The policy basis of group sentiments

Abstract

Although influential models of public opinion hold that group sentiments play an important role in shaping political beliefs, they often assume that group attitudes stem from socialization and are thus exogenous to politics. We challenge this assumption, arguing that group attitudes may themselves be the consequence of political views. Across three survey experiments that each uses a unique social group–issue pair, we consistently demonstrate that attitudes toward groups are influenced by information about the groups' policy views. These findings persist even when accounting for potential partisan signaling. Altogether, these results show that group sentiments should not be regarded as wholly exogenous to policy concerns and suggest that the use of group-based heuristics can be consistent with instrumental models of public opinion.

Group attitudes have long been described as central objects in political belief systems that shape and constrain political attitudes. Scholars have proposed a variety of group-centric models of public opinion (Miller et al., 1991; Wlezien and Miller, 1997; Green et al., 2002; Kane et al., 2021) and offered a wealth of evidence that group attitudes shape issue attitudes. For example, there is evidence that views of Muslims influenced support for the War on Terror (Sides and Gross, 2013), feelings toward African Americans affect support for welfare (Gilens, 1996), and that attitudes toward Latinos affect support for immigration (Ramirez and Peterson, 2020). Common in this literature is a tendency to assume, either explicitly or implicitly, that group attitudes “are acquired early in life and represent long-standing predispositions that are then capable of shifting political attitudes” (Elder and O'Brian, 2022, 1409). Thus, in many common theoretical and statistical models, group attitudes are assumed to be exogenous to policy attitudes and other instrumental concerns.

We, however, challenge this notion. We argue that just as policy preferences shape opinions about individuals (Rogowski and Sutherland, 2016; Webster and Abramowitz, 2017; Clifford, 2020; Orr and Huber, 2020; Lelkes, 2021; Goren, 2022; Simas, 2023), they should also drive perceptions of social groups. Across three experimental studies using three different groups and issues, we consistently show that information about a group's support for an issue significantly affects the favorability of that group. These results hold even when accounting for potential partisan signaling. So while group attitudes can play an important role in shaping opinions and behaviors, our work shows that the causal arrow also runs the other way. Thus, our findings challenge the idea that contemporary American politics is mere tribalism and suggest that emphasizing areas of issue agreement may be a fruitful path for ameliorating inter-group conflict and animosity.

1. Theory

Our basic argument is that people evaluate social groups, in part, based on the perception of shared political interests. For this to happen, citizens must (1) know where social groups tend to stand on issues; and (2) hold meaningful attitudes on those same issues. The first claim is crucial to the group-centrality literature itself, as that knowledge is necessary for group attitudes to affect political attitudes. For example, Elder and O'Brian (2022, 1422) state that “[m]any people are knowledgeable about the types of social groups that support or oppose policies; this knowledge has historically exceeded knowledge of where parties or ideological groups stand on those same issues.” And indeed, there is growing evidence of individuals' capabilities to accurately connect various groups to parties, policy preferences, and vote choices (Rothschild et al., 2019; Orr and Huber, 2020; Kane et al., 2021; Titelman and Lauderdale, 2021). Thus, people do seem to hold the knowledge required for the reverse causal process.

The second premise, that people hold meaningful issue attitudes, is more contentious. Although many citizens lack ideologically constrained belief systems (Converse, 1964; Kinder and Kalmoe, 2017), they often hold strong and meaningful attitudes on specific issues. A wide range of evidence supports the existence of “issue publics,” or groups of respondents who care deeply about a particular issue (Krosnick, 1990). Citizens may come to hold particularly strong views on an issue due to perceived self-interest (Boninger et al., 1995), or their moral beliefs and values (Ryan, 2014; Skitka and Morgan, 2014). Moreover, some types of issues are particularly “easy” in that they require little knowledge or awareness to make a connection between an issue position and core values (Carmines and Stimson, 1980; Johnston and Wronski, 2015; Ryan and Ehlinger, 2023). Thus, most people seem to hold some meaningful policy attitudes, even if they are not ideologically constrained.

Taken together, there is ample reason to expect that group attitudes may, in part, be shaped by the perception of shared political interests. To be sure, we are not the first to advance such a hypothesis. A number of recent studies have challenged the presumed causal role of group attitudes, primarily through the use of panel data. For example, partisanship may cause racial attitudes as much or more than the reverse process (Engelhardt, 2021). And although issue attitudes are often seen as a consequence of partisan identity (Barber and Pope, 2019; Freeder et al., 2019), views on culture war issues can cause partisan identity (Goren and Chapp, 2017). Finally, turning a classic finding on its head, Goren (2022) finds that views on welfare influence attitudes toward African Americans, a group that is strongly associated with the policy.

Extending this line of literature, we are the first to explicitly test how the favorability of a social group is influenced by whether an individual shares the policy views of that group. We do so across multiple social groups and political issues, demonstrating the broad applicability of our theory. Moreover, instead of using panel data, we use pre-registered experimental designs that allow us to isolate the causal effects of shared policy attitudes and show that they operate even when accounting for partisanship. This aspect of our work is especially important given debates about whether individuals actually care about policy attitudes or just the partisan identities they signal (Dias and Lelkes, 2021; Orr and Huber 2021; Orr et al., 2023).

2. Evidence from three studies

2.1 Design and measures

Between February 2022 and March 2023, we conducted three experimental studies among three different samples.¹ All three studies were approved by the institutional review board at the University of Houston. Studies 2 and 3 follow a pre-registration plan.² Each study featured a different social group and a different issue. Although it is extremely difficult to separately estimate the effects of identity and policy preferences (Fowler, 2020; Rogers, 2020; Orr and Huber 2021; Orr et al., 2023), all our experimental groups are explicitly shown the partisanship of the group, while only the treatment groups receive the additional information that the group holds an opinion that runs counter to the party. Since the policy information conflicts with what should be inferred from the party label, this setup reduces problems with pre-treatment and addresses concerns that any effects of issue information are simply due to the fact that it is signaling partisan identity.

To avoid providing misinformation, we thus selected three groups (Vietnamese Americans, Catholics, and Mormons) for whom public opinion data show to have a policy preference (support for gun control, opposition to transgender rights, or support for environmental regulation) that conflicts with the party's stance. While chosen for this more pragmatic reason, these groups offer variation in the strength of their stereotypical partisan association, as Kane et al. (2021) show that Mormons are clearly perceived as Republican, perceptions of Catholics are more mixed, and Vietnamese Americans actually counter the stereotypical association of Asians with the Democratic Party. Having this variation will allow us to speak to the generalizability of our results, and findings of consistent effects across all three groups should help allay potential concerns that the effects of issue information are contingent on the strength of the association between a social group and a party.

Each study follows the same basic structure. First, respondents reported their position on the focal issue used in the treatments (see Table 1 for wordings), followed by measures of attitude strength (studies 2 and 3 only), which make up the moderating variable. Respondents also reported feelings toward several social groups, including the target group,³ and their partisan identification. These measures are used as pre-treatment control variables to increase the precision of our estimates (Clifford et al., 2021). In all three studies, respondents then completed unrelated content prior to the experiment.

Table 1. Summary of experimental designs

To introduce the experiment, respondents were told that researchers are interested in their opinions on a social group that plays an important role in politics. As noted above, respondents in both conditions were told about the partisan distribution of the focal group, while only those in the treatment condition received the additional information about the group's party-inconsistent position on the featured issue. Following the treatment, respondents completed several questions capturing their attitudes toward the target group, as well as two items serving as manipulation checks.

Even though our experiments are designed to minimize the potential for issue information to only operate by signaling partisan identity, this is again a concern we cannot completely eliminate (Dias and Lelkes, 2021). Likewise, the issues themselves may evoke separate group sentiments (Conover, 1988), and the partisan stereotypes of those groups may also influence responses. We further address these concerns by assessing manipulation checks (Dafoe et al., 2018), and by estimating alternative models that account for partisan inferences.

2.2 Measures

In study 1, respondents reported their position on gun control on a five-point scale. In studies 2 and 3, respondents reported their issue position on a seven-point scale, then how important the issue is to them personally and their level of moral conviction on the issue (Skitka, 2010; Ryan 2014).⁴ For these studies, following our pre-registration plan for study 2, we rescale the attitude position variable to range from −1 to 1, average the two measures of attitude strength, then multiply the attitude position measure by the attitude strength measure to construct a single measure of issue attitude (for a similar approach, see Taber and Lodge 2006).

To measure group attitudes, respondents were asked how well the social group shares their values (five-point scale), how close they feel to the group (Mason and Wronski, 2018) and their favorability of the group (seven-point scale). For our primary outcome, we follow our pre-registered plan and recode these three variables to range from 0 to 1, then averaged them (αs > 0.71). Finally, as a manipulation check, respondents were asked to estimate the percentage of the focal social group favoring the target policy and the breakdown of partisan identification among that group.

3. Results

Following the study 2 analysis plan, we predict the group attitudes index as a function of treatment assignment, issue attitudes, and an interaction between the two. Additionally, we control for pre-treatment measures of partisan identity and pre-treatment feelings toward the featured group. Figure 1 shows the marginal effects of the treatment as moderated by the issue attitude (full model results are available in the Appendix). As expected, in all three studies there is a significant interaction between the treatment and the respondent's issue attitude (p's < 0.004).

Figure 1. Marginal effects of group issue information by respondent issue attitudes. Main plots show the effects of the group issue stance treatments and the 95 percent confidence intervals for those estimates. Minor plots show the distribution of issue attitudes in our samples. See the Appendix for full model results.

Starting with the left-hand panel, among respondents who strongly favor stricter gun control laws, the treatment increases the favorability of Vietnamese Americans by 0.07 (p < 0.001), or about 0.35 standard deviations. Among those who strongly oppose stricter gun control laws, the treatment decreases the favorability of Vietnamese Americans by 0.08 (p < 0.001), or about 0.42 standard deviations. Moving to the middle panel, among respondents who oppose a transgender bathroom bill, the treatment decreases the favorability of Catholics by 0.08 (p = 0.001), or about 0.36 standard deviations. Among those who favor the bathroom bill, the treatment increased favorability by 0.05, or about 0.23 standard deviations (p = 0.026). Turning to the right-hand panel, effects are similarly strong at either end of the attitude scale. Among those who strongly favor (oppose) environmental regulation, the treatment increases (decreases) the favorability of Mormons by 0.12 (p < 0.003), or about 0.28 standard deviations. We find substantively similar results, though weaker in magnitude, when we examine only the favorability outcome (see the Appendix for details). Notably, all of these effects are substantially larger when accounting for non-compliance (Harden et al., 2019; see the Appendix for details), which likely arises due to pre-treatment and satisficing.

An alternative explanation is that the treatments are affecting group attitudes largely because they are sending signals about the group's partisanship (Dias and Lelkes, 2021).⁵ We minimized this concern by design by providing information about group partisanship in all conditions. Manipulation checks available in the Appendix show that in studies 2 and 3, there was no evidence that the policy treatment moved perceptions of group partisanship (study 2: p = 0.72; study 3: p = 0.37). In study 1, however, the treatment shifted perceived support for the Republican Party by about four percentage points (p = 0.002).⁶ So to address this possible confound, we re-estimate the initial models while also including an interaction between respondent partisanship and the treatment. If the treatment works primarily by affecting perceptions of group partisanship, then we should see a strong interaction between the treatment and partisanship, which eliminates the interaction between the treatment and issue attitudes.

The results of these models (see the Appendix) reveal some evidence of partisan signaling, though our core findings are unchanged. In both studies 1 and 2, the interaction between the treatment and partisanship is significant (p's < 0.029). In study 3, however, the interaction between the treatment and partisanship is both substantively and statistically indistinguishable from zero (b = 0.00, p = 0.849). Most importantly, in all three studies, the interactions between the treatment and issue attitudes remain statistically significant (p < 0.01) and show little change in magnitude. Our experimental design does not allow us to estimate the relative importance of party and policy cues, but the persistence of the effects of policy agreement support our claim that partisan signaling cannot explain away sizable effects of issue attitudes on group attitudes.

4. Discussion

Scholars have long acknowledged the role that group attachments and sentiments play in shaping political attitudes and identifications. We contribute to this line of research not by disputing the role of those sentiments, but by probing deeper into their roots. Across three samples and three different group–issue pairs, we consistently show that feelings about social groups are shaped by agreement with the policies those groups endorse. These findings show that the common assumption that group sentiments are exogenous to policy concerns needs to be reconsidered and suggest the implications of some prior studies should be revisited. For example, recent work showing a relationship between group affect and partisan identification argues that this connection “suggests that political decisions can often be made on the basis of liking or disliking groups, rather than purely rational self-interest” (Kane et al., 2021, 1784). But since our evidence suggests that like or dislike of those groups is at least partially due to shared political interests, it appears that the public may in fact be more rational than previously assumed.

We have focused on the USA, as group theory plays a prominent role in explaining shifts in party coalitions over time (Achen and Bartels, 2016), but we expect our findings to generalize beyond the USA. But, of course, our experiments are somewhat limited. Although we find consistent results across multiple groups and issues, future work should expand the design and further test (1) the conditions under which policy information may be more or less informative, (2) the relative importance of policy and non-policy factors, and (3) the generalizability across groups and issues. Still, our work makes an important contribution by highlighting the need to better explore the more instrumental component of group sentiments. Group attitudes surely play a causal role in politics, but as works on partisan intoxication show (Fowler, 2020; Rogers, 2020), identity and policy explanations should not be treated as mutually exclusive. As such, failure to adequately acknowledge the potential role of policy agreement can lead to mischaracterization of the nature of contemporary partisan politics in the USA. While any attempts to address affective polarization will undoubtedly be confronted with elements of pure “teamsmanship,” the possibility of appealing to common group interests does open up broader avenues for dealing with the negative consequences of the growing divide.