Selecting Out of “Politics”: The Self-Fulfilling Role of Conflict Expectation

Abstract

In recent decades, the term “politics” has become almost synonymous with conflict. Results from eight studies show that individuals averse to conflict tend to select out of surveys and discussions explicitly labeled as “political.” This suggests that the inferences researchers draw from “political” surveys, as well as the impressions average Americans draw from explicitly “political” discussions, will be systematically biased toward conflict. We find little evidence that these effects can be attenuated by emphasizing deliberative norms. However, conflict averse individuals are more willing to discuss ostensibly political topics such as the economy, climate change, and racial inequality, despite reluctance to discuss “politics” explicitly. Moreover, they express greater interest in politics when it is defined in terms of laws and policies and debate is deemphasized. Overall, these findings suggest the expectation of conflict may have a self-fulfilling effect, as contexts deemed explicitly “political” will be composed primarily of conflict seekers.

INTRODUCTION

Many readers will be familiar with Aristotle’s statement that a person is “by nature a political animal,” but some may not be aware of what he actually meant by it. Against the backdrop of contemporary American politics, one might interpret Aristotle to mean that people are by nature conniving, power-hungry, and self-interested. But Aristotle’s intent was different, explaining that what distinguishes people is their “perception of good and bad and right and wrong and the other moral qualities, and it is partnership in these things that makes a household and a city-state” (Aristotle 2017). When Aristotle claimed that people are by nature political, he meant that they are naturally social, cooperative, and community-oriented. To Aristotle, politics was not the art of manipulation (Riker 1986), but rather a collective effort to attain the good life through the construction of a just society.

The etymology of the word politics is long and complicated, with theorists including Aristotle (2017), Arendt (1958), Rousseau (2018), and Schmidt (1996) emphasizing varying degrees of cooperation versus conflict. But, since Lasswell (1936), most introductory American politics texts have defined politics as “who gets what, when, and how,” imparting a notably conflictual flavor to the term. Since then, the literature has tended to focus on conflict (see, e.g., Del Ponte, Kline, and Ryan 2020). Thus, while politics might reasonably be understood as the act of collective problem-solving, akin to the way Aristotle originally saw it, both textbooks and personal experience teach people to think of politics in more conflictual terms. It should then come as no surprise that studies have shown people associate the term “politics” with conflict (Fitzgerald 2013).

Perceptions aside, neither definition is objectively correct or incorrect since the definition of politics is a social construct. As such, it only exists intersubjectively within our collective imagination. This realization is far from revolutionary, but it has important consequences: the norms and expectations we associate with politics are likely to play a major role in determining who selects into and out of contexts explicitly deemed political.

Studies suggest that people often avoid political discussion and other forms of engagement due to concerns about conflict (Conover, Searing, and Crewe 2002; Mutz 2006; Sydnor 2019; Testa, Hibbing, and Ritchie 2014; Ulbig and Funk 1999; Wolak 2022) At the same time, however, people frequently encounter ideas that are ostensibly political (Fitzgerald 2013) and engage in ostensibly political conservations (Cramer Walsh 2004; Eliasoph 1998) without recognizing them as “political.” How, then, should we interpret evidence suggesting certain people dislike politics (Hibbing and Theiss-Morse 2002; Ulbig and Funk 1999)? If a person’s goal is to avoid conflict, will they also avoid the range of topics we typically categorize as political or is the term “politics” the main deterrent?

With these concerns in mind, this paper examines how the association between politics and conflict influences people’s willingness to share their opinions in contexts deemed “political.” We show that the mere mention of “politics” leads conflict seekers to select in while leading conflict avoiders to select out. Contrary to our a priori expectations, conflict avoidant individuals tended to opt out of explicitly political contexts even when deliberative norms are emphasized. On the other hand, these people are more willing to discuss ostensibly political topics, such as the economy, religion, climate change, and racial inequality. Finally, we show that while conflict averse individuals report lower levels of interest in “politics,” they report more interest when we define “politics” in more substantive terms. This suggests conflict avoiders are not averse to the substance of politics—the topics we typically categorize as political—but instead turned off by the idea of “politics”—the expectation of conflict that accompanies this label.

These findings have important implications. If, for example, surveys recruit by highlighting that their subject matter is “political,” this labeling may influence compliance (e.g., Esterling, Neblo, and Lazer 2011) potentially leading to widespread overestimation of the public’s proclivity for conflict. Additionally, while reticence to engage in politics may be interpreted as indifference, our results reinforce the idea that people often avoid politics intentionally (Carlson and Settle 2022; Eliasoph 1998). This, it appears, is due to the tendency to associate politics with conflict, not a lack of interest in addressing social problems, potentially changing the way we think about political disengagement.

Our findings also hold implications for the types of voices we are most likely to hear in the public sphere. One direct outcome is the possibility that conflict seekers are more likely to be represented in political spaces—but the implications may still be broader. Conflict orientation, for example, is associated with gender, which suggests another possible pattern to whose voices are especially diminished. Overall, these findings suggest expectations of conflict may lead to a self-fulfilling prophecy whereby self-selection effects continually reinforce these expectations, turning them into reality.

ENTERING POLITICAL SPACES

We are not the first to suggest that self-selection may influence who participates in surveys (e.g., Cavari and Freedman 2023; Heckman 1979) and political discussions (e.g., Mutz 2006). Research shows that conflict orientation affects this self-selection: people who are averse to conflict are especially likely to avoid political interactions (Conover, Searing, and Crewe 2002; Mutz 2006; Sydnor 2019; Vraga et al. 2015; Wang et al. 2017; Wolak 2022). What can bring these conflict averse individuals back into political spaces?

One possibility is that deliberative norms may help to diversify the range of voices heard in the public sphere. The goal of deliberation is consensus-building via the open-minded exchange of ideas (e.g., Jacobs, Cook, and Delli Carpini 2009; Guttman and Thompson 2004). Thus, a political event described as having genuinely deliberative goals could be appealing to many people who might otherwise be turned off by “politics” (Neblo et al. 2010). Yet research suggests that formal deliberation may be subject to the same self-selection issues as other political contexts (e.g., Karpowitz, Mendelberg, and Shaker 2012; Karpowitz and Raphael 2014; Luskin, Fishkin, and Jowell 2002; Ryfe 2005; Sanders 1997). Indeed, Neblo et al. (2010) find some evidence that those who are conflict avoidant are less interested in deliberation.

We believe that this research on deliberation carries implications beyond the deliberative context to the relationship between conflict avoidance and selection into a variety of political spaces (including informal conversations and even surveys). The key, we believe, is to recognize the centrality of the “pictures in our heads” (Lippman 1922). While many people may be averse to conflict, this will only lead them to avoid political contexts if they picture these contexts as conflictual. Deliberation works when a shared expectation of civility and open-mindedness can be forged, but this is not always easy; the promise of deliberation may not be enough to overcome expectations once a context has been deemed to be about “politics.”

At the same time, when left to their own devices people do discuss topics one might categorize as “political” (Cramer Walsh 2004). As Eliasoph (1998) argues, people may avoid explicit mention of politics (and explicitly political spaces), even as they are willing to address topics that are ostensibly political (outside of explicitly political spaces). This suggests a possibility: perhaps leading people back to political contexts requires more than a promise of open-mindedness and civility, but rather a focus on the substance of politics—issues, policies, and problems to be solved. We consider these ideas in the pages that follow. As we will show, some people do avoid politics, yet they grow more engaged when we highlight its substance.

EMPIRICAL APPROACH

Our empirical approach relies on eight studies outlined in Table 1, as well as additional studies.¹ Given the number of studies, we have to leave some details out of the main manuscript. Those details can be found in one of two online appendices—the Supplementary Material (SM) available on the journal’s website and a Dataverse Appendix (DA) available on the APSR Dataverse (Groenendyk et al. 2024). Data and replication code are also available at the Dataverse (Groenendyk et al. 2024).

Table 1. Study Samples and Designs

Note: Study pre-registrations are here: Survey Selection-Dialogue: (we made a small typo in the description of one of the treatments in our original pre-registration, therefore, we filed a second pre-registration prior to fielding as a correction):

Original: https://aspredicted.org/QFF_LWS; Correction: https://aspredicted.org/59F_JCN; Dinner Conversation 2: https://osf.io/m6q52/?view_only=9d88a309654d4479a5ac7f1273b6eaa6; Conversation on Issues: https://aspredicted.org/H8V_N4S; Mechanism Check: https://aspredicted.org/1YT_K8V.

For Interest in Politics, we include our original application for Time-Sharing Experiments for the Social Sciences (TESS) which funded this study; the application is also here: [https://tessexperiments.org/study/groenendykS79]. This information is also included in DA 10.

ResearchNow has since merged with Survey Sampling International (SSI) to form Dynata. It was a non-probability sample that aimed to reflect the U.S. on demographic characteristics.

Study Conversation Topics is part of Pew’s American Trends Panel—meaning that it was not initiated, designed, nor fielded by the authors of this manuscript.

The initial studies were conducted in an exploratory manner with the goal of tracking who is most likely to select into and out of political contexts. Drawing on these initial results, we then designed the remaining studies to test specific expectations based on the previous patterns, which we pre-registered prior to fielding. Our studies are also conducted across a variety of survey companies and online platforms (Table 1)—some with convenience samples (e.g., MTurk), others with more nationally representative samples (e.g., NORC). We also rely on a study designed and fielded by Pew.² We list the sample demographics for each study in DA Section 3, where we also address the generalizability of our results given our reliance on convenience samples.³

We present the results from these studies in a series of steps. First, we show that people tend to select out of surveys and conversations labeled “political.” Second, we turn to conflict orientation, first presenting descriptive and correlational patterns and then moving on to experimental treatment effects, all showing that conflict avoiders select out. Finally, we consider how to bring conflict averse individuals back into politics. We first attempt to bring them back to politics by focusing on deliberative discourse. Then, we shift to a different approach—focusing on separating the term “politics” from the substance of politics. Full details for each study appear in DA Section 1. In the remainder of the manuscript, we use the descriptors in Table 1 (e.g., “Survey Selection Experiment 1”) to distinguish between studies.

We note that our studies are not without limitations. Our work is conducted using survey experiments, yet how people report their willingness to discuss politics in surveys may differ from the way they behave in social contexts (e.g., Cramer Walsh 2004; Morey, Eveland, and Hutchens 2012). Our work cannot speak to what participation looks like within actual conversations. Still, our studies do speak to participants’ willingness to enter explicitly political spaces. Understanding whose voices are most likely to be found in these types of contexts is also important for understanding political expression in a democracy.

SELECTING OUT OF POLITICS

We begin with a basic question: do people avoid politics? This follows from the premise that people associate politics with conflict (Fitzgerald 2013), and, perhaps, find politics stressful (Blanton, Strauts, and Perez 2012). Therefore, we begin by examining treatment effects in studies that randomly assign people to either a political or non-political context and gauge their interest in participating. We first examine how this affects participation in surveys and then broaden our focus to examine willingness to attend social gatherings where politics will be discussed.

Politics Versus Entertainment: Survey Context

Our first study—Politics versus Entertainment — is a 2 $ \times $ 2 survey experiment. The primary manipulation of interest is the lead screen of the survey, and we use this study to consider whether the expectation of politics in the survey lowers study completion rates. All participants were recruited in the same way by ResearchNow, using the survey company’s standard invitation (which does not include the word “politics”). The randomization occurred after entry into the study: those assigned to the politics condition were told that they would answer questions about the media “as it relates to politics,” while those in the entertainment condition were told that they would answer questions about the media “as it relates to entertainment.”

We crossed this labeling manipulation with a manipulation of the survey’s content. Although each respondent was asked the same questions in total, we manipulated whether respondents were asked contentious policy questions at the start of the survey or later in the survey. The purpose of this manipulation was to disentangle the effect of merely labeling the survey as “political” from the effect of priming people with thoughts about substantive policy conflicts and debates from the start.

Participants assigned to the contentious policy condition were asked these questions (e.g., abortion, healthcare policy), immediately after the labeling manipulation—even if they had been told the study was about entertainment.⁴ These contentious policy questions were followed by a battery of non-contentious questions about politics (e.g., voter registration status, political knowledge items) and an entertainment battery (e.g., favorite TV shows, favorite musicians), and these batteries were administered in random order. Participants assigned to the baseline condition (i.e., not primed to think about contentious policies at the immediate outset of the study) received these batteries in an order consistent with the labeling of the survey. If they were told the study was about politics, the survey started with the non-contentious political questions, and the entertainment questions came later. If they were told the study was about entertainment, the survey began with a series of entertainment questions, and the political questions came later. In sum, all participants in the survey received the same questions, but the order was manipulated. It is likely that questions asked earlier in the study will have more influence over completion rates, since likelihood of completion should rise as participants near the end of the study.⁵

As shown in Figure 1, being in the “politics” condition as opposed to the “entertainment” condition increased defection rates. Priming thoughts about contentious policies, by asking those questions early in the study, however, had no significant direct effect or interaction with the political treatment. People were less likely to complete the survey in the “politics” condition regardless of whether the survey began with contentious or less contentious questions. This initial result suggests that it is the association people have with the word politics itself, that shades their experiences and expectations and perhaps not the policies under discussion.

Figure 1. Survey Defection Rates by Condition

Note: Figure shows percent defection across the four experimental conditions—estimates derived from a logit model interacting treatments. Treating low and high contentiousness conditions as separate tests, differences between the entertainment and politics conditions are significant when contentiousness is low (p = 0.03) and approach significance when contentiousness is high (p = 0.13). The product term on the interaction effect is not statistically significant (p = 0.63). When analyzed in a single logit model without interacting treatments, politics is significant (p = 0.01), while contentiousness is not (p = 0.67). See Supplementary material S1.1 for full results table.

Political Versus Non-Political Conversations

Given that the label of politics led people to drop out of the survey in the first study, we next consider whether this pattern extends to people’s willingness to share their opinions in other contexts, such as more casual discussions of politics. To do this, we conducted an experiment in which we manipulated the attributes of a hypothetical dinner invitation. We use this invitation design twice, but, for now, we focus on Dinner Conversation 1. In this study, participants were randomly told a dinner was being organized to discuss either movies or politics.

The politics/movie conditions were crossed with an encouragement prompt: some people were merely invited to attend, while others were randomly assigned to a prompt noting that their host was especially interested in their thoughts. The idea behind this second manipulation was to counteract individual expectations of politics as conflict (Fitzgerald 2013). If the host was interested in the participants’ thoughts, people might be more likely to believe that their voices would be valued, leading to a better discussion experience (e.g., Tyler, Degoey, and Smith 1996). In total this experiment had four groups.⁶

We examine willingness to attend a dinner, because dinner conversations are, perhaps, the most classic forum for informal political discourse.⁷ This study also broadens the potential implications beyond those of our first study: while selection bias in surveys can certainly cause trouble for researchers, selection bias into and out of politics more broadly would raise concerns about the health of the public sphere and democracy. We acknowledge limitations; our “dinner invitation” happens in the context of a survey experiment, rather than a social, interpersonal interaction. At the same time, we believe people’s (un)willingness to accept or decline what is a generally costless invitation within an experiment can shed light on their (un)willingness to enter explicitly political contexts.

In this study, the outcome measure is simple: would you attend this event? Note, the participants are not choosing between movies and politics, rather, they are answering yes or no to a dinner invitation in front of them. Therefore, we begin by presenting the percentage of participants, by condition, who would agree to attend. As illustrated in Figure 2, we see the same pattern regardless of whether participants received an encouraging prompt. In each case, study participants were significantly more likely to accept the dinner invitation if the topic of discussion was movies rather than politics—reflecting the patterns in the Politics versus Entertainment study. We do not see evidence that telling people that their contributions are valued and encouraged leads to greater participation. These patterns offer suggestive evidence that people are avoiding politics—although this study cannot capture motivations for doing so; we delve into these patterns more directly in the next sections.

Figure 2. The Effect of Politics on Dinner Attendance

Note: Each bar represents the percentage of people in an experiment condition who selected the dinner invitation. Estimates are calculated from a logit model with the treatments and an interaction between the treatments. Significantly fewer people accepted the dinner invitation when the topic of discussion was politics (p < 0.01). The prompt condition never reaches statistical significance (p = 0.84 in movie condition and p = 0.85 in politics condition). See Supplementary material S2.1 for full coefficient estimates.

Who Selects In, and Who Selects Out?

Two studies thus far show that the expectation of “politics” leads people, on average, to select out of surveys and informal discussions. Our next step, then, is to track more systematically who selects in, and who selects out of political contexts. Studies show that people tend to associate politics with disagreement, incivility, and fighting (Fitzgerald 2013; Klar and Krupnikov 2016). This suggests that people who seek conflict should be more willing to engage in politics while conflict avoiders should be more likely to opt-out (Mutz 2006; Sydnor 2019; Ulbig and Funk 1999; Wolak 2022). If this is the case, it may help to explain why politics often feels so conflictual: it’s not necessarily the case that people are becoming more conflict-oriented, but that conflict-oriented people are selecting into politics while conflict-avoiders are selecting out.

To test this possibility, we turn to studies that include a pre-treatment measure of conflict orientation. There are different ways to capture conflict orientation, and there are differences in measures across our studies. In most of the studies, we rely on a dichotomous measure asking participants whether they enjoy debating with family and friends or avoid disagreement with family and friends (see also Mutz 2006; Ulbig and Funk 1999). However, the NORC study and the Pew study rely on different measures; NORC uses a two-item measure (Wolak 2022) and Pew employs a three-item index measure of feelings toward disagreement, conflict, and challenging other people’s opinions. All items appear in DA 2. In studies that did not include the dichotomous measure, we present our results across all values of the scale.

We analyze the patterns in two steps. First, we set aside our various experimental manipulations and focus on who selects in and out of explicitly political surveys and discussions: are those who select in more conflict-oriented than their counterparts who select out? As a second step, we turn to treatment effects.

Selecting Political Surveys: Correlational Results

Unlike our first study, where we measured survey retention, in both Survey Selection studies (both conducted with MTurk samples) we measured which type of survey participants would rather take. In Survey Selection Experiment 1, participants were asked to select a survey type at the start of the study and were then routed into surveys that matched these descriptions. In Survey Selection Experiment 2, participants were asked which type of survey they would prefer to take in the future. In both cases, this choice constitutes our measure of survey selection.

In both studies, participants could choose between a survey related to politics and one about a different topic. The alternative topics were deliberately selected to be somewhat “boring.” In both studies, the main manipulation was the title of the political survey. In Survey Selection Experiment 1, participants could choose between a survey about politics and consumer products. One condition referred to a survey about politics, a second condition referred to a survey about political debates and a third condition attempted to reduce the expectation of conflict by altering the survey title to emphasize deliberation. The title of the consumer products survey did not vary. After making their choice, those who selected the political survey answered a series of questions about politics, while those who selected the consumer survey answered a series of questions about consumer products. Afterward, the survey content converged.⁸

In Survey Selection Experiment 2, participants were asked about their willingness to participate in a future study. Again, participants could select which survey they would be most interested in taking, this time selecting between surveys about consumer products, health behaviors, politics, and “none of the above.” We once again randomized the labeling of the political survey. In one condition the political survey was listed as “Survey about political issues;” in another, we noted that it was political issues on which there is contentious debate, and in a third condition the title noted that it was a survey on political issues on which there is respectful dialogue.

We will return to the manipulations—both for Survey Selection Experiments 1 and 2—later in this manuscript. For now, we set aside the experimental variation in the title of the political survey and consider the conflict orientations among those who selected the “political survey” (regardless of wording) compared to those who selected another survey type (or “none of the above”, in the case of the second study). Since in these studies participants selected between politics and other options—rather than reporting only whether they would participate in a political survey—we can distinguish between a general unwillingness to take surveys, and an unwillingness to take political surveys.

We present the results in Figure 3. Both studies use the dichotomous measure of conflict avoidance, coded 1 if the participant is conflict avoidant, 0 if they are conflict seeking. Across the two studies, we see nearly identical patterns: there are significantly fewer conflict avoidant people among those who selected the political survey compared to those who selected the non-political survey options.⁹

Figure 3. Conflict Avoidance Levels Among Those Who Choice Political/Non-Political Survey

Note: Bars represent percentage of respondents who were conflict avoidant among those who selected a particular survey. Those who select “none of the above” are included among those who chose the non-political survey, but the results are substantively similar if “none of the above” is excluded. The title of the politics survey was experimentally manipulated, but those conditions are combined for this analysis. Due to self-selection, the composition of the political group was significantly less conflict avoidant/more conflict seeking compared to the non-political group in both studies (p < 0.01). Additional results are in Supplementary materials S3.1 and S4.1.

Selecting into a Political Dinner: Causal Test

The results in Figure 3 suggest that people who select into political surveys tend to be more conflict-oriented than those who do not, but what about people who participate in informal political discussions? Using our second dinner invitation experiment, which measures conflict avoidance pre-treatment, we examine whether the same pattern holds. We have thus far shown that people who select into politics tend to be more conflict-oriented, but those results have been correlative. As a next step, we turn to an experiment where the topic of discussion is randomly assigned.

The design of Dinner Conversation Experiment 2 followed that of the first dinner conversation study. Whereas the survey selection experiments allowed study participants to choose between political or non-political options, the dinner studies manipulated whether the dinner conversation would focus on politics or movies, and study participants were simply given the option to accept or decline the invitation. We note that, much like our first dinner conversation study, this study contained varying invitation prompts—with one, again, encouraging participation more directly.

As a first step, just as we did with the Survey Selection Experiments, we consider levels of conflict avoidance among participants by their responses to the dinner invitation. Once again, we use a dichotomous measure of conflict avoidance, where 1 means someone is conflict avoidant and 0 means otherwise. We find that people who accepted the politics invitation are significantly less conflict avoidant than those who accepted the movie invitation. Half of the respondents who accepted the movie invitation were conflicted-avoidant compared to about 37% of the respondents who accepted the politics invitation.¹⁰

These initial results are suggestive, but since conflict avoidance is measured pre-treatment, a more direct test of our model is to examine how conflict avoidance interacts with the treatment to influence the likelihood of invitation acceptance. Figure 4 shows the percentage of people who accepted the invitation to the dinner when told politics would be discussed compared to those who were told movies would be discussed. We see no effect among conflict seekers: about 70% of them accepted the invitation, regardless of the topic of discussion. However, we see a significant effect among conflict avoiders, only 31% of whom accepted the invitation to the dinner where politics would be discussed compared to 53% of those who accepted the invitation to the dinner where movies would be discussed.

Figure 4. The Effect of the Politics Treatment on Dinner Composition

Note: Acceptance rates for two experimental conditions among conflict avoiders/seekers; the other experimental factor, which we discuss in the next section, is treated as a nuisance parameter. Estimates from a logit model which includes an interaction between politics treatment and conflict avoidance. Treatment differences: conflict avoidant: 22 percentage points, p < 0.01; conflict seeking: 0.04 percentage points, p = 0.327. See Supplementary material S2.3 for full coefficient estimates.

Robustness Checks

Since conflict orientation cannot be randomly assigned, we follow Kam and Trussler (2017) and use pre-treatment measures to control for potential confounds in both the Survey Selection Experiments and Dinner Party Experiment 2. In particular, we use education to proxy knowledge across all three studies and account for interest in movies in Dinner Party Experiment 2. We see that the inclusion of these controls does not change our findings, even when interacted with the treatment variable. We present these results in Supplementary material S2.5 (Dinner Party Experiment 2) and Supplementary materials S3.3 and S4.3 (Survey Selection studies).

SELECTING BACK INTO POLITICS

Thus far, we have shown that people who avoid conflict tend to select out of explicitly political contexts. We now turn to the question of how such effects might be overcome. Specifically, if “political” contexts are expected to be conflictual, might characterizing these contexts in ways that suggest non-conflictual discourse attenuate the effect? One way to address this concern is to underscore that, while opinions may differ, the exchange will follow the norms of deliberative democracy. Indeed, this was the a priori expectation guiding the design of one set of our studies, and we pre-registered this prediction.

We tested this prediction using a series of treatments in which we manipulated the description of the survey or dinner party. In the Survey Selection Experiments, we randomly assigned participants to different survey choice sets. In Survey Selection 1, one option was always a “consumer products survey” and the other was either a “political survey,” a “political debate survey,” or a “political deliberation survey.” In Survey Selection 2, they chose between five different options, including “none of the above.” Again, the only option that varied was the political option: a survey about “political issue,” “political issues on which there is contentious debate,” or “political issues on which there is respectful dialogue.” In Dinner Conversation 1, we randomly assigned whether participants received a prompt explaining that the host was especially interested in their voice: “I contacted you because I’d really like to hear your thoughts.” We presented the results of this manipulation in Figure 2. In Dinner Conversation 2, we randomly assigned respondents to a treatment in which respondents were told the host encourages open-minded deliberation: “I want you to know that I make it a point to encourage open-minded deliberation, not arguing at these dinners.” None of these efforts increased participation among the conflict avoidant.¹¹

When we emphasized debate in our Survey Selection Experiments, we saw generally inconsistent results. Although mentioning debate decreased selection of the politics survey among conflict avoiders in Survey Selection Experiment 1 (p < 0.01), mentioning the word “contentious” had null effects in Survey Selection Experiment 2 (p = 0.359). We see these shifts as consistent with the possibility that terms like “debate” and “contentious” reinforce conflict avoiders’ fears about politics—which is why they have negative and null effects on the selection of the political survey.

More surprising, perhaps, is that emphasizing deliberative norms (e.g., respectful dialogue, emphasizing that participant’ voice will be valued, etc.) did not increase conflict avoiders’ willingness to participate. Across three studies, these conditions either produced null results or decreased willingness to discuss politics. These patterns do not match our pre-registered expectations. One possibility is that by emphasizing the lack of contentiousness, we may actually be drawing attention to the possibility of contentiousness. Another possibility is that our use of “deliberation” was unusual for participants—for example, some may have wondered what we meant by a discussion of movies that was deliberative—therefore we conducted a post hoc check.

Our post hoc check examined how people perceive various forms of political discussion (see full description in DA 1 [design] and DA 7 [results]).¹² In this check, we focused on a hypothetical conversation to emphasize either that it is “deliberative” or “open-minded.” We also manipulated whether the conversation was about politics or “policy”, though we found no difference. What mattered seemed to be the terms about the nature of the conversation. We found that among conflict avoiders, referring to the discussion as “open-minded” had no significant effect on expectations of contentiousness, but referring to the conversation as “deliberative” led people to expect it to be more contentious.

In short, despite our a priori expectations, using abstract words like “deliberation” in our treatments, or even describing a more deliberative interaction, did not help to overcome the perceptions of conflict associated with politics in our studies. In retrospect, we see this as a result that speaks more to how political contexts are described—rather than deliberation more broadly. Our post hoc check suggests people associate the term “deliberation” with high levels of contentiousness. It is possible, however, that people may welcome deliberation in practice. As Sydnor, Tesmer, and Peterson (2022) conclude, “simply promising” that a discussion will have deliberative elements “is not enough to bring the conflict-avoidant into the political arena” (9). This promise is not the same as creating an environment that is genuinely deliberative and makes conflict avoiders comfortable, which is something that occurs during deliberation but cannot be manipulated with a term in a study.

In sum, our results suggest that people who are conflict avoidant are more likely to select out of politics, and emphasizing deliberative norms does not seem to bring them back. Therefore, as a next step, we consider politics from a different perspective. As Cramer Walsh (2004) finds, Americans think of politics, not as a topic, but as a descriptor indicating “impasse and petty griping” (39). As a result, topics that appear ostensibly political to researchers may not appear so to the average American. The question then, is whether people who are averse to conflict might be willing to engage with ostensibly political topics even as they avoid explicit discussion of “politics” (Eliasoph 1998). We consider these ideas in the studies that follow.

Explicit Discussion of “Politics” Versus Discussion of Ostensibly Political Topics

Our final three studies examine how conflict orientation affects willingness to engage with the substance of politics, even if one dislikes the idea of politics. To do this, we use three different approaches. The first two studies focus on comfort with discussions about politics versus other topics. Here, we first turn to data collected by Pew’s American Trends Panel, which provides several advantages. This is a large, nationally representative sample of Americans, and relies on a design that allows us to examine a broad set of conversational scenarios, beyond politics and movies. However, in the Pew study, all participants were asked about each conversation topic, which means we cannot be sure whether responses are independent. To address this potential concern, we conducted an experiment in which people were randomly assigned to only one conversation topic. We then compared willingness to participate in the conversation across topic conditions. The benefit of this experimental approach compared to Pew’s approach is that it ensured responses to the various discussion topics could not affect one another.

Our third study relies on a different approach. In our final experimental study, we asked people about their interest in politics, but rather than focusing on specific issues (like in the two conversation studies) we randomly assigned what we meant by politics. In other words, rather than letting people imagine what a political context might entail, we define it for them. Through this design, this final study also allows us to better disentangle aversion to conflict in politics from aversion to politics as a topic.

Comfort with Different Conversation Topics

We begin with data from Pew. In 2019, Pew used a wave of its American Trends Panel (N = 10,170) to consider people’s comfort with political contexts. A Pew report found that people were generally dissatisfied with political discourse, in part because of its conflictual nature (Pew Research Center 2019). Here, we delve more deeply into these patterns and rely on a series of six questions, presented in random order. Each question began with the same introduction: “Thinking about a conversation you might have with someone you don’t know well, how comfortable would you feel discussing each of the following?” Then, respondents were asked to mark how comfortable they would be with the following topics: movies and television, sports, the weather, religion, the economy, and finally, politics. The final topic—politics—randomly assigned respondents to one of two options, a discussion of “politics” explicitly or a discussion of Donald Trump, famous for inciting conflict and violence.¹³

Most of these topics (e.g., movies and television, sports, and weather) are not ostensibly political. However, a topic like the economy has a clearer link to politics—indeed “the economy” is often cited in political surveys as the most important problem facing the country, and economic performance is typically among the best predictors of election outcomes.¹⁴ Likewise, as cultural issues have come to the forefront of American politics, religion—another included topic—has become a major political fault line (Margolis 2017). Another benefit of the Pew study is that conflict avoidance is measured in the American trends panel, making it possible to examine its relationship with these various topics of discussion; we note, however, that Pew measures conflict avoidance somewhat differently than we have so far (see DA 2).

We estimate a series of models that examine the relationship between conflict avoidance and comfort discussing each topic, controlling for potential confounds. For ease of interpretation, we dichotomize the dependent variables (discussion comfort) coded 1 if the respondent is comfortable and 0 if the respondent is not comfortable regardless of level of comfort/discomfort.¹⁵ Given that the measure of conflict avoidance is not binary, in Figure 5, we show the predicted probability of comfort across the full range of values in the conflict avoidance measure. Those higher in conflict avoidance (as we would expect) are generally less likely to engage in any of the conversations. Importantly, however, the effect of conflict orientation varies by discussion topic. While conflict avoiders are nearly as comfortable as conflict seekers discussing topics like movies, sports, and weather, patterns differ when we turn to explicit discussion of politics compared to discussion of ostensibly political issues.

Figure 5. The Effect of Conflict Avoidance on Comfort with Conversation

Note: Results are predicted levels of comfort from logit models each using comfort with a different topic as the dependent variable. Each model also controls for partisanship, ideology, religious attendance, age, gender, income, race, and education. Note that the Trump and politics models are based on a split sample design. Full coefficient estimates are in Supplementary material S5.1.

While the predicted probability of expressing comfort talking explicitly about politics is quite high among conflict seekers (0.764), conflict avoiders show far less comfort discussing politics (0.188). Turning to a topic one might categorize as ostensibly political, the economy, we again see gaps between conflict seekers and avoiders, but the differences are smaller. Even in discussions of a topic like religion, differences between conflict seekers and avoiders are lower than they are for politics. Importantly, this difference is driven not primarily by conflict seekers, but by conflict avoiders’ greater willingness to discuss the economy (0.596) and religion (0.410) compared to talking about “politics” specifically (0.188).

These patterns suggest that those who are conflict avoidant are most likely to avoid conversations where “politics” is mentioned explicitly. The only discussion topic that exerts a comparable effect is Donald Trump, who is among the most divisive figures in American history. We note, however, conflict avoidant individuals do express slightly more comfort talking about Donald Trump than talking about “politics.”¹⁶

Comfort with Different Issues, Experimental Approach

Building on the Pew results, we conducted an additional study using the same approach—asking people how comfortable they would feel discussing various topics—but randomly assigning them to only one of four discussion topics: politics, climate change, racial inequality, or the role of lobbying in government. ¹⁷ It is not our intention to suggest that conversations about race or climate change are unconditionally simple or non-conflictual. Indeed, we selected these issues as they address important and difficult conversations in American politics (e.g., Cagle and Herndl 2019; Takahashi and Jefferson 2021). Rather, our goal is to compare comfort in these conditions to one labeled simply “political.” Participants (N = 891) were recruited via MTurk.

Once again, we present our results by level of conflict avoidance, measured prior to treatment, again using a dichotomous measure. Figure 6 shows average comfort by conversation topic and conflict orientation.¹⁸ Again, we find that across all topics, conflict avoiders are always less comfortable with conversation.

Figure 6. Comfort Level by Topic and Conflict Avoidance

Note: Y-axis is the percentage reporting comfort by condition and conflict orientation. Estimates are from logit models. Differences by conflict orientation are as follows: politics = 46.6 pp, p < 0.001; climate change = 9.4 pp, p = 0.086; racial inequality = 18 pp, p = 0.004; lobbying the government = 40.0 pp, p < 0.001. Differences in comfort level by condition, relative to the political baseline are as follows. Among the conflict seekers: climate change = −5.3 pp, p = 0.222; racial inequality = −13.2 pp, p = 0.001; lobbying the government = −12.8 pp, p = 0.002. Among conflict avoiders: climate change = 31.3 pp, p < 0.001; racial inequality = 12.7 pp, p = 0.096; lobbying the government = −8.1 pp, p = 0.302. Full coefficient estimates in Supplementary material S6.

In the condition where the conversation is described only as “political,” we see that 90.4% of conflict seekers report that they would be comfortable having a discussion. This level for politics is similar to climate change (85.0%, p = 0.222), and significantly higher relative to racial inequality (75.4%, p = 0.001) or lobbying (75.3%, p = 0.002). In contrast, conflict avoiders generally show a disinclination to discuss politics. While only 44.1% of conflict avoiders said they would be comfortable discussing politics, 75.5% of avoiders expressed comfort in discussing climate change (p < 0.001). Conflict avoiders are also somewhat more willing to discuss racial inequality, though the increase relative to politics does not reach conventional levels of significance (57.0%, p = 0.096).

The only issue that does not seem to fit the expected pattern, perhaps surprisingly, is the role of lobbying in government. In this case, both conflict seekers and conflict avoiders express less comfort discussing the topic, compared to politics, although the reduction is only statistically significant for conflict seekers (p = 0.002), not conflict avoiders (p = 0.302). This does not follow our pre-registered expectations. In hindsight, we suspect the phrasing of the topic may have intimidated people, regardless of conflict orientation, since both groups rate it as the least comfortable issue to discuss. Framing it as “the role of lobbying” may have suggested the discussion would constitute a test of their technical knowledge of lobbying rather than a discussion of opinions about lobbying.¹⁹

In considering the results of this study, we want to underscore a limitation. Our study utilized a generic title (“participate in a survey”), following best practices for minimizing self-selection bias in MTurk recruitment (Litman and Robinson 2020). However, as we noted earlier—and in Supplementary material S1—the word “politics” was inadvertently included in the study description. Given the generic title, this inclusion of “politics” in the description is much less central in the MTurk interface relative to the title—and, indeed, likely more subtle than our deliberate manipulations in other studies.²⁰ It is important to acknowledge this limitation, and we work through its implications in DA 6, which suggests it is a limitation that is unlikely to have changed our results. In the next section, we further assuage concerns by conducting a closely related experiment.

Robustness Checks

Results from Pew and the Conversations on Issues experiment suggest that conversations explicitly presented as being about “politics” are less attractive to those higher in conflict avoidance compared to conversations about other topics—even ostensibly political topics (e.g., “the economy”). Given the association between politics and conflict (Fitzgerald 2013), it certainly makes sense that this relationship would hold. But it is also possible that these results are driven by knowledge. Perhaps conflict avoidant individuals are less knowledgeable about politics, and this is driving down their willingness to discuss politics. We address this possibility using Pew’s American Trends Panel using a 13-item measure of political knowledge, which we present in DA 5. We note however that this involves merging waves, which does decrease our N. ²¹ Nonetheless, see similar results.²²

In the Conversations on Issues study, we draw on research into factors that may affect willingness to engage in conversations about race (e.g., Takahashi and Jefferson 2021) and climate change (e.g., Cagle and Herndl 2019).²³ We find that our results are robust to the inclusion of additional controls. We include these additional results in Supplementary material S6.2.

Redefining Politics

The previous two studies suggest that the term “political” functions as a unique descriptor. People who are conflict avoidant seem to be especially uncomfortable with conversations that are explicitly about politics compared to ostensibly political issues. In contrast, conflict seekers seem to be drawn in, expressing more comfort in discussing politics compared to other topics. In our next study, we take a different approach by defining for people what we mean by “politics.” This approach allows us to consider what happens when we try to revise people’s images of what it means to engage in a political context.

In this final study, we rely on an outcome measure that differs somewhat from the others in this manuscript. Specifically, we ask people whether they are interested in politics. Although this measure may not directly speak to opinion expression in the way our previous measures do, we believe it nonetheless speaks to the same ideas. Across all of our measures—whether one takes a survey, participates in a dinner, takes part in a conversation, or reports being interested—we are capturing a willingness to engage with politics (Huckfeldt and Mendez 2008). Thus, employing a variety of measures to tap this construct is beneficial.

Our study relies on a sample recruited via the NORC AmeriSpeak Panel (N = 2,005).²⁴ In this study, we randomly assigned participants to one of five groups. In the first group participants were asked about their interest in politics; in the next four groups, we clarified what we meant by “politics:”

1. 1. …politics, by which we mean laws and policies,

2. 2. …politics, by which we mean laws and policies to address problems facing the country,

3. 3. … politics, by which we mean people debating laws and policies,

4. 4. …politics, by which we mean people debating laws and policies to address problems facing the country.

This design allows us to consider two ideas. First, drawing attention to politics as laws and policies may shift people’s attention away from the “impasse and petty griping” they typically associate with politics (Cramer Walsh 2004, 39). Emphasizing that these laws and policies are meant to address problems facing the country could suggest a still more positive goal. Second, however, this study more directly considers the power of bringing contentiousness to mind by adding the term “debate” while holding other aspects of the definition constant. In this study, we also include a pre-treatment measure of conflict avoidance, which we will use to consider whether people who are more conflict avoidance report more engagement when we define what is meant by “politics”.

We present our results in Figure 7.²⁵ Since this study (like Pew) does include a binary measure of conflict avoidance, we show results across the full range of the scale. Here we see several patterns. First, people who are conflict avoidant are much more drawn to politics when it is defined as “laws and policies” or “laws and policies to fix problems.” This reinforces our previous results: conflict avoiders are avoiding the label politics, rather than the substance. When substance is emphasized, they become more interested. Lower levels of interest in politics, such as we see among conflict avoiders in the baseline “politics” condition, are typically interpreted as apathy. However, given that they show more interest when laws and policies are emphasized, such conclusions may need to be revised.

Figure 7. Specifying the Meaning of “Politics” Influences Reported Interest in “Politics”

Note: Y-axis represents the marginal effect of the various meanings of politics compared to the baseline. Estimates are from an OLS model with interest as the dependent variable coded from 1 to 5. For the lowest level of conflict avoidance, the predicted mean interest in the baseline condition is 3.7 compared to 3.6, 3.6, 3.4, and 3.7 for the law and policies, law and policies to address problems, debating law and policies, and debating law and policies to address problems treatments respectively. For the highest level of conflict avoidance, the predicted mean interest in the baseline condition is 2.2 compared to 2.7, 2.9, 2.1, and 2.2 for the law and policies, law and policies to address problems, debating law and policies, and debating law and policies to address problems treatments respectively. The model producing these effects is the second model in the table found in Supplementary material S7.

Second, inclusion of the term “debate” produces results that are largely equivalent to the original “politics” group. Given that we can compare the patterns in the “debate” groups to the patterns in a group with an identical definition without that term, our results suggest conflict avoiders report lower levels of interest in politics because they have concerns about conflict, not because they find the substance of politics to be uninteresting.

This last study reinforces the previous results, and jointly our three final studies underscore a reticence among the conflict avoidant to engage in politics. This reticence, however, may not reflect genuine disinterest in laws and policies or substantive issues, but rather the nebulous idea of politics. Indeed, our final study reinforces the idea that it is the term “political”—and the associated pictures in our heads (Lippman 1922)—which lead many people to self-select out.

WHO LEAVES?

Our results suggest that contexts that are explicitly political lead some people to select out of politics. This self-selection is not random. It is people who are conflict avoidant who are most likely to self-select out of explicitly political contexts. This, our results suggest, is due to concerns about the association between conflict and politics. As Cramer Walsh (2004) observes, “talking about politics is ‘opinionated’ talk; unless a person holds controversial opinions…the conversation is not political” (38). This self-selection out of politics may have additional implications.

Conflict avoidance is not randomly distributed (Sydnor 2019; Wolak 2022). Research suggests a strong association between conflict avoidance and gender (Deckman 2022; Mendelberg and Karpowitz 2016; Sydnor 2019; Wolak 2022). Indeed, we see a similarly strong association in all our studies that measure conflict avoidance (see DA 9 for a focus on gender). This association is important because it suggests that women’s voices are more likely to be underrepresented or absent from explicitly political spaces (Sydnor 2019). We find this pattern in our own data as well. Across our studies, for example, we see that women are less likely to select into political conditions (DA Table 9.2).²⁶ There is also research to suggest similar gender gaps in other public political forums (e.g., Van Duyn, Peacock, and Stroud 2021).

We also see a suggestive pattern in our Politics versus Entertainment study. In our first analysis of this study, we find that people are most likely to defect from the survey when politics is made explicit. Tracking the post-treatment responses of those who remain, we find significantly larger gender gaps in issue positions in the explicitly political conditions than in the entertainment conditions (full results in Supplementary material S1.2).²⁷ Keeping this in mind, however, we note that we are not the first to suggest that survey defections may be shifting the shape of public opinion (Cavari and Freedman 2023). Our results, however, raise the possibility that there is a gendered pattern to these shifts.

Although aside from gender we do not see other consistent correlations between conflict orientation and other group identities, our results do raise implications for future research about people’s willingness to enter political spaces. Existing research suggests, for example, that experiencing and observing harassment leaves people less likely to express themselves (Nadim and Fladmoe 2021). Further, researchers show that members of marginalized groups—especially those with multiple intersecting marginalized identities—are more likely to experience harassment when expression occurs (Francisco and Felmlee 2022). In the context of our work, these patterns open the possibility that unequal distribution of negative experiences may leave marginalized groups especially reluctant when it comes to politics. While our data and designs are not suited to do justice to this question, considering the role of political marginalization in shaping the distribution of voices in political spaces (e.g., Collins 2021) is an important direction for future research.

DISCUSSION

We believe the results presented in this paper have important implications for public opinion research as well as our broader understanding of what is happening to the American public sphere. Across the results presented in this paper, we find that the association between politics and conflict leads many Americans to disengage from explicitly political contexts. Moreover, those who make this choice are not a random subset of individuals, but rather those who typically seek to avoid conflict.

The selection effects we find in this manuscript suggest a series of escalating implications. First, researchers may overestimate the representativeness of conflict and incivility they see in explicitly “political” surveys. Second, citizens may overestimate the representativeness of conflict and incivility they experience in explicitly “political” interactions. Third, as more people seek to avoid this conflict, some voices may become less present in the public sphere. Fourth, this all suggests a feedback loop whereby expectations about “politics” lead to selection effects, and selection effects feed back into expectations, turning the expectation of conflict into a reality.

Our results, of course, are not without limitations. We rely on survey experiments to measure, not just survey participation, but also discussion and engagement in politics. In actual social settings with friendly interpersonal interaction (Cramer Walsh 2004; Morey, Eveland, and Hutchens 2012: Sydnor, Tesmer, and Peterson 2022), even a conflict avoidant person might feel more comfortable engaging with explicitly political ideas. At the same time, elites track public opinion through the voices heard in settings that are often explicitly political, like surveys, public meetings, and social media (McGregor 2020), which makes our results consequential.

How might we address these challenges? Our results suggest several courses of action. The first is survey recruitment: survey researchers should consider how the term “political” affects participation.²⁸ The term “politics” is quite ambiguous, and it may not be understood by study participants the same way that it is understood by researchers. Study participants are likely to associate the term with conflict, whereas the researcher likely means to indicate that the study pertains to laws and government policies. As our results suggest, many study participants will avoid “politics” even though they are quite interested in “laws and policies.” This also carries implications for larger public surveys. As a baseline, researchers would benefit from knowing the recruitment language used by the surveys. This information, however, is not always clear or available; it is, for example, not part of the Roper Archive Transparency Project as of this writing.

Second, people should be cognizant of the possibility that labeling contexts—discussions, meetings, etc.—explicitly political will lead some people to self-select out, thereby shaping the conversation before it ever starts. Our results suggest that avoiding politics may not always signal a lack of interest in laws, policies, and pressing social issues (Eliasoph 1998). Some people may deliberately avoid explicitly political contexts because they envision politics to be highly conflictual and antagonistic. Since members of marginalized groups are more likely to experience antagonism in political settings, thereby strengthening this association, and women are more likely to avoid conflict, it is their voices that are most likely to be underrepresented in contexts labeled “political.”

Third, it is critical to consider how we might break the feedback loop these processes create. As more and more people avoid explicitly political settings, the reality of politics comes to resemble the stereotyped images we have in our heads. Rather than a venue for deliberation and problem solving, politics is quickly becoming a verbal sparring ring where those who relish “owning” their opponents go to debate. Our results suggest this feedback loop may be difficult to break, since people have such a strong expectation of conflict in situations deemed “political.” On the other hand, our results also point to a promising avenue for future research. Expectations may be difficult to overcome once a situation is deemed “political,” but it may be possible to change people’s beliefs about what it means for a situation to be political in the first place.