the psychological literature (e.g., Examining the data from a panel study spanning the 2004 elections through the lenses provided by foundational psychological theories of the attitude-behavior relationship, we hope to shed greater light on processes of reciprocal patterns of causality between efficacy attitudes and various political and communicative behaviors. Specifically, we hypothesize that within the more strategic context of campaigns and elections, attitudes tend to drive behavior and not the other way around, whereas we anticipate a reciprocal attitude-behavior linkage in civic or community-based contexts. In one instance, we even anticipate a mutual reinforcement-with a closely related attitude and behavior, each exerting a causal influence on the other. By theorizing about the deeper mechanisms by which such causal patterns work, and then testing those hypothesized mechanisms through a panel survey design, we hope to contribute to a better integration and the further development of theory and research on the role played by communication in the facilitation of political efficacy and civic engagement, as well as other attitude-behavior connections in the wider communication research literature. Assumptions beneath the surface A survey of the literature on communication and civic engagement with an eye toward attitude-behavior assumptions reveals a variety of assumptions about these linkages lurking behind the scenes. The dominant assumption that attitudes drive behavior is evident in most research on political efficacy and trust. Here, the central reference points come from the political science research on participation These studies helped forge assumptions that attitudes toward political participation drive actual behavior, and that we should study political communication because of its potential to influence these attitudes. In the classic account, the mass media promote a ''strategic'' interpretation of political actors and the electoral process, which systematically undermines citizen attitudes toward the political system, thereby reducing their active participation in it Another prime example of this approach in the context of political communication research is seen in Aarts and Smetko's (2003) examination of media effects on political 319 Of Attitudes and Engagement J. Gastil & M. Xenos attitudes and involvement. Citing conflicting strands of research, one pointing to media malaise and the other pointing to media mobilization, Aarts and Semetko convincingly show (as do The reverse assumption of behavior driving attitudes, particularly those of efficacy and trust, can be found in the growing literature on public deliberation and discussion The implicit linkage in such a research entertains the alternative assumption that certain behaviors are important causal antecedents to feelings of efficacy and attitudes toward fellow citizens. For example, in his research on the relationship between various structures of decision making, including forms of deliberation, Even research that takes a much more skeptical approach to deliberation operates from much the same assumption at the level of the attitude-behavior linkage. Describing what they see as Americans' desire for a Stealth Democracy, Hibbing and Theiss-Morse (2002) suggest that as any deliberation worthy of the name involves prolonged discussion and conflict, and as most citizens dislike conflict and a sizeable proportion find themselves relatively less loquacious than those who often come to dominate such discussions, the likely experience of deliberation for most citizens is negative, if anything possibly leaving them with less confidence in the system and themselves. To be sure, the assumptions that guide the lines of research reviewed here are undoubtedly more complex than this simplified or even caricatured discussion might suggest. Most if not all communication researchers would likely refuse to be categorized as assuming attitudes always or exclusively drive behavior or vice versa. Of Attitudes and Engagement Nevertheless, we contend that it is useful to consider the predominant assumptions in various strands of research, so that their implications may be better understood. General accounts of the attitude -behavior linkage Theorizing more broadly than the particular context of civic engagement, the psychological literature provides general explanations for both directions in the attitude-behavior relationship. Indeed, a vast literature takes these issues as a point of central concern, rather than simply as background to more specific questions. We begin, however, with consideration of two archetypical accounts of the attitude-behavior linkage that can provide us with an initial set of tools for theorizing possible relationships between civic attitudes and behaviors. The strongest foundation for the dominant assumption of attitudes-to-behavior comes from the work of In the case of the general narrative of research on the effects of media on political efficacy, we may interpret the process as follows. Information from media sources, accounts of the candidates and their campaigns, for example, lead citizens to develop a set of beliefs about the political system and their place within it (e.g., ''politics is a game''). On the basis of these beliefs, negative (or positive) attitudes toward politics form, thereby creating corresponding sets of intentions toward participation, which in turn drive actual participation. The other side of the attitude-behavior dynamic is buttressed by Bem's (1972) self-perception theory. This theory elegantly stipulates an opposing narrative of the attitude-behavior linkage that can be summarized in two simple postulates. The first is that ''individuals come to 'know' their own attitudes . . . partially by inferring them from observations of their own overt behavior and/or the circumstances in which this behavior occurs'' (Bem, 1972, p. 5). Thus, individuals derive their attitudes from a deductive reasoning process in which the input is their behavior; judgments about how one feels are determined by observations of how he or she acts. The second postulate is that ''to the extent that internal cues are weak, ambiguous, or uninterpretable, the individual is functionally in the same position as an outside observer . . . who must necessarily rely upon those same external cues to infer the (Bem, 1972, p. 5). Thus self-perception theory offers us a disarmingly simple, yet quite accurate account of the formation of many attitudes, including political attitudes (e.g., In the case of civic engagement and behavior, citizens may ask themselves, ''What must my attitude toward participation be if I am willing to participate in these ways, under these circumstances?'' On this account, we may then think about processes by which either positive or negative experiences with various forms of deliberation and other kinds of participatory behavior may translate into feelings of efficacy and trust. Reciprocal relations between behaviors and attitudes Given the research discussed earlier on the antecedents and consequences of feelings of efficacy and trust, it is natural to assume that both of these causal processes are likely at work in the public sphere at any given time. The possibility of reciprocal causal structures surrounding political efficacy is not new. Nearly a decade ago, Bandura (1997) called for a ''comprehensive research effort'' to understand the ''determinants, mechanisms, and outcomes'' of efficacy (p. 428). Indeed, while noting the effects of efficacy on political behavior, Bandura further theorized that the greatest benefits to research on efficacy would come from explorations of its causal antecedents, especially those found in individuals' experiences with various group and communicative processes. A handful of researchers have undertaken specific exploration of reciprocal relationships between attitudes toward civic engagement and political participation However, this small body of research has produced mixed findings and examined a very limited number of variables. Perhaps the first to explore this relationship fully with a panel design was More recently, Stenner-Day and Fischle (1992) conducted a similar investigation, using Australia's National Social Science Survey data from 1984 and employing a slightly wider array of variables including not only conventional and partisan forms of political participation, but also measures of extremist political participation and community activism. Leaving aside some of the more complex relations uncovered in this study, it is notable that with respect to conventional participation and partisan activism, the findings are a mirror image of A reciprocal model of civic attitudes and behaviors The earlier theoretical discussion provides plausible accounts for many potential connections-including some reciprocal relationships-between civic attitudes and behaviors. At this juncture, we believe that research on these relationships could substantially benefit from a more comprehensive approach to exploring the circumstances under which different kinds of connections are more, or less, plausible. To do so, researchers need to investigate a broader set of attitudes, reaching beyond political efficacy to consider other beliefs and evaluations potentially linked to a wide range of civic behaviors. With this larger set of attitudes and behaviors in hand, we then need to develop a systematic theoretical account for the causal relationships between civic attitudes and behaviors. Such a model would explain, parsimoniously, why some attitudes shape behavior, why some behaviors shape attitudes, and why some attitudes and behaviors mutually reinforce one another. Widening the range of attitudes and behaviors Before advancing a tentative theoretical model, we suggest the addition of two attitudes to complement internal and external efficacy. The two efficacy concepts refer to personal confidence in one's own political acumen and faith in the system's responsiveness to competent strategic action. In the less agonistic world of community involvement, we suggest two somewhat parallel concepts-a personal sense of pride in one's civic responsibility along with one's faith in the responsibility of fellow citizens. We call these civic pride and civic faith, respectively, and they provide a community-oriented parallel to the conventional notions of internal and external political efficacy. These concepts are an original formulation, though they bear resemblance to Civic pride is confidence that one takes one's community and civic responsibilities seriously, as opposed to believing that one has no civic duties or fulfills them only out of reluctant obligation. Instilling (and drawing on) civic pride is often the aim of public engagement processes, such as when Denhardt and Denhardt The companion of civic pride is civic faith, the belief that others' will, in the same way, act responsibly. Its opposite is the belief that other community members and 323 Of Attitudes and Engagement J. Gastil & M. Xenos fellow citizens will not make any necessary effort, let alone sacrifices for the public good. Whereas external efficacy concerns one's perception that political institutions are responsive, civic faith represents the belief that individual citizens are responsible and can be counted on to vote during elections and otherwise participate in public life for the sake of their community or nation. In addition to complementing political efficacy with civic attitudes, we look beyond voting and more traditional conceptions of political participation to consider a broader set of civic counterparts. As communication scholars, we have particular interest in looking at the wider range of ways in which we communicate in politics and community life. First, the range of behaviors we consider herein includes a set of three strategic behaviors: voting (likely the most formal, institutionalized means of expression), political/campaign involvement (attending rallies or other events and/or volunteering on behalf of political causes), and public affairs media use. We call these the set of strategic behaviors because they are all oriented toward effective political engagement-from gathering information through the media, to influencing others' votes, to casting one's own ballot. As a contrast to these behaviors, we wish to also consider highly interactive civic behaviors celebrated in the literature on deliberative democracy Hypotheses With the aid of the psychological concepts explicated earlier, we may now think more specifically about how the variables in this more inclusive set of political and civic engagement attitudes and behaviors may relate to one another, qua attitudes and behaviors. In doing so, we draw on Fazio's contributions to scholarship on attitude-behavior relationships The first set of hypotheses (H1) we advance grows from the empirical and theoretical kinship between the classic political efficacy variables (external and internal efficacy) and those political behaviors we have earlier described as strategic (voting, political action, and media use). As noted earlier, there is ample empirical support for political efficacy as a predictor of political participation. What is of concern to us, however, is not so much that these variables share an orientation to the same political context (typically campaigns and elections), but rather the nature of participation in that context. Specifically, H1 derives from the nature of participation in campaigns and elections, where individuals are typically expected to collect information on various options, form a private opinion, and then express it through partisan campaigns and, ultimately, at the ballot box. As Fazio and Williams (1986) argue, this is ''most likely the result of a controlled process in which individuals reflect and arrive at a behavioral intention'' (p. 512). Moreover, because this enterprise culminates in the aggregation of private ballots by neutral agents, and the process of opinion formation is largely private, we contend that the strategic behaviors considered here are experienced relatively less directly than political talk or civic group involvement. Thus, opportunities for behavioral input into the attitude formation process are likely fewer H1: External and internal efficacy at time 1 (T1) will be positively related to strategic political behavior (voting, political action, and media use) at time 2 (T2). The second set of hypotheses (H2) principally concerns those variables that we have incorporated to broaden the range of civic attitudes and behaviors that researchers consider. Specifically, H2 focuses on the more community-oriented attitudes (civic pride and civic faith) and behaviors (participation in political talk, community talk, and group involvement). Again, our concern is less that these variables share a common context, but it is about the nature of that context. In this case, the defining feature of the more community-oriented context is the sense in which behavior is more communicatively anchored in collective processes-plausibly quasi-deliberative processes H2a: Civic pride and civic faith at T1 will be positively related to political/community talk and group involvement at T2. H2b: Political/community talk and group involvement at T1 will be positively related to civic pride and civic faith at T2. The study that follows will test each of these hypotheses, but we will also explore all other possible relationships among the attitudes and behaviors described herein. In this way, our study will not only confirm or disconfirm the particular predictions above but also explore the other potential reciprocal relationships. Study design and methods To test these hypotheses, we designed a two-wave panel study Participants Two waves of survey data were collected from 2,872 residents of King County who reported for jury service between February 10, 2004 and August, 20, 2004. 1 Fifty-four percent of the participants were female, and 90% were White, with 5.5% Asian-American, 1.7% African American, 1.5% Native American, and 1.2% Hispanic. The median educational level was a college degree (B.A., B.S., or A.B.), with 29.7% having less formal education and 37.4% having more. The median Of Attitudes and Engagement age was 50, with 50% of respondents between 40 and 58 years of age. The work status of the respondents included 62.7% working full-time, 14.8% retired, 9.9% working temporary or part-time jobs, 6.2% self-identified as ''homemaker,'' 3.4% unemployed, and 3.0% enrolled as students. For the structural equation models (SEMs), we reduced the sample size by removing 489 cases that were missing a substantial portion of their data. (Most of these were missing exact-matching voting records, as described below.) The remaining 2,383 cases had the same general characteristics as the full sample but were also ideal for data imputation, with only one of the variables (ethnicity) missing in more than 5% of all cases. Missing values were replaced using the expectation maximization (EM) method in SPSS with normal distribution. The large sample used in this study was necessary to ensure adequate statistical power. Recall that Procedures Wave 1 survey During the study period, researchers attempted to approach every person who signed in for jury duty to request their voluntary participation in a study on ''community life.'' This first survey was administered after jury orientation but before jurors had been sent to courtrooms, with a response rate of 78% and a cooperation rate of approximately 81% (4% of those reporting to service were sent to courtrooms before research staff could administer the survey). Wave 2 survey Those who responded in Wave 1 were contacted again in November 2004, after the completion of the 2004 primary and general elections. Respondents were invited to complete a follow-up survey on paper or online. A repeated-contact design (Dillman, 1999) sent postcards, up to two survey booklets, reminders, and thank you cards to potential respondents. Seventy-three percent of those who were contacted again via a valid e-mail or mailing address returned a completed survey. The median lag time between receiving and returning the survey was 2 weeks. 2 Measures Political attitudes and behaviors Standard survey items from the National Election Study and General Social Survey were used to measure internal efficacy, external efficacy, and political participation. Original items were developed to measure civic pride and civic faith, along with public affairs media use, political discussion, and active participation in a wider array of community groups and associations. In total, each of the two surveys included 12 attitude items (three for each of four attitudes) and 16 participation items (four for each of four scales). A confirmatory factor analysis with varimax rotation yielded results consistent with this set of four attitudinal and four behavioral variables, and all item wordings are in Appendix. Control variables Additional survey items measured were ethnicity, sex, age, educational level, partisanship, and political knowledge. The partisanship measure was constructed in two steps: We first averaged scores on a standard 7-point liberal/conservative item with responses to a 7-point scale created by merging items asking for party identity, then the strength of identification (interitem r = .76); then, we recoded this scale such that it measured one's degree of partisan strength (i.e., the absolute distance from the scale midpoint). Political knowledge was measured as the number of correct answers to a Results Preliminary analyses consisted of partial correlations to confirm the presumed existence of cross-sectional associations among the attitudes and behaviors measured herein. Main hypothesis tests Our main test assessed the hypothesized attitude-behavior relationships in a single SEM linking Wave 1 attitudes with Wave 2 behaviors and Wave 1 behaviors with Wave 2 attitudes, as explained in Hypotheses 1 and 2. Additional structural relationships in this first model included covariance among all Wave 1 variables, including the aforementioned controls, and paths to each Wave 2 variable from each control plus its Wave 1 counterpart (i.e., linking Wave 1 external efficacy to Wave 2 external efficacy). Using SPSS-Amos 17 SEM software, all of the relevant attitudes and behaviors were treated as latent variables with multiple operational indicators, with the exception of Wave 1 and 2 voting rates. This latent variable approach was particularly helpful in the present study, in which only three or four items could be used to measure key variables, because SEM takes into account measurement error when estimating causal relationships. When measures are taken over multiple panel waves, it is possible to remove considerable bias from estimated paths between latent variables by incorporating correlations between each item's measurement errors at two points in time while simultaneously specifying the measurement model and structural effects The fit of this initial model was assessed following the recommendations of Stephenson (2002, 2008). Besides reporting the overall model χ 2 , they recommend using the Standardized Root Mean Square Residual (SRMR) in conjunction with the Root Mean Square Error of Approximation (RMSEA) and a comparative fit index (CFI). Ideally, a model should have an SRMR score below .09, an RMSEA below .06, and a CFI above .95. The hypothesized model met these standards sufficiently to allow interpretation (SRMR = .050, RMSEA = .035, and CFI = .920), with a χ 2 = 6610.8 (df = 1656, p < .001), though the CFI suggests a less-than-desirable incremental model fit. Some-but not all-of the specific hypothesized paths, however, reached significance. With regard to Hypothesis 1, external efficacy had no effects on strategic political behaviors (voting, political action, and media use). Wave 1 internal efficacy, however, did have significant positive relationships with political action (b = .073, SE = .022, p < .001) and media use (b = .225, SE = .050, p < .001). The oft-cited causal path from internal efficacy to voting (e.g., The second set of hypotheses posited reciprocal relationships between civic pride and faith, on the one hand, and political/community talk and civic group involvement, on the other. This pattern emerged for one pairing of variables-civic pride and political/community talk. In this case, Wave 1 pride promoted greater Wave 2 talk (b = .065, SE = .039, p = .008), and Wave 1 talk positively reinforced Wave 2 pride (b = .139, SE = .023, p < .001). The only other significant causal path was from Wave 1 group involvement to Wave 2 (b = .127, SE = .020, p < .001). The other five paths in Hypothesis 2a and 2b were nonsignificant. An alternative model: Full reciprocation Although the hypothesized model fits the data, in terms of overall statistical measures, we thought it prudent to also generate an alternative model. Rather than incrementally adjusting our hypothesized model, we chose to move directly to one that included Of Attitudes and Engagement the full set of possible paths from attitudes to behaviors and vice versa. As we advised earlier, there exists little theoretical work on the potential reciprocal relationships in civic and political engagement, so exploring the full range of such connections could prove fruitful to future theoretical work. This second model tested all of the potential associations simultaneously in a single SEM linking Wave 1 attitudes with Wave 2 behaviors and Wave 1 behaviors with Wave 2 attitudes. As earlier, additional structural relationships included covariance among all Wave 1 variables, paths from each Wave 1 variable to its Wave 2 counterpart, and the same paths for the control variables. The fit statistics for this model were nearly identical to those of the hypothesized one (SRMR = .050, RMSEA = .036, and CFI = .921), with a χ 2 = 6538.2 (df = 1630, p < .001), again with the incremental fit indicator (CFI) falling below the .95 threshold. Formally testing the change in overall fit between this model and the hypothesized one nested within it, the χ 2 -difference test shows a reduction in χ 2 estimates of 72.6, which exceeds the p = .001 critical value associated with 26 degrees of freedom (i.e., the difference in df between the models; see Holbert & Stephenson, 2008, p. 196). From this model, Discussion The most obvious general insight from the findings presented here is that civic and political attitudes, as hypothesized, have causal effects on behavior, though the reverse appears true even for the efficacy variables that we expected to shape behavior, rather than vice versa. Attitudes about one's competence in political and community arenas are important predictors of civic and political participation. Moreover, individuals' experiences with various elements of public engagement, including the most communicative aspects of public life, exert tangible force on the same attitudes that are also believed to predict their participation in the first place. Thus, our findings can be viewed as a confirmation and extension of Stenner-Day and Fischle's (1992) reciprocal findings and a qualification of A second implication of our study concerns the utility of classic models of the attitude behavior linkage in understanding contemporary questions of political communication research. Given the dominance of the attitude-to-behavior model, we will focus our discussion here on the value of Bem's (1972) self-perception theory. Although relatively simple and uncomplicated, we believe that this often overlooked theoretical model provides a compelling narrative explanation for not only the relationships between participation and efficacy observed in our data Of Attitudes and Engagement but also potentially a variety of other phenomena of interest to communication researchers as well. In particular, the self-perception model appears particularly suited to understanding the effects of a participatory communication medium like the Internet, where depending on the level of interactivity present in a given web environment, exposure to communication content is very often essentially a behavior in which the individual participates him or herself The specific substantive findings of our analysis not only offer some validation of our initial hypotheses but also go farther. In our original model, internal efficacy did prove predictive of strategic political behavior, and there was the expected reciprocal relationship between civic pride and community/political talk. The first of these findings was consistent with our expectation that individual, strategic political engagement would depend on one's sense of efficacy as a political actor. The second finding fits with our notion that highly interactive, communal behavior would have a mutually reinforcing relationship with a sense of one's own and other community members' civic responsibility. The lack of significant findings for the other variables in these hypotheses (external efficacy, civic faith, and group involvement) may suggest the need to generate more refined models, which may limit the connections to these variables or, particularly in the case of civic attitudes, reassess their measurement. With an eye toward future theory and research, the findings of our secondary SEM may suggest some fruitful lines of inquiry. The former paths fit with general expectation of mutual reinforcement between civic/political behaviors and attitudes, whereas the latter results require explanation. If these findings prove robust in future research, it might suggest that strategic political engagement and media use depend on and reinforce a degree of skepticism about one's fellows and public institutions. The negative findings in If future research validates these revised groupings of attitudes and behaviors, it might ultimately prove fruitful to pursue second-order factors Doing so, however, will require more confidence than we have at present in how to group the wide array of civic, community, and political attitudes and behaviors. Even with the addition of the results presented herein, we believe that there are simply not enough studies concerning these causal structures. Just as Bandura (1997) has called for a more comprehensive understanding of the causes and effects of efficacy, communication researchers should continue to advance our theoretical knowledge Of Attitudes and Engagement of how civic attitudes and communication behaviors interrelate. Our results suggest that it is not enough to simply say that they are mutually reinforcing, as We do not mean to suggest that all political communication research must adopt the panel design or undertake other efforts to tease out causal directions between attitudes and behaviors. In those cases where tests of directionality are not possible (e.g., Taken together, we believe these insights demonstrate the importance of future research investigating, in the greatest detail possible, the mechanisms that underlie the complex causal forces at work in the relationship between citizens, each other, and processes of communication. Even as communication itself changes through the advent of new technologies, behavior and attitudes remain relatively stable constructs, and basic theories of how the two interrelate continue to be of use in understanding how our civic beliefs and behaviors mutually influence one another. In the end, such knowledge will help us understand the communicative practices and attitudes that sustain-or potentially undermine-engaged democratic citizenship.