In the last 30 years, attitudes toward different sexual orientations¹ have rapidly become more tolerant around the world. Some attribute this to a bi-product of economic growth and the expansion of post-materialist values (Andersen and Fetner, 2008) and declining religiosity (Sherkat et al., 2011), increasing education levels in society (Loftus, 2001) and the media visibility of gays and lesbians (Schiappa et al., 2006).

There is clear evidence that younger people are more tolerant of gays and lesbians than older people (Halman and van Ingen, 2015; Dotti Sani and Quaranta, 2020), a pattern that is common regarding shifting social issues. For this reason, including age as a demographic variable in cross-sectional research without distinguishing the meaning of age from other temporal phenomena obscures important social dynamics in attitudes. Attitudinal differences between the old and the young are not necessarily due to aging, in other words, the process of becoming older. Rather, the age of a person holds different substantive meanings (Riley, 1973, 1987; Kertzer, 1983). In addition to their placement along the lifecycle, a person's age also indicates their birth year, which carries its social significance. Birth cohorts, individuals born around the same time, can experience different socializations through a unique sequence of events and circumstances which have long-lasting impacts on their socio-political attitudes (Inglehart, 1977).

There is an existing debate about whether this growing liberalization of attitudes toward gay and lesbian people is due to a period effect or a cohort effect. A period effect refers to an underlying general trend toward more tolerant attitudes for all of society which are often driven by events or institutional change. In the case of a period effect, a trend is observable if there is a consistent worsening or improvement of attitudes vis-à-vis a given issue. This can occur for two reasons: either because people change their opinions or because the composition of the population changes, such as through the mechanism of cohort replacement. Thus far, there is evidence of both period and cohort effects (Treas, 2002; Ruel and Campbell, 2006; Andersen and Fetner, 2008; Sherkat et al., 2011; Sheoran et al., 2012; Pampel, 2016). Baunach (2011) decomposes this trend and finds that while two-thirds of the changes in the United States are due to intracohort aging (individuals changing their views), one-third are due to a cohort effect.

Scholars have been more recently interested in explaining what appears to be a non-linear trend of attitudes across cohorts. Recent research shows a fluctuating or inconsistent trend of cohort effects, implying that this is not simply due solely to a period or aging effect. Younger cohorts tend to be less politically tolerant than baby boomers and what seems to be a general increase in approval of same-sex relationships may simply be an artifact of baby-boomers making up a larger portion of the general population (Schwadel and Garneau, 2014). In a recent study, Ekstam (2021) finds discontinuities in the pattern of adjacent cohorts and suggests that this is due to the effects of different political socialization of cohorts, who come of age in different formative contexts. At present, much of the existing attention in the socialization of attitudes toward same-sex relationships has been toward micro-socializing factors such as religion or education or macro-socializing factors such as legislation for same-sex relationships (for a discussion see Dotti Sani and Quaranta, 2020).

This article addresses the question of why cohorts differ in their attitudes toward gays and lesbians by investigating the role of contextual values during formative years. It is well established that ideas and values that are part of the zeitgeist of a cohort's youth have a formative impact on their worldviews throughout their adulthood (Ryder, 1965). According to the “impressionable years” argument, socio-political orientations are acquired during young adulthood. During youth, people experience a window of “plasticity” while they transition from adolescence to young adulthood as they engage for the first time with the social and political world around them (Marsh, 1971; Niemi and Sobieszek, 1977; Hanks, 1981; Sapiro, 2004; Neundorf et al., 2013). According to this view, a person's orientations are then crystallized and remain remarkably persistent as the person grows older. As a result, attitudes and values become more deeply entrenched and become more stable over the lifetime (Sears, 1981; Jennings and Gregory, 1984; Stoker and Jennings, 1995; Visser and Krosnick, 1998; Lewis-Beck, 2009). Following this reasoning, symbolic predispositions toward different sexual orientations formed in young adulthood are then solidified throughout the individual's life course. So, after the critical period of young adulthood, we would then not expect a person's biological age to be a factor in underlying their attitude and preferences.

While much emphasis has been placed on “landmark events” (in other words, exogenous shocks) and how they leave a mark on political socialization (see Sears and Valentino, 1997), there is now growing attention to more subtle shifts in the political values can have lasting effects (Grasso, 2014). For instance, in their study of attitudes toward immigration, Jeannet and Dražanová (2023), analyze the role of the political climate during formative years as a socializing agent in a person's political life. They define political climate as “the normative principles, beliefs, ideals, and values that prevail in the political zeitgeist” during a person's impressionable years [Jeannet and Dražanová, 2023, p. 6].

Similar to attitudes toward immigration, attitudes toward different sexual orientations are strongly related to human values. Values represent a person's fundamental priorities and as such underlie and motivate attitudes (Davidov et al., 2008). Schwartz (1994) offers a framework of certain basic values which are universal and generally explain the inevitable human tension between an openness to pursuing change and an inclination toward conservation. Kuntz et al. (2015) examine the relationship between attitudes toward gays and lesbians and Schwartz (1994)'s human values, finding a positive association between Schwartz's values of universalism and acceptance, while values of tradition lead to the opposite outcome. According to Schwartz (1994) tradition is “respect, commitment, and acceptance of the customs and ideas that traditional culture or religion provide” whereas universalism is “understanding, appreciation, tolerance and protection for the welfare of all people and nature” (Schwartz, 1994, p. 35). While Schwartz names his value concept “universalism”, it is clear from its definition that it is tapping into the importance of equality in society.

What is the mechanism between these values and attitudes toward gays and lesbians? The value of tradition underlies disapproval of different sexual orientations which are often perceived as a threat to “traditional family values” and accepting them would imply a perceived abandonment of tradition that such individuals hold dear. Therefore, individuals who hold tradition as a strong value can be more motivated to oppose the disruption of the status quo as sexual mores shift (Kuntz et al., 2015). On the other hand, universalism values, such as egalitarianism, foster the approval of different sexual orientations by prioritizing “tolerance and acceptance of those who differ from oneself, understanding rather than rejecting those with unconventional lifestyles”. Universalism values emphasize equal opportunities for all (Kuntz et al., 2015, p. 122).

While the work of Kuntz et al. (2015) examined the relationship between a person's contemporary values and their attitudes, our argument does not dispute this. Instead, this article offers a different investigation and contribution. We consider the role of the formative value context, during a person's coming-of-age period, and test whether this has an impact on the attitudes toward different sexual orientations when they are adults.

The analysis relies on the European Social Survey (ESS), rounds 1 (2002) to 10 (2020). We use 13 countries that have participated in at least five rounds of the survey: Austria, Belgium, Switzerland, Germany, Denmark, Finland, France, Great Britain, Ireland, Italy, Netherlands, Norway, and Sweden.² Following standard practice in cohort analysis, we group individuals into cohorts based on five-year birth intervals. We, therefore, have 14 cohorts and 206,906 respondents in total.

The dependent variable measures the extent to which respondents agree with the statement that “gay men and lesbians should be free to live their own life as they wish”. The response categories are ordinal and range from 1 (agree strongly) to 5 (disagree strongly). We recorded the variable so that higher numbers mean more positive attitudes.³ This survey item has been frequently used as an indicator of attitudes toward gays and lesbians in extant research (van den Akker et al., 2013; Abou-Chadi and Finnigan, 2019; Dotti Sani and Quaranta, 2020, 2022).

For our two independent variables, we rely on data from The Manifesto Project, a widely used data set that includes a coded content analysis of party manifestos since 1945. Using this data, we measure the two values of equality and tradition as the share of quasi-sentences calculated as a fraction of the overall number of allocated codes per manifesto (Volkens et al., 2018). We conceive of the principle of equality as a positive view of social justice and the need people to be treated fairly. “This includes references to topics such as special protection for underprivileged social groups, removal of class barriers, need for fair distribution of resources and the end of racial or sexual discrimination” (Volkens et al., 2018, p. 17). On the other hand, the principle of tradition is coded as positive or favorable mentions of traditional and/or religious moral values. “This includes references to topics such as prohibition, censorship and suppression of immorality and unseemly behavior, maintenance and stability of the traditional family as a value, and support for the role of religious institutions in state and society” (Volkens et al., 2018, p. 19).

Borrowing a similar approach by Jeannet and Dražanová (2023), we compute annual measures by country from the manifesto data. We do this by weighting the share of votes that the party has received in the country's last elections.⁴ As we are interested in the individual's formative political environment, rather than the environment at the time of the survey, we assign individuals the values of tradition and equality when cohorts were between the ages of 18 and 23 by averaging the five years of each cohort.

We also include a series of controls using the European Social Survey based on individual-level characteristics which have been shown to be associated with attitudes toward sexual orientation in prior research. Individuals who are older, less educated and male tend to hold more negative attitudes (Schwadel and Garneau, 2014; Halman and van Ingen, 2015). We control for religiosity since individuals who are more religious tend to oppose same-sex relationships (Adamczyk and Pitt, 2009), living in an urban/rural area (Jakobsson et al., 2013), socio-economic position (Anderson and Fetner) by perceived income difficulties (Andersen and Fetner, 2008) and left-right self-positioning (Baunach, 2012).

Empirical evidence also shows that attitudes toward sexual orientation depend on context (Adamczyk and Pitt, 2009) and can vary cross-nationally (Dotti Sani and Quaranta, 2020). We, therefore, control for policy regulations such as gay marriage or the possibility for registered partnership based on Waaldijk (2020) data tracking the years when either registered partnership or same-sex marriage were introduced at the cohort and period level within each country, the percentage of university-educated people at the cohort and period level within each country and the level of unemployment at the cohort and period level within the country (Inglehart, 2008).

Our empirical strategy employs hierarchical age-period-cohort regression analysis (HAPC) using repeated cross-sectional surveys (Zheng et al., 2011). HAPC analysis separates the three temporal phenomena of age, period (year of survey) and birth cohort (year of birth) effects using micro survey data (Yang and Land, 2013). It allows us to construct synthetic cohorts based on age groups where individuals are cross-classified by both country-period and country-cohort. By constructing cohorts, the age, period and cohort is no longer perfectly collinear. This is because one cannot derive the exact age of the respondent from the age and year, but only a range of possible ages. Similar approaches have been used in the study of cohort effects in public opinion (Smets and Neundorf, 2014; Gorodzeisky and Semyonov, 2018).

Following Jeannet and Dražanová (2023), we apply hierarchical three-level age-period-cohort models. In these models individuals are nested both within two second-level variables (country-cohort and country-period) as well as nested within countries since possible clustering at the country level might still occur.

The models also include random effects for cohorts and periods. We assume that while age is related to biological processes of aging, cohort and period effects are the result of external influences and are to be considered as macro-level variables (Yang, 2008).

where, within each country-cohort j, country-period k and country c, respondents' attitudes to sexual orientation (Y) are a function of their individual characteristics (vector X). β_{0j kc} is the mean of attitudes of individuals in country-cohort j, country-period k, and country c, β₁ is the level-1 fixed effects and e_ijkc is the random individual variation.

where Z is a vector of country-cohort characteristics and T is a vector of country-period characteristics, μ_0jc is the residual random effect of country-cohort j, ν_0kc is the residual random effect of country-period k. The level-3 model is:

where ω_0c is the residual random effect of country c. In all three models (1), (2) and (3) μ_0j, ν_0k and ω_oc are assumed normally distributed with mean 0 and variance τμ, τν and τω respectively.

We have grand-mean centered all continuous individual, country-cohort and country-period level variables.