Increasing American Political Tolerance: A Framework Excluding Hate Speech

Abstract

According to prior research, political tolerance has either stagnated since the 1970s (if to be tolerant one must be tolerant of every group in all circumstances) or steadily increased (if tolerance is measured using an index, averaging across groups). Using General Social Survey cross-sectional and panel data on civil liberties, this article proposes a new framework: separating out the groups that use hate speech from those that may be only controversial. The United States is unique among Western liberal democracies in not having a prohibition against hate speech. By applying a dichotomous hate speech framework to measuring political tolerance, this article finds that the proportion of Americans who are always tolerant has increased by 8 percentage points from 1996 to 2018. Meanwhile, tolerance of groups that use hate speech has remained flat and even decreased among groups that historically were more tolerant of such groups, including the college educated.

The First Amendment to the U.S. Constitution enshrines the right to freedom of expression, and particularly since the latter half of the twentieth century, the Supreme Court has erred on the side of allowing more rather than less dissenting speech, including that which could be hateful (Stone 2008). The General Social Survey (GSS), inspired by Samuel Stouffer’s (1955) work in the 1950s to investigate Americans’ tolerance of far-left groups, has collected data since its inception in 1972 on Americans’ tolerance of potentially controversial target groups in society exercising their rights to freedom of expression. This rich stream of survey data on a fundamental principle of American democracy has attracted the interest of numerous scholars over the years (Bobo and Licari 1989; Davis 1975; Fischer and Hout 2006; Gibson 1992; Mondak and Sanders 2003; Sullivan, Pierson, and Marcus 1979). According to some interpretations of the GSS data, political tolerance, defined as the willingness to extend civil liberties to groups one dislikes, has been steadily increasing in the United States. According to others, the level of political tolerance has not changed significantly in more than 40 years, if to be tolerant one must be permissive toward every controversial group. But what if the political tolerance measurement instrument did not include certain groups whose hateful ideologies were deemed outside the realm of freedom of expression protection?

In this article I examine recent changes in political tolerance in the United States using the GSS civil liberties data and posit a novel formulation of political tolerance. Many European legal systems forbid hate speech that denigrates persons or groups on the basis of their gender, race or ethnicity, religion, or national origin. Two of the potentially controversial target groups about which the GSS asks would likely be considered sources of hate speech: the racist (“someone who believes that Blacks are genetically inferior”) and the Muslim extremist (“a Muslim clergyman who preaches hatred of the United States”). Thus, in this article I distinguish between groups that could be controversial and groups that espouse the kind of hate speech that would be illegal in a parallel system to the United States.

For example, in the French legal system, actress Brigitte Bardot was fined for inciting racial hatred by writing a public letter that said, “I am fed up with being under the thumb of this population [the Muslim community] which is destroying us, destroying our country and imposing its acts” (Poirier 2014.) On the other hand, the satirical paper Charlie Hebdo’s 2007 publication of cartoons of the prophet Mohammed, which went against the Islamic religious precept of not depicting the prophet, was ultimately not deemed hate speech by the French courts because of the satirical nature of the caricatures, which by definition should be provocative (Mbongo 2009). By examining political tolerance in a way that distinguishes hate speech from controversial speech, new patterns emerge over the past two decades in how Americans’ attitudes toward allowing the exercise of civil liberties have morphed, calling into question assumptions about which groups are more or less tolerant.

Additionally, two innovations the GSS introduced in the past 20 years allow a more thorough investigation of the civil liberties items. First, the GSS collected three sets of panel data between 2006 and 2014, each consisting of three waves of interviews conducted over 4 years (e.g., respondents in the first panel were interviewed in 2006, then reinterviewed in 2008 and 2010). The addition of these longitudinal data to existing cross-sectional data allows analyses of within-individual changes in addition to tracking societal-level changes. Such within-individual analyses provide better evidence for causal inference, as each respondent acts as his or her own “control,” thus eliminating generally unchanging attributes (such as gender, race, etc.) from the model. Second, the GSS added a new target group to the survey in 2008: a Muslim extremist. In the wake of the September 11, 2001, attacks and the subsequent war on terror, the willingness to allow a Muslim preaching hatred of the United States in the style of Osama bin Laden was likely to be a difficult test of tolerance, especially in a society in which the media were overtaken by far-right fringe accounts of the danger presented by Muslims in the United States (Bail 2014).

Scholarly work on the GSS civil liberties items has not yet addressed the introduction of the Muslim extremist items and the newly available panel data. By taking up the study of political tolerance with these new data, and reframing the investigation in a way that takes into account the normative differences determining the balance between free expression and prevention of harm from hate speech, this article contributes analyses of unexplored recent trends and a framework that can explain these developments. The article proceeds as follows. First, I review the scholarship that has used the GSS civil liberties items, as well as the ways other scholars have proposed to measure political tolerance. I discuss the different legal frameworks regulating hate speech in the United States and most other liberal democracies and present a new framework for measuring political tolerance. Second, I detail the set of civil liberties items and the GSS panel data. Third, I present the results from analyses of political tolerance using both the cross-sectional and longitudinal GSS data sets. Finally, I conclude with a discussion of the results and reflections for future directions.

Measuring Tolerance in the United States

Why is political tolerance an important concept? Political scientists have long argued for free speech as part of a democratic culture throughout the discipline’s history of research on political tolerance. Such tolerance is considered crucial for the functioning of a liberal democracy with a diverse populace. The first major effort to measure political tolerance began in 1954, with the debut of Samuel Stouffer’s (1955) questions about allowing communists, anti-religionists, and socialists to exercise their civil liberties. Although these questions reflected specific contemporary concerns about intolerance of leftist ideologies, this approach, called the “fixed group” method, was adopted in 1972 by the GSS and expanded in 1973 to ask about other kinds of potentially controversial groups, such as racists, militarists, and gay men. The racist items and militarist items were added to offer ideologically “right wing” target groups, and the items about a gay man were added for their “topicality” (Davis 2007).

Later scholars argued that the GSS fixed-group measurement was too narrow and did not truly capture political tolerance, as respondents vary in how much they like or agree with the various groups the GSS asks about. Sullivan et al. (1979) introduced the concept of having respondents select which group they liked the least and then testing their tolerance of that “least liked” group. More recently, Gibson (1992) argued that general questions about civil liberties, abstracting away from questions about tolerance of specific groups, should instead be used to measure political tolerance. Finally, Mondak and Sanders (2003) contended that to be truly tolerant, respondents must support civil liberties in every instance. They proposed that the GSS civil liberties items were still valuable and should not be abandoned, as Gibson and Sullivan et al. had done, but that tolerance should be measured as a dichotomous outcome: respondents were either tolerant for every target group in every instance, or they were not. Mondak and Sanders did concur with Gibson and Sullivan et al. that being tolerant is contingent on disliking the group one supports being allowed to speak or assemble, and they cited separate survey results demonstrating that Americans report disliking at least one of the target groups which the GSS asks about. Thus, the GSS fixed-group civil liberties items’ comparative advantage to other political tolerance measures lies in their longevity, capturing more than 40 years of American public opinion data, and the assumption that all respondents will dislike at least one of the six target groups, producing a meaningful dichotomous measure of tolerance.

Whether one measures tolerance as a dichotomous outcome or using the “least liked” approach, substantively, the level of political tolerance in the United States does not appear to have increased that much since the 1950s. However, the target groups of intolerance have changed, with the most commonly disliked groups moving from far left ideologies to far right, for example, changing from communists to white supremacists (Mondak and Sanders 2003).

Recent work on GSS civil liberties items has emphasized generational trends. Davis’s (2012) most recent findings demonstrate increasing tolerance through 2006 (if not measured dichotomously), driven by higher levels of formal education and the growing liberalism of newer cohorts such as millennials and Generation X compared with baby boomers and the silent generation. However, the effect of education and cohort appeared to have plateaued, and the tolerance gains for the racist target group were much smaller than the other target groups. Davis raised the possibility that a ceiling may exist for Americans’ political tolerance that is significantly below the constitutional standard of free expression for all. However, Chong and Levy (2018) updated Davis’s finding with another generational change: examining GSS tolerance from 1976 through 2016 using indices, they found that although tolerance of the racist target group was positively correlated with support for racial equality among older cohorts in earlier waves of the GSS, for younger cohorts, the relationship flipped direction. That is, the more members of younger cohorts supported racial equality, the less likely they were to be tolerant of a racist’s expression. Additionally, the authors found that Americans were increasingly differentiating between tolerance of the racist target group and all others, with tolerance for the racist group stagnating.

Two GSS innovations to the core civil liberties items have recently arisen that merit new investigation. First, the GSS added a new target group: an anti-American Muslim cleric who preaches hatred of the United States. As with the other target groups, respondents are asked whether such a person should be allowed to give a speech, teach at a college, and have a book in the public library. Second, the GSS conducted three-wave panel surveys covering four-year periods between 2006 and 2014. The civil liberties items were included in the panels, which allows longitudinal, within-subject analysis of changes in political tolerance for the first time.

National Approaches to Hate Speech

In the United States, speech that does not fall into so-called low-value categories (including threats, child pornography, fighting words, commercial advertisement, obscenity, and incitement to criminal behavior) is protected legally, including hate speech. Not all countries include tolerance of hate speech in their legal commitment to civil liberties. In fact, the United States is unique among Western liberal democracies. France and Germany have had legal bans on hate speech for many years, dating back to the legacy of racist propaganda and the Holocaust in World War II, and now all European Union member countries have comparable policies. Canada also prohibits hate speech, though this provision arose out of the long and deliberative process of the 1982 Constitution and official policy of multiculturalism (Sniderman et al. 1996). Additionally, many Western democracies are signatories to international human rights covenants that ban hate speech (Rosenfeld 2002).

There is no universally agreed upon definition of hate speech. Generally, it refers to public speech that denigrates an individual or group on the basis of perceived characteristics or identities, such as gender, religion, race, sexuality, or nationality. Although European countries differ in the details of their hate speech laws, almost all are signatories of the European Convention on Human Rights dating back to 1959. Under this convention, freedom of expression is explicitly protected as a crucial element of democracy, but speech that incites racial hatred, among other categories, is deemed incompatible with the convention’s fundamental values of social peace, tolerance, and nondiscrimination. What this means in practice is that after a hate speech case in one of the signatory countries exhausts the national appeal process, the European Court of Human Rights—the supranational court responsible for adjudicating cases related to the convention—determines whether speech is hateful, and it has the power to overturn state rulings. Some cases the court has deemed to be hate speech, and not protected by freedom of expression, over the past 40 years include the distribution of leaflets that pushed for all non-White people to leave the Netherlands (Glimmerveen and Hagenbeek v. the Netherlands 1979), Holocaust denial in a book titled The Founding Myths of Israel (Garaudy v. France 2003), and displaying a poster with New York City’s twin towers aflame and the words “Islam out of Britain—Protect the British People” (Norwood v. the United Kingdom 2004).

In places where hate speech is illegal, legal prosecution is relatively rare, and the penalty is generally a fine rather than jail time. Germany has one of the strictest approaches to hate speech. These laws directly stem from postwar efforts to prevent another atrocity such as the Holocaust, and perhaps they are due in part to a desire to protect the country’s national image from its history of Nazism (Kahn 2005). Additionally, the conceptual and legal framework for hate speech is turned on its head in Europe: whereas in the United States, speech is considered a negative freedom—freedom from government censorship of one’s speech—in Europe it is a positive freedom: freedom from hate through the government’s involvement in regulating speech (Rosenfeld 2002). Although the hate speech prohibitions in Europe represent a societal ideal, it must be noted that their strictness does not necessarily correspond to public opinion of what should count as hate speech.

Racism is a particularly odious manifestation of the protections afforded by the freedom of expression in the United States. However, historically liberals and elites have been more tolerant of racist expression, along with almost all other kinds of controversial expression, such as flag burning (McClosky and Brill 1983). In their book on democratic theory and conflict between values, Sniderman et al. (1996) noted that the clash between racial tolerance and liberty tested the former elite consensus on the absolute value of freedom of expression. Presciently, the authors predicted that in the years ahead, liberty would be even more heavily contested, and now from both political poles: “from the right in the name and on behalf of order; from the left in the name and on behalf of tolerance” (p. 244). In other words, support for hate speech prohibitions, especially as it related to race, seemed to be increasing, even among those who overall were very supportive of protecting civil liberties, especially the highly educated (Chong 2006). This is contrary to the broader view of the liberalizing effect of attending college on students’ political attitudes, particularly regarding civil liberties (Bryant 2003; Campbell and Horowitz 2016; Ohlander, Batalova, and Treas 2005).

A European Framework for Measuring Americans’ Tolerance

Given that the historically staunch defenders of free speech such as liberals and the college educated have begun to question the desirability of allowing racist speech and that countries with similar governmental systems to the United States explicitly prohibit hate speech, thus normatively excising it from the fabric of legitimate and desirable discourse, I propose using a more “European” framework to evaluate recent trends in political tolerance. Such a framework separates target groups with explicitly hateful rhetoric from those that espouse potentially unpopular ideas. Separating the extreme Muslim and racist target groups from the GSS civil liberties items leaves four other target groups: a militarist, an anti-religionist, a communist, and a gay man (see more details in the following section). The anti-religionist would not qualify as hate speech because of the broadness of the target of the speech (“all churches and religion”): if one religion were singled out for denigration, then it might qualify as such. Although three of these other four target groups are generally more palatable to leftist ideologies, the militarist remains as a more conservative target group, and thus one that liberals would be more likely to dislike. I contend that this framework, which examines hate speech target groups separately from the others, more accurately conveys the state of political tolerance in the United States and is essential to understanding changes in Americans’ attitudes on the subject.

Data and Methods

To examine how Americans’ political tolerance has evolved, I use two data sets from the GSS: the cross-sectional data set from 1996 to 2018 and the three-wave, four-year panel data set fielded between 2010 and 2014. Consistent with the other core items on the GSS, the civil liberties questions are presented to two thirds of the total sample. In the panel data set, respondents were reinterviewed in person two and four years after the first interview, which constitute the second and third follow-up waves. The retention rate of respondents was 64 percent by the third wave, and the analyses presented here consist of those individuals who responded in all three waves of the panel (n = 778).

There are currently 18 items in total in the GSS regarding civil liberties: respondents are asked about civil liberties for each of six target groups in three different contexts. The six target groups asked about are (ordered as they are asked on the questionnaire and including the GSS variable name suffix in the brackets):

• an anti-religionist (“someone who is against all churches and religion”) [*ATH];

• a racist (“someone who believes Blacks are genetically inferior”) [*RAC];

• a communist (“a man who admits he is a communist”) [*COM];

• a militarist (“someone who advocates doing away with elections and letting the military run the country”) [*MIL];

• a homosexual (“a man who admits he is a homosexual”) [*HOMO]; and

• a Muslim extremist (“a Muslim clergyman who preaches hatred of the United States”), which was added to the GSS in 2008 [*MSLM].

Although the words communist and homosexual are used during the interview, the terms used for the other items—racist, militarist, anti-religionist, and Muslim extremist—are not said to respondents, but instead such a person’s views are only described as quoted above. The wording for these items has been little changed since the GSS’s beginning in 1972, because of concerns about maintaining comparability over time, but this has also continued the outdated and offensive wording describing the gay man target group as an “admitted homosexual.” ¹ Also, when the gender of any of the target groups is mentioned, it is always male. These wording idiosyncrasies likely influence changing understandings of the items, even as the wording of the items does not change.

The three contexts are (including the variable name prefix in brackets) as follows:

• teacher at a college: “should such a person be allowed to teach in a college or university or not?” (The wording for the communist target group is different: “should he be fired, or not?”) [COL*] ² ;

• library: “If some people in your community suggested that a book he wrote _____ [about target group ideology] should be taken out of your public library, would you favor removing this book, or not?” [LIB*]; and

• speech: “If such a person wanted to make a speech in your [city/town/community] _____ [about target group ideology] should he be allowed to speak, or not?” [SPK*].

The questionnaire form asks about one target group in all three situations (college, library, speech), then moves on to the next target group. The response options are “yes” or “no,” providing a binary outcome measure. All the groups are introduced by emphasizing that some in society find their ideas objectionable: “There are always some people whose ideas are considered bad or dangerous by other people. For instance, somebody who is . . . .”

In these analyses, I use dichotomous “always tolerant” outcomes, in which respondents must be tolerant in every instance for a given set of items. The GSS civil liberties data have been also analyzed by researchers both as summary indices of responses to all 18 items—or 15 items, prior to 2008 (Chong and Levy 2018; Davis 1975)—or grouped by target group or context. I extend the “always tolerant” approach used by Mondak and Sanders (2003) because, theoretically, being tolerant in every situation models the free speech paradigm championed by recent First Amendment jurisprudence (Stone 2008), and empirically, being tolerant in every situation is the modal response. Indeed, although only about 20 percent of Americans in 2018 were tolerant for every civil liberty item, tolerance in 18 of 18 items is by far the most common pattern, see Figure 1.

Figure 1.

Distribution of total number of “allow” responses to General Social Survey civil liberties items, out of 18 possible.

To construct the dichotomous measures of tolerance, all the civil liberties items are coded as 1 for “yes” and 0 for “no.” These responses are then added, and sums equal to 18 (for the years the including and after 2008) or equal to 15 (for the years prior) are coded 1 for “always tolerant.” Any sum less than 18, or 15 for the earlier years, is coded “not tolerant.” For groupings of only the racist target group or only the Muslim extremist target group, respondents must be tolerant in all three contexts (giving a speech, having a book in the library, teaching at a university) for that target group in order to be coded as 1 for “tolerant” or 0 for “not tolerant.”

In addition to documenting trends using the cross-sectional data, I also use the GSS panel data to track individuals’ opinions over time from 2010 to 2014. In this case I use the panel data as a supplemental analysis (see Morgan and Kang 2015) to investigate tolerance of hate speech trends over time within individuals, comparing those with and without bachelor’s degrees. By tracking the same individuals, and thereby somewhat “controlling” for unchanging characteristics such as race, sex, and personality (Allison 2009), this ameliorates the issue of comparing the tolerance of today’s more demographically diverse college graduates compared with earlier cohorts.

Results

Changes in the GSS civil liberties items using weighted data from 1996 to 2018 are presented in Figure 2. Over the past 22 years, tolerance has significantly increased (tested by separately regressing each civil liberties item on year) for the anti-religionist, militarist, communist, and homosexual target groups across contexts by 5 to 10 percentage points. Meanwhile, tolerance of the racist target group was relatively flat until 2010, when it began to significantly decline. Tolerance of the Muslim extremist is much lower than all other groups, averaging only about 40 percent, and its levels have not greatly changed since the group’s introduction in 2008. Teaching at college is the civil liberty least likely to be extended to all target groups but the homosexual target group, for whom having a book in a library is more contentious. Consistent over time, tolerance is highest for the homosexual target group and lowest for the Muslim extremist target group. When all 18 items are added to form a summary variable, the mean number of items American think should be allowed, as of 2018, is 12.08 (SD = 5.04).

Figure 2.

General Social Survey civil liberties items, percentage that should be allowed (1996–2018 cross-sectional).

On the surface, Americans’ political tolerance, measured dichotomously using the weighted GSS data as allowing expression for all groups in all contexts or not, does not appear to have shifted much over the past 20 years: if anything, it appears to have decreased, represented by the solid line in Figure 3. ³ In 1996, nearly a quarter of Americans were always tolerant, but by 2018 that proportion had dropped to 19 percent. However, much of this drop can be explained by the addition in 2008 of the Muslim extremist items. If those items are separated out from the calculation (Figure 3, short dashed line), then the proportion of Americans who are always tolerant hardly changes from 1996 to 2016.

Figure 3.

Always tolerant, removing hate speech groups.

Finally, if a hate speech framework is applied, and both the Muslim extremist and racist items are removed from the calculation of absolute political tolerance, a much different picture of political tolerance trends emerges (Figure 3, long dashed line). The hate speech–excluded always tolerant grouping remains 5 to 7 percentage points higher than the Muslim extremist–excluded line from 1996 to 2008, and beginning in 2012, it experiences a marked increase, while the other tolerance formulations decline slightly. The hate speech–excluded always tolerant group expands from 29 percent of the population in 1996 to 37 percent in 2018.

While dichotomously measuring political tolerance is the closest conceptualization of support for free speech, I also used an index approach to measure the extent of change of the mean number of “allow” responses for each grouping. These means were standardized (so that the mean is 0 and the standard deviation is 1) to allow comparisons between groupings with different numbers of items. The groupings are as follows: all 18 items (or all 15 in the years before 2008, when the Muslim extremist items were introduced), all items except the extremist Muslim group, and all items except the extremist Muslim and racist groups. For all 18 or 15 items, tolerance increased from 1996 to 2018 by 0.062 of a standard deviation. Excluding the Muslim extremist from the calculation of tolerance resulted in an increase of 0.163 of a standard deviation. Finally, excluding the hate speech groups (Muslim extremist and racist) showed an increase in political tolerance of 0.250 of a standard deviation. The results of this indexed approach support the robustness of the findings from the dichotomous framework I employ.

Shifts among the Educated and Political Party Affiliates

How have predictors of tolerance changed over the past 22 years as both the set of civil liberties items expanded and attitudes changed? From the descriptive figures above, the decline in tolerance for the racist target group seems to be crucial, but the aggregate may obfuscate heterogenous effects. Table 1 presents the predicted marginal probabilities from two weighted logistic regressions predicting tolerance of the racist target group (in all contexts, measured as a dichotomous outcome) in 1996 and in 2018 using demographics previously identified by researchers as predictive of political tolerance (Bobo and Licari 1989; Davis 2012; Golebiowska 1999; McClosky and Brill 1983; Sniderman et al. 1989). See Appendix B for the full regression models.

Table 1.

Predicted Probabilities of Being Always Tolerant of Racist Target Group (in All Contexts: Teach, Give Speech, Book in Library), Weighted, Presented as Percentages.

	1996	2018
Party identification
Democrat	37%	30%
Independent	39	37
Republican	36	34
Ideology
Liberal	42	35
Moderate	35	32
Conservative	35	33
Sex
Male	42	35
Female	33	32
Race
White	39	35
Black	30	33
Education
Bachelor’s or higher	51	36
No college	32	32

Some predictors of political tolerance of a racist shifted over 22 years, while others witnessed little change. Notably, a gap emerged between Republicans/Independents and Democrats, with Democrats going from a 37 percent predicted probability of tolerance in 1996 to 30 percent in 2018. Similarly, liberals dropped from 42 percent tolerant in 1996 to 35 percent in 2018, while conservatives dropped only 2 percentage points, from 35 percent to 33 percent. Men became less tolerant over the past 22 years, dropping to tolerance levels nearly equal to those of women. Although Blacks increased their willingness for racists to exercise their civil liberties by 3 percentage points over the 22-year period, Whites’ tolerance slightly dropped, leaving the two racial groups at roughly the same level.

The most striking change is the 15 percentage point drop in predicted probability of political tolerance of a racist for those with a bachelor’s degree or more: in 1996, a majority, 51 percent, supported allowing a racist to speak, teach, and have a book in the library, but by 2018 that group had dropped to 36 percent. Confirming this finding, Figure 4 shows the interaction between the survey year and having a college degree or more (see Appendix C for the full logistic regression model). Controlling for other demographics, the predicted effect of a college degree on tolerance of the racist target groups has decreased over time. By the same token, the percentage of college-educated Americans who are always tolerant, excluding the target groups that use hate speech, has risen by 5 percentage points in the same period, from 48 percent in 1996 to 53 percent in 2018.

Figure 4.

Predicted probability of tolerance of racist target group by year and education, 95 percent confidence limits.

A Period of Change for the College Educated: 2010 to 2014

The demographics of Americans going to college have diversified. Cross-sectional data on the college educated can muddle whether the decrease in political tolerance of racists, plateau in tolerance of extremist Muslims, and increase in tolerance of all others from 1996 to 2018 are due to different groups of people going to college in larger numbers (such as women and racial and ethnic minorities) who are associated with lower levels of tolerance or to a new attitudes among those with bachelor’s degrees, making distinctions between hate speech and potentially controversial speech. Also, it is difficult to disentangle liberalizing cohort effects from increased education, as the millennial generation is the most educated America has ever produced. However, leveraging within-respondent change using the GSS panel data can bolster the understanding of the relationship between political tolerance and education.

Figure 5 below presents the tracked mean levels of tolerance of the racist target group for those without and with at least bachelor’s degree from 2010 to 2014. The sample for these graphs consists of the same people, first interviewed in 2010 and then reinterviewed in 2012 and 2014. This time period corresponds to the divergence in 2012 noted in Figure 3 between the absolute tolerance levels of all target groups and excluding those that use hate speech.

Figure 5.

Proportion tolerant of racist target group, General Social Survey panel 2010 to 2014.

For the those without bachelor’s degrees (n = 527), shown in the left panel of Figure 5, the proportion always tolerant of the racist target group increased during the panel, rising from 31 percent to 42 percent, a statistically significant change. Although higher overall than those with less education, the tolerance of those with bachelor’s degrees or more (n = 251), shown in the right panel, did not significantly increase, as demonstrated by the overlapping confidence intervals for all three years. Thus, during the period from 2010 to 2014, the panel respondents with more education did not become more tolerant of the racist target group, which supports the findings of the cross-sectional analyses.

Discussion and Conclusion

Americans, especially highly educated Americans, are seemingly adopting a more European approach to tolerance. Although over the past 20 years, Americans have become more tolerant of most groups asked about in the GSS, especially gay men, attitudes have stalled and reversed their liberalizing course in the cases of the Muslim extremist and racist, respectively. The description of these two target groups, a person who thinks Blacks are genetically inferior and a Muslim clergyman who preaches hatred of the United States, would meet the commonly accepted criteria for hate speech, that which denigrates a person or group of people on the basis of gender, race, ethnicity, national origin, or religion. In almost any other Western liberal democracy, including most of Europe (all signatories to the European Convention on Human Rights) and Canada, this speech would be illegal. Such hate speech could result in fines, paying damages, or even jail time if in the speech there were a particularly strong incitement to violence.

By separating out these two hate speech target groups, the racist and extremist Muslim, from the others in the GSS civil liberties items, I show that the share of Americans who are always tolerant of groups that do not use hate speech has dramatically increased, from 29 percent of Americans in 1996 to 37 percent in 2018. Thus, what appears to be a plateau of political tolerance levels over time (Mondak and Sanders 2003; Sullivan et al. 1979) in fact obfuscates the changing tolerance of specific groups. Target groups that espouse hate speech have experienced significant decreases in tolerance over the past 20 years, especially by those who historically were the defenders of such freedom of speech. Consistent with other research that used an index score of tolerance and found decreased tolerance of the racist target group (Chong and Levy 2018), tolerance of hate speech groups is stagnating or dropping. Rather than the contention that the set of GSS items is largely immune to questions of group “palatability” (Davis 2012), the specific target groups chosen to constitute the test of support for civil liberties has a direct effect on whether political tolerance appears to be increasing or decreasing. The original GSS items reflect cold war–era fears of leftist agendas: communists, socialists, even atheists were perceived as potential threats to the American democratic “norm.” Changing norms today about which groups deserve speech protections influence the inclusion or exclusion of target groups for measuring political tolerance, as seen in the 2008 addition of the anti-U.S. Muslim cleric target group to the GSS civil liberties items after September 11, 2001, and the war on terror.

We may have hit a ceiling of political tolerance, if groups that use hate speech are included in tolerance measurements, and tolerance for hate speech groups decreases while tolerance for other groups increases. Those with the most education are increasingly becoming less tolerant of the racist target group, and their tolerance of the Muslim extremist group has not increased. College-educated Americans may be becoming less tolerant of hate speech because of a greater awareness of the harm hate speech can cause from reading scholarship on the subject, and/or they may be socialized at college to not tolerate racism in any form, often highlighted in campus student speech codes prohibiting offensive speech (Chong 2006). On the other hand, tolerance of groups that do not use hate speech has increased substantially, and particularly notable is the increased tolerance of the gay man target group (see also the dramatic and unprecedented change in Americans’ attitudes toward same-sex marriage; Rosenfeld 2017). The fact that tolerance of the militarist has also continued to increase alongside the gay target group suggests that this phenomenon is not simply a conservative-versus-liberal divide, as espousing support for military dictatorship is quite illiberal.

The time period of this study includes the election of President Barack Obama, the first Black person to hold that office. His election had a profound impact on politics but also on attitudes. Exposure to Obama’s campaign and his election as president decreased White Americans’ racial prejudice toward Blacks, especially among conservatives and Republicans (Goldman 2012; Welch and Sigelman 2011). However, this diminution of racial prejudice did not correspond to a large decrease in tolerance of the racist target group. Conservatives’ and Republicans’ tolerance decreased by only 2 percentage points from 1996 to 2018, a smaller decrease than the 7 percentage point drop among Democrats. Of course, Obama’s election also resulted in the surfacing of ugly racial stereotypes, which may have alerted Americans to the harm racial bigotry can still accomplish and its tenacious roots in the twenty-first century. These events perhaps contributed to highly educated and more liberal Americans in particular changing from privileging the value of liberty and freedom of expression to privileging freedom from harm in the context of hate speech (Chong and Levy 2018).

The Muslim extremist target group is extremely difficult for people to tolerate. There has been little change in tolerance levels since the target group was introduced in 2008, and even the college educated have not become significantly more tolerant of the group as they have for all the other groups that do not use hate speech during the same period. Although some of the reticence to extend civil liberties to this group may be an aversion to hate speech, Americans likely also feel directly threatened by the description of the Muslim extremist clergyman as someone who “preaches hatred of the United States.” Such a person may not even seem “American” to GSS respondents, as Muslim Americans are sometimes not counted as such by others (Dahab and Omori 2018) and thus perhaps less deserving of civil liberties. Recent work suggests that if the wording of the GSS item were changed, with more a gender-neutral “Islamic cleric” replacing “Muslim clergyman,” tolerance would be even lower, perhaps because of Americans’ implicit association of the phrase “Islamic cleric” with terrorism and violence (Morgan 2019). But rather than fear of Muslim terrorism as an exception to general civil liberties, Mondak and Hurwitz (2012) demonstrated that Americans are just as likely to sacrifice civil liberties to prevent serious crimes as to prevent terrorism.

By reframing the GSS civil liberties items as those that are controversial and those that constitute hate speech, a more “European” lens, trends in political tolerance over the past 20 years are diverging. Tolerance of the hate speech items is flat: it is falling among groups that are associated with liberalism, such as the college educated, men, and political liberals, and it is slightly increasing among those associated with illiberalism, such as Blacks and those without college degrees. Meanwhile, tolerance of the hate speech–excluded items is rising among all groups and seems to have accelerated since 2012. A recent Cato Institute survey found that although 59 percent of Americans oppose the government’s taking action to legally ban hate speech, a strong majority, 79 percent, say that hate speech is morally unacceptable (Ekins 2017). What this study’s results suggest is that Americans may be increasingly applying such a moral framework to civil liberties, drawing a line in the sand against hate speech.