Introduction
Recent research has documented a remarkable increase in self-censorship by ordinary people. For instance, at the height of the 1950s Red Scare, Stouffer’s legendary survey (Reference Stouffer1955) found that only 13 per cent of the American people felt less free to ‘speak their mind’ than they used to; nearly 85 per cent said they did feel free to speak their minds. Comparable 2020 survey results are sobering: fully 46 per cent of the American people in 2020 reported being less free to speak their minds (Gibson and Sutherland Reference Gibson and Sutherland2023). Similar findings have been reported from Germany (Menzner and Traunmüller Reference Menzner and Traunmüller2023) and Sweden (Widmalm et al. Reference Widmalm, Persson and Casselbrant2024). Something major – and likely highly consequential for democracy – seems to have happened to levels of perceived personal political freedom over the many decades.
There are, however, several obvious and consequential limitations to extant research. For instance, we know from research on perceived personal freedom from a few decades ago that substantial differences existed in the United States across racial groups in the perceived freedom to speak out, with African Americans reporting enjoying much less freedom than white people (for example, Gibson Reference Gibson1992, Reference Gibson1995). Nevertheless, contemporary research in the United States gives short shrift to inter-racial differences and practically no attention at all to intra-racial variability (for example, Gibson and Sutherland Reference Gibson and Sutherland2023 and others) – and especially to how these differences might have changed over time (but see Gibson Reference Gibson2012).
Moreover, questions about the nature of the changes documented in earlier studies remain. For instance, that individual-level affective polarization is unrelated to self-censorship in 2020 (which has been established) does not necessarily indicate that the variables have been unrelated over the course of recent history. It could easily be that affective polarization today is at such an extreme level that no meaningful variance remains, even if such variance did exist in the past and was consequential both in previous times and for changing levels of self-censorship. Contemporary descriptive findings are tantalizing, but unanswered analytical issues remain.
Consequently, the purposes of this article are several. First, using self-censorship questions asked in both a 2022 survey (a nationally representative survey, known as the ‘Political Freedom Survey’, conducted in July 2022 using NORC’s AmeriSpeak® Panel) and a 1987 nationally representative survey (the General Social Survey and Gibson’s GSS Reinterview Survey, which I will refer to as the ‘RGSS’ – see Online Appendix A for more details on both surveys), I am able to present a more comprehensive comparison of whether and how perceived political freedom has changed over time. Second, both the 1987 RGSS survey and the 2022 Political Freedom Survey included extremely high-quality representative oversamples of black Americans, making it possible to examine both inter-racial and intra-racial differences, cross-sectionally and over time. Finally, I consider how the correlates of self-censorship have evolved, within race, focusing in particular on the earlier finding of parallel aggregate-level trends between lost freedom and growing polarization. Thus, this article’s goals are specifically to examine how perceived political freedom varies by race, to investigate how perceived political freedom has changed among black and white Americans, and to determine whether and how the correlates of self-censorship have varied over time.Footnote 1
Two tantalizing empirical findings in particular stand out from my analysis. First, while black people expressed more reluctance to speak out than white people in 1987, the relationship reversed in 2022, with white people now engaging in much more self-censorship. Second, while older people were more likely to self-censor in 1987, it is younger people who are more likely to self-censor in 2022. These reversals of the signs of relationships are unusual in public opinion research, and most likely reflect fundamental changes in the nature of pressures towards conformity in the contemporary era. I conclude that the social mechanisms encouraging self-censorship have evolved rather dramatically in the last several decades, and with speculation about the consequences of declining freedom to speak for Western democracies.
The Importance of Studying Variability in Self-Censorship
While the longstanding consensus on the value of free and open debate for democracy may have weakened a bit in recent times (for evidence from the United States, see Chong et al. Reference Chong, Citrin and Levy2022; for evidence that young people hold weaker commitments to democratic values, see Frederiksen Reference Frederiksen2024), most democratic theorists continue to believe that democracy profits when citizens feel free to speak their minds. Mutz (Reference Mutz2002, 122) put it well when she wrote: ‘Exposure to dissimilar views has been deemed a central element – if not the sine qua non – of the kind of political dialogue that is needed to maintain a democratic citizenry’. If citizens are fearful of expressing their views, the essential rough-and-tumble of democratic politics becomes hampered and circumscribed. Consequently, considerable earlier research on perceived political freedom has been reported, in the United States, in Europe, and elsewhere.
Extant Research
Research on self-censorship and the perceived loss of political freedom has been of interest to scholars for decades. Recent research on the topic generally falls into three categories, with investigations of (1) change in levels of self-censorship over time, (2) aspects of the context of political conversations that correlate with self-censorship, and (3) cross-sectional, individual-level variability in self-censoring.
As to the first category, research has found a large increase in perceived loss of freedom of speech over time. For instance, German research found that ‘Whereas in 1971 some 83% of the German population felt they could “freely express their political opinion,” 50 years later in 2021 this share had shrunk to a mere 45%’ (Menzner and Traunmüller Reference Menzner and Traunmüller2023, 2).Footnote 2 As already noted, Gibson and Sutherland (Reference Gibson and Sutherland2023) identify a very large boost in the percentage of people reporting that they feel less free to speak their mind; at 46 per cent in 2020, up from 13 per cent in 1954 (see also Gibson Reference Gibson2008).
In the second research category, Carlson and Settle (Reference Carlson and Settle2023) found that expected disagreement with a conversation partner was the most influential factor of several tested in terms of whether people would opt to withhold their true beliefs in a hypothetical discussion. The conversation not being face-to-face or the conversation partner not being someone close also negatively affected the likelihood of the person sharing true opinions. Carlson and Settle’s 2022 book What Goes Without Saying goes on to conclude:
People were significantly more likely to list affiliative motivations when they described conversations they avoided, and significantly more likely to list affirmation motivations to describe conversations they pursued … Individuals tend to avoid discussions when they anticipate that they will damage their social relationships or when they sense that they will be negatively evaluated for their political opinions, but they choose to engage in discussions when they feel a desire to express their opinions to others. (Reference Carlson and Settle2022, 119)
Walsh’s Talking About Politics additionally found that a sense of collective identity was a key factor in groups being able to comfortably converse about politics: ‘Collective identities can go a long way toward facilitating ideas of what is acceptable to say in a group’ (Reference Walsh2010, 87).
As to the third category, in Germany, Menzner and Traunmüller found that three factors were ‘significantly, consistently, and substantively related to subjective free speech in Germany: political preferences, populist attitudes, and identification with the AfD’ (Reference Menzner and Traunmüller2023, 20).Footnote 3 Gibson and Sutherland (Reference Gibson and Sutherland2023) observed that levels of self-censorship correlate with ideology, with conservatives being more self-censoring than liberals. Weak correlations were also found between self-censorship and personal intolerance as well as perceived freedom constraints emanating from the government.
Research on cancel culture is also relevant to the renewed conversation about self-censorship.Footnote 4 Loosely speaking, cancel culture can be described as the phenomenon of public backlash against a person for expressing allegedly insensitive words or actions, although the term is rarely objectively defined (and trying to define it can be contentious).Footnote 5 Right-wing media and pundits often claim that liberals want to silence the free speech of their opponents through cancel culture (Fahey et al. Reference Fahey, Roberts and Utych2023). Whether that sentiment is true or not (and Dias et al. Reference Dias, Druckman and Levendusky2025 show that it is not), the perception that those who hold conservative opinions are more susceptible to ‘cancelation’ could cause more self-censorship among conservatives, contributing to raising overall levels of self-censorship.
A Simple Theory of Self-Censorship
In research as ambitious as this, it is perhaps not surprising that I rely upon several different theoretical frameworks. As to variability over time, I draw from Gibson and Sutherland’s (Reference Gibson and Sutherland2023) conclusion that self-censorship has grown in close parallel to rises in polarization in the United States.Footnote 6 As to inter-racial differences in affective polarization, Enders and Thornton (Reference Enders and Thornton2022) provide most useful empirical findings and insights regarding inter-racial differences in polarization. Intra-racial differences require a different body of theory. White and Laird (Reference White and Laird2020) put forth a theory, dubbed ‘racialized social constraint’, that seeks to understand the evolution of pressures towards conformity within the black community (see also White et al. Reference White, Laird and Allen2014).
From these bodies of research a simple theory of why people engage in self-censorship can be outlined. The elements of the theory are as follows:
-
Those living in a completely homogeneous community would feel no need to withhold their views since all views are shared.
-
Self-censorship requires recognized diversity of views in a community. Potential speakers must be aware that not all share their views.
-
Diversity itself provides only small incentives to self-censorship. When diversity becomes weaponized – especially through affective polarization – the incentives to self-censor can increase significantly. Affective polarization means not just that people hold conflicting views on many if not most political issues, but that they also hold hostile assessments of those committed to opinions contrary to theirs.
-
When disagreement is expressed, speakers can be punished for speaking out. In the most visible forms, this can produce a ‘cancel culture’. For many ordinary conversations, conflict results, sometimes heated, producing unpleasant animosities, and even the break-up of friendships and families. Shutting up may not be that costly; speaking out can be.
-
Not all individuals are equally susceptible to sanctions for expressing their views. Susceptibility depends in part on the nature of the in-groups with which the individual identifies, and in part on individual characteristics and attributes. Individuals holding more extreme and/or unrepresentative views compared to their group (‘outliers’) are more likely to receive sanctions for making their views public.
-
To the extent that groups have ‘sorted’ by relevant political and ideological characteristics they have more incentive and ability to sanction ‘deviant’ viewpoints. ‘Unsorted’ groups are more likely to be characterized by cross-cutting cleavages that work against strong affective polarization.
In my analysis, a key driving force for increasing self-censorship is the rise in affective polarization.Footnote 7 Many scholars are concerned that this increased polarization has amplified the risk of freely sharing one’s opinion, therefore causing people to engage in self-censorship more frequently (Carlson and Settle Reference Carlson and Settle2022). Gibson and Sutherland (Reference Gibson and Sutherland2023) report a strong correlation between aggregate-level measures of affective polarization and self-censorship in the United States over the last several decades. But the rise of affective polarization has not affected black and white in exactly the same way.
As to polarization in the United States, the story for white people is well-known: over the last several decades, polarization has increased dramatically, in no small part owing to the sorting of partisanship and ideology (Levendusky Reference Levendusky2009). That is, among white people, partisanship and ideology came into greater alignment, with the most obvious example being the switch of conservative white Southerners from the Democratic Party to the Republican Party. Without cross-cutting ties, disdain for one’s opponents became easier to embrace and in fact materialized.
Among black Americans, the story is more complicated. Obviously, comparable sorting has not taken place as there has been no noticeable exodus of black conservatives (or especially moderates) to the Republican Party. However, as Enders and Thornton (Reference Enders and Thornton2022) have explained, increasing polarization has nonetheless taken place, but not to the degree that it has among white people. They argue that the mechanism behind increasing black polarization is the enforcement of strong social norms favoring the Democratic Party, and making it clear who the ‘enemy’ is. If so, then one obvious response to pressures for conformity is self-censorship, especially regarding intra-racial interactions (but not necessarily inter-racial interactions).
Still, what is perhaps most noteworthy of the various findings on black polarization is that partisan and ideological identifications among black people remain only very weakly correlated, quite in contrast to the correlation among white people (see especially chapter 0, White and Laird Reference White and Laird2020; see also Philpot Reference Philpot2017). This means that a black Democrat is far more likely to interact with those of varying ideological persuasion than a white Democrat (or white Republican).Footnote 8 In this sense, polarization is much less complete among black folks, suggesting that pressures towards self-censorship may also be weaker among black people as compared to white people. In terms of change over time, this also implies that self-censorship has likely increased among both black and white people, but has increased considerably more among white folks. To state the hypothesis more generally, those more closely connected to a polarized political environment are more likely to engage in self-censorship because the perceived costs of expressing discordant views have increased over time.
As to individual-level within-group variability, the increasing cost of expressing ideas perceived to be unpopular has most likely been felt more keenly by particular sectors of the American population. For instance, young people may have been exposed to greater polarization and stricter norms today about what should and should not be said. As Tyler and Iyengar discovered (Reference Tyler and Iyengar2023, 352),
the absence of age differences in our 2019 results suggests that the learning curve for polarization plateaus by the age of 11. This is very unlike the developmental pattern that held in the 1970s and 1980s, when early childhood was characterized by blanket positivity toward political leaders and partisanship gradually intruded into the political attitudes of adolescents before peaking in adulthood.
Moreover, young people are more engaged with social media, and social media can be an effective means of enforcing conformity (for example, Powers et al. Reference Powers, Koliska and Guha2019; see also Das and Kramer Reference Das and Kramer2013). Thus, good reasons exist to expect that the relationship between age and self-censorship may also have changed over time, with young people becoming more likely to engage in self-censorship today.
These bits and pieces of a theory suggest the following hypotheses:
-
H 1 : To the extent that affective polarization has increased, self-censorship has become more likely.
-
H 2 : Levels of affective polarization can vary by groups; groups characterized by cross-cutting cleavages are less likely to experience affective polarization.
-
H 3 : Within any group or time period, those holding views more distant from the dominant norms are more likely to self-censor, but only to the extent that they realize they are deviant and are susceptible to sanctions for expressing their views. The ability to effectively enforce an orthodoxy can vary by group and time period.
These hypotheses are not fully tested in the analysis that follows. However, the evidence I consider is directly relevant to many aspects of these expectations.
Data Sources
My analysis in this article compares two surveys. The 2022 Political Freedom Survey was conducted by NORC using its nationally representative probability-based AmeriSpeak® panel. The sample for that survey included an oversample of 628 non-Hispanic African Americans.
The RGSS data are taken mainly from Gibson’s reinterview of the 1987 General Social Survey respondents. The main 1987 GSS included a probability-based oversample of 488 African Americans (see Bobo and Gilliam Reference Bobo and Gilliam1990 for an example of reliance upon this oversample). The appendix in Gibson (Reference Gibson1992, 350–51) provides considerable detail on the reinterview survey and the black oversample. For additional details see Online Appendix A.
Most importantly, both surveys included valid, nationally representative, and sizable subsamples of African Americans, allowing an investigation of how race connects to perceived personal political freedom in the United States.
Racial Differences Using the Stouffer Measure of Freedom
Using the Stouffer measure, Figure 1 reports the levels of self-censorship among black and white people in the 2022 Political Freedom Survey and the 1987 RGSS.Footnote 9 The figure reports the percentages of respondents who do not feel as free to express themselves as they used to (see Online Appendix B for the wordings of all questions used in this research).
Differences in levels of self-censorship, between races, and across time.
Note: the Stouffer question asks: ‘What about you personally? Do you or don’t you feel as free to speak your mind as you used to?’ 95 per cent confidence intervals around the means are shown.
The data reveal stark inter-racial differences indeed in 2022, with about six of ten white Americans saying they do not feel free, while for African Americans, less than one-half do not feel free. That white Americans feel substantially less free than black Americans – by about 20 percentage points – is a notable finding.
The 2022 survey can be compared to the 1987 RGSS.Footnote 10 The differences in the results are stunning in every respect. First, in 1987, relatively small minorities of both racial groups said they did not feel free to speak their minds. The differences in levels of self-censorship in 1987 compared to 2022 are striking. Second, in 1987, white Americans felt somewhat less constrained when it comes to speaking their minds than did African Americans. The inter-racial difference in 1987 is not as stark as it is in 2022, but perhaps the most important finding is the reversal in the relationship’s direction, from black people feeling less free than white people in 1987 to feeling more free than white people in 2022. Finally, the increase in self-censorship between 1987 and 2022 among white people is nearly 45 percentage points; for black people, the increase is only 14 percentage points. This variability in the magnitude of change between the two groups is striking.
While the findings from these data are so stark that it seems unlikely that they are some sort of by-product of the specific measure of self-censorship, the Stouffer item is certainly not perfect: it is a single-item, dichotomous indicator that purports to summarize behavior across a variety of contexts. It would be useful to have a measure of self-censorship that is more discriminating. Fortunately, such a measure is available, at both points in time.
A More Finely Tuned Measure of Self-Censorship
The Stouffer measure of self-censorship can be improved upon. In the 2022 survey, the respondents were asked: ‘How worried are you about expressing your political views to … ?’, regarding five different contexts.Footnote 11 Figure 2 reports their replies.
Reluctance to express one’s views, across contexts, 2022.
Note: the percentages reported for each item are those who are to any degree worried about expressing themselves in the particular context. For each item, the percentage not at all reluctant is equal to 100 per cent minus the percentage shown. See Online Appendix B for the question wording. 95 per cent confidence intervals around the means are shown.
The figure details the percentages of respondents reluctant to at least some degree to express their political views to others. People are least worried about speaking freely to their immediate families and are most worried about expressing themselves publicly in their community. On every item, very roughly one-half of the American people express some degree of unwillingness to express their views.
The questions used in 1987 to measure reluctance to express oneself differ slightly from the questions used in 2022. The 1987 query referenced five contexts, as shown in Figure 3, with some minor deviation from the exact wording of the groups in 2022.Footnote 12
Reluctance to express one’s views, across contexts, 1987.
Note: the percentages reported for each item are those who are reluctant to express themselves in the particular context. For each item, the percentage not at all reluctant is equal to 100 per cent minus the percentage shown. See Online Appendix B for the question wording. 95 per cent confidence intervals around the percentages are shown.
Several things stand out in this figure. First, reluctance to express one’s views is dramatically less common in 1987 than in 2022. Second, I observe much less variation across contexts in 1987 than in 2022. Finally, the rank order of the contexts is not the same in 1987 as it is in 2022, although this conclusion must be understood within the framework of relatively little variability in the responses to the five contexts in 1987.
In 2022, only 23.5 per cent of the respondents claimed to be free to express themselves in all five contexts. In 1987, in five contexts, the comparable figure is 72.0 per cent. In 2022, 27.4 per cent were unwilling to express themselves in all five contexts; in 1987, 5.6 per cent were unwilling to speak freely in all five contexts. These differences across time are dramatic and comport with the conclusions drawn from the simple Stouffer indicator of self-censorship.
As a summary measure of the latent construct ‘self-censorship’, I created an index from the responses to being reluctant to express one’s views in the various contexts. The measure ranges from feeling unwilling to self-censor in any of the contexts to being unwilling to express views in all the contexts about which the respondents were asked. In 2022, the index takes into consideration the breadth of contexts in which the respondent is worried as well as the degree of reluctance within each context. I scored the summary measure to range from 0 to 1. In terms of the hypothesized latent construct, the five-item set is strongly reliable (Cronbach’s alpha = 0.84; mean inter-item correlation = 0.52) and, in a Common Factor Analysis (CFA), the items are strongly unidimensional (eigenvalue2 = 0.70). All indicators load strongly on the first extracted factor (minimum factor loading = 0.60, for the item about expressing oneself to one’s immediate family). Thus, the psychometric properties of the measure are strong indeed (and do not differ by race – see Online Appendix B).
I also created an index of reluctance to express one’s views in the 1987 survey. The five-item set is quite reliable (Cronbach’s alpha = 0.90; mean inter-item correlation = 0.64), and the set is strongly unidimensional (eigenvalue2 = 0.53). As reflects the apparent reality of 1987, the index is skewed, with a majority of the respondents receiving a score of 0.
Figure 4 reports the inter-racial differences in perceived freedom using the indices of self-censorship. In 2022, white people engaged in more self-censorship than did black people. In 1987, the relationship between race and perceived freedom is negative, with black people perceiving less freedom than white people. For both black people and white people, self-censorship increased significantly from 1987 to 2022, although the increase among white people is dramatically greater than the increase among black people.
The changing relationship between race and self-censorship, 1987–2022.
Note: inter-racial differences (full index): 1987 – p < 0.001; r = −0.14; N = 1,213; 2022 – p < 0.001; r = +0.18; N = 1,224.
To more concretely illustrate these findings, in 1987, black people were nearly twice as likely than white people to be fearful of expressing their views around ‘co-workers’ (23.1 per cent versus 12.8 per cent, respectively), but in 2022 the relationship is reversed. Among white people, 57.3 per cent were worried about voicing their views to their co-workers, while that was true of only 43.9 per cent of black people. As these percentages illustrate, both black and white Americans seem to have lost some freedom between 1987 and 2022, although the rate of loss among white people was quite a bit higher than the rate of loss among black people.
Multivariate Analysis
1987 Results
Table 1 reports the results of regressing the self-censorship index on a basic set of predictors, ranging from partisan and ideological identifications to demographic variables. As presented, the table reports separate regressions for black people and white people. I also detail, however, the results from a single integrated equation including the predictors, the race dichotomy, and the interaction of each of the predictors with race. I indicate the statistical significance of the interaction terms with a system of asterisks attached (or not attached) to the predictor names.
Predictors of self-censorship, 1987
Note : all variables are scored to range between 0 and 1.0.
b = unstandardized regression coefficient
s.e. = standard error of the unstandardized regression coefficient
r = Pearson correlation coefficient
Significance of unstandardized OLS regression coefficients: *** p ≤ 0.001 ** p ≤ 0.01 * p ≤ 0.05
See Online Appendix D for details on the distribution of each variable.
For predictors for which the scoring might be ambiguous, I include in parentheses the meaning of a ‘high’ score.
For both the black and white people equations, no Variance Inflation Factor (VIF) coefficients exceed (or even approach) 5.0. In the black equation, VIFs slightly greater than 2.0 (but still considerably less than 5.0) are observed for only the Party Identification and Strong Partisanship variables, as expected since one is derived from the other. This is not true, however, for the Ideological Identification pair. The next largest VIF is 1.5 for Income.
In the white equation, no VIFs exceed (or even approach) 5.0. The Party Identification and Ideological Identification variables have VIFs only slightly larger than 1.0. The largest VIF is 1.4 for Income.
The asterisks next to the variable names indicate the statistical significance of an interaction between the predictor and the race dichotomy in a single integrated equation including both black people and white people with all predictors interacted with race (and, of course, including the race dummy variable). Note that the probability that the two subsamples have a different coefficient for respondent age is 0.093 (p > 0.05).
In the 1987 survey, neither of these equations provides particularly powerful predictions of self-censorship. For white people, the R 2 is distinguishable from zero; for black people, it is not.Footnote 13 For African Americans, only a single predictor achieves statistical significance: age, with older black folks engaging in more self-censorship. The same is not true of white folks. Generally, black people less than about 50 years old engage in less censorship than those 50 or older (data not shown).
For white people, only a single predictor achieves statistical significance: whether the respondent claims a strong ideological self-identification. Those holding strong ideological self-identifications (either liberal or conservative) are more reluctant to express their views. The same is not true for black people – and this is the only instance in which the difference between the black and white regression coefficients is statistically significant.
It is also noteworthy that ideological affective polarization is unrelated to self-censorship for both black and white people. Those who are more polarized are neither more nor less likely to engage in self-censorship. This finding comports with that of Gibson and Sutherland’s (Reference Gibson and Sutherland2023) analysis of their 2020 cross-sectional data, suggesting that individual-level polarization is connected to self-censorship neither when polarization is low (1987) nor when it is high (2020). This also reinforces the conclusion that polarization seems to be important for self-censorship mainly as an attribute of the political environment, not as an attribute of the individual citizen.
Perhaps the most substantively significant conclusion supported by the results in Table 1 is that in 1987 those who tended to self-censor did not differ much from those who did not tend to self-censor. For instance, women did not engage in self-censorship at different rates than men. Partisanship did not distinguish those likely to self-censor from those not. Indeed, one possibility emerging from this analysis is that self-censorship may have been more influenced by idiosyncratic contextual factors – those contexts surrounding the individual respondent – rather than the predispositions of the respondent. In general, Democrats in 1987 did not differ from Republicans in their propensity to engage in self-censorship, but when Democrats were associating with certain types of Republicans, perhaps they withheld their views, just as Republicans, when associating with certain types of Democrats, tended to withhold their views. The available measures cannot pick up this variability. At the same time, however, it must be remembered that the tendency to self-censor does not vary very much across the contexts asked about in 1987. It is also worth reiterating that the index of self-censorship is both valid and highly reliable, so random measurement error is most likely not the cause of these results.
Two findings from the analysis consolidating black and white people within a single equation deserve emphasis. When each predictor is interacted with the race dichotomy, I first find that the change in R 2 from adding the interaction terms does not achieve statistical significance. While the race dichotomy approaches being significant (p = 0.057) in the linear (non-interactive) equation (affecting the intercept), it is not at all in the full interactive equation. Only a single interaction term achieves statistical significance: holding a strong ideology. As can be seen in the coefficients from the two-equation results (split by race), white people with a strongly held ideology are more likely to engage in self-censorship (p < 0.05), while for black people, there is no significant relationship (p > 0.05). Also noteworthy from my 1987 analysis is that party and ideology by themselves seem to have had few consequences for self-censorship.
A Closer Look at Partisanship and Ideology, 1987
As Table 1 reports, neither party nor ideological identifications are significantly related to self-censorship. To examine these relationships in more detail, Figure 5 depicts the bivariate relationships between party identification and self-censorship and between ideological identification and self-censorship for both black and white people in 1987.
Party and ideological identification and self-censorship, 1987.
Among black people, it is easy to see the total lack of relationship between party identifications and self-censorship. For white people, the relationship is slightly stronger but still extremely weak and only marginally statistically significant (at the bivariate level, which is what this figure depicts).
Although caution must be exercised owing to the small Ns in some categories, for nearly every category of party attachment, the black respondents engage in marginally more self-censorship than the white respondents. However, these within-category differences are generally not statistically significant across race – the inter-racial difference in average levels of reluctance to speak is only significant for independents and for RepublicansFootnote 14 – even if they are consistent with the general finding that, in 1987, black people perceived themselves to have less freedom to speak than white people perceived themselves to enjoy. In sum, partisanship is not much of a predictor of self-censorship in 1987.
Very similar conclusions emerge from my analysis of ideological self-identifications. At most levels of ideology, black people are more reluctant to speak out than white people, but the difference is only statistically significant for moderates.Footnote 15 For neither black nor white people is there any substantial relationship between ideology and self-censorship, even if the weak relationship among white people is statistically significant. The data in this table also illustrate the conclusion from Table 1 that, among white people, those with the strongest ideological identifications (liberal or conservative) are most likely to withhold their political views.
2022 Results
Table 2 reports a similar multivariate analysis of the 2022 data.Footnote 16 I first note that both the equations for black and white people produce a statistically significant R 2, although both equations have only minimal predictive capacity.
Predictors of self-censorship, 2022
Note: all variables are scored to range between 0 and 1.0.
b = unstandardized regression coefficient
s.e. = standard error of the unstandardized regression coefficient
r = Pearson correlation coefficient
Significance of unstandardized OLS regression coefficients: *** p ≤ 0.001 ** p ≤ 0.01 * p ≤ 0.05
See Online Appendix D for details on the distribution of each variable.
For predictors for which the scoring might be ambiguous, I include in parentheses the meaning of a ‘high’ score.
For both the black and white equations, no Variance Inflation Factor (VIF) coefficients exceed (or even approach) 5.0. In the black equation, VIFs slightly greater than 2.0 (but still considerably less than 5.0) are observed for the Party Identification and Strong Partisanship variables, as expected since one is derived from the other. This is not true, however, for the Ideological Identification pair. The next largest VIF is 1.6 for Income.
In the white equation, no VIFs exceed (or even approach) 5.0. The Party Identification and Ideological Identification variables have VIFs less than 3.0. The next largest VIF is 1.4 for Age.
The asterisks next to the variable names indicate the statistical significance of an interaction between the predictor and the race dichotomy in a single integrated equation including both black people and white people with all predictors interacted with race (and, of course, including the race dummy variable).
For African Americans, the strongest observed relationship, as with the 1987 data, is with the respondent’s age, but with older black people expressing less reluctance to express their views. Also notable is the connection between self-censorship and level of education, with better educated black people being more reluctant to express their views. Finally, whether one holds one’s partisan self-identification strongly is connected to reluctance to express one’s views. The equation indicates that, ceteris paribus, black people who are strong Democrats or strong Republicans self-censor more. This relationship, while statistically significant, is not very strong. Nor is there a significant relationship between withholding one’s views and simple partisanship. Black men are slightly more likely to self-censor than black women.
The equation for white self-censorship reveals several significant coefficients. Like black people, older white people are more likely to self-censor less. Level of education is also related to self-censorship, with those white people with more education self-censoring more. The data also reveal a small but significant relationship with gender, with white men being less likely to withhold their views than women. Finally, weak relationships are observed among white respondents between more self-censorship and not holding a strong ideological self-identification and with more frequent church attenders being more reluctant to speak out.
As Figure 6 illustrates, in the 2022 analysis, age is a significant predictor of self-censorship among both black and white people. Recall that, in 1987, age is related to black self-censorship but not white self-censorship. However, a critically important difference between the findings in 1987 and 2022 exists. In the earlier period, older black people were more likely to withhold their views; in the later period, self-censorship is most common among both younger black and white people. (The sign on the age coefficient flips.) The dynamics of self-censorship seem to have changed markedly from 1987 to 2022.Footnote 17
The relationship between age cohort and self-censorship, 1987 and 2022.
Partisanship and Ideology, 2022
Figure 7 strongly supports the conclusion that levels of self-censorship in 2022 are not connected to one’s party identification. Nor is there much of a relationship with the strength of one’s party identification, a finding that characterizes both black and white people. Within the types of identifications, black/white differences are statistically significant only among those with some sort of Democratic Party identification. For all the types of Republican identifications, the levels of self-censorship among black and white people do not differ significantly.
Party and ideological identifications and self-censorship, 2022.
The findings for ideological self-identifications are similar: self-censorship varies little across the ideological continuum. In two instances – those who are very liberal and those who are somewhat conservative – the black/white differences are not statistically significant, but generally black people engage in less self-censorship than white people (with the most significant difference found among those who identify as moderates).
Discussion and Concluding Comments
This article’s analyses only begin to scratch the surface of the question of why some are reluctant to express their views and others are not. Some tantalizing findings have been uncovered here – especially the reversal over time in the difference between black and white people in their levels of self-censorship. In 1987 in the United States, black people felt less free than white people to express themselves; in 2022, it is white people who withhold their views more than black people. The levels of self-censorship among both black and white people increased markedly from 1987 to 2022, but among white people the explosion of self-censorship was far more dramatic, eclipsing that among black Americans.
The age/self-censorship connection also reversed itself between 1987 and 2022. In the earlier time period, older black people were less willing to speak out; in the later time period, it is younger people, black and white, who are more reluctant to express themselves. This finding is reinforced by the observed relationship between level of education and self-censorship, among both black and white people, but more strongly so among white people. The better educated actually self-censor more, most likely because they are more aware of thought norms and are more easily sanctioned for deviations (for example, through social media mechanisms). In connection with the youth correlation, these findings may imply that self-censorship is likely to continue to grow as norms punishing the unorthodox become stronger and more entrenched, norms that many educational institutions often seem to inculcate.
While I have no measures of group norms or of the degree of homogeneity of one’s micro-environment, these data are compatible with the hypothesis that growing polarization has caused more self-censorship among white people, but less so among black people. Regarding the latter, while black people overwhelmingly identify with the Democratic Party, they are remarkably diverse in terms of ideological identifications, thereby producing cross-cutting cleavages (Philpot Reference Philpot2017). For white people, perhaps social networks have become more homogeneous and more rigid, exacerbated by polarization that has fortified norms of what is and is not acceptable.
In the earlier analysis of 2020 data by Gibson and Sutherland, affective polarization was not found to be a predictor of cross-sectional variability, even if, at the aggregate level, the growth of polarization closely mirrored the growth of self-censorship. Here, I have supplemented their analysis by showing that cross-sectional variability in self-censorship in 1987 was not related to individual (relatively low) levels of affective ideological polarization. While the degree of polarization in one’s environment may affect willingness to speak out, it seems that the degree of polarization of an individual has little connection to self-censorship. Perhaps this is not surprising. On at least some issues (for example, abortion), clear majorities and minorities exist but without the weaponization of affective polarization of partisan politics. One can be reluctant to speak out without necessarily believing that those holding opposing views are the ‘enemy’.
This article has not been able to address the consequences of self-censorship. To the extent that those who are reluctant to express their views are not just a random sample of those holding views on any given issue – suppose that those reluctant to speak out are less trusting of government than those not reluctant to speak out – then expressed public opinion can differ from true public opinion. While the size of the bias is a function of the strength of the relationship between self-censorship and policy views, in some instances the bias could be substantial and politically meaningful. If self-censorship is not randomly distributed across policy views, then the policy views expressed are different from the true views of the people.Footnote 18
As reported in Figure 2, a majority of Americans in 2022 were reluctant to speak out in three of the five contexts about which they were asked. Indeed, 49.0 per cent of the respondents said that they were even to some degree reluctant to express themselves to close friends, and 45.3 per cent were reluctant to express themselves to immediate family members. These are worrisome figures indeed. How can the robust discussion and debate so critical to democratic deliberation be sustained under these conditions? Perhaps some of this is nothing more than frivolous political correctness, but some might also include intimidated minorities on serious issues such as abortion rights, gun ownership, support for aid to Israel, etc. Noelle-Neumann’s Spiral of Silence (Reference Noelle-Neumann1974) becomes a deadly threat to democracy when it robs minority opinion – for example, on issues of gay rights, climate change, affirmative action, and the death penalty – of the chance to become majority opinion. Public opinion – observed public opinion, that is – can then become ossified. And taking a page from the work of J.S. Mill, this also robs majority opinion of the chance to strengthen itself, to re-prove its values and assumptions, and to reinforce certainty about the rightness of its own views.
Perhaps self-censorship does in some instances help avoid interpersonal conflict; but avoiding such conflict never has been a central objective of democratic deliberation (see Luskin et al. Reference Luskin, Sood, Fishkin and Hahn2022). In any and every large political system, people differ in their interests and their views. While it may rarely be clear what it means to ‘respect’ differing positions, democracies seem to profit when differences are tolerated, and all feel as free to express their views as they used to. Consequently, rising levels of self-censorship in the United States, Germany, and elsewhere cannot but be a bad omen for the health of Western democracy.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0007123425101002.
Data availability statement
Replication materials for this article can be found in Harvard Dataverse at https://doi.org/10.7910/DVN/DQEJ7P.
Acknowledgements
I am indebted to Professor Jacob M. Montgomery and The American Social Science Survey (TASS) for support for my 2022 survey. My analysis makes use of the 1987 GSS and especially the Gibson GSS Reinterview. I am grateful to Katie O’Quinn for her assistance in preparing an earlier version of this article. I am also beholden to Taylor Carlson and Sten Widmalm for comments on an earlier version of this article.
Financial support
This work was supported by The American Social Science Survey (TASS), Weidenbaum Center at Washington University in St. Louis; and by the National Science Foundation for the 1987 GSS and the Gibson GSS Reinterview (SES 86-06642).
Competing interests
None to disclose.
Ethical standards
This research was approved by the Washington University in St. Louis Institutional Review Board. The IRB judged this project to be in the ‘exempt’ category owing to the fact that participation in the survey was voluntary, no harm was inflicted on the respondents, and no identifiers were connected to the database generated, among other factors.
Open access
