Time Period, Generational, and Age Differences in Tolerance for Controversial Beliefs and Lifestyles in the United States, 1972–2012

Abstract

Americans have become increasingly tolerant of controversial outgroups in results from the nationally representative General Social Survey (1972–2012, N = 35,048). Specifically, adults in the 2010s (versus the 1970s and 1980s) were more likely to agree that Communists, homosexuals, the anti-religious, militarists, and those believing Blacks are genetically inferior should be allowed to give a public speech, teach at a college, or have a book in a local library. Cross-classification hierarchical linear modeling (HLM) analyses separating the effects of time period, cohort/generation, and age show that these trends were driven by both a linear time period effect and a curvilinear cohort effect, with those born in the late 1940s (Boomers) the most tolerant when age and time period were controlled. Tolerance of homosexuals increased the most, and tolerance of racists the least. The increase in tolerance is positively correlated with higher levels of education and individualistic attitudes, including rejecting traditional social rules, but is negatively correlated with changes in empathy.

Suppose someone with controversial views came to your community. Would you favor allowing them to give a speech? Would you want them to teach at a local college or university? Should a book expressing their views be banned from your local library? The General Social Survey (GSS), a nationally representative sample of adult Americans collected since 1972, asked these questions about five different controversial views or lifestyles: homosexuals, Communists, anti-religious atheists, militarists, and racists. These questions get to the heart of recent debates around tolerance, acceptance, individual rights, and freedom within liberal democracies. In this paper, we examine whether tolerance for controversial views and lifestyles has changed over time. That is, are people today more tolerant than they were in the 1970s, and of which groups? For the purposes of this paper, we define tolerance as agreeing that controversial outgroups should be allowed public expression. Tolerance is an important indicator of how societies treat people with views and lifestyles divergent from their average members.

If tolerance has changed, a second, equally important, question is the mechanism behind that change. Populations change over time in three ways: time period (a cultural change that affects people of all ages), birth cohort/generation (a cultural change primarily affecting young people that is retained with age), and age (developmental effects; see Yang [2008]). That is, has tolerance changed because people of all ages and generations change at the same time (a time period effect), because new generations enter the survey and older generations exit (a generational or cohort effect), or because the American population has aged (a developmental effect)? (Note that birth cohort refers to everyone born in a given year, and generation to those born within a specified period. Both refer to the effects of being born during a certain era and are thus somewhat interchangeable; we will use generation most of the time but will refer to birth cohort when we are specifically referring to birth year). Until recently, it was difficult to separate the effects of time period, generation, and age, as each variable is a function of the other two and thus cannot be simultaneously entered into a regression equation. However, hierarchical linear modeling (HLM) techniques now allow the separation of the three effects (e.g., Yang 2008). Teasing apart the influence of these three variables allows for a more precise determination of the origins of social change. If the differences are due to generation, tolerant attitudes formed during childhood are likely the main driver of social change. In contrast, if the differences are due to time period, the entire culture and those of all ages change, not just the young. Finally, developmental effects suggest that any cultural change is due to an overall shift in age of the population, such as a “graying” population or “youth bulge.”

Third, why does tolerance change—which demographic and attitudinal changes co-occur with tolerance? Several possible models could explain an increase in tolerance, including increasing education and contact across groups (Kozloski 2010; Ohlander, Batalova, and Treas 2005; Williams, Nunn, and Peter 1976), differential mating strategies linked to greater independence after formal education (Rosenfeld 2007), genetic changes from migration or differential birth/mortality rates or environmental changes (Rentfrow 2010), drug use (Compton et al. 2006), or even decreasing parasite load (Murray, Trudeau, and Schaller 2011). Finally, rising individualism may be linked to increasing tolerance. We discuss this possibility in more detail below, as it was the impetus for this research. However, we acknowledge that it is very difficult to answer the “why” question adequately—the data over time are limited, those that exist are correlational, and multiple intertwined causes are likely at work at the same time.

To summarize, we have three primary goals. First, we seek to determine whether tolerance changed over time. Second, we explore the nature of this effect—is it primarily a time period effect, a generational effect, or an age effect? Third and finally, we examine the correlates of tolerance to determine if they are consistent with a rise in individualism and/or other factors.

Placing Tolerance in the Broader Context of Individualism

Changes in attitudes and values over time periods and generations are rooted in cultural change (Elder 1974; Twenge, Campbell, and Gentile 2012a). Cultures and individuals mutually influence and constitute each other (Markus and Kitayama 2010). Much of the work on cross-cultural differences has focused on individualism (a cultural system that favors the needs or desires of the individual over those of the group) versus collectivism (which favors the group over the individual; e.g., Markus and Kitayama [1991]). This model can also be applied within cultures to explain cultural shifts over time and generations (Twenge 2014). For example, several authors have argued that Western cultures have become more focused on the individual self over time, beginning in the Renaissance (Baumeister 1987) and accelerating in the second half of the 20th century (Fukuyama 1999; Myers 2000). Growing individualism appears, for example, in the more individualistic and less collectivistic language used in American books (Greenfield 2013; Oishi et al. 2013; Twenge, Campbell, and Gentile 2012b, 2013). Among individual people, outcomes have included more positive self-views (Twenge, Campbell, and Gentile 2012a) and lower collectivistic traits, such as empathy (Konrath, O'Brien, and Hsing 2011) and concern for others (Twenge, Campbell, and Freeman 2012), among recent generations (e.g., the Millennials, born after 1982) compared to their predecessors (Boomers, born 1946–1964, and GenX, born 1965–1981). Parents from more recent generations are less likely to value collectivistic traits such as obedience, responsibility, and religious faith in children, and more likely to value individualistic traits such as independence and imagination (Park, Coello, and Lau 2014; Trifan, Stattin, and Tilton-Weaver 2014).

Here, we explore whether the growth in individualism has extended to tolerance for outgroups. Cross-culturally, more individualistic countries are more tolerant of difference and more accepting of broader roles for individuals regardless of race or gender (e.g., Brandt 2011; Chia et al. 1994; Gibbons, Stiles, and Shkodriani 1991; Hadler 2012; West and Hewstone 2012). However, it is unclear whether this tolerance extends to the more marginalized outgroups examined in the GSS survey (e.g., the anti-religious, Communists, racists). Certainly, the changes of the past few decades—such as increasing public support for the legalization of gay marriage—suggest that tolerance has risen. However, the GSS items focus on tolerance unconnected to marriage rights, including accepting outgroup members as public speakers, authors, and teachers who may influence young people, and includes groups other than gays and lesbians.

On the other hand, more individualism and less collectivism may not automatically mean more tolerance. Perhaps less empathy and less concern for others could mean less tolerance for others. More self-focus and individualism could lead people to more readily reject views with which they disagree. That has led some to propose the converse of this idea, reasoning that because Millennials are more tolerant of others, the findings showing they are lower in empathy and concern for others must be wrong (e.g., Arnett 2013). In addition, those who embrace individualistic values may be more likely to have negative attitudes toward minority groups, perhaps because they believe in individual self-sufficiency and thus oppose programs such as affirmative action (Katz and Hass 1988). In other words, if individualism promotes the idea of “every man for himself,” outgroups may be perceived negatively if they are seen as asking for “special treatment.”

A contrasting hypothesis that we endorse posits that individualism should lead to more tolerance even if concern for others has declined. First, individualistic systems value self-expression, so members should generally be tolerant of others' expression and choices (e.g., Kim and Sherman 2007). For example, individualistic systems place fewer restrictions on sexual behavior. Second, when people are treated as individuals—rather than members of distinct groups—tolerance based on group membership should be higher. For example, individualistic cultural systems usually reject social rules restricting the actions and opportunities of groups such as racial minorities and women (e.g., Brandt 2011). Third, individualistic systems allow more contact with outgroup members, which should increase tolerance (e.g., Binder et al. 2009; Madon et al. 2001).

This individualism hypothesis also suggests a model for correlates of tolerance. If tolerance is related to individualism, it should correlate positively with indicators of individualism such as the rejection of traditional social rules (e.g., support of organized religion, taboos against drug use, and restrictions on premarital sex). That also suggests tolerance will be linked to less empathy, not more, as lower empathy is linked to individualistic personality traits (Watson, Biderman, and Sawrie 1994). Empathy (usually defined as understanding the feelings of another) is not the same as tolerance (granting rights to others, which does not necessarily include understanding). Tolerance should also correlate with years of education, as education—especially at the college level—promotes greater individualism (Ohlander et al. 2005).

Economic factors may also be important. If economic prosperity leads to greater individualism and postmodernism (e.g., Greenfield 2009; Inglehart and Welzel 2005), the annual unemployment rate should be negatively correlated with tolerance. Tolerance may also be linked to income inequality, though the direction of the effect is not clear a priori. A society with more income inequality may be a more individualistic, and thus more tolerant, one. Conversely, income inequality could promote less tolerance, with more competition among groups for scarce resources.

Previous Research

Several studies have documented increases in support for equal rights for racial minorities and women (e.g., Carter 2010; Koenig et al. 2011; Twenge 1997). However, it is not clear if this increased tolerance extends to the outgroups examined in the current paper, such as Communists or atheists, who remain very small minorities and thus may be seen as more marginalized and controversial (more “out” as an outgroup). According to recent polling data, atheists are only about 2 percent of the population in the United States, with an additional 3 percent identifying as agnostics. The number of self-identified Communists in the United States is likely to be just as small or smaller (e.g., Lipset and Marks 2001). With questions about five different outgroups, these questions provide a broad view of tolerance of diverse opinions and lifestyles.

Previous research on this topic is intriguing but incomplete. Persell, Green, and Gurevich (2001) found that tolerance for homosexuals increased in the GSS between 1972 and 1994, but it is unclear how responses to these questions have changed in the past 18 years. They also did not examine whether the change was due to time period, generation, or age. Danigelis, Hardy, and Cutler (2007) examined the GSS items on tolerance through 2004, finding considerable changes in tolerance even among those in their 60s and older. However, their focus was on intra-cohort aging; they specifically note that they did not separate time period, generation, and age effects. Schafer and Shaw (2009) reported increases on a wide range of tolerance markers from 1990 to 2006, including some items from the GSS, though their report was descriptive and did not include effect sizes or secondary statistical analyses. Hadler (2012) examined tolerance in the World Values Survey and found few differences between 1989 and 2010.

Furthermore, the idea that attitudes change over time periods and generations remains an area of debate. Some have questioned the idea that generational or time period differences exist at all, arguing that any perceived changes are due to non-representative sampling or the biased perceptions of older generations (Arnett 2010; Trzesniewski and Donnellan 2010). Where differences do exist, these authors contend, they are usually too small to matter. Therefore, they argue, the idea that generations or time periods are significantly different in their attitudes, values, or personality traits is a myth (Trzesniewski and Donnellan 2010). Thus, it is important to determine whether attitudes change over time and the size of any effects.

The Current Research

Our focus in the present research is on changes in tolerance in the United States, including whether trends are due to time period, cohort/generation, or age. The GSS is a useful sample in which to examine these questions. In addition to being nationally representative, the GSS includes respondents of many ages, allowing the separation of time period, generation, and age effects. We accomplish this through the use of HLM techniques.

Given the increase in cultural individualism in the United States, we predict increased tolerance for controversial groups (Hypothesis 1). Further, if cultural individualism is a key driver of increasing tolerance, we would expect this change to be fairly linear and occur at the cultural level (a time period effect; Hypothesis 2a) and as a linear generational effect (Hypothesis 2b). Finally, we expect that education, religion, political beliefs, income inequality, empathy, cultural individualism, and the rejection of social rules will correlate with tolerance, both at an individual level and when matched by year (Hypothesis 3).

Method

Sample

The GSS is a nationally representative sample of Americans over 18, collected in most years between 1972 and 2012 (N = 56,859; for the questions in the current survey, N ranges from 29,631 to 35,048). The GSS data and codebooks are available online (Smith et al. 2013). As suggested by the GSS administrators, we weight the analyses by the weight variable WTSSALL to make the sample nationally representative of individuals rather than households and correct for other sampling biases. Also as suggested by the administrators, we excluded the Black oversamples collected in 1982 and 1987.

Measures of Tolerance

A section of the GSS asks 15 questions about tolerance for people with controversial views. It begins: “There are always some people whose ideas are considered bad or dangerous by other people. For instance, somebody who is against all churches and religion … If such a person wanted to make a speech in your (city/town/community) against churches and religion, should he be allowed to speak, or not?” with the two response choices “Yes, allowed” and “Not allowed.” The next question was “Should such a person be allowed to teach in a college or university, or not?” with the same two response choices, followed by “If some people in your community suggested that a book he wrote against churches and religion should be taken out of your public library, would you favor removing this book, or not?” with the response choices “Favor” and “Not favor” (with “favor” meaning favoring removing the book). The next questions use the same format to ask about other groups, including “consider a person who believes that Blacks are genetically inferior,” “a man who admits he is a Communist,” “a person who advocates doing away with elections and letting the military run the country” (a militarist), and “a man who admits that he is homosexual.” The question about a Communist teaching in a college is worded slightly differently, asking instead about whether he should be fired. The questions about Communists and the anti-religious were asked since 1972, homosexuals since 1973, and the questions about those believing Blacks were inferior and those advocating military rule were asked since 1976.

Measures of Individualism, Rejection of Social Rules, and Other Characteristics

The GSS does not include a direct measure of individualism. However, in one year (1984) it included an item tapping individualistic and anti-collectivistic beliefs: “In our society everyone must look out for himself. It is of little use to unite with others and fight for one's goals in politics or in unions.”

The GSS more consistently included other items that indicate rejection of traditional social rules restricting individual freedom. These included believing marijuana should be legalized, rejecting traditional roles for women (a composite of three items: “A working mother can establish just as warm and secure a relationship with her children as a mother who does not work,” “A preschool child is likely to suffer if his or her mother works,” and “It is much better for everyone involved if the man is the achiever outside the home and the woman takes care of the home and family”), likelihood of voting for a Black president, choosing “None” for religious affiliation, and expressing approval of premarital sex. Based on past research, we also examined correlations between tolerance and years of education, political party (e.g., Glenn and Weaver 1979), and self-rating as liberal versus conservative.

We also gathered annual statistics on economic factors, cultural individualism, and empathy from publicly available sources and previous research. For economic indicators, we examined the unemployment rate and the GINI index of income inequality. Individualistic words and phrases are from the Google Books database, American books corpus (Twenge, Campbell, and Gentile 2012b). Uncommon names, an indicator of need for uniqueness (a facet of individualism), are from the Social Security Administration database of names; we used boys' names, as they change more linearly over time (Twenge, Abebe, and Campbell 2010). We obtained mean scores by year on empathy among college students from Konrath, O'Brien, and Hsing (2011). We matched these social indicators with the tolerance coefficients by year.

Data Analysis Plan

As a first step, we report descriptive statistics, inferential statistics, effect sizes, and cross-time comparisons by age and birth cohort groups in a table. Data collected over time can be analyzed in many ways, including grouping by 20-year generation blocks, by decades, or by individual year. We felt that separating the data into five-year intervals provided the best compromise between specificity and breadth. We report the effect sizes (d, or difference in terms of standard deviations) comparing the first group of years to the last, but also (1) provide the means and SDs for the five-year intervals between these endpoints, so fluctuations at other times are apparent; and (2) provide a figure with the year-by-year results. We also examine whether the trends are moderated by gender, race, and education level.

In describing the trends in the text, we will occasionally employ labels for the generations such as the GI or “Greatest” generation (born 1900–1924), Silent (1925–1945), Boomers (1946–1964; some argue 1943–1960), GenX (1965–1981 or 1961–1981), and Millennials (1982–1999 or 1977–1999; for reviews, see Strauss and Howe [1991]; Twenge [2014]). These generational birth year cutoffs are arbitrary and are not necessarily justified by empirical evidence, but are common labels for those born in certain eras.

To separate the effects of time period, generation, and age, we perform HLM analyses. Specifically, we use a cross-classified HLM successfully applied to period-cohort-age data in the past (e.g., Yang 2008). This approach involves the estimation of an overall regression line of IRT-estimated tolerance onto the age of respondents; the intercept and slope coefficients of these regression lines are “cross-classified” by birth cohort and survey year in the sense that differences in these coefficients between birth cohorts and survey years are estimated separately. Significant differences in intercepts indicate differences in overall tolerance by birth cohort and/or survey year, whereas differences in slopes indicate that the trajectory of tolerance over the life span differs by birth cohort and/or survey year (for further details, please see the appendix).

Latent Variable Analyses

We also employed several statistical techniques to explore whether the tolerance items could be combined, including factor analysis and item response theory (IRT). Principal axis factor analysis scree plots suggested that only one factor was necessary. A varimax-rotated, two-factor model showed a less comprehensible solution, and factor plots suggested that all items clustered together on one factor. Therefore, the IRT assumption of unidimensionality (i.e., only one latent variable underlies the survey scores) was upheld.

The two-parameter logistic (2PL) IRT model was fitted to the data using the IRTPRO software program. This model takes into account the extremity/popularity of items (i.e., b, referred to in ability testing as the “difficulty” parameter) and the discriminating power of the item (i.e., the a parameter, which is analogous to an item-total correlation). The 2PL provides a “purified” estimate of the underlying variable that does not include the item-specific idiosyncrasies that would be included in a simple sum-score. The model showed good fit to the tolerance items. Of the 15 tolerance items, none were above the χ²/df ratio cutoff of 3, ranging from < .001 to .38, with a mean of .09, SD = .11. These results suggest that item and person parameters are interpretable. In addition, the Cronbach's alpha for the 15 items was .92. Thus, for the HLM analyses, we relied on the IRT estimate for the composite of the 15 tolerance items.

Results

Change over Time in Mean Tolerance (Hypothesis 1)

In support of Hypothesis 1, tolerance for those espousing controversial views was markedly higher among American adults in the early 2000s and 2010s compared to the 1970s and 1980s, showing a fairly linear increase over time (see table 1 and figure 1). Increases in tolerance for homosexuality were particularly large, with the percentage of those saying a homosexual man should be allowed to teach at a college increasing from 52 percent in 1972–1974 to 85 percent in 2010–2012 (d = .75). However, tolerance for racists did not increase very much (d = .05 averaged across the three items).

Table 1.

Changes in Tolerance for Controversial Beliefs and lifestyles in the United States, 1972–2012

	n	72–74	75–79	80–84	85–89	90–94	95–99	00–04	05–09	10–12	d
Anti-religionist
Allow to speak	35,048	66% (.47)	65% (.48)	67% (.47)	70% (.46)	74% (.44)	75% (.43)	77% (.42)	78% (.42)	77% (.42)	.24*
Allow to teach	34,215	43% (.49)	42% (.49)	47% (.50)	49% (.50)	55% (.50)	60% (.49)	61% (.49)	62% (.48)	63% (.48)	.41*
Allow book in library	34,540	63% (.48)	61% (.49)	64% (.48)	67% (.47)	71% (.46)	72% (.45)	73% (.45)	73% (.44)	76% (.43)	.28*
Homosexuals
Allow to speak	32,947	65% (.48)	65% (.48)	69% (.47)	72% (.45)	80% (.40)	83% (.37)	84% (.37)	83% (.37)	87% (.33)	.52*
Allow to teach	32,706	52% (.50)	53% (.50)	58% (.49)	61% (.49)	70% (.46)	77% (.42)	80% (.40)	80% (.40)	85% (.36)	.75*
Allow book in library	32,905	57% (.50)	58% (.50)	60% (.49)	61% (.49)	70% (.46)	72% (.45)	74% (.44)	76% (.43)	79% (.41)	.48*
Communists
Allow to speak	34,611	59% (.49)	57% (.50)	58% (.49)	62% (.49)	69% (.47)	67% (.47)	69% (.46)	68% (.47)	66% (.47)	.15*
Allow to teach	33,468	41% (.49)	42% (.49)	46% (.50)	50% (.50)	58% (.49)	61% (.49)	63% (.48)	62% (.48)	64% (.48)	.47*
Allow book in library	34,292	59% (.49)	58% (.49)	60% (.49)	62% (.49)	69% (.46)	69% (.46)	70% (.46)	70% (.46)	72% (.45)	.27*
Militarist
Allow to speak	30,251	–	54% (.50)	57% (.50)	58% (.49)	65% (.48)	66% (.47)	67% (.47)	67% (.47)	69% (.46)	.31*
Allow to teach	29,707	–	37% (.48)	41% (.49)	42% (.49)	48% (.50)	52% (.50)	53% (.50)	54% (.50)	58% (.49)	.43*
Allow book in library	29,960	–	57% (.50)	59% (.49)	60% (.49)	67% (.47)	68% (.47)	69% (.46)	71% (.46)	73% (.45)	.33*
Racist
Allow to speak	30,252	–	61% (.49)	61% (.49)	61% (.49)	64% (.48)	63% (.48)	62% (.48)	62% (.49)	58% (.49)	−.06
Allow to teach	29,816	–	42% (.49)	43% (.50)	44% (.50)	45% (.50)	48% (.50)	49% (.50)	48% (.50)	48% (.50)	.12*
Allow book in library	29,970	–	63% (.48)	64% (.48)	64% (.48)	68% (.47)	66% (.47)	66% (.47)	65% (.47)	65% (.47)	.04
Tolerance composite	25,439	–	56% (.34)	59% (.34)	60% (.33)	66% (.31)	68% (.31)	69% (.30)	69% (.30)	71% (.29)	.47*
Age groups (time-lag design = birth cohort + time period)
18–29	6,226	–	70% (.30)	66% (.31)	67% (.30)	71% (.29)	73% (.26)	73% (.26)	71% (.26)	72% (.24)	.07
30–39	5,470	–	64% (.33)	72% (.31)	69% (.30)	72% (.29)	71% (.29)	71% (.28)	72% (.27)	75% (.26)	.37*
40–49	4,997	–	53% (.33)	59% (.35)	66% (.32)	72% (.30)	71% (.30)	72% (.29)	72% (.29)	75% (.28)	.72*
50–59	3,900	–	48% (.33)	51% (.34)	53% (.33)	63% (.32)	67% (.31)	71% (.29)	72% (.30)	71% (.30)	.73*
60–69	2,813	–	38% (.32)	43% (.33)	46% (.33)	51% (.32)	59% (.33)	63% (.32)	64% (.32)	68% (.30)	.97*
Over 70	2,250	–	30% (.29)	34% (.30)	38% (.30)	47% (.31)	49% (.34)	52% (.33)	53% (.32)	54% (.31)	.80*
Birth cohort groups (quasi-longitudinal design = age + time period differences)
Born 1919 or before	2,000	–	37% (.31)	38% (.32)	36% (.30)	44% (.30)	45% (.32)	–	–	–	.26*
Born 1920s	2,330	–	49% (.33)	49% (.33)	48% (.33)	50% (.31)	50% (.34)	50% (.34)	52% (.33)	–	.09
Born 1930s	2,854	–	56% (.33)	58% (.34)	55% (.33)	60% (.32)	61% (.33)	61% (.33)	56% (.33)	55% (.29)	−.03
Born 1940s	4,364	–	66% (.32)	68% (.32)	68% (.32)	70% (.30)	69% (.31)	71% (.30)	67% (.32)	68% (.31)	.06
Born 1950s	5,777	–	69% (.30)	70% (.30)	70% (.30)	74% (.29)	70% (.30)	73% (.28)	73% (.30)	71% (.29)	.07
Born 1960s	4,460	–	–	63% (.31)	67% (.29)	70% (.29)	73% (.28)	70% (.28)	70% (28)	73% (.29)	.33*
Born 1970s	2,557	–	–	–	–	71% (.27)	73% (.26)	73% (.28)	71% (.27)	75% (.26)	.15*
Born 1980s–1990s	1,254	–	–	–	–	–	–	72% (.25)	70% (.26)	73% (.25)	.04

Note: Cells with dashes indicate either that the question was not asked or that the group had less than 100 respondents during that time period. d = difference in standard deviations. At most levels, d = 2r. * = p < .05 or less, t-test comparing earliest years to latest years available.

Figure 1.

Americans' tolerance (by percentage) for marginalized outgroups, 1972–2012

Women (d = .56, from 54 percent in 1975–1979 to 71 percent in 2010–2012) increased in total tolerance slightly more than men (d = .41, 57 to 70 percent). Whites (d = .54, 56 to 74 percent) increased slightly more than Blacks (d = .42, 48 to 62 percent). Respondents who did not attend college increased more in tolerance (d = .41, from 51 to 64 percent) than those who did (d = .19, from 78 to 83 percent). Thus, the largest changes appeared among White men who did not attend college. In both eras, those who attended college were more tolerant.

Time Period, Generation, and Age Differences (Hypothesis 2)

Next, we examined whether the trends were due to time period (survey year), generation (birth year/cohort), or age. Consistent with Hypothesis 2a, the cross-classified HLM analyses revealed that time period (i.e., survey year) was the main driver of variability in total tolerance (see tables 2 and 3 and figure 2a). Thus, Americans of all ages and generations increased in tolerance for outgroups between the 1970s and the 2010s. Consistent with Hypothesis 2b, there was also a significant generational effect. However, somewhat inconsistent with the hypothesis, this effect was curvilinear rather than linear, with the first wave of Boomers born in the 1940s expressing the most tolerance, especially compared to the Silent generation born 1925–1945 (see figure 2b). More recent generations, such as Generation X and Millennials, were about average in tolerance when time period and age effects were removed. This can also be seen in the means in table 1; among those 18–29 years old, tolerance did not differ significantly between the mid-1970s (Boomers) and the 2010s (Millennials).

Table 2.

Regression Coefficients of Level-1 Age Predictors and Their 95% Confidence Intervals

		95% CIs
Fixed effect	Coefficient	Low	High
β0jk	.073	−.001	.151
β1jk, Age	−.277	−.333	−.222
β2jk, Age2	−.104	−.124	−.086
β3jk, Age3	.035	.021	.050

Table 3.

Variance Components for the Effect of Survey Year and Birth Cohort on Level-1 Regression Coefficients

Random effect		σ2	χ2	df	p
Variance components for:
Birth cohort effects	t00k (β0jk)	.003	215.84	97	.007
	t10k (β1jk)	<.001	114.12	97	.113
	t20k (β2jk)	<.001	108.05	97	.208
Time period effects	c00k (β0jk)	.035	449.84	24	<.001
	c10k (β1jk)	.016	181.43	24	<.001
	c20k (β2jk)	.001	55.77	24	<.001
	c30k (β3jk)	<.001	68.72	24	<.001

Figure 2.

Americans' tolerance by: (a) survey year (time period), (b) birth year cohort (generation), and (c) age; predicted values estimated by the cross-classified HLM model

When time period and generation are controlled, tolerance declines with age, with younger respondents more tolerant than older respondents (see figure 2c). Time period also had a significant effect on the linear slopes, β_1jk, and quadratic slope, β_2jk, of the regression of tolerance onto age. As figure 3 shows, the age trajectory of the tolerance decline is less steep in more recent years. In the 1970s, 18-year-olds were considerably more tolerant than 60-year-olds. In the 2010s, however, these age groups differed little in their level of tolerance (see figure 3). Thus, the time period increase in tolerance shown in figure 1a is primarily caused by tolerance levels not declining over the lifespan as sharply as they once did.

Figure 3.

Age trajectories in tolerance by decade of survey

Predictors of Tolerance (Hypothesis 3)

Consistent with Hypothesis 3, tolerant respondents were more likely to be individualistic, including believing people need to look out for themselves, not affiliating with a religion, believing marijuana should be legal, voting for a Black president, approving of premarital sex, and supporting working mothers (see table 4). Education was the strongest predictor of tolerance. These correlations show that, at an individual level, those who embrace individualistic, egalitarian, and nontraditional views, and who are more highly educated, are also more tolerant.

Table 4.

Predictors of Time Period Changes in Tolerance

	With total tolerance	With year	With total tolerance, matched by year
GSS items
Everyone should look out for themselves (1984)	.20	–	–
No religious affiliation	.19	.13	.87
Favors legalization of marijuana	.28	.14	.70
Would vote for a Black president	.23	.14	.90
Supports working mothers	.31	.15	.80
Premarital sex not wrong	.32	.11	.62
Years of school completed (education)	.42	.21	.96
Political party affiliation (higher = Republican)	.04	.06	.75
Liberal vs. conservative (higher = conservative)	−.14	.02	.56
Social indicators
GINI index of income inequality	–	.98	.97
Unemployment rate	–	−.10	−.20
Empathy (college students), 1979–2009	–	−.52	−.36
Need for uniqueness (uncommon names)	–	.99	.98
Individualistic words	–	.71	.63
Individualistic phrases	–	.95	.97

Note: Total tolerance is the mean tolerance composite estimate by year, adjusted for cohort and age in the HLM analyses. r = .02 significant at p < .01; all r > .03 significant at p < .001.

A second question is whether trends in these attitudes might help explain the increase in tolerance over time. Educational attainment and rejection of social rules also increased over time (see table 4), suggesting that they co-occurred with the increase in tolerance and may be among its possible causes. As a second step, we examined the correlations between mean tolerance (using the coefficients for each year from the HLM analyses) and the means on the other variables matched by year (see, e.g., Twenge, Abebe, and Campbell 2010), as well as other indicators of individualism and economic conditions available at the group level (individualistic language, need for uniqueness, empathy, unemployment, and income inequality). These analyses provide a view of how closely the change in each variable followed the change in tolerance over time at the group level. (These are ecological correlations, but are appropriate, as we are analyzing at the group level for cultural change). These analyses showed that education and all of the variables measuring rejection of social rules and individualism were significant predictors of tolerance over time (see table 4). Thus, as adult Americans rejected traditional social rules and embraced individualism, tolerance also increased.

Years with more tolerance were also years with less empathy among college students, suggesting that the two constructs—at least at the cultural level—are distinct. Liberal political views were correlated with more tolerance. However, tolerance was only weakly linked to political party, and neither variable was strongly correlated with year. Thus, it is unlikely that changes in political views per se caused the increase in tolerance. Unemployment was weakly but significantly linked with tolerance; years with less unemployment were years with higher tolerance. Income inequality, which has risen steadily over the years, was highly correlated with tolerance. Thus, years with more income inequality were years with higher tolerance.

We also examined the relationship between income inequality and the generational differences in tolerance by matching the birth cohort coefficients with income equality 20 years later (when respondents were young adults; with the GINI index available beginning in 1967, this included only those born in 1947 and later). The correlation between a birth cohort's tolerance and income inequality was r(38) = –.52, p < .01, suggesting that tolerance was high among birth cohorts (such as the Boomers) who experienced less income inequality when they were young.

Discussion

Americans have become increasingly tolerant of controversial beliefs and lifestyles (i.e., marginalized outgroups). They are more likely to believe that homosexuals, Communists, militarists, and the anti-religious have the right to give speeches, teach at a college, and have a book in a local library. Smaller increases appeared in tolerance for a person who claims that Blacks are genetically inferior (commonly labeled a racist). Increases in tolerance were larger among respondents who did not attend college. HLM analyses separating the effects of time period, generation, and age (based on Yang [2008]) showed that the increase in tolerance was caused by a combination of time period and generational effects, suggesting that the increase in tolerance is a broad cultural trend. At the generational level, Boomers were more tolerant than other generations when time period and age effects are controlled.

The increase in tolerance co-occurred with increases in individualistic beliefs such as rejecting traditional social rules around gender, race, religion, sexuality, and drug use. At the group level, tolerance was higher in years with more individualistic language in books and a higher need for uniqueness. These analyses cannot infer causation, but these results are consistent with our hypothesis that increasingly individualistic attitudes may be one cause of increasing tolerance for outgroups. Tolerance was also strongly correlated with educational attainment, which has also increased over time. This suggests that as Americans have completed more years of formal education, more have learned tolerance for outgroups.

However, years with higher tolerance were also years with lower empathy among college students. This is not as contradictory as it may seem: tolerance and low empathy are both linked to high individualism (Brandt 2011; Konrath, O'Brien, and Hsing 2011; Watson, Biderman, and Sawrie 1994), which may be the third variable causing both to increase over time. These results suggest that tolerance and empathy are distinct constructs and may even oppose each other. For example, at Rutgers University in 2010, a young man (Dharun Ravi) expressed tolerance of his roommate (Tyler Clementi) being gay, but displayed a lack of empathy by filming Clementi's sexual encounter with another man, leading to Clementi's suicide (Foderaro 2010). This is an extreme example, but it illustrates that tolerance and empathy can be distinct.

Tolerance was also higher in years with more income inequality. This may be simply a product of both increasing in recent years, or it could be that the rise in individualism underlies both trends. By generation, tolerance was highest among Boomers, who experienced less income inequality when they were young. This may be one reason why Boomers broke new ground in tolerance while younger generations such as GenX and Millennials continued the trend toward tolerance but did not break from the general time period trend.

If one generation always continues the trends of the previous generation, one might ask how social change ever occurs at all. In this case, the significant time period effects suggest that some Americans, perhaps primarily the first wave of Boomers, rejected their parents' lack of tolerance during adolescence or young adulthood and then raised their own children with these attitudes. This may have had a multiplier effect if tolerant parents always lead to tolerant children, but non-tolerant parents do not always lead to non-tolerant children. Thus, the time period and generational effects worked in tandem to produce a faster increase in Americans' tolerance for marginalized outgroups.

The increases in tolerance for gays and lesbians occurred at the same time that government policies (such as same-sex marriage and anti-discrimination laws that included sexual orientation) were beginning to change. Thus, as shown by Lax and Phillips (2009) when comparing variation among US states, voter opinion and state policies are fairly congruent. In other words, when public opinion shifts, so do government polices, at least around issues of tolerance for gays and lesbians.

The results could be interpreted as indicating a growing liberalization of attitudes among the US population, as tolerance for groups associated with liberal causes (homosexuals, Communists, atheists) has increased while tolerance for a non-liberal viewpoint (believing that Blacks are genetically inferior; Leonard [2005]) has increased to a smaller extent. However, tolerance for people with a belief in military rule, which is also not a traditionally liberal view, has also increased. In addition, self-identifying as a liberal or Democrat has decreased in the GSS since the 1970s, over the same time that tolerance increased. This suggests that the results are not due to the US population becoming more politically liberal. It is plausible that these views correlate with each other because they are associated with a more progressive rather than strictly liberal perspective, but the GSS does not separate liberal from progressive values.

Many of the effect sizes are moderate (around .50) to large (more than .80; Cohen [1988]; note that Cohen did not intend these values as cutoffs, but as a rough guide for interpretation). Tolerance for the civil liberties of controversial groups nearly doubled among many age groups; for example, Americans in their 60s were nearly a standard deviation more tolerant in the 2010s than their predecessors were in the 1970s (see table 1). Thus, the argument that time period and generational differences are too small to matter (Trzesniewski and Donnellan 2010) does not seem to apply to the attitudes and generations examined in the current paper. These larger differences in tolerance supplement other findings on the growth of individualism, such as the moderately sized changes in positive self-views (Twenge, Campbell, and Gentile 2012a).

Danigelis, Hardy, and Cutler (2007) examined these GSS items until 2004, focusing on intra-cohort aging. They concluded that people do not grow more conservative in their attitudes with age but instead grow more tolerant, with a surprising amount of change among people in their 60 s. The current analysis instead found that tolerance decreases with age when time period and generation are controlled. In addition, changes with time period and generation (those of the same age over time) are much larger than changes with age and time period (following cohorts over time; see table 1). Thus, it seems likely that respondents in their 60 s did not show increased tolerance because they grew older, but because respondents in their 60 s underwent time period and generational shifts. Thus, our analyses suggest that the change observed by Danigelis, Hardy, and Cutler (2007) was not due to changes in attitudes during middle age, but to time period effects and generational replacement.

Limitations

One limitation of the current analysis is that the HLM coefficients were based on the available data. Those born in the 1920s and before were already in their 40 s and older when GSS data collection began in the 1970s. Similarly, as of 2012, those born in the 1980s had not yet reached their late 30 s or beyond, and those born in the 1950s had not yet reached their mid-60 s. Thus, it is possible that the apparent decline in tolerance with age may be partially due to generation, as Boomers, GenXers, and Millennials have not yet reached older ages. If the age trajectory of tolerance is different for these groups (and figure 3 suggests it might be), then future analyses incorporating more comprehensive life-span data may find that generation explains more of the increase in tolerance than suggested here.

Another limitation, as mentioned in the introduction, is that it is difficult to determine a precise role for individualism in changing tolerance. We provide some correlational data consistent with individualism, but the other mechanisms detailed earlier may also contribute (e.g., differential birth/mortality rates, parasite prevalence). Furthermore, these factors may interact in ways that we did not examine. Cultural change is complex, and while these data provide a clear picture of increasing tolerance and a relatively clear picture of the importance of a time period effect in this process, the role of individualism and other proposed mechanisms is primarily suggestive.

Finally, other variables may also contribute to increasing tolerance. For example, the communicability of stereotypes plays a role in their transmission (Schaller, Conway, and Tanchuk 2002). Perhaps the communication of negative stereotypes of and intolerance toward these groups has declined, resulting in increased tolerance. This possibility, however, cannot be addressed with the current data.

Conclusion

Overall, these results demonstrate considerable growth in the acceptance of the public expression of people with controversial beliefs or lifestyles. Americans are increasingly likely to believe that people with minority opinions—those substantially different from the larger group—have the right to speak, teach, and have their books in a community library. Both time period and generation are behind the increase in tolerance, possibly driven by concomitant increases in education and individualistic views rejecting traditional social rules. The increase in tolerance is consistent with an American culture that has become markedly more individualistic over the past few decades.

About the Authors

Jean M. Twenge, Professor of Psychology at San Diego State University, is the author of more than 100 scientific publications and the book Generation Me: Why Today's Young Americans Are More Confident, Assertive, Entitled—and More Miserable Than Ever Before and (2nd ed., 2014).

Nathan T. Carter is Assistant Professor of Psychology at the University of Georgia. His research interests include understanding the use of psychological measures in organizational settings, the accuracy of psychometric models, and the role of human judgment and decision-making in employee selection.

W. Keith Campbell is Professor and Head of the Department of Psychology at the University of Georgia. He is the author of more than 120 scientific publications, coauthor (with Jean M. Twenge) of The Narcissism Epidemic: Living in the Age of Entitlement, and coeditor (with Joshua D. Miller) of The Handbook of Narcissism and Narcissistic Personality Disorder.

Appendix

Details of the HLM Analyses

We estimated a regression equation that tested linear, quadratic, and cubic terms to determine the trajectory of tolerance over the lifespan:  

$$Y_{ijk}={\beta }_{0jk}+{\beta }_{1jk}(Age)+{\beta }_{1jk}(Ag{e}^2)+{\beta }_{1jk}(Ag{e}^3).$$

(1)

This model takes into account the inherent nesting in cross-sectional data involving persons of different ages, nested within different birth year cohorts, which are in turn nested within the year the survey was conducted. We use the IRT-based estimate of the dependent variable, Y (i.e., total tolerance), for every person, i, in each birth year cohort, j, for every survey year, k.

The model in Equation 1 incorporates quadratic and cubic effects of age, denoted by Age² and Age³. To separate age effects from the effects of birth year cohort membership and the year the survey was conducted, a level-2 model was created that finds the unexplained variance in Y that is attributable to birth cohort and survey year. The level-2 model is stated:  

$$\begin{array}{l}{\beta }_{0jk}={\pi }_0+t_{00j}+c_{00k}\\ {\beta }_{1jk}={\pi }_1+t_{10j}+c_{10k}\\ {\beta }_{2jk}={\pi }_2+t_{20j}+c_{20k}\\ {\beta }_{3jk}={\pi }_3+c_{30k}\end{array}$$

(2)

The level-2 model treats each regression coefficient as an outcome variable. β_0jk is the mean of Y across birth cohorts and survey years at the average age, and this mean is partitioned into “rows” and “columns.” More specifically, the deviations from the mean β_0jk are taken across the “rows,” represented by t_00j, to estimate the effects for each birth year cohort. Effects away from β_0jk based on the survey year are represented by c_00k. Considerable effects for slopes, such as β_1jk, would be indicated by large variance components for t_10k and c_10k, and suggest that the linear slope of age in Equation 1 would differ by birth cohort or survey year, respectively. That is, for each coefficient, row and column effects on that coefficient are evaluated by the size of the variance component, σ², one for each t and c coefficient in Equation 2. Notably, β_3jk is only predicted by c_30k, and t_30j is set to zero. This was done because including the weight variable caused the HLM7 program to require a parameter to be fixed. We chose to fix t_30j to zero because in unweighted analyses, this parameter had a near-zero (< .001) variance component (with p > .50).