WHAT DRIVES PEOPLE TO EXPRESS OR HIDE THEIR TRUE OPINIONS?

In a nation in which free political expression is valued, why should we focus on people’s hesitation to exercise that right in daily conversations? Because many Americans report being uncomfortable expressing their political opinions: in 2019, 40% reported that they feel less free to speak their mind than they used to be (Gibson and Sutherland Reference Gibson and Sutherland2021). This hesitancy also has been observed in laboratory experiments (Carlson and Settle Reference Carlson and Settle2016, Reference Carlson and Settle2022a; Levitan and Verhulst Reference Levitan and Verhulst2016). Following the 4D Framework of political discussion (Carlson and Settle Reference Carlson and Settle2022a), we consider political discussion to be a social process through which people make many decisions. After first detecting another’s political views and then deciding to engage in the discussion, individuals choose what to say. To better understand why people hide their true opinions, this article focuses on this third stage.

One answer is that people seek to avoid conflict; therefore, in the face of expected political disagreement, they will be less likely to express their true opinions. Previous research has shown that people are more likely to avoid disagreeable conversations (Settle and Carlson Reference Settle and Carlson2019), have more homogeneous discussion networks (Mutz Reference Mutz2006), and silence or censor their opinions when they expect disagreement (Carlson and Settle Reference Carlson and Settle2022a). We expect to observe that participants would be less likely to express their true opinions to someone who disagrees with them. Based on prior work, we operationalized expected disagreement as a discussant with whom someone does not share their partisan identity, as follows:

Whereas most political discussion research focuses on the effects of disagreement, more recent scholarship identifies other dividing lines in American politics that go beyond the rift between Democrats and Republicans. Krupnikov and Ryan (Reference Krupnikov and Ryan2022) argue that Americans are divided into those who are deeply involved in politics and the majority who are not. Indicators of how involved potential discussants are in politics—such as how knowledgeable about and engaged in politics they are—could contribute further to how likely someone else would be to express their true opinions in a conversation. For example, people might be hesitant to express their true opinions to someone who is deeply invested in politics for fear of getting into a heated conversation (Connors and Howell n.d.) or of feeling foolish for expressing potentially ill-informed opinions. Indeed, there is evidence that political knowledge asymmetries can affect political expression (Carlson and Settle Reference Carlson and Settle2022a). Yet, the research designs of previous studies make it difficult to assess how these imbalances stack up against the role of expected disagreement, leading to our next two preregistered hypotheses, as follows:

In a media environment as fractionalized as that in the contemporary United States, we also expected that the origin of how people choose to become informed about politics might be a relevant signal. Settle (Reference Settle2018) showed that people make inferences about a person’s political identity based on the sources of information they share on social media. Divisions over which news outlets are trustworthy fall along partisan lines (Jurkowitz et al. Reference Jurkowitz, Mitchell, Shearer and Walker2020), but we suspected that intraparty cleavages about preferred news sources also are important. For example, even if two people share a partisan identity, if one receives information from mainstream sources and the other turns to partisan or fringe sources, the pair might recognize that the other’s opinions are shaped by a vastly different information environment—one that they do not value. In turn, this could make them less willing to express their true opinions, an expectation that we tested as follows:

Finally, consistent with recent work demonstrating the social considerations that people make while navigating political discussions (Carlson and Settle Reference Carlson and Settle2022a), we examined the strength of the social relationship between discussants. Research on political discussion suggests that the strength of social relationships is a primary driver of whether political discussions occur (Ahn, Huckfeldt, and Ryan Reference Ahn, Huckfeldt and Ryan2014; Mutz Reference Mutz2006; Walsh Reference Walsh2010). We expect that people would be less forthcoming with their opinions in conversations with people that they do not know well to avoid damaging a potentially fragile relationship. We tested the following hypothesis:

RESEARCH DESIGN

Our research design allowed us to assess each of these five characteristics (i.e., expected disagreement, political knowledge, political engagement, news consumption, and strength of relationship) as a contributing factor relative to the contribution of the other factors. We conducted a conjoint experiment (see, e.g., Bansak et al. Reference Bansak, Hainmueller, Hopkins, Yamamoto, Druckman, Green, Druckman and Green2021) in which we presented participants with a randomized profile of a political discussant and asked them to report how likely they would be to express their true opinions in a conversation with that person. We fielded our experiment as a module on the Knight Foundation Free Expression Survey, which was conducted by Ipsos. The survey was administered to approximately 5,000 US adults from the online, probability-based Ipsos KnowledgePanel in English and Spanish in July 2021. Our module was administered at random to 2,802 respondents. Because participants repeated the task five times, with new randomized attribute levels in each task, the result was 14,010 decisions to analyze; however, most analyses depend on 13,803 decisions after excluding observations according to our pre-analysis plan.

Before the conjoint experiment began, participants were shown the following prompt to provide instructions on how the study would proceed:

Table 1 summarizes the attributes shown to study participants. We manipulated the characteristics needed to test Hypotheses 1–5, along with additional features that were included for exploratory purposes. All attribute levels were presented with equal probability, with the exception of preferred news source. If the discussant was randomly presented as preferring mainstream partisan news and a Republican, the example source was Fox News; if the discussant was a Democrat, the example source was MSNBC; and if the discussant was an Independent, the example source was randomized.

Table 1

Conjoint Design

After viewing each profile, participants were asked to report how likely they would be to express their true opinions in a political conversation with the person described. We measured their response on a five-point scale, ranging from 0 (very unlikely) to 1 (very likely).

We recoded variables for copartisanship, coethnicity, and shared gender. If participants identified with the same party as the discussant shown in the profile, they were considered copartisans. Independent respondents were only considered to be copartisans with discussants who were described as Independents.Footnote ²

As is standard in analyzing conjoint experiments, the unit of analysis was the profile rather than the respondent. We estimated the average marginal component effect (AMCE) for each attribute in a linear regression in which the dependent variable is the likelihood of expressing their true opinions, and right-hand side variables include indicators for each attribute level. We used cluster-robust standard errors at the respondent level.