The Social Stratification of Choice in the Transition to Adulthood

The occurrence and timing of major demographic decisions in the transition to adulthood is strongly stratiﬁed, with young adults with a high socio-economic status (SES) background usually experiencing many of these events later than young adults with a low SES background. To explain this social stratiﬁcation, we outline a theoretical framework in which social stratiﬁcation affects choice in the transition to adulthood through three, potentially reinforcing, pathways: stratiﬁed socialization, stratiﬁed agency, and stratiﬁed opportunity. We test our framework against longitudinal data from two waves of the Generations and Gender Surveys for Austria, Bulgaria, and France. We ﬁnd evidence for the importance of all three pathways. Furthermore, processes differ little by gender, age and country context.


Introduction
Young adulthood is 'demographically dense' (Rindfuss, 1991). The transition to adulthood, with a series of events that are concentrated in a relatively short age span, shapes life courses in a crucial way. In their classic paper, Modell, Furstenberg and Hershberg (1976) defined the transition to adulthood as a process marked by five events: leaving school, entering the workforce, leaving the parental home, marriage, and parenthood. The first two events are connected to educational and occupational attainment, and foster economic independence. Economic independence, in turn, is closely linked to major demographic markers of the transition into adulthood (Furstenberg et al., 2004;Spéder, Murinkó and Settersten, 2014), including residential moves as well as union and family formation.
The transition to adulthood is strongly socially stratified. Significant attention has been paid to stratification in the status attainment literature, which documented that children of parents with higher socio-economic status (SES) tend to achieve higher educational levels, i.e. leave school and start work later (and with better paid, more secure, and higher status jobs) than children from lower SES backgrounds (Settersten and Ray, 2010;Buchmann and Kriesi, 2011) Though fewer studies have examined how demographic outcomes are socially stratified, children from an advantaged family background have-on average-been found to leave home earlier, but enter a union, marriage and parenthood later than their peers from a more disadvantaged family background (Hogan and Astone, 1986;Rindfuss, 1991;Buchmann and Kriesi, 2011). This social stratification of demographic behaviour has been explained by differences in economic and cultural resources that shape children's opportunities during the transition to adulthood (Marini, 1985;Avery, Goldscheider and Speare, 1992;Heinz et al., 1998;Shanahan, 2000;Shanahan and Macmillan, 2008;Lareau, 2011;Hitlin and Johnson, 2015). However, the specific pathways by which this social stratification comes about have not received much attention.
Recent contributions to the general sociological debate on social structure and its impact on social behaviour could be useful to fill this gap (Adkins and Vaisey, 2009;Vaisey, 2009Vaisey, , 2010Hitlin and Johnson, 2015). Building on these perspectives, the present study focuses on how social stratification in demographic behaviour comes about by examining three pathways. First, in the socialization process, parents influence the values, attitudes, and intentions concerning demographic decisions that young adults develop. Children of high-and low-SES parents may develop different intentions about demographic events. The former may for instance plan to postpone family transitions that are costly and hard to reverse, in favour of individual autonomy and investments in human capital. Building on Barber (2000), we label this stratification of intentions stratified socialization. Second, parental SES (PSES) may influence young adults' ability to realize their behavioural intentions. High-SES parents may foster their children's life-course agency more efficiently (Elder, 1994;Elder, Johnson and Crosnoe, 2003;Hitlin and Elder, 2006;Macmillan, 2006;Hitlin, 2007;Hitlin and Johnson, 2015) by stimulating planful competence (Clausen, 1991) and selfefficacy (Bandura, 1977). As a result, young adults from advantaged and disadvantaged family backgrounds may differ in mental processes underlying human agency, leading to differences in the ability to realize behavioural intentions. Also, more advantaged families of origin may provide economic resources to realize costly transitions. We label this stratification of the intention-behaviour link as stratified agency. Third, social stratification not only operates via explicitly agentic processes of goal setting (stratified socialization) and goal realization (stratified agency), but also via structural processes that run on top of stated intentions. Structural factors often lead to an earlier transition to adulthood among young adults with a low-SES background than among young adults with a high-SES background (Marini, 1985;Furlong and Cartmel, 2007). We label this pathway stratified opportunity. Thus, our central research question is to what extent the stratification of demographic events in the transition to adulthood results from the stratified socialization, stratified agency, and stratified opportunity pathways, and whether these pathways are important for each demographic event.
Our study contributes to the literature mostly in three ways. First, we develop a theoretical framework that views the social stratification of demographic events in the transition to adulthood as occurring through three complementary and potentially reinforcing pathways. These pathways are rarely distinguished theoretically, and have never been distinguished empirically for demographic behaviour in young adulthood.
Second, the existing literature on the social stratification of demographic behaviour predominantly focuses on specific events, such as leaving the parental home or the transition to parenthood. Our study examines empirically four key demographic events, thereby providing a holistic perspective on the social stratification of the transition to adulthood. In addition, to acknowledge the increasing diversification of the union formation process, we explicitly distinguish between the formation of a first co-residential union (marriage or cohabitation) and first marriage, using novel data from two waves of the Generations and Gender Survey (GGS) (Vikat et al., 2007).
Third, we explore the comparative validity of our empirical results, by examining whether the pathways are the same for men and women, for those who make these demographic choices relatively early in young adulthood and those who postpone them to later ages, and for people in different societies. The importance of specific stratification channels could differ across societies, as institutional, economic and cultural factors influence both socialization processes and the opportunities and constraints that parents and children are facing (e.g. Régnier-Loilier and Vignoli, 2011;Kapitá ny and Spéder, 2013).

Theoretical Background and Hypotheses
The Social Gradient of Demographic Behaviour in the Transition to Adulthood The transition to adulthood in contemporary industrialized societies has undergone significant changes that can be summarized by first, a general delay of nearly all events Furstenberg, 2010b), and second, the de-standardization of the timing and order of events (Shanahan, 2000;Brü ckner and Mayer, 2005;Elzinga and Liefbroer, 2007). Latemodern societies are characterized by a high valorization of individualism, autonomy, self-fulfillment, and flexibility suggesting that the influence of the family of origin on young adults' lives has become less relevant (Kohli, 1986;Giddens, 1991;Beck and Beck-Gernsheim, 1994). A more structurally oriented literature, however, has stressed that socio-economic origin is still a major determinant in shaping the transition to adulthood (Brannen and Nilsen, 2002;Furlong and Cartmel, 2007;Settersten and Ray, 2010;Furstenberg, 2010b).
One influential approach in the literature on social inequality in family formation emphasizes 'diverging destinies' by social origin (McLanahan, 2004;Furstenberg, 2010a). This approach explicitly views early family transitions as potential signs of disadvantage, with cumulative negative consequences across the life course. This view is supported by findings that for instance, early (and often non-marital) childbearing, and teenage pregnancies in particular, occur more often among the socio-economically disadvantaged strata of society and have negative consequences for the future employment and partnership career of women (Upadhya and Ellen, 2011). Findings on an inverted relationship between educational enrollment and union formation (Blossfeld and Huinink, 1991), as well as between education and the timing of births (particularly for women, see Rindfuss, Bumpass and John, 1980;Martin, 2000;Mills et al., 2011) suggest that educational aspirations and opportunities vary by social origin and have consequences for demographic events in the transition to adulthood.
The story of the social stratification of leaving the parental home is usually told differently. Leaving home has not been significantly postponed during recent decades in all societies (Breen and Buchmann, 2002;Billari and Liefbroer, 2010). Given the connection between leaving home and the continuation of (tertiary) education (Billari, Philipov and Baizá n, 2001;Mulder and Clark, 2002;Buchmann and Kriesi, 2011), in advanced societies early home leaving is often viewed as a marker of advantage. However, the relationship between parental resources and the timing of leaving the parental home differs depending on whether young adults consider to marry or intend to live independently (Avery, Goldscheider and Speare, 1992). Leaving home too early might have negative consequences for educational attainment (Goldscheider, 1997;White and Lacy, 1997). Nevertheless, in societies characterized by extremely late home-leaving, later timing is associated with worse lifecourse outcomes (Billari and Tabellini, 2011). In a study on age deadlines for leaving the parental home in the Netherlands, Liefbroer and Billari (2010) found that these are lower for the highly educated than for the intermediate and lower educated.
Choice in the Transition to Adulthood: Stratified Socialization, Stratified Agency, Stratified Opportunity In the introduction, we distinguished three pathways linking parents' socio-economic background and young adults' demographic decisions. Below, we discuss these pathways and formulate hypotheses.
First, children from low and high SES family backgrounds differ in their expectations and intentions concerning the occurrence and timing of major demographic events (Keijer, Nagel and Liefbroer, 2016). Socialization is a key process through which parents influence the expectations and intentions of their children (Bengtson, Biblarz and Roberts, 2002). Parents' preferences thus shape their children's intentions regarding which demographic events they (do not yet) want to experience, for instance through parenting practices that are socially stratified (Lareau, 2011). Much work has been inspired by Kohn (1969), suggesting a greater appraisal of self-direction values among high-SES families and of conformity values among low-SES families. High-SES parents are thus likely to transmit to their children the embracement of values of self-exploration and self-focus linked to the postponement of life transitions that imply strong interpersonal commitment and are hard to reverse, such as union formation, marriage, and childbearing (Arnett, 2000). Therefore, young adults with a higher socio-economic background might be more likely to intend to leave the parental home relatively early, because they are more likely to internalize the idea that residential autonomy defines personal autonomy (Spéder, Murinkó and Settersten, 2014). At the same time, high-SES parents will value educational attainment very highly, sharing a concern about downwards mobility (Breen and Goldthorpe, 1997) and making them more likely to transmit preferences for the postponement of family formation. Low-SES parents, by contrast, may favour the early adoption of adult roles and lead their children to internalize preferences to commit to union formation, marriage, and parenthood earlier. Young adults from low SES background also tend to leave education and enter the labour market earlier, making them economically independent from their parents earlier, too. Building on Barber (2000), we label this pathway of parental influence on young adults' intentions as stratified socialization. Our first hypothesis is therefore: H1 (stratified socialization): Ceteris paribus, young adults from higher socio-economic background are more likely at a given time point to have intentions to leave the parental home, and less likely to have intentions to start living with a partner and enter parenthood than their counterparts from lower socio-economic background.
Children with low-and high-SES background not only differ in their intentions concerning demographic behaviours, but also in their ability to translate these expectations and intentions into actual behaviour. Recently, the concept of 'life-course agency' (Hitlin and Elder, 2006;Hitlin, 2007;Hitlin and Johnson, 2015) received increased attention. Life-course agency implies planful competence, hence the ability to pursue plans effectively and with perseverance (Clausen, 1991;Shanahan, Hofer and Miech, 2003), which is also related to the perceived ability to achieve a specific behavioural goal (Ajzen, 1991). We transpose this concept to young adults' capability to realize intended, and avoid unintended, demographic behaviour. We expect life-course agency to be socially stratified for two reasons. First, young adults from higher socio-economic background are better able to develop and stay committed to what they perceive to be advantageous long-term plans, i.e. develop planful competence (Hitlin and Elder, 2006). Second, more generous transfers from high-SES parents may increase high-SES young adults' agency and therefore better enable them to realize their intentions (Furstenberg, 2010a;Hitlin and Johnson, 2015). For instance, a young adult from a high-SES background aged 18 who intends to leave the parental home by age 21 might be more likely to realize this intention than her peer with a low-SES background, as growing up in an affluent and supportive family equips her with greater planning skills to organize the tedious search for appropriate and affordable housing, and makes her more resilient to cope with setbacks. In addition, her parents may activate their social network to find suitable accommodation or fund the costs of setting up an independent household. We label this pathway stratified agency. As it is based on general differences between children with different SES backgrounds in the capability to realize their intentions and expectations, we do not assume stratified agency to vary with the type of event. Therefore, our second hypothesis is: H2 (stratified agency): Ceteris paribus, young adults from higher socio-economic background are more likely to realize their intentions with regard to home leaving, union formation, marriage and entry into parenthood than their counterparts from lower socio-economic background.
The two pathways discussed so far emphasize differences between young adults from low-and high-SES backgrounds in agentic processes, with the latter group being better capable to set realistic goals (stratified socialization) and to realize such goals (stratified agency). In addition, though, stratification may partly be generated by differences in the opportunities and constraints that young adults from a low and a high SES background face. Such factors may generally lead to differences in the speed with which the transition into adulthood occurs within both groups. Children from low-and high-SES families face different constraints that often lead to the early occurrence of demographic events for the former compared to the latter group. Structural elements of the social context in which they grow up (e.g. crowded housing, unemployment rates, scarce career options, limited access to effective contraception) may push children from low SES background to a faster transition into adulthood, particularly in the sphere of union formation and parenthood. For instance, children from low-SES background may more easily drift into early parenthood than their high-SES counterparts, with a vast variety of explanations provided in the literature (Dribe, Oris and Pozzi, 2014). The opposite may be true for children from high-SES background, who may face structural constraints which could lead to the postponement of family formation events. Examples are that children from high-SES background often are enrolled in the educational system longer than children from low-SES background and that women from high-SES background face higher opportunity costs in combining family life and career than women from low-SES background (Blossfeld and Huinink, 1991). As a result, structural factors may lead to a relatively early occurrence of events among those from a low-SES background, and to a relatively late occurrence of events among those from a high-SES background. We label this third pathway stratified opportunity. Generally, we expect that this pathway will make it more likely for children from low-SES background to make steps in the family formation process earlier than those from high-SES background. As we outlined earlier, leaving home may be an exception, as children from high-SES background often need to leave the parental home early for educational purposes. Therefore, our third hypothesis is: H3 (stratified opportunity): Ceteris paribus, young adults from higher socio-economic background are more likely to leave the parental home earlier, and less likely to start living with a partner and enter parenthood than their counterparts from lower socio-economic background.
Figure 1 summarizes how we expect social background to influence the timing of demographic events in young adulthood. First, parental socio-economic background is expected to influence the intentions of young adults concerning the timing of demographic events (Path 1: Stratified Socialization). Second, parental socioeconomic background may influence the extent to which young adults are able to realize these intentions 1 (Path 2: Stratified Agency). Third, parental socio-economic background may influence the timing of the event via differences in the structural opportunities that children encounter during their transition to adulthood (Path 3: Stratified Opportunity).

Age, Gender, and Context Differences
The framework outlined above is expected to be generic. However, we explore whether the process differs between relatively young and relatively old young adults, men and women, and across societal contexts. At younger ages, intentions may be less firmly developed. This may be particularly true for intentions concerning marriage and parenthood, as these events often occur relatively late during the transition to adulthood (Bachrach and Morgan, 2013).
Whereas we expect that the framework outlined applies equally to men and women, the social stratification of choice in the transition to adulthood could differ between societal contexts. In countries that are more individualized and equal, both high and low SES parents may embrace views concerning demographic events marking the transition to adulthood that emphasize individual autonomy and postponement of committal life events (Adkins and Vaisey, 2009). As a consequence, differences in their children's scripts during young adulthood may be smaller than in less individualized countries. Moreover, the welfare state aims to reduce the role of social origin in the opportunities individuals have to shape their own life courses (Esping-Andersen, 1990), and in particular to compensate for social inequalities that young adults face in the transition to adulthood. In countries with a strong welfare state, young adults depend less on the economic resources of their parents to realize their demographic plans than in countries with a weak welfare state. As a consequence, one could expect differences between young adults from lower and higher socio-economic backgrounds in their ability to realize their intentions to be smaller in countries with a strong welfare state than in countries with a relatively weak welfare state.
Our exploratory country-comparative analyses will focus on three societies (France, Austria, and Bulgaria) that, despite being members of the European Union, show different patterns of welfare coverage. Bulgaria, after the transition from state communism, has experienced an economic crisis, low fertility and a collapse of social capital on the verge of anomie (Philipov, Spéder and Billari, 2006). In 2006, a year before the second wave of the national GGS was conducted, Bulgaria's Total Fertility Rate (TFR) was 1.44 (Eurostat, 2019a), with 51 per cent of births out-of-wedlock (Eurostat, 2019a), and women's mean age at first marriage being 25.7 (Eurostat, 2019c). Its expenditure on social protection was 14 per cent of the Gross Domestic Product (GDP) (Eurostat, 2018), and its GDP per capita in Purchasing Power Standards was 72 per cent lower than the European Union (28 countries, EU-28) average (Eurostat, 2019b). Its Gini Coefficient was 35.7, indicating the highest inequality among the three countries in this study (The World Bank, 2019). Austria and France have solid welfare states that provide support for young adult transitions, although the focus of welfare provision is on families, in conformity with a 'continental' model of welfare (Esping-Andersen, 1990). However, France is well-known for having a constantly higher fertility compared to its Western European neighbours. In 2011, the year before their second GGS wave, Austria's TFR was 1.43, with 40 per cent of births out-ofwedlock, and a women's mean age at first marriage of 30.3. Its expenditure on social protection was 30 per cent of the GDP, and its GPD was 28 per cent above the EU-28 average. Its Gini Coefficient was 30.8. In 2007, the year before their second GGS wave, France's TFR was 1.96, with 52 per cent of births out-of-wedlock, and a women's mean age at first marriage of 30.0. Its expenditure on social protection was 31 per cent of the GDP, and its GDP of 32.6 implying that it was 7 per cent higher than the EU-28 average.

Data
We use data from the GGS (Vikat et al., 2007). The GGS is a panel survey, conducted among nationally representative samples of the 18-79-year-old resident population in a large set of participating countries. In the first wave, the overall sample sizes by country was about 10,000 cases. Fokkema et al. (2016) present additional information on data collection procedures, nonresponse rates, and data quality.
The GGS implemented questions on intentions regarding demographic events measured in the first interview and its longitudinal design offers the opportunity to examine behavioural outcomes about 3 years later. The three countries we selected, Austria, Bulgaria, and France, had data for two waves. In these countries, questions on intentions concerning leaving home, entry into a union, entry into marriage and entry into parenthood were posed in exactly the same manner. the average interval between Waves 1 and 2 differed somewhat between countries. In Bulgaria it was 2.5 years, in France 3 years and in Austria 4 years. Retention rates were 73 per cent in Bulgaria, 65 per cent in France, and 78 per cent in Austria. For the large majority of Europeans, particularly from high-SES background, demographic events like marriage and childbearing typically occur between age 20 and 35 Buchmann and Kriesi, 2011).We therefore selected a sufficiently large age range, 18-35 years, in Wave 1.

Demographic Events: Intentions and Event Occurrence
Information on whether or not respondents had experienced the demographic events of interest were routinely collected as part of the reconstruction of respondents' life course. Respondents who had not yet experienced an event by Wave 1 were asked whether they intended to experience the event during the next 3 years (this roughly coincides with the interval between panel waves). Respondents who lived in the parental home were asked 'Do you intend to start living separately from your parents within the next 3 years?' Respondents who did not live with a partner were asked 'Do you intend to start living with a/your partner during the next 3 years?' The exact wording (a/your partner) depended on whether respondents at the time of Wave 1 were in a relationship or not. Respondents who were not married were asked 'Do you intend to marry somebody/your partner during the next 3 years?' In order to grasp marital intentions, we only selected those respondents who were currently in a steady partnership (either living apart together or in unmarried cohabitation), thus excluding respondents who had no current partner. Finally, childless respondents were asked 'Do you intend to have a child during the next three years?' For all these questions on intentions, answer categories were 1 ¼ 'definitely not', 2 ¼ 'probably not', 3 ¼ 'probably yes', and 4 ¼ 'definitely yes'. A higher value thus indicated a stronger intention to experience the event. In Wave 2, information on whether respondents experienced any of these events in between waves was collected.

Parental SES
We measured PSES by combining information on father's and mother's highest level of educational attainment as well as their occupational status when the respondent was 15 years old. This information was collected in the first interview. Parental education was measured by converting information on education into the International Standard Level of Education (ISLED) coding, which is a one-dimensional (continuous) score combining information on highest school level and highest vocational classification that is comparable across countries (Schrö der and Ganzeboom, 2014). Occupational status was measured by converting occupational codes into the continuous International Socio-Economic Index of Occupation (ISEI) (Ganzeboom, De Graaf and Treiman, 1992;Ganzeboom and Treiman, 1996). 3 The upper part of Table 1 presents the distribution of father's and mother's education and occupation for each of the countries in our sample. In Austria, fathers were somewhat higher educated than mothers, whereas no gender differences in level of education were observed in Bulgaria and France. The level of occupation also was slightly lower for mothers than for fathers in Austria. Once again, no gender differences were observed in Bulgaria and France.

Analytical Approach
We used Structural Equation Modeling (SEM) in MPlus 7.3 (Muthén and Muthén, 2014) to test our hypotheses. Using SEM was appropriate for several reasons (Kline, 2011). First, SEM allowed us to build an elaborate path model that examined different pathways of the social stratification of demographic behaviour. Second, using SEM we could examine additive and multiplicative relationships between observed and latent variables. PSES was measured as a latent variable, with father's and mother's ISLED and ISEI as indicators. Intentions and event occurrence were measured as manifest (or observed) variables.
We estimated one structural equation model (SEM) for each key demographic event to test our hypotheses. In these models, data from all three countries were pooled. Empirical models closely followed the logic of Figure 1. Intentions at Wave 1 were regressed on a set of control covariates (age, age 2 , gender, country dummies) and a latent variable indicating PSES. The effect of PSES on intentions indicated the existence of a stratified socialization effect. Whether or not an event occurred by Wave 2 was regressed on the same set of control covariates, intention in Wave 1, PSES and the interaction between intention and PSES. In addition, the time elapsed between the two waves was added to the analysis. The main effect of the standardized latent PSES variables indicated the existence of a stratified opportunity effect, whereas the interaction between PSES and intention showed whether the effect of intention on event occurrence differed by PSES, and thus indicated the existence of stratified agency.
We specified the intention variables as continuous and the event occurrence variables as dichotomous, and we fitted a linear model on the intention to experience an event and a binary logit model on whether or not an event occurred at Wave 2. 4 In a second step, we explored whether the effects differed by age, gender, and country. To do so, we ran three sets of multi-group SEM-models. In these models, we explored whether the socialization, agency, and opportunity effects (and the strength of the intention-behaviour link) differed between age groups, gender, and countries. Separate models were estimated in which one of these effects (i.e. pathways) was allowed to vary (1) between respondents 25 years and younger and respondents 26 years and older 5 ; (2) between men and women; (3) among respondents from Austria, France, and Bulgaria. Model fit was evaluated using the Bayesian Information Criterion (BIC) (Raftery, 1993). For the type of models that we estimated, other fit statistics have not been developed in the literature and we thus follow the developer's recommendation to compare models using the BIC. 6 An advantage of using BIC for model fit comparisons is that it puts a rather heavy penalty on complex models, and thus offers a natural way to only select a more complex model if it provides a significant gain in model fit.

Descriptive Findings
We first present descriptive information on intentions and realization of intentions among respondents of our analytical sample in the lower part of Table 1. Three main patterns are visible from these results. First, fewer respondents experienced events than intended to experience them. This was true for all events. However, this discrepancy was larger for events that require more rather than less commitment and agreement of a partner, for example marriage and parenthood compared to leaving home. For instance, in France 66 per cent of young adults still living in the parental home intended to leave home in the next 3 years, and 55 per cent actually left home. The percentage of childless young adults who intended to have a child within the next 3 years was 42 per cent, but only 8 per cent actually had a child 3 years later. So, the discrepancy between the percentage of respondents who intended to experience an event and the percentage that actually did so was much larger for having a first child than for leaving home.
Second, just as in studies on fertility behaviour (e.g. Testa and Toulemon, 2006), the intention-event occurrence relationship appears stronger for respondents with negative intentions. For instance, in Austria 94 per cent of respondents did realize their intention not to marry within the next 3 years, whereas only 35 per cent realized their intention to marry. The same is true for having a child: 94 per cent of childless young Austrians who intended not to have a child realized their intention, compared to only 31 per cent among those who intended a child. The discrepancy was smaller for leaving home and union formation, though. For instance, 69 per cent of Austrians realized their intention not to leave home, compared to 64 per cent realizing their intention to leave home. These findings support the notion that realizing a positive intention may require more resources and effort than realizing a negative intention, as the former implies change whereas the latter implies sticking to the current course of action.
Third, young Bulgarians seemed less likely to realize their intentions than respondents in France and Austria. For example, whereas 51 per cent of Austrians and 52 per cent of French respondents who wanted to start a union within 3 years did so, this was only true for

Multivariate Findings
To test our hypotheses on the social stratification of the transition to adulthood, we estimated separate SEM models for each of the four demographic events. Results are presented in  Notes: Pooled data from three countries (Austria, Bulgaria, France), all paths of the SEM constrained to be equal across countries. Estimates are unstandardized b coefficients. The dependent variable 'intention' is treated as a continuous variable. The dependent variable 'event occurrence' is treated as a dichotomous variable. *P < 0.05; **P < 0.01; ***P < 0.001.

PSES and intentions: stratified socialization
If the social stratification of demographic behaviour operates through stratified socialization, we would expect to find the intention to experience events in the transition to adulthood to be socially stratified (H1). Results on stratified socialization are presented in the panel where intention is the dependent variable in the line starting with PSES (stratified socialization). For three out of the four processes, evidence of stratified socialization was found. The higher PSES, the higher the likelihood that young adults intended to leave the parental home in the next 3 years and the lower the likelihood that they intended to enter marriage or have a first child within the next 3 years. These results were statistically significant at the 5 per cent level and thus clearly in line with Hypothesis 1. PSES did not show an association with the intention to start a first union though, indicating no evidence for stratified socialization in that process.

PSES and the realization of intentions: stratified agency
Life-course agency is expressed in the ability to realize one's intentions concerning major family life events. We expect this ability to be socially stratified, with higher PSES young adults being more likely to realize their intentions (H2). Please recall that realizing an intention may not only mean that the respondent succeeded in experiencing a desired demographic event between waves, but also that (s)he avoided an unintended event.
Before discussing the results of stratified agency, it is worth noting that a clear effect of intentions on behaviour was found for all four events. This is shown in Table 2 in the panel where event occurrence is the dependent variable in the line starting with Intention. Given that an interaction between intention and PSES was also included in the model, the estimate for intention can be interpreted as the effect of intention on event occurrence for respondents whose parents have an average SES score. A one-point increase in the intention score (that could run from 1 to 4) led to an increase in the odds that a respondent experienced an event by 56 per cent (exp(b) ¼ exp(0.44) ¼ 1.56) for leaving home, 92 per cent for union formation, 180 per cent for marriage and 161 per cent for first birth. This shows that intentions are strong predictors of subsequent behaviour.
Results on stratified agency are presented in Table 2 in the panel where event occurrence is the dependent variable in the line starting with PSES*Intention (stratified agency), indicating that we specified an interaction between PSES and the path from intention to event (Figure 1). The coefficient thus shows the change in the effect of intention on event occurrence if the PSES score increases by 1 unit. To facilitate interpretation, we show in Figure 2 the odds ratio of a one-point increase in intention for three groups of respondents; those whose parents have an average level of SES, those whose parents have a SES level one standard deviation below the average and those whose parents have a SES level one standard deviation above the average. For three of the four demographic behaviours (leaving home, union formation, and parenthood), a higher PSES was positively associated with the respondent's odds ratio to realize an intention towards event occurrence. Hence, young adults with a high-SES background were more likely to experience events that they intended to experience (and more likely not to experience events that they did not intended to) than young adults with a low-SES background with one exception. For the realization of marriage intentions, although the pattern is in the expected direction, PSES differences in the effect of intention on event occurrence were not statistically significant. These findings thus support Hypothesis 2 for home leaving, union formation, and first birth, but not for marriage.

PSES and event occurrence: stratified opportunity
We finally hypothesized that PSES influences the opportunity structure that young adults face leading young adults with a high SES background to postpone life transitions that require more commitment and are less (or not at all) reversible (H3). The corresponding results are reported in Table 2 in the panel where event occurrence is the dependent variable in the line starting with PSES (stratified opportunity). Findings suggest that the higher PSES was, the less likely were respondents to experience three (leaving home, union formation, and first childbirth) out of the four events within the next 3 years. For union formation and parenthood, these findings support Hypothesis 3. The opportunity structure facing young adults encouraged particularly young adults with a high SES background to postpone demographic events in the transition to adulthood.
Age, gender, and context differences We ran additional models to examine whether key effects differed between respondents below age 26 and those aged 26 and over, between men and women, and between countries. Results are presented in Table 3 (and  in Supplementary Tables S3-S5). We examined whether models in which estimates were allowed to vary across groups had lower BIC values than models where effects were constrained to be equal across groups. Three differences between age groups were observed (see Supplementary Table S3). At young ages, the likelihood to intend to marry and have a child within the next 3 years decreased by PSES, implying a higher likelihood of postponement among young adults with high-SES parents. However, above age 25, no differences in marriage and parenthood intentions were observed by PSES. In addition, both 'younger' and 'older' young adults were more likely to intend to leave the parental home the higher their PSES, with this gradient being larger at older than at younger ages. No differences at all between age groups in the process of union formation were observed.
Just one difference was found for gender (see Supplementary Table S4), with the intention-behaviour link for leaving home being somewhat stronger for women than for men. This suggests that generally the same models hold for both genders.
Generally, effects were comparable across countries as well, with one exception: the intention-behaviour link was weaker for leaving home in Bulgaria than in France and Austria (see Supplementary Table S5). No differences between countries in stratified socialization, stratified agency, or stratified opportunity were observed.

Conclusion and Discussion
Demographic events in the transition to adulthoodleaving the parental home, starting to live with a partner, marrying, having a first child-crucially shape individual life courses, and are socially stratified. Building on recent advances in life course theory, this study examined three pathways through which demographic choice in the transition to adulthood may be socially stratified. First, the stratified socialization pathway suggests that PSES shapes young adults' intentions regarding demographic events. Second, the stratified agency pathway expects young adults with different family backgrounds to differ in their resources to realize their intentions. Third, the stratified opportunity pathway suggests that general structural prescriptions and opportunities influence demographic behaviour in less reflective ways.
Our first hypothesis (stratified socialization) was that PSES influences the intentions of young adults to experience key demographic events in the near future. This hypothesis was confirmed. Young adults from advantaged family backgrounds were more likely to intend leaving the parental home within the next 3 years, but they were less likely to intend starting a union, marrying, or becoming a parent within the next 3 years. Thus, the intentions of young adults were clearly socially stratified.
Our second hypothesis (stratified agency) was that children with high-SES parents are more likely to realize their intentions than children with low-SES parents. We found support for this hypothesis as well. The higher PSES, the better young adults were able to realize their intentions concerning leaving home, union formation, and parenthood. No differences in stratified agency were found for entry into marriage. One reason for this could be that deviations from original marriage plans mainly depend on changes in the quality of the partner relationship, and such changes may be more or less equally likely among young adults from all different types of social backgrounds.
Finally, we hypothesized that children from low and high SES background were exposed to different life circumstances that influence their behaviour via structural processes. Our results showed that stratified opportunity was indeed important for all four outcomes. Controlling for differences in intentions and in the ability to translate intentions into behaviour, children were less likely to experience these events within 3 years (and thus are more likely to postpone them), the higher their PSES was.
Age differences were only observed for the stratified socialization pathway. At young ages, respondents with high-SES parents were less likely to intend to marry and have a child within 3 years than respondents with low-SES parents, suggesting that higher-SES children are socialized into postponing demographic events that require substantial commitment. No differences were found after age 25. This suggests that compared to their counterparts with low SES, high-SES parents not only transmit norms regarding the timing of major demographic events more successfully to their children, but also more strongly favour postponement of the transition to adulthood of their offspring.
We found hardly any evidence of gender differences. The only exception was that women were better able to realize their intention to leave home than men. Overall, the lack of gender differences suggests that the stratification model outlined here represents the processes among both genders equally well.
Overall, we found little indication for substantial country differences in the pathways studied. Having high SES parents led to stronger marriage intentions in Bulgaria, and to weaker ones in France and Austria. Furthermore, the link between intentions and behaviour was generally weaker in Bulgaria than in France or Austria. This suggests that the relatively bad economic and housing conditions in Bulgaria may affect the ability of young adults with both high and low SES backgrounds to realize their plans concerning major demographic decisions.
Some limitations of this study and a number of possible avenues for future research are worth mentioning. The validity of the three pathways of social stratification could be examined only among respondents who had not yet experienced the demographic events marking the transition to adulthood before Wave 1. This may be a selective group. Not surprisingly, additional analyses reported in Supplementary Table S1 show that female Notes: Models in which freeing up parameters improves model fit (based on BIC) are presented in bold. Models include the same variables and the same estimation procedures as in Table 2. and older respondents are more likely to already have experienced these demographic events prior to the first interview. Analyses reported in Table 3 (and  Supplementary Tables S3-S5), however, showed that there were few differences in the stratification channels comparing relatively young and relatively old respondents and men and women. To the extent that they exist, they suggest that the link between social background and intentions concerning marriage and parenthood was more pronounced among younger than among older respondents, but that the link between social background and intentions to leave the parental home was stronger at older ages. Furthermore, women were slightly more successful than men to realize their intentions regarding parental home leaving, and the same was true for respondents in France and Austria compared to those in Bulgaria. In light of the restricted number of age, gender and cohort differences in the pathways studied, we conclude that sample selection processes may only have a relatively weak effect on our results. In addition, additional analyses reported in the online Supplementary Materials revealed that young adults from low-SES family background were also more likely to already have experienced first union formation, marriage and childbearing prior to the first interview. The children of high-SES families are consequently overrepresented in three of our four analytical samples. Another concern may be the relatively large age range of our sample. The strong postponement and large cross-national variation in the duration of completing the transition to adulthood in Europe makes it necessary to avoid setting too low upper age limits because this may introduce bias in the comparative findings. Again, the fact that we found only small differences in the processes between young adults aged 18 to 25 and 26 to 35 suggest that processes operate in comparable ways at young and older ages.
Although our findings suggest that the social stratification of the timing of demographic events operates via three pathways, more research on their relative importance is necessary. First, intentions are expected to be volatile when opportunity structures change (Johnson, 2002). Thus, the shorter the time horizon of intentions, the better intentions will predict subsequent behaviour. The short-term horizon of the intentions questions is certainly a main advantage of the GGS data but especially in young adulthood, a lot may change even within the relatively short period of 3 years between two waves of data collection. In particular children from low-SES families are potentially exposed to external shocks, such as unemployment, that may lead them to revise their initial intentions.
Second, the different years of GGS data collection across countries potentially imply that the populations studied here were subject to different period effects that may affect all three pathways of parental influence.
Third, we used a unidimensional measure of PSES. It could be that stratified socialization depends more on cultural aspects of PSES, whereas stratified agency and opportunity depend more on economic aspects of PSES, and that there is a differential role for the mother's and father's SES. Distinguishing the relative importance of the transmission of cognitive skills via cultural resources vis-à -vis the provision of economic resources for the realization of intended behaviour, for instance, would require a decomposition of the multidimensionality of PSES. Only few data sets meet the necessary requirements. One recent study that does meet them suggests that this is a promising approach (Keijer, Nagel and Liefbroer, 2016). Although our data allowed combining information on both parent's SES based on observable characteristics, namely education and occupation, we could not firmly distinguish economic resources from cultural resources as we lack information on parents' income, an indicator that in the literature is agreed upon as the best measure of parents' economic resources (Kalmijn, 1994).
Fourth, while our empirical strategy based on structural equation models is appropriate to model statistically the various pathways we hypothesized, and to test the empirical hypotheses we generated, it does not automatically warrant a causal interpretation of our findings, i.e. that SES differences cause differences in the transition to adulthood. Our approach has been substantially driven by a 'causes of effects' strategy (Goldthorpe, 2016), where we start from the actual social stratification of the transition to adulthood to outline a model of the underlying social mechanism that implies testable hypotheses. However, given that PSES differences are temporally and logically antecedent of youth's transition to adulthood, we are confident that what we find on SES differences is unlikely to be spurious. The formulation of intentions at the individual level, however, may be affected by factors that are possibly correlated with the likelihood of realizing these intentions-and in this sense our findings on the link between intention and behaviour should be interpreted as documenting associations rather than evidence of causal relationships.
It would also be interesting to study SES differences in the possible destinations of leaving the parental home. Although we can identify some of the possible destinations (i.e. living independently vs. living with a partner), we cannot test our model as we only have a generic measure of the intention to leave the parental home. Another interesting avenue for future research would be to apply our framework to the sequencing of transitions in the transition to adulthood. This would imply collecting more detailed information on preferences of young adults about the sequences of transitions during young adulthood.
Another worthwhile idea is to examine our stratification model in a larger set of countries, preferably with a broad representation of different types of welfare state regimes and cultural backgrounds. This would allow generalizing our results to a broader societal context. As the number of countries with multiple waves of the GGS is increasing, the opportunities to approach this ideal design will improve.
The key contribution of this article is that it has demonstrated that the social stratification of the demographic life course can be linked to three complementary pathways: stratified socialization, stratified agency, and stratified opportunity. Creative use of a rich panel dataset allowed us to examine these pathways and find evidence for all of them across three different social contexts and for four demographic events. Future research should pay attention to all three pathways. In particular, emphasis on the stratified agency pathway seems warranted, given the emphasis on agency in narratives about changes in young adults' lives.
Notes 1 Our conceptualization of agency differs from the one in the psychological literature, where agency often is viewed as a personal characteristic that can be measured. We view agency as the ability to act in line with one's expectations and intentions. This ability can result both from personal characteristics (e.g. self-efficacy) and from resources (e.g. parents' financial contributions). We expect that agency is larger among children from high SES background. 2 In other countries with two waves of the GGP survey, with the exception of Germany, either the wording of the intentions or the answer categories differed. Germany was not included, because of the very low response rate in the second panel wave (33 per cent). 3 Missing data in the educational and occupational indicators was handled using Full Information Maximum Likelihood in SEM. 4 As a sensitivity analysis, we also estimated three alternative models; one model in which intentions were estimated with ordered logit and event occurrence with a binary logit, one in which both intentions and event occurrence were estimated with a binary logit, and one in which both intention and event occurrence were estimated with a linear model. Results from the first two estimation strategies are very similar to the one presented in the text. The model in which both intention and event occurrence were estimated with a linear model gave deviating results for the marriage and parenthood model. We decided to present results from the model with intention as a continuous and event occurrence as a dichotomous variable as this allows for a relatively simple interpretation of model estimates. Results of the all models are presented in the online Supplementary Material. 5 The decision to compare 18-26-year-old adults with those up to age 35 was driven by the aim to compare two subsamples that are about equally large. We experimented with other cut-off points to distinguish multiple age groups that did not change the main conclusions we drew from the comparison of the two age groups presented in the paper.
Oxford. His main research interests are fertility and family change, the transition to adulthood, life course analysis, population forecasting, digitalization, and comparative surveys.
Nicole Hiekel is a Post-doc Researcher and Lecturer at the University of Cologne and affiliated to the Netherlands Interdisciplinary Demographic Institute. Her research interests are at the intersection of sociology and demography. She studied the diversity of unmarried cohabitation, the consequences of cultural shifts in the meaning of marriage, the social stratification of the transition to adulthood, the impact of family structure in childhood on adult partnership behaviour and later life outcomes of women s family trajectories.