Social Ties and the Influence of Public Policies on Individual Opinions: The Case of Same-Sex Marriage Laws

Abstract

1. Introduction

Since the early 2000s, laws protecting and promoting gay rights¹ have been progressively adopted and populations seem to have become overall more accepting of gay rights. In the United States, only 39% of people would vote against same-sex marriage (SSM) in 2016, against 68% in 1996 (see Roy and Abaid 2017). In the European Union (EU), 61% of people agreed that “SSMs should be allowed throughout Europe” in 2015, while they were only 44% in 2006 (Eurobarometer 2015). Still, attitudes toward homosexuality vary greatly across countries and in all countries some remain strongly hostile to gay rights, as exhibited by the rise of homophobic hate crimes in a number of countries. According to the FBI’s annual Hate Crime Statistics Report, the number of hate crime incidents targeting gays, lesbians, and bisexuals in the United States in 2018 increased by nearly 6% compared with the previous year. In the United Kingdom, 11,600 crimes motivated by a bias toward gender identity were recorded by the police in 2017–2018, an increase of 144% since 2013–2014 (Marsh et al. 2019). In France, 231 assaults were reported in 2018, accounting for an increase of 66% in one year (SOS Homophobie).

In this paper, we aim to evaluate the extent to which SSM laws, which were passed in several countries in the 2000s and 2010s, contributed to gay-friendly opinions in the population. There are direct and indirect effects of a law change. In addition to direct adjustments of opinions or behaviors to the new law, there is an indirect effect that comes through information transmission among people, and this transmission is affected by the nature of the social ties through which the information transits. This paper focuses on the indirect effect and investigates how social interactions act as a factor facilitating or, on the contrary, preventing the acceptance of gay rights that SSM laws officialized. We first propose a theoretical framework with social interactions to model how law and social norms interact in the opinion formation process. We then test our theoretical predictions in a dynamic difference-in-difference setting relying on progressive adoption of SSM laws in European countries to control for the concomitant evolution of other factors that may also have helped the development of gay-friendly opinions.

Social ties convey information through observation of others’ decisions as well as through conversations and the sharing of opinions. Individuals belonging to a social network are influenced in their opinions and choices by those of other members of the social network. However, different opinions may be advocated in an individual’s social group, and some peers may be more or less influential.

To understand how individuals form opinions in such a context, we adapt the dyadic theoretical setup proposed in Giulietti et al. (2018). We assume that individuals develop both weak and strong ties. They engage in close social interactions with a partner (the other member of the dyad—the strong tie), that may or may not accept the dominant social norm and they also have random encounters (the weak ties), who are assumed to create and foster the social norm.² Individuals can only receive information on the social norm through social ties. They are even more exposed to the social norm if their partner agrees with it. They are said to be integrated when they adopt the dominant social norm, non-integrated otherwise.

Our theoretical setup predicts that, if a law does not clash with the dominant social norm and then reduces the rejection rate of this norm, individuals with dominant strong ties will be on average less responsive to the law, as they may be less exposed to the social norm it conveys (if their partner is non-integrated). On the contrary, if the law clashes with the dominant social norm, it may lead to an increase in the rejection rate of this norm. In this case, the role of weak ties as diffusors of the officialized social norm is reduced and the relative influence of strong ties in promoting the norm is improved. Applied to the particular case of gay rights, we expect SSM laws adopted by European countries during the 2000s and 2010s to reduce the rejection rate of gay rights in these countries, since the average opinion was favorable to these rights. However, in communities (within countries) where the local³ social norm was homophobia, SSM laws may have rather promoted a rejection of gay rights since opinions radicalize.

We test our theoretical predictions using the 2002–2016 European Social Surveys (ESSs) and exploiting variations in the timing of SSM laws across European countries. More precisely, we adopt a pseudo-panel (cell) approach to implement a difference-in-difference model that exploits sequential approval of SSM across countries to assess the impact of the law on the cells’ average opinion about gay rights. Following Wolfers (2006) and Stevenson and Wolfers (2006), we estimate the dynamic response to the law to account for possible time-varying effects on opinions.⁴ To control for the endogenous timing of the law, we control for country- and cohort-specific trends (in addition to the usual country-, year-, and cohort-fixed effects). To control for socio-demographic changes over time, which may also participate in changing overall opinion about gay rights, cells are weighted so as to maintain constant the socio-demographic structure of the sample.

To assess the extent to which the effect of a law legalizing the dominant social norm differs depending on social interactions and on the alignment between the law and the local social norm prevailing in the group, we distinguish between cells with dominant strong ties and cells where strong ties are not dominant, as well as between cells where the dominant local social norm aligns with the social norm promoted by the law and cells where it does not. We assume that an individual has a strong tie if she has a partner, which is in line with our dyadic theoretical framework. To address potential biases from reverse and simultaneous causation between the nature of social relations and the opinion about gay rights, we fix the ties’ composition of each cell to its level in 2002, that is, the first year observed in the data. We consider that the dominant local social norm aligns with the social norm promoted by the law when the majority of people in the cell is favorable to gay rights. Again, to address reverse and simultaneous causation we will consider the majority opinion of the cell in 2002. Furthermore, we propose an original strategy exploiting the homophobic or non-homophobic origin of immigrants to analyze the role of social ties in shaping individual’s opinions depending on whether the law aligns or not with the local social norm.

Consistently with our theoretical framework, we find that when considering the whole sample or a subsample composed by a majority of gay-friendly individuals, cells with dominant strong ties are less likely to positively change the opinion about gay rights following SSM approval. This result is reversed when we focus on a community of people for whom homophobia is the local social norm, in spite of living in a country where the aggregate social norm is the full recognition of gay rights. In this case, people with dominant strong ties react less negatively following the approval of SSM law.

We combine and contribute to three streams of literature. The first body of literature we relate to has its main representation in the recent paper of Acemoglu and Jackson (2017), which focuses on how social norms are shaped by laws while, at the same time, social norms constrain the effectiveness of laws.⁵ The interplay between social norm and law is complex. On some societal issues the question is repeatedly asked whether the law can precede the norm and change behaviors and opinions (Benabou and Tirole 2011). As pointed out in Acemoglu and Jackson (2017), in some circumstances, we observe a rapid and profound change in the social norm following the passing of the law—for example, the impact of anti-discrimination and anti-racist laws on norms and practices in the US South (Wright 2013)—while in other cases, the law appears ineffective in changing behaviors and opinions of a large share of the population—for example, the (un)enforcement of British laws in Afghanistan (Barfield 2010), in India, and Kenya (Parsons 2010)—or may even have countervailing effects (Parisi and von Wangenheim 2006; Bishin et al. 2016). Acemoglu and Jackson (2017) underline that a conflict between prevailing norms and new laws often renders such laws ineffective. Because authorities cannot check if everyone is respecting the law, they must rely on whistle-blowing by private citizens or firms to enforce the law. When the law goes against the prevailing norms, that is, restricts the behavior excessively relative to the distribution in the society, then most people prefer breaking the law, which reduces whistle-blowing and hence the effectiveness of laws. Moreover, as laws are broken by more people, whistle-blowing becomes even less likely, and law-breaking snowballs. The authors conclude that “gradual imposition of laws that are more in accord with prevailing norms can successfully change behavior and thus future norms.”

In the present paper, we consider the effect of the law on the opinion of individuals about a particular societal issue and we do not refer to law enforcement, so that the mechanism of whistle-blowing cannot apply here. In our paper, the snowball effect is at play depending on the local norm, the nature of the individual’s social ties and opinions of the partner and encounters. In a study close to ours, Tankard and Paluck (2017) use a controlled experimental setting to study reactions to the June 2015 US Supreme Court ruling in favor of SSM. They found an increase in perceived social norms supporting gay marriage after the ruling but no average change in personal attitudes. Tankard and Paluck (2017) do not consider potential heterogeneous changes in attitudes depending on the nature of social interactions. The idea of our work is that it is the social norm prevailing in the individuals’ social network that matters to shape opinions. Agents tend to imitate choices and opinions of their ties (conformity within groups was first studied by Banerjee [1992] or Bikhchandani et al. [1992] and has also been recently analyzed by Bursztyn et al. [2020]). This imitating behavior is likely to affect the potential influence of the law on individuals’ choices and opinions.

The second stream of literature we contribute to is about networks and the strength of interpersonal relationships (weak versus strong ties), which has already shown quite extensively how strong and weak ties affect individuals’ decisions and behaviors in a number of areas, such as employment, migration, or criminality (see, e.g., Calvo-Armengol et al. 2007; Hemet 2015; Zenou 2015; Giulietti et al. 2018).⁶ We investigate here how the nature of social ties may matter in an area where evidence is still scarce, that is opinion formation and the diffusion of a social norm (see French [1956], Harary [1959], DeGroot [1974], for seminal theoretical references, and Golub and Sadler [2016], for a detailed review on learning in social networks).

On a close subject, Broockman and Kalla (2016) analyze the influence of social interactions on individuals’ opinions about antitransgender prejudice, using a field experiment with a door-to-door canvassing intervention in South Florida. They show that these conversations substantially reduced transphobia (this effect persisting for 3 months), thus pointing to the importance of social interactions as drivers of individual opinions. While Broockman and Kalla (2016) focus on weak ties (i.e., canvassers did not previously know the interviewee), in our paper, we analyze the differentiated influence of weak ties and strong ties in the formation of opinions.

Last, our work relates to the specific literature about gay-friendly attitudes. Existing studies show that attitudes toward gays and lesbians are influenced by a number of factors, including religious and moral values, political affiliation, contacts and social interactions with gays, lesbians and same-sex couples, and demographics such as gender and age (Baunach 2012; Keleher and Smith 2012; Garretson 2015; Broockman and Kalla 2016; Roy and Abaid 2017). In this respect, recent studies show that historical events, by creating a variegated spatial pattern of sex ratios, help to explain the spatial distribution of current attitudes toward homosexuality. Baranov et al. (2020a  2020b) provide such evidence for Australia by exploiting the British convict transportation to the country in the 18th and 19th centuries. Brodeur and Haddad (2018) show similar consequences from the US gold-rushes during the mid-to-late 19th century. Moreover, Fernández et al. (2019) show that the AIDS epidemic in the United States promoted greater political organization and mobilization of the gay community in the late 80s and 90s. Mainstream political parties had to take positions on gay-related issues for the 1992 presidential election, and there were heated political debates and large media coverage. After the election, the authors find that states that were more exposed to the AIDS epidemic experienced a greater change in the positive opinions about gay people.

The positive influence of SSM laws has been documented by Flores and Barclay (2016), Tankard and Paluck (2017), Ofosu et al. (2019), and Sansone (2019) on US data, and by Aksoy et al. (2020), Takacs and Szalma (2011), and Hooghe and Meeusen (2013) on European data. Similarly to these last three studies, we use the ESS and exploit variations in the timing of SSM laws across European countries to evaluate the effect of SSM laws on people’s opinions about gay rights. In contrast, we provide two types of contributions. First, we focus on the heterogeneous effects of the law with respect to the nature of social ties and investigate to what extent the nature of social ties may favor or, on the contrary, limit the influence of laws on people’s opinions about gay rights depending on the alignment between the law and the dominant social norm at the aggregate and local community levels. Second, we adopt a slightly different empirical strategy that allows for dynamic reactions to the passing of a law. The formation of opinions and the diffusion of social norms within the population are not an instantaneous process. In the specific context of SSM, its adoption is likely to have time-varying effects: the debates surrounding the law passing may exacerbate extreme opinions and once the law is passed the social norm conveyed by the law may progressively be adopted or may foster an increased rejection as individuals interact with more and more same-sex married couples.⁷ Our approach specifically addresses such dynamics.

The remainder of the paper is organized as follows. The theoretical setup is presented in Section 2. Data and variables are described in Section 3 and the empirical strategy is detailed in Section 4. Estimation results are analyzed in Section 5 and Section 6 concludes.

2. The Model

The objective of this section is to illustrate the economic rationale behind the econometric results proposed in Section 5. We adopt the same framework as in Giulietti et al. (2018) but, instead of analyzing migration decisions, we analyze the adoption of a social norm by the individual. The social norm considered here is the acceptance of gay rights. As shown in the following section, between 2002 and 2016, the average share of people in European countries agreeing or strongly agreeing that gays and lesbians have the right to live their life as they want progressed from 70% to 80%. The sequential approval of SSM by European countries during this period of time is likely to have directly modified the rejection rate of the social norm. We show here how the impact of the law on individuals’ opinions may also be indirectly shaped by the nature of ties depending on the dominant local norm.

2.1 Dyads and Social Interactions

We consider an economy composed by a population whose size is normalized to one. Time is continuous and individuals live forever. As in Calvo-Armengol et al. (2007), Zenou (2015), Hemet (2015), or Giulietti et al. (2018), we assume that individuals have two types of social interactions. First, they belong to mutually exclusive two-person groups, referred to as dyads. Two individuals belonging to the same dyad hold a strong tie with each other. We can thus think of strong ties as marriage or civil partnerships, where two individuals agree to sign a social and economic contract to stay together and the contract is enforced by law.⁸ Second, individuals may have weak ties, which are encounters outside the dyad partnership (or random encounters). These relationships will correspond to colleagues or friends met time to time. There is no emotional or socio-economic contract between weak ties, which can be easily broken up.

We assume repeated random pairwise meetings over time. Meetings can take place between dyad partners or not. At time t, each individual meets a weak tie with constant probability ω and her strong tie partner with constant probability α. We assume that these probabilities are exogenous and independent of the integrated nature or not of the individual and of the social tie.¹⁰ In line with Giulietti et al. (2018), we assume that individuals spend some time with their weak ties (captured by ω) and some time with their strong ties (captured by α) and impose neither complementarity nor substitutability between these two types of interactions.

2.2 Information Transmission

Information on the social norm can only arrive through social ties. The pace at which individuals receive this information depends on the status of the dyad partner and on the time spent with strong and weak ties. By definition weak ties always promote acceptance of the social norm but, as shown below, their relative influence is increased when the social norm is accepted by the majority of people in a society, that is, i_t > u_t. If the individual’s dyad-partner is integrated, the influence about the social norm will arrive through both strong ties and weak ties. Conversely, if the partner is non-integrated, information about the social norm can only arrive through sporadic contacts outside the dyad (i.e., weak ties).

Quite naturally, we assume that individuals are more willing to accept the social norm when the information comes from a strong tie rather than from a weak tie. This difference is explained by the fact that a strong tie who has a long-term relationship with the non-integrated individual is considered to provide more reliable information. To simplify, the probability of accepting the social norm is set to 1 when the information is received from the strong tie and is equal to $0<p<1$ when it comes from a weak tie.

We assume that individuals that already accept the social norm will only change their mind and start rejecting the social norm following an exogenous shock that arrives with probability δ—a hypothesis which is also consistent with the assumptions in Giulietti et al. (2018).¹¹ In our case, the approval of SSM is likely to modify this probability of rejection, δ.¹² We can reasonably argue that the approval of SSM was aimed at reducing the rejection rate of gay rights. However, the law was so controversial among some people that it could have eventually increased this rejection rate. We will consider the consequences of both scenarios.

This information transmission protocol defines a Markov process. The state variable is the relative size of each type of dyad. Because of the continuous-time Markov process, the probability of a two-state change is zero during a small interval of time t and t + dt. This means, in particular, that both members of a dyad cannot change their status at the same time.

2.3 Flows of Dyads between States

Entry and exit flows of dyads from each state between t and t + dt are given by Equations (A1)–(A3) in Appendix A. Figure 1 illustrates the different types of dyads we encounter in our model as well as the flows between the dyads.

Figure 1.

Flows across Dyads.

Open in new tabDownload slide

A non-integrated individual being in a dyad d_UUt can only be influenced to change her mind about the social norm through random contacts with integrated individuals she might meet or hear. Time spent with weak ties equals ω, the probability of being influenced by a weak tie equals p and the number of integrated individuals equals i_t. Therefore, outflows from the d_UUt equal the number of d_UU-dyads in which one of the two members has accepted the social norm through her weak ties, $pωit$⁠.

Non-integrated individuals who belong to d_IUt dyads, and hence have a dyad partner who accepts the social norm, have a probability $(α+pωit)$ of accepting the social norm. They are more likely to be influenced by the social norm than individuals from a d_UUt dyad, since they interact with both strong ties and weak ties accepting the social norm.

2.4 Steady-State Population Composition

As shown in Appendix A, a steady-state dyad equilibrium flow is a quintuple $(i*,u*,dUU*,dIU*,dII*)$ such that $dUUt.=dIUt.=dIIt.=0, 12=dII*+dIU*+dUU*,i*=2dII*+dIU*$ and $u*=1−i*$ are satisfied.

Following the same procedure as in Appendix A1 of Giulietti et al. (2018), we can establish the equivalent proposition as their Proposition 1:

1. There always exists a steady-state equilibrium $N$ where no one accepts the social norm and only d_UU-dyads exist. In such equilibrium, $dII*=dIU*=i*=0, dUU*=1/2$ and $u*=1$⁠.

2. If

   $δ<ωp+ωp(4α+ωp)2,$

   (3)

   there exists an interior steady-state equilibrium $I$⁠, where $0<i*<1$ is defined by
   $i*=4δα+α2+2αωp+ω2p2−2δ−α+ωp2ωp.$

   (4)

If the rate at which individuals reject the social norm δ is sufficiently low (condition (3)), then an interior equilibrium exists. Otherwise, individuals reject the social norm with a very high probability and mostly become non-integrated, so that the steady-state equilibrium $N$ prevails. In the interior steady-state equilibrium $I$⁠, we can determine the share of people $i*$ that accepts the social norm, as well as the dyad composition (see Appendix A).

As in Proposition 2 of Giulietti et al. (2018) we find that:

1. If ω = 0 (i.e., no weak ties), then Condition (3) does not hold and we will be in $N$ with $i*=iω=0*=0$⁠.

2. If α = 0 (i.e., no strong ties), then Condition (3) holds and we will be in $I$⁠. From Equation (4), we find $i*=iα=0*=1−δωp$⁠.

3. If we compare the share of integrated people in each case, we find that $iω=0*<iα=0*<i*$⁠, where $i*$ denotes the share of integrated people when both types of interactions are allowed, that is, $ω≠0$ and $α≠0$⁠.¹³

Moreover, from Equation (4), we can also draw the following conclusions:

• The share of integrated people increases with time spent with weak ties $∂i*∂ω>0$¹⁴;

• The share of integrated people decreases with the exogenous rate at which integrated people reject the social norm $∂i*∂δ<0$⁠.¹⁵

In this framework, non-integrated individuals can only hear about the social norm through their ties. Then, social ties always influence positively the number of integrated individuals. If i_t > u_t, the relative influence of weak ties is increased compared with the situation where u_t > i_t. In the absence of weak ties, non-integrated individuals with non-integrated partners never have the opportunity to hear about the social norm. Non-integrated individuals with integrated partners will always accept the social norm since information comes from their partner. However, given the constant rejection rate δ, the economy with no weak ties eventually converges to an equilibrium $N$ with $iω=0*=0$⁠. In contrast, in the presence of weak ties, when time spent with strong ties equals zero, all non-integrated individuals, whatever the status of their partner, have the opportunity to hear about the social norm through weak ties. The larger the share of integrated people, the larger the relative influence of weak ties. The economy converges to an interior equilibrium $I$ where $iα=0*=1−δωp$⁠. Weak ties then play a relatively more important role in the acceptance of the social norm. This is consistent with findings in Calvo-Armengol et al. (2007) and Zenou (2015).

2.5 SSM Approval

In our specific case, SSM approval is assumed to modify the rate δ at which people reject the social norm, that is, gay rights. We now seek to analyze how the impact of this δ-change on individuals’ opinions is altered by the nature of ties when we are in the interior equilibrium $I$⁠.

The positive cross derivative implies that following a decrease in the rejection rate of gay rights, the increase in the share of integrated people will be smaller the larger the time spent with strong ties. Similarly, following an increase in the rejection rate δ, the decrease in the share of integrated people will be smaller the larger the time spent with strong ties. In this case, strong ties promote more integration.

The sign of $∂i*2∂δ∂ω$ is in principle undetermined. Following an increase in the rejection rate of gay rights, the decrease in the share of integrated people should be smaller the larger the time spent with weak ties, since these ties promote acceptance. However, if δ is very high, the number of integrated people will be so small that the positive effect induced by weak ties on the number of integrated people will be dominated by the negative impact of the rejection rate. The sign of the cross derivative $∂i*2∂δ∂ω$ is then undetermined.¹⁶

2.6 The Local Social Norm

Until now, we have interpreted the adoption of SSM law as a legalization of the dominant social norm at the country level, that is, at the aggregate level i_t > u_t. There are though some communities (or social networks) within countries having approved SSM where the dominant opinion is unfavorable to gay rights. For these communities, the local social norm is opposed to the aggregate social norm. In these communities, the share of non-integrated people overcomes the share of integrated people, u_jt > i_jt where j stands for community. Following the adoption of SSM law, did weak and strong ties shape differently the opinion about gay rights within these communities?

Given that the static comparative results presented in the previous section do not depend on whether $ut≶it$⁠, the answer would be negative. Weak and strong ties do not shape in a different way the opinion of individuals when u_t > i_t. However, the impact of SSM law on the rejection rate may differ. In communities where the majority of people is favorable to gay rights, the adoption of SSM laws simply makes legal the already dominant social norm and therefore it should contribute to reduce the rejection rate δ or, at least, it should not increase it. In contrast, in communities where the majority of people is unfavorable to gay rights, the direct impact of SSM laws may rather be negative and favor an increase in the rejection rate δ. Positions in these communities may become radicalized (at least in the short run) following the approval of SSM.

If within the communities unfavorable to gay rights the δ is so high that Condition (3) does not hold, we will converge to the steady-state equilibrium $N$ where only non-integrated people exist in the community. If the rejection rate satisfies Condition (3), the community will converge to an interior steady-state equilibrium $I$⁠, where the share of integrated people is given by Equation (4). In this equilibrium, comparative static results from the previous section still hold: $∂i*∂α>0$ and $∂i*∂ω>0$⁠. Non-integrated individuals can only change their mind if they are influenced by weak or strong ties. However, the influence of weak ties is reduced with respect to strong ties since now the share of integrated people is smaller (i.e., u_t > i_t) and non-integrated individuals continue to accept the norm with probability p when information arrives through weak ties. In contrast, non-integrated individuals always accept the norm whenever the information arrives from the dyad-partner (strong tie).

Within these communities, the aggregate social norm (conveyed by SSM law) clashes with the local social norm. SSM adoption may increase the rejection rate of the aggregate social norm, δ. Since results from the previous section do not depend on $it≷ut$⁠, we still have $∂i*2∂δ∂α>0$⁠. This implies that, following a rise in δ, the decrease in the share of integrated people will be smaller the longer the time spent with strong ties. In these communities, strong ties should then favor relatively more integration.

If SSM is still associated with a decreased rejection rate even within these communities, conclusions from the previous section will hold and strong ties should promote relatively less integration.

The way social ties shape individuals’ opinions is going to depend on whether the law induces a decrease or an increase in the rejection rate, and this will depend on whether the law is in line with the local norm. Our empirical analysis is operated in a backward manner since we deduce the impact of the law on the rejection rate depending on the influence of strong ties on personal attitudes toward LGBT rights.

2.7 From Theory to Econometrics

The sequential approval of SSM laws in European countries represented a progressive legalization of what was already becoming a dominant social norm: “gays and lesbians have the right to live their life as they want.” The model provides the following predictions concerning the impact of SSM approval and the role of weak/strong ties in shaping this impact.

At the aggregate level, if the legalization of the dominant social norm (favorable to gay rights) through the approval of SSM is associated with a decrease in the rejection rate δ, we should find that the increase in the number of integrated people should be smaller the longer the time individuals spend with strong ties, that is, $∂i*2∂δ∂α>0$⁠. We will refer to this first prediction of the model as Scenario G (general scenario).

The second prediction of the model focuses on the role of the dominant local social norm, which stands for the social norm shared by the majority of people belonging to a particular community within the society. When the aggregate and local social norms are similar, predictions from Scenario G are not modified. SSM approval stands for a legalization of the local social norm. As far as this decreases the rejection rate δ we should still find that, within the community, the increase in the number of integrated people is smaller the longer the time spent with strong ties. We refer to the situation where the aggregate and local social norms are similar as Scenario NH (non-homophobic scenario).

When the aggregate and local social norms differ, the approval of SSM clashes with the local norm. This may foster an increased rejection of the aggregate social norm leading to an increase in δ. In this case, the model predicts that, following the increase in δ, the number of integrated people should be larger among individuals spending more time with strong ties. The scenario where SSM approval clashes with the local social norm and induces an increased rejection rate is referred to as Scenario H (homophobic scenario).

The econometric section proposes to test the three predicted scenarios: (i) Scenario G applies to the whole sample of individuals; (ii) Scenario NH applies to any subsample including individuals whose dominant opinion is favorable to gay rights and (iii) Scenario H applies to any subsample including a majority of homophobic people.

3. Data

3.1 The Data

We use the ESS, an academically driven cross-national survey which has been conducted every 2 years since 2001 on more than 30 countries. In each participating country, face-to-face interviews are conducted with newly selected, cross-sectional representative samples of residents aged 15 years and above. The aim of the ESS is to chart stability and change in social structure, conditions, and attitudes in Europe and to interpret how Europe’s social, political, and moral fabric is changing. Individuals are asked about their attitudes, beliefs, and behavior patterns in various areas. In particular, they are asked about their opinion regarding the statement “Gay men and lesbians should be free to live their own life as they wish.” They report whether they “1. Strongly agree; 2. Agree; 3. Neither agree nor disagree; 4. Disagree; and 5. Strongly disagree” with this statement. We consider that people agreeing or strongly agreeing with this statement have gay-friendly opinions. This is our main outcome variable.

According to Acemoglu and Jackson (2017), the dominant social norm in a country is that which is accepted by the majority of the individuals. On the basis of this definition, we consider that in European countries where SSM laws have been adopted by a democratically elected government the dominant social norm is a positive recognition of gay rights.¹⁷ We collect information on SSM legalization dates from the International Lesbian, Gay, Bisexual, Trans, and Intersex Association. In every country, the year the law is passed is also the year it becomes effective. We restrict our analysis to SSM laws and do not extend our analysis to the recognition of unregistered cohabitation or to civil unions because for most countries such changes occurred before the beginning of our observation period.¹⁸

To test our theoretical predictions, we need to determine what is a strong tie. There is no unanimous definition of strong ties in the empirical literature. Granovetter (1973) measures the strength of the ties by the frequency at which individuals met. Lin et al. (1981) include relatives, friends, and neighbors in strong ties, as did Giulietti et al. (2018), who, however, excluded members of the household as strong ties so as to avoid endogeneity issues. In this paper, we consider that individuals being married or having an official partner have dominant strong ties. While being married or having an official partner may be an imperfect measure on the nature of ties, a marriage or partnership involves two individuals in a couple (or dyad), who have chosen to be together and have agreed to sign a social and economic contract which can be enforced by law. This makes the link stronger than other social contacts. Furthermore, in most occidental countries, around 50% of married couples do not divorce and stay together “for life.”

To test our Scenarios H and NH, we do not need to know the opinion of the partner (which is anyway not available in the data), as our theoretical predictions do not depend on the opinion of the strong tie, but we need to know the social norm that is dominant within the individual’s social network. To define this local social norm, we will adopt two strategies, one of which requires to gather data about homophobia around the world. To do so we use the “Gay Voyageur” (https://www.gayvoyageur.com), which provides advice to homosexual travelers on both gay-friendly and homophobic countries. According to this site, recommendations are based on culture, religion, and history of a country, city, or town. We also use the World Value Survey (waves 2010–2014),¹⁹ which asks individuals if they would not like to have as neighbor a homosexual person.

3.2 The Sample

We keep countries who participated in the ESS for at least seven out of the eight waves from 2002 to 2016 and that have not passed SSM laws by 2002.²⁰ We end up with the following 17 countries: Belgium, Switzerland, Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, Great Britain, Hungary, Ireland, Norway, Poland, Portugal, Sweden, and Slovenia. Nine of these countries legalized SSM between 2002 and 2016 (see Table A1 in Appendix B).

We use all the individual observations of these countries and apply design/post-stratification weights in combination with population size weights to our estimations.²¹  Table A2 in Appendix B describes the main variables used in the analysis for the whole sample and the immigrant sample.

3.3 Descriptive Analysis

Over the 2002–2016 period, the average opinion on gay rights improved overall (see Figure 2): the share of individuals agreeing or strongly agreeing with the statement “Gay men and lesbians should be free to live their own life as they wish” increased from 70% to 80%. This increase is more marked starting from 2006 and is mostly driven by an increase in the share of individuals strongly agreeing with the statement.

Figure 2.

“Gay Men and Lesbians Should Be Free to Live Their Own Life as They Wish”: Average Opinion over Time. Sample: All countries together. Source: ESS (2002–2016).

Open in new tabDownload slide

The situation is far from being homogeneous across countries and it strongly differs with respect to the legal framework (see Figure 3). If the share of people agreeing with the statement that gay men and lesbians should be free to live their own life as they wish is fairly the same in countries with and without SSM law, the share of the population that strongly agrees on this statement is much larger in countries that passed SSM laws than in those which did not, with 40% versus 23%. Additionally, the share of individuals with no opinion or (strongly) disagreeing with the statement is also greater in countries where SSM is illegal than in countries where it is legal. The gap on the average opinion about gay rights between countries with legalized SSM and the rest of the countries has remained fairly constant around 10–15 ppt since 2002, as shown by the panel (b) in Figure 3.

Figure 3.

“Gay Men and Lesbians Should Be Free to Live Their Own Life as They Wish”: Average Opinion by Legal Status of SSM and by Year. Note: Gay-rights supporters are individuals who agree or strongly agree with the statement: “Gay men and lesbians should be free to live their own life as they wish.” Source: ESS (2002–2016).

Open in new tabDownload slide

Legal frameworks concerning SSM and population opinions about gay rights are then strongly correlated. The causality between laws and average opinions remains undetermined though. This positive correlation between SSM laws and gay-friendly opinions may reflect a positive impact of SSM approval on gay-friendly opinions, but it may also be due to the fact that countries adopt SSM laws when a majority of the population is in favor (or is expected to be soon in favor) of such laws. Both explanations seem to be supported by Figure 4, which shows, for each country that passed SSM laws between 2002 and 2016, how the share of gay-friendly opinions evolved before (dashed line) and after (solid line) the adoption of the law. The approval of SSM was preceded in all countries, except for France, by a rising path in the share of people agreeing or strongly agreeing that gay men and lesbians should be free to live their own life as they wish. After legalization this positive trend in gay-friendly opinions has remained and even accelerated in almost all countries. These pre-existing changes in opinions would suggest that the approval of SSM was the legalization of the dominant social norm, meaning that law approval stands for a good indicator of the dominant social norm at the aggregate level. As for the post-law changes in opinions, it would suggest that SSM approval promoted a reduction in the rejection probability of gay rights, δ, at the country level. The empirical analysis developed in the rest of the paper aims to control for these pre-existing changes in opinions and neutralize the reverse causality problem to determine the extent to which SSM approval has a causal impact on opinions and contributes to explaining the positive post-law trends documented here.

Figure 4.

“Gay Men and Lesbians Should Be Free to Live Their Own Life as They Wish”: Average Opinion by Country and over Time. Notes: Gay-rights supporters are individuals who agree or strongly agree with the statement: “Gay men and lesbians should be free to live their own life as they wish.” For each country, the dashed (solid) line indicates the share of supporters before (after) the legalization of gay marriage in the country. Source: ESS (2002–2016).

Open in new tabDownload slide

One main goal of this paper is to test whether people are less likely to adopt the social norm made legal depending on the nature of social ties. To get a first descriptive picture of the possible relationship between law, opinion, and social ties, Figure 5 shows the evolution of the average gay-friendly opinion depending on the legal status of SSM in the country of residence and on the nature of individuals’ dominant ties (see Online Appendix²² for a decomposition by country). While there are strong differences in the share of gay-rights supporters depending on the legal status of SSM, the nature of dominant ties does not seem to fundamentally modify the observed trend in the share of gay-rights supporters. Any further interpretation of this figure is prevented by the endogenous nature of gay-friendly opinions to the type of ties. Individuals with a partner may indeed have specific attributes that make them also be more or less in favor of gay rights. The econometric analysis will deal with these issues.

Figure 5.

“Gay Men and Lesbians Should Be Free to Live Their Own Life as They Wish”: Average Opinion of Europeans by Legal Status of SSM and Nature of Dominant Social Ties. Notes: Gay-rights supporters are individuals who agree or strongly agree with the statement: “Gay men and lesbians should be free to live their own life as they wish.” Source: ESS (2002–2016).

Open in new tabDownload slide

4. Econometric Analysis

4.1 A Dynamic Difference-in-Difference Model

We exploit cross-country differences in the timing of adopting SSM and implement a difference-in-difference strategy on a pseudo-panel created from the repeated cross-sectional individual ESS data. The diff-in-diff estimator compares changes in the share of gay-rights supporters between cohorts which are subject to the policy at different moments of time and those which are not subject to it yet or that are not at all subject to it.

We adopt a cell approach to introduce a temporal dimension—as we can follow cells over time—and consider fixed effects models as with genuine panel data.²³ Our cells are defined by country, year, and individual demographics, including age (15–25, 26–35, 36–45, 46–55, and more than 55) and gender.²⁴ Considering age and gender as the demographic characteristics driving social interactions seems consistent with the empirical evidence and the existing literature. As revealed by Panel A in Figure A1 in Appendix C, whatever the age range we consider, around 70% of the population in the age range declares a “Strong sense of belonging to age group.” Moreover, the ESS (wave 2008) also asks the individual if she has friends (outside the family) younger than 30 years old and older than 70 years old. Panel B of Figure A1 in Appendix C considers the answers to these questions by age range. We find that among people between 15 and 25 years old, 80% declare having at least two friends younger than 30. This percentage is above 60% when considering the age range 26–35, and it falls to 37%, 30%, and 21% when considering the age ranges 36–45, 46–55, and more than 55, respectively. When individuals are asked about their friends above 70 years old, around 5% of people in the age ranges 15–25 and 26–35 declares having at least two friends. This percentage equals 13% and 18% for the age ranges 36–45 and 46–55, respectively, and it raises above 40% for people above 55 years old. Friendship seems then strongly driven by age. It is also driven by gender. As remarked by the psychological literature (see Kovacs et al. [1996] or Rose [2002]), friends in childhood and in youth are typically same-sex peers and consistent gender differences are present in these relationships. During childhood the two sexes engage in fairly different kinds of activities and games (see Huston [1985] or Maccoby [1990]). As children move into adolescence and adulthood, despite the powerful attraction to members of the opposite sex, young people continue to spend a good proportion of their social time with same-sex partners, which contributes to the development of sex-typed peer relationship processes that influence the emotional and behavioral development of girls/women and boys/men (see Rose and Rudolph [2006] or Giulietti et al. [2020]). In adulthood, there is extensive gender segregation in workplaces and in leisure time even after marriage (see Maccoby 1990).

In addition to age and gender, we add a geographic dimension to define our cells. We work at the country level which, in the case of the ESS, also captures the fact that people that interact together speak the same language.²⁵ The construction criteria for our cells are left constant for every year of the considered period. Most cells are composed of more than 100 individuals.²⁶

Our setup departs from the standard two periods—two groups diff-in-diff model, as we consider multiple time-periods and multiple groups that are treated at different dates. Moreover, we need to consider time-varying effects, as the spread of a new norm may take time. Such a diff-in-diff framework with multiple treatment times and dynamic responses has recently been a source of controversy in the literature. Wolfers (2006) pointed out that when treatment effects vary over time, a major difficulty in diff-in-diff analyses involves separating out preexisting trends from the dynamic effects of the policy shock. A single-coefficient diff-in-diff estimator reflects the difference between the actual path of the outcome variable and a systematically biased estimate of its counterfactual.²⁷ To remedy this identification issue, Wolfers (2006) and Stevenson and Wolfers (2006) propose a specification that explicitly models the dynamic response function, thanks to the inclusion of dummy variables for the different couples of years following the adoption of the law (1–2, 3–4, 5–6, and so on) instead of a single before/after dummy. These dummy variables identify the entire response function allowing the estimated country-specific time trends to identify pre-existing trends. The inclusion in our sample of countries that have not adopted SSM laws over the period also helps us to identify trends and separate the effects of passing on absolute time and relative time.

Identification of causal effects in a difference-in-difference model relies on the common trend assumption. In our context, this means that one needs to be concerned with the timing of the reform. If more progressive countries or countries that experience (or expect) a faster improvement in gay-friendly opinions pass SSM laws first, estimates would be upward biased. Time-invariant country characteristics are captured by country-fixed effects. To account for possible country time-variant differences (and therefore country-specific pre-trends), we add country-specific trends. We allow for linear and non-linear trends. Identification then comes from the discontinuity surrounding the adoption of SSM. Controls for country-specific (pre-)trends prevent estimated coefficients to reflect the action of other possible changes, such as changes in religious or political thinking or other LGBT specific laws like anti-discrimination, registered partnership, and parent rights legislations, that were adopted at different times in different countries,²⁸ that are potentially correlated to the adoption of SSM and that may affect the evolution of opinions about gay rights independent to any SSM laws—see Autor et al. (2007b),²⁹  Wolfers (2006), or Bailey (2006).³⁰

The diff-in-diff strategy also relies on the no-anticipation assumption, according to which individuals should not change their opinions due to the expectation of upcoming law adoption. Obviously, SSM law adoptions are preceded by debates that may influence opinions about gay rights before the passing of the law. Our assumption here is that such pre-existing effects may occur in all countries and are captured by our country-fixed effects and country-specific trends.

A last concern comes from structural socio-demographic changes. The inclusion of a cohort-specific trend controls for (pre-)trends at the cohort level which are orthogonal to SSM approval. Still, in diff-in-diff frameworks applied on repeated cross-sections as the ESS, the compositions of treated and control groups need to be constant over time and must not be influenced by treatment. Otherwise estimates would capture changes in gay-friendly opinions due to structural changes (as aging) and to possible increasing growth of socio-demographic groups that would have more or less gay-friendly opinions. To control for socio-demographic changes over time, and remove such confounding composition effects, we assign to each age–gender cell in a given year and a given country the average yearly weight of the cell in the corresponding country over the whole observed period (as in Autor et al. 2007b). Put differently, we weight observations so as to keep the socio-demographic structure of the sample constant.

4.2 Estimating the Role of Social Ties on the Effects of SSM Laws

To test our theoretical predictions, we need to augment Equation (6) to allow for heterogeneous effects with respect to the nature of ties. Our main challenge here is to account for the endogenous nature of ties. If SSM laws affect the structure of the population in terms of ties, for instance if people’s opinion about gay rights influences the probability to have strong ties, then this would threaten the identification of causal effects.³¹ To remedy this problem we first consider all types of partnerships to define strong ties and do not restrict in particular to married couples. Moreover, results in Table A6 in Appendix D reveal no significant relationship between the approval of SSM and the share of people living in partnership, suggesting that our measure of strong ties (including all partnerships) is not endogenous to SSM approval.

Second, and most importantly, we compute the composition of cells in terms of the nature of ties (i.e., the share of individuals with dominant strong ties and without dominant strong ties) in 2002, when SSM was illegal in all European countries (other than The Netherlands, which is not included in our sample). Therefore, people’s opinion on the topic was very unlikely to influence their social tie formation (by that time SSM was not a controversial issue). For the whole period of analysis, we leave constant the tie composition at its 2002 level. Such a strategy also protects us from a contamination bias: we expect that SSM adoption in a country may help to change mentalities within the population beyond the country’s borders. With the sequential adoption of SSM in most European countries, the institutional context is likely to further shape individual opinions, and then the composition of social ties in terms of strong and weak ties is more likely to be endogenous. Such a contamination between countries can neither be excluded nor tested and may lead to (unsigned) biases in the estimation of the average effect of the law on opinion if it is related to the composition of social ties (see Equation (6)).³² By leaving constant and equal to their 2002 levels the internal composition of cells in terms of strong/weak social ties, we control for this bias.

4.3 Estimating the Role of Local Social Norms on the Effects of SSM Laws

According to the model’s predictions, the law has heterogeneous effects depending on the local norm prevailing among social ties of a given community. The law can actually decrease or increase the rejection rate δ of the aggregate social norm depending on whether the law is in line with the local social norm or not.³⁴ Depending on the impact on the rejection rate, the relative role of strong ties in shaping individuals’ opinions will vary. If the SSM law leads to a reduction in the rejection rate δ of the aggregate social norm, we expect individuals (i.e., cells in our pseudo-panel approach) with dominant strong ties to experience a smaller progression in their opinion about gay rights than the other observations. In contrast, if the SSM law fosters an increase in the rejection rate of the aggregate social norm, we expect individuals (i.e., cells in our pseudo-panel approach) with dominant strong ties to experience a more positive progression in their opinion about gay rights with respect to other individuals.

How can we measure the local norm? A first strategy relies on the dominant opinion about gay rights in the cell, as computed in 2002. We consider that the local social norm is non-homophobic (i.e., in line with SSM laws) when more than 60% of individuals in the cell declare gay-friendly opinions in 2002. We chose the threshold of 60% so that we are not at the margin. The categorization of cells according to their gay-friendly nature or not is computed for 2002 and left unchanged for the whole period so as to control for the potentially endogenous progression in the share of favorable opinions about gay rights. We consider that in cells that are classified as gay-rights supporters according to this criterion, the average dominant opinion among weak ties will be favorable to gay rights. The estimation of Equation (7) on the subset of gay-rights supporters cells provides a test for Scenario NH. In spite of setting the threshold of gay-rights supporters to 60% to classify a cell as gay-friendly, there are not enough cells classified as non-gay-rights supporters to test for Scenario H.

The second strategy focuses on immigrants and distinguishes those coming from homophobic countries from those coming from non-homophobic countries. For the former the local social norm is considered to be homophobia, while for the latter it is non-homophobia. This strategy allows us to test the theoretical predictions corresponding to both Scenarios NH and H. It stems from the literature that network effects are of major importance on immigrants’ choices and behaviors (see, e.g., Waldinger 1996; Munshi 2003; McKenzie and Rapoport 2010; Patel and Vella 2013). Immigrants tend to interact more frequently with people coming from the same country of origin. Immigrants in a given host country are likely to be part of a social network or community made up by people coming from the same country of origin. Therefore, immigrants coming from a country classified as homophobic (i.e., the dominant social norm in the country of origin is homophobia) are likely to have social ties who are mostly homophobic. This guess is confirmed by Figure A2 in Appendix C. Among immigrants coming from non-homophobic countries, more than 75% (on average, over the whole period) agree that gay men and lesbians should be free to live their own life as they wish. This share does not exceed 52% among immigrants coming from homophobic countries. From a dynamic perspective (panel (a) of Figure A2), opinions of immigrants coming from the two categories of countries have followed a similar path, implying that the difference of 20 percentage points in opinions has remained fairly stable over time. We only observe a slight decrease in this difference in 2016.³⁵

To ensure a sufficient number of observations, cells are now redefined for immigrants by host country, type of country of birth (i.e., homophobic versus non-homophobic), and year. Using this strategy requires dealing with the endogeneity of location choices with respect to opinions about gay rights. In particular, immigrants may migrate to countries that are more or less favorable to gay rights depending on their own opinion about homosexuality. To control for endogenous location choices, we rely on the past settlement of immigrants across European countries. We specifically use data from Docquier et al. (2007) on the stock of immigrants by European country in 1990.³⁶ We compute the share of immigrants coming from $r=(homophobic, non-homophobic)$ countries in the host country c in 1990.³⁷ For each country of birth we define an indicator variable Homophobic, which equals 1 if the corresponding country is classified as homophobic by the website “Gay Voyageur” or if more than 60% of individuals in the country declare in the World Value Survey (2010–2014) that they would not like to have a homosexual person as a neighbor.³⁸ The combination of the information provided by both data sources ensures to have more reliable classification of homophobic countries. We find that around 39% of the 192 origin countries included in our sample for foreign born people are classified as homophobic (see Table A2 in Appendix B).

In period t, the estimated number of immigrants in host country c from r countries, $Immigrantscrt^$⁠, is obtained by multiplying the 1990 share of immigrants coming from r countries that were living in host country c by the total number of immigrants in our sample coming from r countries in period t. This sharing rule ensures exogeneity of the distribution of immigrants. The implicit assumption is that the 1990 sharing rule of immigrants across host countries is exogenous to any current shock, such as the approval of SSM.

To control for population composition changes, we consider a constant weight for each cell defined by host country and country of origin (homophobic/non-homophobic) over the whole period.³⁹ These constant weights result from multiplying the share of immigrants from r countries in host country c in 1990 $γcr,1990$ times the average stock of immigrants from r countries during the whole considered sample period. As previously, 2002 tie composition is held constant for all subsequent years in our sample.

To test Scenarios NH and H from the theoretical model, we re-estimate Equation (7) for people coming from, respectively, non-homophobic and homophobic countries. Again, we include year- and country-fixed effects as well as a country-specific trend. $ψk0$⁠, ψ₂, and $ψk1$ keep the same interpretation as in Equation (7).

5. Results

The sequential approval of SSM by European countries over the period 2002–2016 is considered as a legalization of the dominant favorable opinion about gay rights, that is, the aggregate social norm. The approval of SSM aimed at decreasing the rejection rate of gay rights (δ) and thus increasing the proportion of people accepting this social norm. With such a decrease in the rejection rate, our theoretical model predicts that people are less likely to accept the social norm when they have dominant strong ties. As a consequence, the adoption of SSM law would induce a larger increase in the share of gay rights supporters among people without dominant strong ties than among people with dominant strong ties. In contrast, if the law induces an increase in the rejection rate (which may happen in communities where the local social norm is not aligned with the aggregate social norm), our theoretical model predicts that people with dominant strong ties are relatively more likely to accept the aggregate social norm. Before testing these heterogeneous effects of SSM approval, we first assess the average effect it has had on people’s opinions about gay rights.

5.1 The Overall Impact of SSM Approval

Parameter estimates from Equation (6) are reported in Table 1. Column (1) reports results from a specification not including fixed effects. Column (2) adds year-, country-, and cohort-fixed effects and the last columns add linear (column (3)) and non-linear (column (4)) country- and cohort-specific trends. Our preferred specification combines yearly shocks with country- and cohort-specific time trends and their squares. This allows to control for the pre-existing trends at the cell level which are orthogonal to SSM approval. In all specifications, observations are weighted using counterfactual weights to control for any population composition changes over the observation period.

In absence of any control, SSM approval is associated with more positive opinions about gay rights (see column (1)). Once year-, cohort-, and country-fixed effects are controlled for, all coefficients of the dynamic response function to the law become non-significant (see column (2)). This confirms the importance of controlling for time-invariant country differences when evaluating the effect of policies on opinions, as highlighted among others by Aksoy et al. (2020) in the context of same-sex relationship recognition policies and attitudes toward sexual minorities. Adding country-specific trends does not change the significance of coefficients (see columns (3) and (4)). These results imply that the observed dynamic response to SSM approval in opinions about gay rights is mostly explained by systematic differences across countries and across cohorts, as well as aggregate yearly shocks. So, on average, there is no impact of SSM approval on opinions about gay rights.⁴⁰

Our results slightly contrast but are not inconsistent with those of Aksoy et al. (2020), who find a significant but rather limited effect of same-sex relationship recognition policies (including SSM) on attitudes toward homosexuality.⁴¹ Our results are in line with those of Bishin et al. (2016) and Tankard and Paluck (2017) who, in response to US Supreme Court rulings on SSM, find no opinion backlash against gay men and lesbians and no significant change in personal attitudes, respectively. Our findings further suggest that the impact of SSM approval on the rejection rate has not been sufficiently large to alter significantly average opinions about gay rights and their dynamics over time. Despite this zero average impact of SSM approval, we could expect different changes in opinions depending on individuals’ social ties and the local social norm of the community to which they belong. We investigate such heterogeneous effects in the next section.

5.2 The Role of Social Ties

In Table 2, we test whether the dynamic response to SSM approval differs between individuals having dominant strong ties and those where strong ties are not dominant. Results are reported according to the same four specifications as in Table 1 and observations are weighted using counterfactual weights. Coefficient estimates reveal that the dynamic response to SSM approval is indeed heterogeneous with respect to the nature of the dominant ties in cells. For both types of cells, including country-, cohort-, and year-fixed effects tends to reduce the influence of SSM approval, but coefficients remain significant. Estimation results are fairly stable across specifications (2)–(4).⁴²

The coefficient on “Strong ties” indicates that the average opinion about gay rights is not significantly different between cells with dominant strong ties and the other cells. However, there are significant differences between these two types of cells in the dynamic response to SSM approval. In cells where strong ties are not dominant, the approval of SSM results in a significant and large increase in the share of gay-rights supporters within the first 3 years following adoption (“First 0–1 years” or “2–3 Years”). More precisely, we find a direct increase by more than 0.09 pp in the share of individuals with favorable opinions about gay rights during the year of or the year after the approval of SSM. Following this first spike, the impact of SSM approval equals 0.07 pp 2–3 years after.⁴³ Dynamic response effects are significantly lower for cells with dominant strong ties from the year of approval to more than 6 years after the approval. The difference between the two groups remains fairly stable over time.

As shown in Figure 6, the lower response identified among cells with dominant strong ties is such that SSM approval actually has had no influence on the average opinion about gay rights in these cells. Consistent with theoretical predictions, these results suggest that SSM approval has reduced the overall rejection rate of gay rights in the population (δ). This reduction is indeed a prerequisite to observe a smaller increase in the share of gay rights supporters (i.e., integrated individuals) among individuals with strong ties. The effect is large enough to influence the opinion of individuals in cells where strong ties are not dominant, but too small to significantly influence cells with dominant strong ties.

Figure 6.

The Influence of SSM Adoption on the Opinion about Gay Rights in Case of Dominant Strong Ties. Notes: The figure plots linear combinations of coefficients from Table 2, column (4)— specification with country- and cohort-specific non-linear trends. Each point is the linear combination of the coefficient associated with the dynamic treatment indicator and the coefficient associated with the corresponding interaction term with strong ties.

Open in new tabDownload slide

This result is though modified when we ignore potential endogeneity issues of social ties and allow the tie composition of cells to change from one year to another. As revealed by Table A8 in Appendix E, in this case the approval of SSM laws displays no significant effect on the opinion of individuals without strong ties and a negative effect on the opinion of individuals with strong ties, underlying the importance of controlling for the endogenous nature of ties.⁴⁴

As previously mentioned, evidence suggests that the way in which women/girls interact with their female peers is very different from the way men/boys interact with their male peers (see Rose [2002], Rose and Rudolph [2006], or Giulietti et al. [2020]). In particular, unlike men, women tend to discuss extensively and revisit problems with their female peers (i.e., co-rumination). These differences in social interactions may drive the way weak and strong ties influence the opinions of women and men about gay rights. Estimates reported in Table 2 could therefore conceal specific dynamics among women and men. Table A10 in Appendix F displays estimation results distinguishing between men and women. Results remain consistent with dynamic response effects being systematically lower for cells with dominant strong ties. Note though that women with dominant strong ties have on average a more positive opinion about gay rights than women without dominant strong ties.

5.3 The Law and the Local Social Norm

To test Scenarios NH and H, we need to control for the local social norm which stands for a proxy of the opinion of weak ties. We do not have this information in our data, but we propose two strategies to approximate it. First, we focus on cells with a majority of gay-rights supporters. In these cells, the dominant local social norm is in line with the aggregate social norm at the time of SSM approval. This allows us to test Scenario NH.⁴⁵ Second, we focus on immigrants, and distinguish those coming from homophobic countries (Scenario H) and those coming from non-homophobic countries (Scenario NH). We assume that the dominant local social norm of the latter is in line with the aggregate social norm in the country at the time of SSM approval, while we assume the opposite for immigrants coming from homophobic countries.

For the first strategy, we replicate the previous analysis on the subsample of gay-rights supporters. Results are presented in Table 3. In line with results from Table 1, columns (1)–(4) in Table 3 reveal that the average impact of SSM approval on the opinion about gay rights is zero among gay-rights supporters. Columns (5)–(8) introduce social ties into the analysis. The coefficient on “Strong ties” reveals that the average opinion about gay rights does not significantly differ between cells with dominant strong ties and the other cells when considering the population of gay-rights supporters. When controlling for country- and cohort-specific linear trends (see column (7)), we find that SSM approval has a positive and significant impact on the opinion about gay rights 0–1 years and 4–5 years after approval among individuals without dominant strong ties. This positive impact is though absorbed by country- and cohort-specific non-linear trends in column (8). Independent of whether these non-linear trends are controlled for or not, we find that the dynamic response for individuals with dominant strong ties is significantly smaller than that of individuals without dominant strong ties 0–1 years, 4–5 years, and more than 6 years after SSM approval.^46,47 As for the whole population, these results suggest that SSM approval has reduced the rejection rate of gay rights among cells with a majority of gay-rights supporters. This reduction is indeed a prerequisite to observe a smaller increase in the share of gay-rights supporters among people with dominant strong ties. This is consistent with predictions from Scenario NH.

For the second strategy, our analysis is based on cells defined at the host country-homophobic origin level—169 cells. Before analyzing the role of social ties, we first propose to test whether the local social norm of immigrants coming from homophobic countries clashes with the aggregate social norm. Table 4 provides coefficient estimates obtained from estimating a rephrased version of Equation (7) where $ψk0$ captures the dynamic response to the approval of SSM on the opinion about gay rights among immigrants coming from non-homophobic countries, ψ₂ captures the average response effect of immigrants coming from homophobic countries (compared with immigrants coming from non-homophobic countries), and $ψk1$ captures the additional dynamic response to SSM adoption for immigrants coming from homophobic countries instead of non-homophobic countries.⁴⁸

Results in Table 4 reveal that immigrants coming from homophobic countries have on average a less favorable opinion about gay rights (see coefficient on “Homophobic”). This finding confirms that the dominant local social norm is less gay-friendly within communities composed of immigrants coming from homophobic countries. The approval of SSM has had no significant effect on the opinion of immigrants coming from non-homophobic countries. In contrast, the opinion of immigrants coming from homophobic countries is more negatively affected immediately after the adoption of the law.⁴⁹ According to the theoretical framework, this result suggests that SSM has increased the rejection rate of gay rights among people from these communities.

We then investigate the role of social ties in the subpopulation of immigrants, considering separately those coming from homophobic countries (Scenario H) and those coming from non-homophobic countries (Scenario NH). We provide estimates from Equation (7) in Figure 7 which represents coefficient estimates from Table A13 in Appendix G. In this table, columns (1)–(3) consider the whole population of immigrants, without distinction with respect to the homophobic nature or not of the country of origin. Columns (4)–(6) allow to test Scenario H thanks to a focus on immigrants from homophobic countries (74 cells). Estimated coefficients are represented on the left-hand side panel of Figure 7. In contrast, columns (7)–(9) focus on immigrants from non-homophobic countries (95 cells) to test Scenario NH, which is represented on the right-hand side panel of the figure. Cells are weighted so as to control for demographic changes.⁵⁰ Among immigrants coming from homophobic countries (left-hand side panel of Figure 7), the local social norm clashes with the aggregate social norm at the time of SSM approval. The role of weak ties in disseminating the aggregate social norm is then strongly reduced and strong ties play a more important role therein. Among these immigrants, we find that those with dominant strong ties have reacted more positively to SSM approval than people without strong ties. For the latter, SSM approval has an immediate negative impact but no lasting effects on gay-friendly opinions. Consistent with our theoretical predictions, these results suggest that SSM approval has induced an increase in the rejection rate of gay rights (δ) among these immigrants coming from homophobic countries. This increase is a prerequisite to observe such a large increase in the share of gay-rights supporters among people with dominant strong ties.

Figure 7.

The Influence of SSM Approval on the Opinion about Gay Rights Over the Immigrant Population Depending on Their Homophobic Origin. Notes: The figure plots linear combinations of coefficients from Table A13, Column (3)—specification with country- and cohort-specific linear trends.

Open in new tabDownload slide

The situation is reversed for immigrants coming from non-homophobic countries, among which the local social norm is likely to be in line with the aggregate social norm at the time of SSM approval. The right-hand side panel of Figure 7 indicates that SSM approval has induced on all subsequent years an increase in the share of gay-rights supporters among immigrants without dominant strong ties. Immigrants with dominant strong ties are on average less likely to be gay-rights supporters (see coefficient on “Strong ties”). According to our theoretical model, such a lower increase in gay-rights supporters among people with dominant strong ties is possible because SSM approval has induced a decrease in the rejection rate of gay rights (δ) in this subpopulation of immigrants coming from non-homophobic countries.

These last results confirm the importance of the local social norm with respect to the impact of SSM approval on the rejection rate of gay rights within a community and the consecutive role that strong ties may play in the acceptance process of the aggregate social norm.⁵¹

5.4 Robustness Analysis

5.4.1 Placebo Tests

Our empirical strategy relies on a number of assumptions, so that identification of the effects we are interested in may be threatened by a number of factors. Do we actually capture the effect of SSM laws and social ties? Our main concern lies in the timing of the law and the credibility of the non-anticipation and parallel trends assumptions that are required in a difference-in-difference approach. As explained above, country-specific trends are included in the model to deal with an endogenous timing of the law, and to ensure that the estimated parameters of interest do not capture the effects of time-varying factors that may affect both the timing of the law and opinions toward homosexuality.

To test the credibility of this identifying assumption and test the relevance of the threat of pre-trends and concomitant progressive changes, we perform two placebo tests, whose results are given in Figure 8. We run both placebos with linear-specific trends and with non-linear-specific trends.

Figure 8.

Placebo Tests. Notes: (a): y=important traditions and customs, actual dates - linear trends; (b): y=gay rights, placebo dates - linear trends; (c): y=important traditions and customs, actual dates - non-linear trends; (d): y=gay rights, placebo dates - non-linear trends.

Open in new tabDownload slide

The first placebo test consists in estimating the effect of SSM laws on an outcome variable which is not (directly) related to the opinion about gay rights and should not be particularly affected by SSM laws, but should be affected by the wide spread of a more progressive way of thinking. If the year-fixed effects and trends failed to capture such a trend, we would get that SSM laws also affect such opinions. A natural candidate for this test is the statement that it is important to follow traditions and customs. Results of this test are reported in column (1) of Figure 8. They reveal that the approval of SSM laws did not have a significant dynamic response effect on the share of individuals who consider that traditions and customs are important. When controlling for a linear trend (first panel in column (1)) no significant impact arises whether we consider individuals with dominant strong ties or without dominant strong ties. When we allow for a non-linear trend (second panel in column (1)), there is only a significant reduction in the share of favorable opinions for individuals without dominant strong ties after 4–5 years and after 6 years.

The second placebo test consists in testing the significance of the effect of SSM laws assuming that laws were passed 2 years before they were actually passed. Given that we have a dynamic model, it is expected that no significant effect is shown within the first 2 years of the placebo years and that the dynamics of the effect mimics the actual one but with a 2-year lag. Results of this test are provided in column (2) of Figure 8. When introducing a linear trend we find that the positive and significant impact that we found in column (3) of Table 2 for individuals without dominant strong ties is delayed by 2 years, while for individuals with dominant strong ties the effect becomes significant only after 4–5 years. As soon as we consider non-linear trends (second and third panel in the column), no significant coefficient arises.

5.4.2 An Event-Study Design

To further assess the importance of the pre-trend issue, we additionally estimate an event-study type specification and include both treatment leads and treatment lags in the model. Recent literature discusses identification and specifications in dynamic event studies (see, e.g., Borusyak and Jaravel 2017; Freyaldenhoven et al. 2019; Schmidheiny and Siegloch 2020; Sun and Abraham 2020) and challenges the ability of tests for joint significance of lead coefficients to test for the presence of pre-trends (Roth 2019). We follow the standard and exclude the pre-treatment indicator just prior to approval and therefore standardize coefficient $ψt0−1$ to zero. Parameters $ψt0+k$⁠, with k now possibly taking negative values in case of leads, are the effect k time periods before or after the event compared with the level one period prior to the event.

Results graphed in Figure 9 correspond to estimation results reported in column (2) of Table A14 in Appendix H. As observed from this column, although not parallel, pre-trends are not different between people without and with dominant strong ties. This suggests that the corresponding effects on individual opinions may be interpreted in terms of causality. Overall, these event-study results confirm our previous findings. Individuals without dominant strong ties positively respond to SSM approval and do so significantly more than individuals with dominant strong ties.

Figure 9.

Event Study for SSM Approval. Notes: The figure plots linear combinations of coefficients from Table A14, Column (2)—specification with country- and cohort-specific linear trends.

Open in new tabDownload slide

Estimation results from columns (4) and (5) of Table A14 are, respectively, represented on the panels (a) and (b) of Figure 10.⁵² Among immigrants coming from homophobic countries (panel (a)), the share of gay-rights supporters is not significantly different after SSM approval compared with its level the year just before the law was passed. Moreover, no significant differences arise between the response of people with and without dominant strong ties in the 5 years following SSM approval. For the subsample of immigrants coming from non-homophobic countries (panel (b)), after the reform people with dominant strong ties are observed to be less gay-friendly, while the others happen to have on the contrary more gay-friendly opinions. These event-study results need to be taken with caution, and causal interpretation is prevented given that pre-trends are significant and positive (see Table A14, column (5)).

Figure 10.

Event Study for SSM Approval: Immigrant Population by Origin (Homophobic versus Non-homophobic). Notes: The figure plots linear combinations of coefficients from Table A14, Columns (4) and (5)—specification with country- and cohort-specific linear trends.

Open in new tabDownload slide

5.4.3 A Continuous Measure of Strong Ties

To simplify economic interpretations all along the analysis, we have considered that cells that had more than 60% of the individuals living in partnerships in 2002 were dominated by strong ties while the other cells were not dominated by strong ties. This threshold was chosen so as to ensure a sufficient number of cells in both categories. In order to test the robustness of our results we consider in this section a continuous measure of strong ties. More precisely, we compute for 2002 the share of individuals living in partnership in each of the cells. This measure is a continuous variable. Again, to account for the endogeneity of ties with respect to SSM, we leave the distribution of tie composition of cells invariant for the whole considered period 2002–2016 at its 2002 level.

Results reported in Table 5 are consistent with main findings from Tables 2 and 3 and Figure 7. At the aggregate level (columns (1) and (2)), there is a decrease in the number of integrated people as the share of strong ties increases, suggesting that SSM approval induced a decrease in the rejection rate δ.