Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Neighbourhood effects on early adulthood family life courses: A trajectory‐based approach

Neighbourhood effects on early adulthood family life courses: A trajectory‐based approach INTRODUCTIONGrounded in Wilson's (1987) seminal treatise on The Truly Disadvantaged, neighbourhood effects on family events have received considerable attention. Researchers have focused on the impact of neighbourhoods on various family events, including teenage pregnancy (Baumer & South, 2001; Brooks‐Gunn et al., 1993), cohabitation (Zito, 2015), nonmarital childbearing (South & Crowder, 1999), sexual activity (Brewster, 1994), marriage (South & Crowder, 1999) and divorcee (South, 2001). Most of these studies have shown that neighbourhood characteristics are correlated with family events above and beyond individual and family attributes.Although the literature to date gains crucial insights into the role of neighbourhoods in family life courses, it has notable limitations. First, as implicated in Wilson's theory (1987, p. 56), neighbourhoods are crucial in keeping alive ‘that family stability is the norm, not the exception’. Moreover, the occurrence, timing and sequencing of family events are known to have crucial consequences for the later life course (Jalovaara & Fasang, 2020; Rindfuss, 1991). Focusing on isolated family events, therefore, provides an incomplete representation of the impact of neighbourhood effects on family life courses and deviations from conventional life‐course trajectories because we lose sight of longer trajectories following specific transitions. Childhood conditions and early life events and experiences can shape later life courses and (dis)advantage can accumulate over the life course (Merton, 1988). This is particularly true in American society, where family behaviour varies widely by social background characteristics (Carlson et al., 2004; Cherlin, 2010). For example, childbearing out of wedlock may be followed by single parenthood or normative age marriage may be complemented by stable and more traditional nuclear families. When individuals from advantaged and disadvantaged neighbourhoods are clustered into particular family formation patterns, traditionally applied approaches give an incomplete representation.Another important challenge in testing the effects of neighbourhood conditions on the life course of individuals is the identification problem (van Ham et al., 2018). Although previous literature has found that neighbourhood conditions are correlated with specific family events, it remains unclear to what extent these associations are due to the causal influence of neighbourhoods. The outcome of interest, in my case family life courses, and neighbourhood conditions might be correlated due to the differential selection of adolescents and their parents into specific neighbourhoods (Sampson et al., 2002). Individuals who are likely to experience specific events could also be nonrandomly allocated into neighbourhoods. Consequently, the observed correlations between neighbourhood characteristics and family trajectories could be driven by selection effects rather than the ‘actual’ influence of neighbourhoods.This study aims to fill these gaps in the literature on neighbourhood effects and family formation using four waves of data from the National Longitudinal Study of Adolescent Health (Add Health). Specifically, I examine typical family trajectories in early adulthood between the ages of 15 and 28 in the United States and relate these trajectories to adolescent neighbourhood conditions.I contribute to the literature in two ways. First, beyond focusing on isolated family formation events as is done in previous studies (e.g., South, 2001; South & Crowder, 1999; Zito, 2015), I consider family formation patterns as ‘process outcomes’ (Abbott, 2005). This strategy avoids the ‘short view on analytical scope’ (Elder, 1985, p. 31) and acknowledges the interdependence of multiple family events. Consequently, using sequence and cluster analysis, I test how neighbourhood conditions affect longitudinal family life courses by looking at single family events in the context of others. Second, I utilize an instrumental variable that is directly related to neighbourhood conditions but not to family life courses to improve the identification of causal neighbourhood effects on family trajectories.THEORETICAL BACKGROUNDUntil the mid‐2000s, studies on neighbourhood effects were predominantly conducted using the US context (e.g., Crane, 1991; Sampson et al., 1997; South & Crowder, 2000; Wilson, 1987), whereas there has been an expansion on studies focusing on the ramifications of neighbourhoods in the European context (e.g., Bråmå, 2006; Hedman et al., 2011; van Ham & Manley, 2012). These studies on neighbourhood effects suggest several mechanisms for how neighbourhoods influence the later life course of adolescents. Perhaps the most influential theoretical explanation comes from Wilson (1987, 1996), who proposes three causal mechanisms that link neighbourhood disadvantage to adverse outcomes. The first mechanism is the epidemic or contagion model, which emphasizes the influence of peers in a neighbourhood. Individuals change or adjust their behaviour according to the benchmarks or social norms set by others in a group to gain the approval of or avoid conflict with others in the group (Asch, 1955; Festinger, 1954; Festinger et al., 1950). Evidence, indeed, indicates that adaptation of behaviour is strongly correlated with the number of relevant others already exhibiting that behaviour in the neighbourhood (Crane, 1991) and that friends' and peers' family formation behaviour are associated with each other (Balbo & Barban, 2014).According to this model, the distinctive normative patterns in disadvantaged neighbourhoods fail to encourage conventional life courses (Brewster, 1994). In contrast, neighbourhoods characterized by concentrated poverty and concomitant social isolation may legitimize alternative subcultures in peer groups that promote nonnormative behaviour such as risky sexual activity, adolescent nonmarital childbearing and less attentive contraceptive behaviour (South & Baumer, 2000).The second mechanism, the collective socialization perspective, emphasizes the positive influence of successful adult role models who serve as an important social control function over adolescents and demonstrate the benefits and viability of stable family structures and the expectation of remaining married. Contrary, poor adults in disadvantaged neighbourhoods do not provide such successful role models increasing the risk of nonconventional family formation patterns such as childbearing outside of marriage (South & Crowder, 2010) and family instability (South, 2001). The relevance of role models in the family life course is also emphasized recurrently in the social interaction effects and family formation behaviour literature: individuals learn from other role models through observation, imitation and modelling, which affects their decision‐making process (Bernardi & Klaerner, 2014; Buyukkececi, 2021; Buyukkececi et al., 2020). The learning process can take different forms, ranging from small observations to intensive discussions with role models about the consequences of certain behaviours.The third is the institutional model, which points to the importance of local institutions, resources, services and other organizations. These include educational institutions, teachers and police officers available in a neighbourhood. Legitimate opportunities to attain adult status and financial security may be scarce in poor neighbourhoods (Brewster, 1994). Consequently, the lack of desirable jobs and the weakness of educational institutions may reduce the costs of nonnormative transitions in disadvantaged neighbourhoods (Jencks & Mayer, 1990). Not only the opportunities but also the perceived opportunities of adolescents may have an impact on family formation behaviour. Individuals with low educational and labour market aspirations and low school attachment might be more likely to deviate from normative life course trajectories, while adolescents from advantaged backgrounds might delay entry into adult family statuses because of high educational and career expectations. Furstenberg (2010) explains that the timing and sequencing of transitions in adulthood, namely cohabitation, marriage and childbearing, become normative for youth from advantaged backgrounds only after completing education and securing employment.In line with these mechanisms, social disorganization theory asserts that delinquency based on structural and cultural factors affects the social order within and across communities (Sampson et al., 1997). Respondents who are victims of crime or fear crime in their neighbourhoods are expected to be less connected to and involved in community life. Neighbourhoods with severe poverty and other disadvantages are often associated with various types of social disorder conditions that may increase a lack of social integration, feelings of mistrust and a sense of insecurity, which could subsequently weaken relationship quality and relationship formation behaviours.Along with statistical advances and burgeoning large‐scale data resources (e.g., Earls & Buka, 1997; Sastry et al., 2006) that have enabled the identification of neighbourhood contexts and their characteristics, empirical research has built on these theoretical innovations such as the epidemic, collective socialization and institutional models (Sampson et al., 2002) in the last decades and examined the relationship between neighbourhood effects and youth outcomes. A vast share of these studies has focused on the role of neighbourhood conditions and family formation behaviour. Most findings, indeed, have shown that neighbourhood disadvantage is positively associated with nonnormative family formation events such as teenage childbearing (Baumer & South, 2001; Brooks‐Gunn et al., 1993; Harding, 2003), nonmarital fertility (South & Crowder, 1999, 2010) and divorcee (South, 2001). Moreover, neighbourhood disadvantage hastens the transition to marriage among white individuals, whereas it has the opposite effect on black individuals (South & Crowder, 2000).These studies primarily considered stable marriages as normative behaviour, whereas childbearing out of wedlock and marital disruption were regarded as nonnormative behaviour. Yet, new ways of living together have emerged in the United States as well as in other western countries. Cohabitation has become widespread and normative (Raley, 2000). Although it used to be more common among people with less education and disadvantageous background, there is evidence that the cohabitation gap has narrowed in recent decades among all educational groups (Cherlin, 2010). Indeed, Zito (2015) found no significant association between neighbourhood‐level economic disadvantage and teenage cohabitation using the National Longitudinal Study of Adolescent Health (Add Health) data and focusing on cohorts born after the 1970s.Although previous literature provides a snapshot of the relationship between neighbourhood conditions and family formation, these studies are generally limited to isolated focal transitions such as out‐of‐wedlock births or divorcees. Focusing on specific aspects of family formation and traditionally applied methods such as event history models have their limitations when it comes to examining the full range of neighbourhood effects on family formation behaviour. It has been well acknowledged that the occurrence, timing and sequencing of these transitions and events are interrelated (Rindfuss, 1991; Wilson, 1987). Individual life courses unfold over time and need to be studied over longer periods (Elder et al., 2003). This is particularly relevant in adolescence and early adulthood, a ‘demographically dense phase’ (Rindfuss, 1991, p. 496) in which a multitude of events take place and decisions are made (Richter, 2006). These early life events and experiences, as well as childhood conditions, shape later life, and inequality can accumulate over the life course as postulated by the cumulative inequality theory (Ferraro & Shippee, 2009; Merton, 1988). In American society, it may be even more important to take a longitudinal life course perspective when testing for neighbourhood effects and to conceptualize family events as a ‘process outcome’ (Abbott, 2005) that unfold from adolescence through early adulthood because family formation patterns vary considerably across the social structure. Nonmarital childbearing, single parenthood and family instability, for instance, are more common in socioeconomically disadvantaged families (Carlson & Furstenberg, 2006; Carlson et al., 2004), resulting in more complex and fluid families (Cherlin, 2010). In contrast, individuals from advantaged backgrounds are more likely to postpone marriage and parenthood and have stable marriages (McLanahan & Percheski, 2008). Accordingly, the consequences of childhood neighbourhood conditions may extend beyond single family events to influence later life outcomes (Elder, 1985; Elder et al., 2003), and it is important to understand how childhood neighbourhood conditions are associated with the social pathways of individuals' family lives.This study considers single family events in the context of others from adolescence through early adulthood. I expect that the likelihood of family trajectories characterized by early and possibly out‐of‐wedlock births and less stable relationships increases for adolescents growing up in disadvantaged neighbourhoods. Contrary, I expect that family life courses outlined by late union formation and childbearing combined with stable marriages are more accessible in advantaged neighbourhoods.Methodological issuesAnother important gap of knowledge in the previous literature focusing on neighbourhood effects and family formation concerns neighbourhood selection, which presents noteworthy obstacles to drawing conclusions on the causal role of neighbourhood effects (Sampson et al., 2002; Sobel, 2006). Most individuals do not choose where to live randomly, and sorting into a neighbourhood is considerably structured (van Ham & Manley, 2012) due to various reasons such as affordable rents, discrimination and availability of housing (Desmond, 2016). Accordingly, individuals living in certain neighbourhoods might have (un)observed characteristics that are correlated with the outcome of interest above and beyond the neighbourhood effects. For instance, individuals might not be poor(er) because they live in a neighbourhood with concentrated poverty, but they might live in a poor neighbourhood because the rents are affordable. ‘Choices are restricted by household preferences, resources, and restrictions, but also by constraints imposed by the structure of the housing market. Likely, poor households do not choose to move to poverty neighbourhoods, but move there because they cannot afford to live anywhere else’ (Hedman et al., 2011, p. 1395).To overcome this bias and examine the consequences of neighbourhood conditions for family formation, previous literature recurrently included multiple controls to account for individual‐ and family‐level characteristics and disentangle neighbourhood effects (e.g., Baumer & South, 2001; South & Crowder, 1999, 2010). However, it is highly likely that this strategy does not fully account for the selection factors that lead to spurious correlations between neighbourhood conditions and family events. In another study, Harding (2003) examined how neighbourhood conditions are related to school dropout and teenage pregnancy by utilizing Propensity Score Matching (PSM) and counterfactual models to improve the identification of neighbourhood effects. Yet, King and Nielsen (2019) indicate that this strategy often accomplishes the opposite of the indicated goal and even increases imbalance, inefficiency and bias under certain conditions building on three main arguments: First, PSM aims to mimic randomized experiments, rather than randomized block experiments, and the latter provides much better precision and control against confounding. Second, PSM introduces the ‘propensity score paradox’ due to further trimming of units, which increases the imbalance beyond a certain point, unlike other matching methods. Third, the effect estimate of PSM is more sensitive to the model specification compared to the other matching methods. For these reasons, the authors conclude that this strategy should not be used for causal inference.Another strand of research has attempted to improve the identification of neighbourhood effects by using social experiments such as the Moving to Opportunity (MTO) programme, in which families in high‐poverty neighbourhoods were randomly offered the opportunity to move out of the neighbourhood and live in a low‐poverty area. These studies have focused predominantly on outcomes such as economic well‐being (Katz et al., 2001), crime (Ludwig et al., 2001), and employment (Rosenbaum & Harris, 2001) and have emphasized that the findings of neighbourhood effects in nonexperimental studies are challenged by the results of the MTO experiment. Studies examining the consequences of neighbourhood conditions for family formation using a similar strategy are scarce. To the best of my knowledge, there is only one study by Chetty et al. (2016) that followed up with MTO participants over the long term using IRS data and found that neighbourhood disadvantage was positively associated with single parenthood. In a more recent study, Harding et al. (2021) emphasized that observational and experimental approaches yield different results for MTO adults. To illustrate that, the authors compared experimental and nonexperimental estimates of MTO with a parallel analysis of Panel Study from Income Dynamics (PSID). The results showed similarities between the nonexperimental estimates of MTO and PSID, but differences from the experimental estimates of MTO suggest that the neighbourhood effects on economic outcomes may be due to selection bias, at least for economic outcomes.Although the nonexperimental estimates of MTO and PSID produced relatively similar findings, it should still be noted that an important drawback of MTO is the population to which the experimental estimates can be generalized (Ludwig et al., 2008). The selected group for the programme was defined as the families with children living in public housing projects in poor neighbourhoods in five major cities in the United States. These families had to meet several requirements, such as having paid their rental payments on time and not having a criminal record (Goering et al., 2003). Moreover, only about a quarter of eligible families applied for the programme and randomization occurred among this group. As a result, Ludwig et al. (2008) emphasize that MTO participants' data are meaningful only for this subset of the population, namely individuals living in disadvantaged neighbourhoods that were interested in moving, and sufficiently completed the application process.In the last decades, a growing body of literature has modelled the likelihood of a family moving to a particular type of neighbourhood based on a set of neighbourhood characteristics (Bråmå, 2006; Clark & Ledwith, 2007) to improve the identification of neighbourhood effects. More recently, van Ham et al. (2018) further developed this approach and proposed an empirical framework in which they modelled neighbourhood choice with a conditional logit model and estimated the probability that each individual would choose a particular neighbourhood from a set of 203 neighbourhoods. In the second step, they incorporated the correction components from the first step into a model of neighbourhood effects, which allowed relaxing the selection bias assumption. This strategy accounts for the unobserved characteristics of individuals that affect the likelihood of selecting particular neighbourhoods based on predetermined observable conditions. Similar to studies using the MTO programme, this line of research has investigated economic outcomes or mobility behaviours and suggested that neighbourhood effects on adult outcomes may be driven by selection bias.In summary, causal evidence on neighbourhood effects on family events is limited to the experimental study by Chetty et al. (2016) who focused on single parenthood events in five large US cities, as discussed above. In this study, I test for causal neighbourhood effects with a nonexperimental design by examining a nationally representative sample using Add Health data and looking at single family events in the context of others longitudinally rather than focusing on isolated events. To improve the causal identification of neighbourhood effects, I propose an instrumental variable approach based on an exclusion restriction that assures that the instrumental variable is directly linked to the outcome (i.e., family trajectories) only through the main predictor (i.e., neighbourhood disadvantage).DATAI used four waves of the National Longitudinal Study of Adolescent Health (Add Health), a school‐based, nationally representative longitudinal study of US adolescents who attended grades 7–12 in Wave I (1994–95). The sampling strategy was based on a systematic random sample of high schools and ‘feeder’ schools (see Harris, 2013). The primary sampling frame was all high schools in the United States with a 11th grade and an enrolment of at least 30 students. From these high schools, a random sample of 80 schools was selected. The sample was stratified by ethnicity, region, school type and size, and urbanicity. Together with feeder schools, the final sample included 134 schools, and the number of neighbourhoods represented in the high schools ranged from 3 to 88 in the original sample. The Add Health cohort was followed for four rounds of in‐home interviews (Wave I in 1994–95, Wave II in 1996, Wave III in 2001–02, and Wave IV in 2008–09). The initial sample consists of 20,745 youth who completed the in‐home interview. During the in‐home interview, 85% of the individuals' parents were interviewed, which provided detailed information about the parents (commonly the mother). Only after the collection of Wave IV, the Add Health cohorts become old enough—aging between 26 and 33 years—to provide comprehensive information on early adulthood family formation.For the analyses, Add Health allowed me to combine two types of information: (i) family formation events and trajectories and (ii) neighbourhood and other childhood background characteristics. Each wave provides detailed information on relationships and fertility histories. These include the timing of each pregnancy and live births, the type of relationship with romantic partners as well as the start and end dates of each relationship.Add Health includes individual‐level information as well as school‐ and neighbourhood‐level information with its multilevel data structure. It provides contextual information on structural neighbourhood characteristics at the county, census tract, and census block group summary levels by linking individuals' communities and neighbourhoods to the preexisting US census measures. These indicators include rates, proportions and population measures ranging from the share of families below the poverty level to the percentage of workers in managerial or professional occupations. Moreover, Add Health provides unique identifiers at the neighbourhood level that allow me to determine individuals living in the same neighbourhood. Combining this information with data on family formation trajectories and neighbourhood characteristics is exceptionally suitable to assess neighbourhood effects on family formation trajectories.Sample selectionI focused on family formation trajectories between the ages of 15 and 28. The choice of the upper age limit was determined mainly by data availability. For instance, approximately 16% and 19% of the original sample would be excluded from the analyses if the age range of the family sequences were extended to 29 and 30 years, respectively. However, additional analyses focusing on these age ranges yielded qualitatively similar results (available upon request). In Wave IV, 92% of individuals from the initial sample were located and approximately 80% of these respondents were re‐interviewed resulting in 15,701 individuals. A total of 3310 respondents younger than 28 in Wave IV were excluded from the analysis. Furthermore, individuals with missing information in the parental questionnaire (n = 2025, i.e., 16% of the sample) were excluded. The proportion of individuals with missing information in the parental was very similar to the baseline sample, supporting the overall representativeness of the sample. A total of 178 respondents (2%) with more than five missing family states in their family trajectories between ages 15 and 28 and 609 individuals (6%) with missing information on independent variables or weights were excluded from the sample. Lastly, I was not able to determine the instrumental variable used to test for causal neighbourhood effects for 539 individuals (6%). Consequently, they were excluded from the analyses resulting in 9040 respondents. Moreover, the final analysis sample comprised 132 high schools and 2316 neighbourhoods. An overview and description, as well as missing information on each variable after sample restrictions, are shown in the Supporting Information: Table A1.METHODSMy empirical analyses proceeded in three steps. First, I used sequence (Abbott, 1995) and cluster analyses to identify and compare typical family formation trajectories. Second, I estimated how sorting into a particular trajectory is related to neighbourhood characteristics by employing multinomial logit models. In the final step, I utilized an instrumental variable estimation to determine whether neighbourhood conditions are causally related to respondents' family trajectories.Sequence analysisI used biographical information on relationship and fertility histories from the four waves of Add Health to identify family trajectories from ages 15 to 28. Family sequences were coded based on the most common 8 states that combined cohabitation, marital status, and the number of children that were also meaningful substantively: 1. ‘Single with no child’; 2. ‘Single with children’; 3. ‘Cohabitating with no child’; 4. ‘Cohabitating with children’; 5. ‘Married with no child’; 6. ‘Married with 1 child’; 7. ‘Married with 2 children’; and 8. ‘Married with 3+ children’.1For individuals whose family state information was missing in a given year, I additionally included another state indicating that the family state was missing.‘Single’ states referred to those who were neither cohabiting nor married. Separated or divorced individuals were captured in the sequential order of the states because these individuals became single following a cohabitation or marriage episode. I distinguished between cohabitation and marriage because of the different selections between cohabitation and marriage. Although cohabitation is becoming more widespread in the United States regardless of social background characteristics, there remains a social divide in expectations and behaviours for forming unions, with the most advantaged more likely to marry and the least advantaged more likely to cohabitate(Cherlin, 2010; Gibson‐Davis et al., 2018; Sassler & Miller, 2017). Moreover, cohabiters comprise a highly heterogeneous group. They may regard cohabitation as a precursor or alternative to marriage or singledom (Perelli‐Harris & Gassen, 2012; Smock, 2000). Cohabiters have less commitment, greater freedoms and a higher risk of separation (Perelli‐Harris & Lyons‐Amos, 2015). Contrarily, marriage is still regarded as the ultimate commitment in most western countries (Perelli‐Harris & Gassen, 2012). Considering cohabitation and marriage as separate states in the analyses is further important because the qualitative meaning of marriage may differ between direct marriages and marriages starting with cohabitation (Wu & Schimmele, 2005).After constructing the family states, I used to sequence and cluster analyses to identify early adulthood family formation trajectories. Life‐course research uses sequence analysis to study processes of life course trajectories—in my case family trajectories from ages 15 to 28—that are an ordered collection of states over time. Sequence analysis is usually combined with cluster analysis to identify the most similar sequences and group them into typologies. A common method used for sequence comparison in social sciences is optimal matching (OM; Abbott, 1995), which measures the distance between two sequences by calculating the ‘cost’ of turning one sequence into another (Macindoe & Abbott, 2004). It proceeds with three transformation operations: (i) substitution of one state with another, and (ii) insertion, or (iii) deletion of a state. The cost of each operation is assigned by the researcher, and the distance between two sequences is defined by calculating the minimum cost of turning one sequence into another. In the sequence and cluster analyses, I used OM, a method that remains the most used measure in sequence analysis (Studer & Ritschard, 2016), and specified constant substitution costs of 2 and indel costs of 1 to ensure that both the timing and order of the family states contribute to the calculation of similarity between sequences (Aisenbrey & Fasang, 2010; Macindoe & Abbott, 2004).To identify the appropriate number of groupings to be extracted, I used ward clustering. These results were combined with portioning around medoids, which allowed me to obtain the most discriminant groups in my sample (Studer, 2013). Guided by the average silhouette width (ASW) cut‐off criteria, which is based on comparing average within‐cluster distances and average between‐cluster distances, I retained the six‐cluster solution as the optimal grouping quantitatively with an ASW value of 0.39. The validation strategy shows that observations closer to 1 are well‐clustered, whereas groups become more heterogenous with smaller silhouette widths, and values smaller than 0.25 are not well‐structured (Kaufman & Rousseeuw, 1990). Accordingly, the highest ASW value indicated that the observations are most similar within their profiles and most distinct from the other profiles. The six‐cluster grouping also provided the most meaningful solution substantively and was supported by other cluster cut‐off criteria as shown in Supporting Information: Figure A1. For the sequence and cluster analyses, I used the R packages TramineR, TraMineRExtras, and WeightedCluster (Gabadinho et al., 2011; Studer, 2013). Moreover, sample weights were utilized to determine the representative family life courses in the United States with cluster and sequence analyses.Multinomial logit modelsI estimated the probability of sorting into a particular family typology with multinomial logit models. In the analyses, longitudinal weights adjusted for clustering and stratification, as well as the design type, were used. The main predictor was neighbourhood disadvantage, which was determined based on census blocks. Following South and Crowder (1999) and Baumer and South (2001), I measured neighbourhood disadvantage by a standardized index constructed from six highly correlated neighbourhood characteristics: (a) percentage of households receiving public assistance, (b) percentage of families below poverty level, (c) percentage of workers not in managerial or professional occupations, (d) percentage of persons aged 25 and older without a college education, (e) percentage of families earning less than $50,000, and (f) percentage of unemployed men (Cronbach's α = 0.88). These data come from the 1990 US census included in the first wave of Add Health, which allowed me to measure neighbourhood disadvantage before age 152Only 49 individuals (i.e., 0.67% of the sample) were 15 years old in 1990., which satisfied the temporal ordering of neighbourhood characteristics and outcomes. Only 49 individuals (i.e., 0.67% of the sample) were 15 years old in 1990. Following the determination of typical early adulthood family life courses, the probability of sorting into a particular cluster was estimated with the following equation:1Pr(ai=j)=exp⁡(β0j+β1jNDIi+β2j′Xi′+β3jφi)1+∑j=15exp⁡(β0j+β1jNDIi+β2j′Xi′+β3jφi), $\text{Pr}({a}_{i}=j)=\frac{\text{exp}{\rm{}}({\beta }_{0j+}{\beta }_{1j}{\mathrm{NDI}}_{i}+{\beta }_{2j}^{^{\prime} }{X}_{i}^{^{\prime} }+{\beta }_{3j}{\varphi }_{i})}{1+\sum _{j=1}^{5}\text{exp}{\rm{}}({\beta }_{0j+}{\beta }_{1j}{\mathrm{NDI}}_{i}+{\beta }_{2j}^{^{\prime} }{X}_{i}^{^{\prime} }+{\beta }_{3j}{\varphi }_{i})},$where Pr⁡(ai=j) $\text{Pr}{\rm{}}({a}_{i}=j)$ was the probability of sorting into a specific cluster for individual i. NDIi ${\mathrm{NDI}}_{i}$ was the endogenous neighbourhood disadvantage index and β1j ${\beta }_{1j}$ was the main coefficient of interest. Xi denoted the set of controls shown in Supporting Information: Table A1 that might be related to both neighbourhood disadvantage and family formation behaviour. This set of variables included sex as neighbourhood effects on adolescent outcomes might be different for men and women (see Leventhal & Brooks‐Gunn, 2000; for a review). I further controlled for race and immigrant status, given that there are notable racial differences both in neighbourhood disadvantage and family formation behaviour (e.g., South & Crowder, 1999, 2010). I also included a set of parental background characteristics that were used in previous literature and are likely to be related to both the main predictor and the outcome. These variables were the number of siblings, parental marital status, education and the natural logarithm of the equivalized household income, and parents' total number of relationships up to Wave I.3As parental education partly accounted for parental SES and including parental income would introduce more missing values, parental household income was not included in the main models. Nevertheless, models were replicated by also considering this variable as a robustness check and the findings are reported in Supporting Information: Appendix. Lastly, φi ${\varphi }_{i}$ represented an unobserved individual‐level component, which was normally distributed with mean 0 and variance 1.As discussed earlier, the main challenge in estimating neighbourhood effects is selection bias as a result of selective sorting into neighbourhoods, indicating that the outcome of interest is not independent of selection into neighbourhoods (Brooks‐Gunn et al., 1997; Duncan et al., 1997). One approach to disentangle neighbourhood effects from selection effects is to utilize an instrumental variable approach. Lundborg (2006), for instance, applied a strategy closer in spirit to the identification strategy used in the present study. The author implemented various average classmate characteristics, such as the share of individuals living in a single parent household as instrumental variables. Similarly, Fletcher (2012) used the availability of alcohol in classmates' households to assess peer effects on alcohol consumption.To have compelling have a compelling instrumental variable, the instrument must be (i) correlated with the main predictor (i.e., neighbourhood disadvantage) and (ii) validly excluded from the outcome of interest (i.e., family life courses). I exploited information on neighbourhoods and parental questionnaires, to have a valid instrument satisfying these conditions. Using neighbourhood identifiers from Wave I, each teenager in the sample was randomly linked to another teenager living in the same neighbourhood. Second, I utilized an item from the matched individual's parental questionnaire that indicated whether the assigned teenager's caregiver lived in that neighbourhood because of its proximity to the workplace. The item was based on a question included in the parental questionnaire of Wave I: ‘You live here because this neighbourhood is close to a place where you (or your spouse or partner) work now’ (see item PA28B in the parental questionnaire codebook).4About 37% of the parents interviewed in Wave I mentioned that they live here because the neighbourhood is close to the workplace in the original sample, whereas this number was 39% for the matched adult neighbours included in the analyses as shown in Supporting Information: Table A1.The strategy builds on the spatial mismatch hypothesis first proposed by Kain (1968) who focused on where individuals live and where jobs are located. The hypothesis posits that low‐income individuals living in disadvantaged neighbourhoods are far from job opportunities, whereas individuals from advantaged backgrounds voluntarily choose to live in neighbourhoods that are closer to their jobs or near employment opportunities. Accordingly, my approach is based on the assumption that a neighbour's parent's preference to live in a neighbourhood closer to work is correlated with neighbourhood disadvantage, whereas it is not (directly) related to the family life courses of the focal person. Moreover, as argued by Hernán and Robins (2020), the instrument must be associated with the main predictor, but the causal direction of this association or other factors such as selection is not important as long as the instrument is not related to the outcome of interest and the error term in the main equation. Accordingly, the instrumental variable employed satisfies the exclusion restriction even when selective individuals are sorted into specific neighbourhoods, as a neighbour's parent's work‐related housing preferences are unrelated to an individual's family trajectory.One way this assumption can be violated is if an adult neighbour's occupational housing preferences are directly related to the focal person's family life trajectory. For example, family formation behaviour and housing preferences are important determinants of short‐distance residential moves (e.g., Clark & Huang, 2003; Feijten & Mulder, 2002), and as family size increases, the likelihood of moving long distances for a job decreases (Kulu, 2008). Moreover, recent evidence shows that flexible working schedules or remote work promote work‐life balance and have a positive impact on fertility (Billari et al., 2019). Consequently, an individual's work‐related housing preferences could be related to his or her own family formation behaviour, which in turn could affect the trajectory of the focal person's family life. Even if this is the case, such an effect would be indicative of neighbourhood effects. This is because it would be consistent with the second mechanism of neighbourhood effects, namely the collective socialization perspective, which emphasizes the influence of adult role models in the neighbourhood.Despite these merits of the analytical strategy, I performed two sensitivity analyses with two additional instruments and compared the results with the main models to further test the reliability of my instrumental variable approach. The first additional instrument used was expected to be related to neighbourhood disadvantage and have no direct effect on the outcome. Contrary, the second instrumental variable was expected to be directly related to both the main predictor and the outcome. The analytical strategy, as well as the findings, are explained in detail in the Supplemental Analyses section.According to the rule of thumb proposed by Staiger and Stock (1997) and Stock and Yogo (2005), the F‐statistic for the significance of the instrument in the first stage regression should be greater than 10 to have a compelling instrumental variable that satisfies the criterion validity. Analyses testing for weak instruments revealed that the F‐statistic was 23.34 as shown in Supporting Information: Table A2. Although the estimated R‐squared value (0.0026) value was low suggesting a weak relationship between the main predictor and the outcome, the F‐statistic was above conventional requirements of instrument relevance, and neighbourhood disadvantage was significantly lower when a neighbour was living in that household because of proximity to the workplace. After validating my instrument, I used the two‐stage residual inclusion (2SRI) estimation strategy developed by Terza et al. (2008) to address the endogeneity in the multinomial logit model. In the first stage, the endogenous variable was regressed on the instrumental variable and exogenous regressors. In the second stage, residuals from the first stage regressions as regressors were included in the multinomial logit regression. The first stage was specified with the following equation:2Ni=α+β1Zi+β2′Xi′+φi+εi, ${N}_{i}=\alpha +{\beta }_{1}{Z}_{i}+{\beta }_{2}^{^{\prime} }{X}_{i}^{^{\prime} }+{\varphi }_{i}+{\varepsilon }_{i},$where Zi ${Z}_{i}$ was a dummy indicating whether the matched neighbour's caregiver was living in the neighbourhood because it is close to the place of work. Xi denoted the set of variables included in Equation 1. φi ${\varphi }_{i}$ was an unobserved individual‐level component and εi ${\varepsilon }_{i}$ was the error term, which was normally distributed with mean 0 and variance σ. All models were estimated with the Generalized Structural Equation Modelling (gsem) command of Stata. Given the sampling frame of Add Health, an alternative strategy might be to use a 2‐level multilevel model in which respondents were nested into schools. However, multilevel models were not used in the main analyses for two reasons. First, in the analysis sample, 58 of 132 schools had fewer than 50 observations, and modelling with unbalanced and small group sizes (i.e., spareness) could suffer from several problems, including upwardly biased estimates of fixed‐effects coefficients and convergence problems (Clarke, 2008; Clarke & Wheaton, 2007). For instance, Schunck (2016) estimated that at extreme parsimony (i.e., only 5–10 observations per cluster), the average relative bias exceeded 93%, and the bias was substantial even at moderate cluster sizes (i.e., 20–40 observations per cluster). Second, estimated effects can differ greatly from actual effects when including an average measure on the right‐hand side of the model constructed by aggregation within the data (Kravdal, 2003), and can also lead to inconsistent IV estimates (Hinke et al., 2019). Nevertheless, non‐IV models were replicated using a 2‐level model that allowed for the grouping of individual outcomes within schools, and the results are reported in Supporting Information: Appendix.RESULTSEarly adulthood family life coursesFigure 1 shows the sequence index plots of the six‐cluster solution: 1. Single childless; 2. Single parent; 3. Cohabitating childless; 4. Cohabitating parent; 5. Married, later parenthood; and 6. Married, early parenthood. Each row represents an individual's family trajectories from age 15 to 28. Descriptive information for each cluster, including the average length of time spent in eight specific family states, is presented in Table 1. Moreover, Supporting Information: Figure A2 shows relative frequency plots sorted by silhouette values (Fasang & Liao, 2014). It illustrates 50 sequences of each cluster with the highest silhouette values that most strongly represent the main features of the profiles.1TableDescriptive information on clustersSingle childless (33.78%)Single parent (5.72%)Cohabitating childless (16.70%)Cohabitating parent (11.38%)Married, later parenthood (16.50%)Married, early parenthood (15.91%)Panel A: State durationSNC12.915.46.784.338.15.25S1C0.045.890.070.560.020.25C0.680.525.942.041.050.86C1C0.071.170.235.830.050.4M0.180.090.240.124.392.03M1C0.040.290.070.240.283.09M2C00.280.010.210.021.59M3C0.010.150.020.190.010.41MIS0.080.220.640.490.080.12Panel B: Main characteristicsWoman0.420.710.470.650.530.63White0.680.440.770.680.840.83Black0.230.540.170.270.110.12American Indian or Alaskan Native0.010.010.010.020.010.01Asian or Pacific Islander0.080.020.050.030.050.04Neighbourhood disadvantage−0.150.35−0.070.28−0.150.1Average silhouette width0.670.280.310.280.270.1Panel C: Health, education and economic indicators in Wave IVGeneral healtha2.252.602.372.572.172.33At least some college0.750.550.650.430.780.58Total HH incomeb8.326.658.397.069.158.3Personal earnings40562.6324904.4437442.4525295.1341575.3732327.61aHigher values indicate less healthy scores.bBased on 12 income categories.Source: National Longitudinal Study of Adolescent Health (Add Health).1FigureSix‐cluster solution. Source: National Longitudinal Study of Adolescent Health (Add Health).The first profile, Single childless (33.78%), was the most common family life course in the sample aged 15–28. Most respondents in this group remained single without forming any union or childbearing. Women were underrepresented and the cluster had the highest ASW (ASW = 0.67) indicating that the individuals' life courses included in this profile were very homogenous. This is to be expected considering that most respondents in this group did not experience any changes in their family states and remained single with no child during the observation period. The second family trajectory identified, namely Single parent (5.72%) also included single individuals, but respondents in this group differently experienced nonmarital fertility. Moreover, this was the least common pathway and more than 70% of the group members were women. On average, individuals remained single without children for 5.4 years and lived with a child but without a partner for more than 5 years between the ages of 15 and 28.The third cluster group was Cohabitating childless (16.70%). This was the second most common family life course and other pathways apart from the Single childless group, which included more men than women, but the sex differences in this group were negligible. After age 15, individuals in this cluster were likely to remain single and childless for about 7 years and to live with a partner for 6 years without having children. The fourth pathway, Cohabitating parent (11.38%), was characterized by early cohabitation and childbearing. After being single and childless around 4.5 years, individuals entered cohabitation toward the end of their teenage years and had children out of wedlock, on average, within 2 years of starting cohabitation. About two out of three respondents in this group comprised women.The last two pathways were characterized by marriage. The Married, later parenthood (16.50%) was the third most common pathway and sex differences were pronounced less in this group similar to the Cohabitating childless profile. On average, individuals started cohabiting in their mid‐20s after living single for 8 years and cohabitated for a year before marriage. Moreover, most respondents remained childless during the observation period, and about one‐fifth of the respondents in this profile had a child at the end of the age range studied (i.e., age 28). Similar to the other profiles characterized by childbearing, women were overrepresented in the last group, namely Married, early parenthood (15.91%). Overall, respondents remained single for 5.2 years and started living with a partner in their early 20s, and married within a year following the entry into cohabitation. They were also likely to have their first and second children within the second and fifth years after the transition to marriage, respectively. About one‐third of those who became parents early but did not have their second child until age 28 were allocated to this group. Moreover, this profile had the lowest ASW (ASW = 0.10), suggesting that the life courses of the respondents were more heterogeneous in this group.Turning to the average neighbourhood disadvantage scores for each cluster, I found that respondents who belonged to the Single parent profile lived in the most deprived neighbourhoods during their childhood, followed by the Cohabitating parent group. Contrary, childless pathways were associated with lower neighbourhood deprivation.Panel C in Table 1 provides further information on race, health, education, and income indicators measured in Wave IV to gain further insight into the clusters. These indicators include self‐reported health status, educational attainment, total household income and personal earnings. Overall, black respondents were overrepresented more in pathways characterized by childbearing out of wedlock, namely Single parent and Cohabitating parent, while white respondents were overrepresented in pathways characterized by marriage. Those whose life course trajectories were characterized by singlehood fared better than their counterparts who were parents similar to the patterns observed when comparing neighbourhood conditions. Respondents sorted into the Married, later parenthood pathway had, on average, better scores on all indicators than the other profiles followed by the Single childless pathway. In contrast, the Single parenthood profile had the lowest scores on all indicators followed by the Cohabitating parent trajectory. The only exception was that the proportion of respondents who at least had some college education was relatively higher in the former group than in the latter.Access to early adulthood family life coursesNext, I estimated how sorting into these six family trajectories was related to neighbourhood disadvantage using multinomial logit models first without and then with the 2SRI estimation method as shown in Table 2 with the Single childless pathway being the reference group. For the comparability and ease of interpretation of the models, I estimated and presented the marginal effects.2TableMultinomial logit models of neighbourhood effects on family trajectories (marginal effects with standard errors in parentheses)Single parentSingle parent (IV)Cohabitating childlessCohabitating childless (IV)Cohabitating parentCohabitating parent (IV)Married, later parenthoodMarried, later parenthood (IV)Married, early parenthoodMarried, early parenthood (IV)(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)Neighbourhood disadvantage0.014***0.054*−0.004−0.0000.029****0.022−0.015−0.0140.034***0.034(0.004)(0.028)(0.009)(0.042)(0.007)(0.040)(0.010)(0.038)(0.010)(0.051)Woman0.026***0.031**−0.016−0.0160.036***0.034***0.0050.0040.073****0.070****(0.008)(0.012)(0.014)(0.014)(0.011)(0.011)(0.015)(0.015)(0.012)(0.012)Race (Ref: White)Black0.083****0.069****−0.023−0.0280.0100.010−0.084****−0.084****−0.089****−0.089****(0.014)(0.015)(0.017)(0.026)(0.016)(0.027)(0.016)(0.021)(0.016)(0.026)American Indian or Alaskan Native0.0040.0030.1170.1150.0080.008−0.022−0.0210.0410.040(0.025)(0.038)(0.077)(0.074)(0.047)(0.045)(0.074)(0.074)(0.067)(0.065)Asian or Pacific Islander−0.016−0.0200.0180.0210.0210.022−0.056**−0.054**−0.023−0.020(0.012)(0.020)(0.047)(0.048)(0.029)(0.027)(0.022)(0.022)(0.036)(0.036)Immigrant−0.025−0.037−0.036−0.035−0.049**−0.046*0.0250.025−0.007−0.006(0.021)(0.027)(0.041)(0.042)(0.024)(0.024)(0.026)(0.026)(0.026)(0.025)Number of siblings0.0030.003−0.019****−0.019****−0.002−0.002−0.001−0.0020.015****0.015***(0.003)(0.004)(0.005)(0.005)(0.005)(0.005)(0.005)(0.005)(0.004)(0.004)Parental marital status (Ref: Single)Married−0.0000.006−0.021−0.019−0.006−0.0060.0240.0240.118****0.115****(0.010)(0.012)(0.030)(0.031)(0.018)(0.019)(0.035)(0.033)(0.021)(0.021)Widowed−0.012−0.009−0.021−0.018−0.0000.000−0.028−0.0260.109**0.107**(0.015)(0.018)(0.041)(0.041)(0.029)(0.028)(0.047)(0.045)(0.044)(0.044)Separated0.0070.0170.0400.0410.0160.014−0.010−0.0090.072****0.070***(0.011)(0.016)(0.041)(0.041)(0.024)(0.024)(0.036)(0.034)(0.021)(0.021)Parental education (Ref: Less than high school)High school0.0140.028−0.002−0.001−0.014−0.016−0.013−0.013−0.029−0.029(0.014)(0.022)(0.018)(0.023)(0.015)(0.015)(0.018)(0.022)(0.030)(0.034)Some college or trade school−0.0050.007−0.013−0.011−0.031**−0.0320.0140.014−0.004−0.003(0.014)(0.022)(0.025)(0.032)(0.015)(0.021)(0.022)(0.030)(0.029)(0.037)Collage graduate−0.0060.010−0.0010.002−0.081****−0.080****0.033*0.033−0.083**−0.080*(0.014)(0.023)(0.026)(0.033)(0.015)(0.023)(0.019)(0.033)(0.034)(0.042)Parent's number of relationships0.0050.0050.022*0.021*0.025****0.024****−0.022*−0.022*0.0120.012(0.005)(0.006)(0.012)(0.011)(0.007)(0.007)(0.012)(0.012)(0.011)(0.010)Equivalized parental HH income−0.014**−0.0050.027***0.028*−0.028****−0.028*0.024***0.024*−0.016−0.015(0.006)(0.008)(0.010)(0.016)(0.007)(0.014)(0.009)(0.014)(0.011)(0.021)N8763876387638763876387638763876387638763*p < 0.10;**p < 0.05;***p < 0.01;****p < 0.001.Source: National Longitudinal Study of Adolescent Health (Add Health).The hazard of sorting into the Single parent pathway in comparison to the Single childless profile increased in disadvantaged neighbourhoods. This is in line with the theoretical expectations suggesting that individuals from advantaged backgrounds postpone starting a family due to high career or educational expectations (e.g., Furstenberg, 2010). This trajectory was also less accessible for black respondents and women. When employing the instrumental variable strategy in column 2, the neighbourhood effects became less significant. Nevertheless, neighbourhood effects were significant at a 10% level suggesting that above and beyond the selection effects, growing up in a more disadvantaged neighbourhood increases the probability of experiencing family life courses characterized by single parenthood.The Cohabitating childless profile was insignificantly associated with neighbourhood disadvantage as shown in column 3. While sorting into this pathway was less likely for black respondents, the number of parental romantic relationships increased the likelihood of being in this cluster. Findings were not altered notably with the inclusion of the instrumental variable in column 4. The chances of being in the Cohabitating parent profile, however, increased with neighbourhood disadvantage. Moreover, this profile was less accessible to men and immigrants. While parents' educational level was negatively related to being in this group, the number of relationships parents had increased the likelihood of sorting into this group. Effects became insignificant with the instrumental variable strategy in column 6, suggesting that individuals likely to experience this pathway characterized by living with a partner and having children out of wedlock were also sorted into specific neighbourhoods.The probability of being in the Married, later parenthood pathway was negatively correlated with neighbourhood disadvantage in comparison to the Single childless trajectory. Similar to the Cohabitating childless profile, there were no significant sex differences in sorting into this pathway. White respondents and those with parents who have higher levels of education and/or less amount of relationships in the past were more likely to be in this family trajectory characterized by cohabiting for a year in the mid‐20s before marriage, albeit remaining childless at least until age 28. Similar to sorting the Cohabitating parent profile, neighbourhood effects became insignificant in the analysis with the instrumental variable. The hazard of sorting into the last pathway, namely Married, early parenthood, increased in disadvantaged neighbourhoods. Moreover, the likelihood of being in this pathway was lower for black individuals and men. Findings also indicated evidence of intergenerational transmission of family formation behaviour: while having more siblings increased the likelihood of sorting into this pathway, respondents with a single parent were less likely to experience this family trajectory. Having a highly educated parent was also negatively associated with being in this profile. Nevertheless, neighbourhood effects became insignificant by the inclusion of the instrumental variables as shown in column 10. Overall, findings indicate that neighbourhood disadvantage is associated with family trajectories in early adulthood. Estimated effects were also similar in magnitude with the IV approach, but the standard errors were larger in these models. Accordingly, most of these associations became insignificant in further analyses using the IV approach suggesting that the relationship between neighbourhood disadvantage and family trajectories was driven by other factors. Only, sorting into the Single parenthood pathway remained significant after the consideration of the instrumental variable strategy to address the endogeneity of neighbourhood effects.Supplemental analysesI performed several robustness checks to strengthen the confidence in my findings. To test the reliability of my instrumental variable strategy, I employed two sensitivity analyses. In the first analysis, I used an item referring to whether the matched adult from the same neighbourhood lived there because he was born there, using a similar strategy as in the main models. The residential satisfaction model explains that families evaluate their housing and neighbourhood conditions based on their life cycle needs, such as entering marriage and childbearing (Speare, 1974), and use their human capital to move to more desirable neighbourhoods. While high‐income families are more likely to move to higher‐income neighbourhoods, socioeconomically disadvantaged individuals as well as nonwhite groups, particularly African Americans, face structural barriers and are less likely to move into more affluent neighbourhoods (South et al., 2005), suggesting that an important share of residential stability is related to mobility constraints rather than residential satisfaction (Coulton et al., 2012). Moreover, disadvantages among individuals staying in the same neighbourhoods may increase when wealthier residents move away (Sampson & Sharkey, 2008) and evidence further indicates that the cumulative burden of neighbourhood poverty is primarily driven by persistent neighbourhood poverty from birth rather than residential mobility (Timberlake, 2007) and parents' tenure in poor neighbourhoods are positively related to exposure to poor neighbourhoods (Li et al., 2019). Accordingly, I expected an adult neighbour's status of being born in that neighbourhood to be related to neighbourhood deprivation, while not being directly related to the focal person's family history. Indeed, criterion validity (F‐statistic = 202.95) was met and the instrument was significantly and positively associated with neighbourhood advantage. As shown in Panel A of Supporting Information: Table A3 in the Appendix, the results were remarkably similar to the estimates of the main instrumental variables. While neighbourhood disadvantage was significantly and positively associated with sorting into the Single parenthood cluster, it had no significant effect on access to the other profiles. The estimated effects were very similar in magnitude to those of the main models.Second, I used an instrumental variable expected to be correlated with both the main predictor and the outcome. To do this, I used the structural educational opportunity, which refers to the proportion of 12th graders enroled in an academic or college preparatory programme as an IV. The costs of nonnormative transitions such as teenage childbearing might be lower when educational opportunities are low. At the same time, legitimate opportunities might be scarce in poor neighbourhoods (Brewster, 1994). Building on this, I expected structural opportunities to be directly related to neighbourhood disadvantage and family life trajectories, violating the assumptions of the IV strategy. In fact, using an IV, directly related outcome, the estimated effects were very similar in significance and magnitude to the main multinomial logit models using no IVs, as shown in Panel B of Supporting Information: Table A2 in the Appendix. Taken together, these two sensitivity analyses provide further support for my IV estimates.In addition to these sensitivity analyses, I conducted several robustness checks. First, I employed multiple imputations instead of listwise deletion of control variables with missing values (see Supporting Information: Table A4). Second, I focused on family trajectories from ages 15 to 26 as shown in Supporting Information: Table A5. I was able to include two additional cohorts and the sample size increased to 11,584 with this strategy. Third, I focused on family trajectories from ages 15 to 30 and employed a longer time horizon to assess the sensitivity of the main findings (see Supporting Information: Table A6). Consequently, 1257 respondents (14%) were excluded from these analyses. The sequence and cluster analyses revealed a six‐cluster solution as the optimal solution in these specifications, and the determined family life courses were qualitatively similar to the main results. After identifying the clusters, I estimated the probability of sorting into these clusters employing 2SRI estimation methods.Fourth, I replicated the models by excluding respondents who weakly represented the main characteristics of the clusters in which they were located using the information on silhouette values, which range from −1 to 1, and denote the distance from other respondents included in the same profile. Low and negative silhouette scores indicate poorly classified individuals who do not reflect or weakly reflect the main characteristics of the cluster (Kaufman & Rousseeuw, 1990). Accordingly, I ran the main regression models by excluding (i) negative silhouette values (see Supporting Information: Table A7) and (ii) silhouette values smaller than 0.15 (see Supporting Information: Table A8). As shown in Supporting Information: Tables A5–A8, the estimated neighbourhood effects were qualitatively robust to all but one of these specifications. Only in the models in which I examined family trajectories from ages 15 to 26 (see Supporting Information: Table A5) were the neighbourhood effects not significantly associated with the risk of transitioning to the Single parenthood profile, in contrast to the main models. Yet, the p‐value was substantially close to the 0.10 threshold (p = 0.103).Sixth, I controlled for intergenerational closure and social cohesion. Previous literature points to two important channels associated with the second (i.e., collective socialization) and third mechanisms (i.e., institutional model) of neighbourhood effects that explain the relationship between neighbourhood disadvantage and nonnormative outcomes, namely, intergenerational closure (Coleman, 1988) and social cohesion (Sampson et al., 1997). Accordingly, as an additional exercise, I tested whether the observed effects in the main models were sensitive to accounting for these factors as shown in Supporting Information: Table A9. The measures were based on the construction of Harding (2009). Intergenerational closure comprised three items converted to a 5‐point scale: 1. if the respondent saw a neighbour's child getting into trouble, would he/she tell the neighbour; 2. if a neighbour saw the respondent's child getting into trouble, would the neighbour tell the respondent; and 3. the number of parents of the adolescent's friends the parent has talked to in the past 4 weeks. The social cohesion scale referred to what extent neighbourhood residents know and look out for each other. It is based on three true/false measures: 1. ‘You know most people in the neighbourhood’; 2. ‘In the past month, you have stopped on the street to talk with someone who lives in your neighbourhood’; and 3. ‘People in this neighbourhood look out for each other’. The inclusion of these items did not alter the main findings and both intergenerational closure and social cohesion were not significantly associated with accessing early adulthood family life courses.Seventh, I replicated the main multinomial logit models without the IV strategy, controlling for ‘whether the respondent's parents lived in that neighbourhood because they were born there’, to test whether this could explain the differences between the models without IV and IV as shown in Supporting Information: Table A10. Indeed, this item was negatively related to the likelihood of sorting into the Single parenthood trajectory, whereas no significant effects were found for access to the other profiles. Nonetheless, the estimated neighbourhood effects remained noticeably the same with the main non‐IV models. In addition, as Add Health used a school‐based design and the primary sampling frame was derived from high schools, I replicated the non‐IV models first, with multilevel models in which students were nested within high schools, and second, using robust standard errors. The results were also robust to these specifications and are reported in Supporting Information: Tables A11 and A12, respectively.Finally, there is the possibility that the timing of childbearing is not well accounted for and that there are differences in the timing of first birth within clusters. As another sensitivity analysis, I compared the trajectory approach to a traditional approach (e.g., Malmberg & Andersson, 2019) that is expected to be more sensitive to the timing of events using survival models. All multiprocess survival models were fitted using the ‘cmp’ command in Stata (Bartus & Roodman, 2014). These models focused on the relationship between neighbourhood disadvantage and timing of first birth. As shown in Supporting Information: Table A13, I used six different outcome measures in these analyses: (i) transition to parenthood referring to all types of births regardless of the type of union; (ii) transition to childbearing within marriage; (iii) transition to childbearing outside of marriage; (iv) transition to single parenthood; (v) cohabitating childbearing; and (vi) teenage childbearing. All models without the IV strategy showed that the likelihood of experiencing any of these six events examined was shorter in disadvantaged neighbourhoods. While estimated effects were strongest in transition to teenage childbearing followed by childbearing in cohabitation and single parenthood, neighbourhood disadvantage was associated weakest with childbearing in marriage. Nevertheless, the insignificant neighbourhood effects obtained with the IV estimation strategy suggest that these associations are due to selection rather than causal effects. Overall, this suggests that in addition to traditional approaches that focus on single events, examining single family formation events in the context of others offers important insights into understanding the consequences of neighbourhood conditions in childhood for the later life courses. For example, estimates from IV found that neighbourhood disadvantage, while not significantly associated with the transition to single parenthood, increased the risk of experiencing a life course characterized by years of single parenthood in which one has children but lives without a partner.CONCLUSIONAdopting a trajectory‐based approach and considering family life courses as a ‘process outcome’ (Abbott, 2005), this study identified typical early adulthood family trajectories in the United States using Add Health data. Subsequently, these typical family profiles were linked to childhood neighbourhood conditions. To disentangle neighbourhood effects from selection effects that lead to spurious correlations between neighbourhood conditions and family life courses, I utilized an instrumental variable that was directly related to neighbourhood conditions but not to individuals' family life courses.Sequence and cluster analyses revealed six typical family life trajectories between the ages of 15 and 28, characterized primarily by partnership formation and parenthood status. Childless life course trajectories without or with subsequent union formation, including marriage and cohabitation, exhibited lower neighbourhood deprivation on average than the pathways with parenthood. Multinomial logit models showed that sorting into the Married, later parenthood pathway was significantly less likely in comparison to being in the Single childless pathway in deprived neighbourhoods. Contrary, being in the pathways shaped by parenthood was more likely in neighbourhoods with higher levels of deprivation. This is in line with theoretical considerations (Furstenberg, 2010) that suggest that the timing and sequencing of family formation become normative for those from advantaged backgrounds only after certain transitions, such as education and job security, have been completed.Importantly, further analyses using the instrumental variables strategy to test for causal neighbourhood effects found that sorting into the Single parenthood and Married, early parenthood trajectories significantly increased with neighbourhood deprivation. In contrast, significant neighbourhood effects in being in the Cohabitating parents and Married, later parenthood pathways disappeared with the instrumental variable approach, suggesting that individuals traversing these family life courses were sorted into specific neighbourhoods. Taken together, findings indicate that while some neighbourhood effects in nonexperimental studies may be driven by selection bias rather than actual neighbourhood conditions as Harding et al. (2021) suggest, not all effects are necessarily due to the selection. Indeed, findings highlight the importance of considering single events in the context of others and family formation trajectories longitudinally over the life course to understand the ramifications of neighbourhood conditions. For instance, I identified two family life courses that included childbearing out of wedlock. Yet, respondents sorted into the Cohabitating parent profile lived in selective neighbourhoods that had unobserved characteristics that correlated with their subsequent family life, whereas neighbourhood disadvantage increased the odds of being in the Single parenthood pathway, as shown by the instrumental variable approach. This is important given that respondents from the latter group also had the lowest average health and income scores. As a result, disadvantaged neighbourhoods may trigger family life trajectories associated with undesirable outcomes.In addition to identifying causal consequences, it is also important for descriptive research to take a holistic perspective when examining the relationship between neighbourhood conditions and family events. For example, the multinomial logit models showed that neighbourhood poverty was not particularly associated with pathways characterized by childless cohabitation, similar to Zito (2015). Yet, the risk of sorting into the cohabitating pathways with childbearing was significantly associated with neighbourhood disadvantage. Similarly, neighbourhood effects had different consequences for the two identified marriage paths. While early marriage combined with multiple children was more common in disadvantaged neighbourhoods, late marriage and living childless until age 28 were less likely in advantaged neighbourhoods.I conclude with limitations and suggestions for future studies. First, I used the neighbourhood information included in Wave I to measure neighbourhood conditions. A drawback of this strategy is that individuals may move to a different neighbourhood and the duration of exposure to the neighbourhood conditions (Wodtke, 2013) as well as neighbourhood attainment (South et al., 2016) might be influential on individuals' life courses. Yet, information on the duration of exposure to the neighbourhoods identified in Wave I was not available. Moreover, the inclusion of neighbourhood information from other waves could be misleading due to the temporal ordering of the family sequences and determinants utilized because the information from other waves is measured after the beginning of the family trajectories. For these reasons, I used neighbourhood information from Wave I to capture neighbourhood effects.Second, there is a large increase in the standard errors for the neighbourhood effects estimates with the IV approach, suggesting that the instrument is weak. Although the F‐tests and additional sensitivity analyses with different IVs further support my IV estimation strategy, note that the insignificant effects found in the main IV model could be due to a weak IV and these results should be interpreted with caution. Third, one way to explain the observed effects using the instrumental variable approach is that a neighbour's parent is in contact with the focal person because they live in the same neighbourhood. Accordingly, the adult person's work‐related housing preferences could be related to other unobserved characteristics that are likely to influence the focal person's family formation patterns. In this case, the outcome and the instrumental variable employed would be directly associated. Yet, such an effect would still demonstrate the presence of neighbourhood effects, as this is one mechanism, namely collective socialization that explains how neighbourhoods can influence adolescents' family formation trajectories.Fourth, my analyses focused on family trajectories in early adulthood. Thus, I was unable to track the complete fertility and relationship formation patterns of individuals. Future studies could examine the role of neighbourhoods in later adulthood. Likewise, it would be interesting to assess the relevance of neighbourhood conditions in work‐life courses. Lastly, focusing on other single family formation events such as the timing of marriage, or divorcee with a similar instrumental variable strategy may shed light on whether the observed neighbourhood effects found in the previous studies were, indeed, causally linked to the family formation events.ACKNOWLEDGEMENTSThis project is financially supported by the Norface Joint Research Programme on Dynamics of Inequality Across the Life‐course, which is co‐funded by the European Union's Horizon 2020 research and innovation programme under grant agreement No. 724363. I further acknowledge funding from the European Research Council (ERC) under grant agreement no. 848861 (KINMATRIX). Open Access funding enabled and organized by Projekt DEAL.REFERENCESAbbott, A. (1995). Sequence analysis: New methods for old ideas. Annual Review of Sociology, 21(1), 93–113. https://doi.org/10.1146/annurev.so.21.080195.000521Abbott, A. (2005). The idea of outcome in US sociology. In G. Steinmetz (Ed.), The politics of method in the human sciences (pp. 393–426). Duke University Press.Aisenbrey, S., & Fasang, A. E. (2010). New life for old ideas: The “second wave” of sequence analysis bringing the “course” back into the life course. Sociological Methods & Research, 38(3), 420–462. https://doi.org/10.1177/0049124109357532Asch, S. E. (1955). Opinions and social pressure. Scientific American, 193(5), 31–35. https://doi.org/10.1038/scientificamerican1155-31Balbo, N., & Barban, N. (2014). Does fertility behavior spread among friends. American Sociological Review, 79(3), 412–431. https://doi.org/10.1177/0003122414531596Bartus, T., & Roodman, D. (2014). Estimation of multiprocess survival models with cmp. The Stata Journal: Promoting Communications on Statistics and Stata, 14(4), 756–777. https://doi.org/10.1177/1536867X1401400404Baumer, E. P., & South, S. J. (2001). Community effects on youth sexual activity. Journal of Marriage and Family, 63(2), 540–554. https://doi.org/10.1111/j.1741-3737.2001.00540.xBernardi, L., & Klaerner, A. (2014). Social networks and fertility. Demographic Research, 30, 641–670. https://doi.org/10.4054/DemRes.2014.30.22Billari, F. C., Giuntella, O., & Stella, L. (2019). Does broadband Internet affect fertility. Population Studies, 73(3), 297–316. https://doi.org/10.1080/00324728.2019.1584327Bråmå, Å. (2006). “White flight”? The production and reproduction of immigrant concentration areas in Swedish cities, 1990–2000. Urban Studies, 43(7), 1127–1146. https://doi.org/10.1080/00420980500406736Brewster, K. L. (1994). Race differences in sexual activity among adolescent women: The role of neighborhood characteristics. American Sociological Review, 59(3), 408. https://doi.org/10.2307/2095941Brooks‐Gunn, J., Duncan, G. J., & Aber, J. L. (Eds.). (1997). Neighborhood poverty (Vol. 1). Russell Sage Foundation. http://www.jstor.org/stable/10.7758/9781610440844Brooks‐Gunn, J., Duncan, G. J., Klebanov, P. K., & Sealand, N. (1993). Do neighborhoods influence child and adolescent development? American Journal of Sociology, 99(2), 353–395. https://doi.org/10.1086/230268Buyukkececi, Z. (2021). Does re‐partnering behavior spread among former spouses? European Journal of Population, 37(4–5), 799–824. https://doi.org/10.1007/s10680-021-09589-xBuyukkececi, Z., Leopold, T., van Gaalen, R., & Engelhardt, H. (2020). Family, firms, and fertility: A study of social interaction effects. Demography, 57, 243–266. https://doi.org/10.1007/s13524-019-00841-yCarlson, M. J., & Furstenberg, F. F. (2006). The prevalence and correlates of multipartnered fertility among urban U.S. parents. Journal of Marriage and Family, 68(3), 718–732. https://doi.org/10.1111/j.1741-3737.2006.00285.xCarlson, M. J., McLanahan, S., & England, P. (2004). Union formation in fragile families. Demography, 41(2), 237–261. https://doi.org/10.1353/dem.2004.0012Cherlin, A. J. (2010). Demographic trends in the United States: A review of research in the 2000s. Journal of Marriage and Family, 72(3), 403–419. https://doi.org/10.1111/j.1741-3737.2010.00710.xChetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the moving to opportunity experiment. American Economic Review, 106(4), 855–902. https://doi.org/10.1257/aer.20150572Clark, W. A. V., & Huang, Y. (2003). The life course and residential mobility in British housing markets. Environment and Planning A: Economy and Space, 35(2), 323–339. https://doi.org/10.1068/a3542Clark, W. A. V., & Ledwith, V. (2007). How much does income matter in neighborhood choice. Population Research and Policy Review, 26(2), 145–161. https://doi.org/10.1007/s11113-007-9026-9Clarke, P. (2008). When can group level clustering be ignored? Multilevel models versus single‐level models with sparse data. Journal of Epidemiology & Community Health, 62(8), 752–758. https://doi.org/10.1136/jech.2007.060798Clarke, P., & Wheaton, B. (2007). Addressing data sparseness in contextual population research. Sociological Methods & Research, 35(3), 311–351. https://doi.org/10.1177/0049124106292362Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, 95–120. https://doi.org/10.1086/228943Coulton, C., Theodos, B., & Turner, M. A. (2012). Residential mobility and neighborhood change: Real neighborhoods under the microscope. Cityscape (Washington, D.C.), 14(3), 55–89. http://www.jstor.org/stable/41958940Crane, J. (1991). The epidemic theory of ghettos and neighborhood effects on dropping out and teenage childbearing. American Journal of Sociology, 96(5), 1226–1259. https://doi.org/10.1086/229654Desmond, M. (2016). Evicted: Poverty and profit in the American city. Broadway Books.Duncan, G. J., Connell, J. P., & Klebanov, P. K. (1997). Conceptual and methodological issues in estimating causal effects of neighborhoods and family conditions on individual development. In J. Brooks‐Gunn, G. J. Duncan, & L. J. Aber (Eds.), Neighborhood poverty, volume 1: Context and consequences for children (pp. 219–250). Russell Sage Foundation.Earls, F., & Buka, S. L. (1997). Project on human development in Chicago neighborhoods. National Institute of Justice.Elder, G. H. (1985). Life course dynamics: Trajectories and transitions, 1968–1980. Cornell University Press.Elder, G. H., Johnson, M. K., & Crosnoe, R. (2003). The emergence and development of life course theory. In J. T. Mortimer, & M. J. Shanahan (Eds.), Handbooks of sociology and social research (pp. 3–19). Springer. https://doi.org/10.1007/978-0-306-48247-2_1Fasang, A. E., & Liao, T. F. (2014). Visualizing sequences in the social sciences. Sociological Methods & Research, 43(4), 643–676. https://doi.org/10.1177/0049124113506563Feijten, P., & Mulder, C. H. (2002). The timing of household events and housing events in the Netherlands: A longitudinal perspective. Housing Studies, 17(5), 773–792. https://doi.org/10.1080/0267303022000009808Ferraro, K. F., & Shippee, T. P. (2009). Aging and cumulative inequality: How does inequality get under the skin. The Gerontologist, 49(3), 333–343. https://doi.org/10.1093/geront/gnp034Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2), 117–140. https://doi.org/10.1177/001872675400700202Festinger, L., Schachter, S., & Kurt, B. (1950). Social pressures in informal groups: A study of human factors in housing. Harper.Fletcher, J. M. (2012). Peer influences on adolescent alcohol consumption: evidence using an instrumental variables/fixed effect approach. Journal of Population Economics, 25(4), 1265–1286. https://doi.org/10.1007/s00148-011-0365-9Furstenberg, F. F. (2010). On a new schedule: Transitions to adulthood and family change. The Future of Children, 20(1), 67–87. http://www.jstor.org/stable/27795060Gabadinho, A., Ritschard, G., Müller, N. S., & Studer, M. (2011). Analyzing and visualizing state sequences in R with TraMineR. Journal of Statistical Software, 40(4), 1–37. https://doi.org/10.18637/jss.v040.i04Gibson‐Davis, C., Gassman‐Pines, A., & Lehrman, R. (2018). “His” and “Hers”: Meeting the economic bar to marriage. Demography, 55(6), 2321–2343. https://doi.org/10.1007/s13524-018-0726-zGoering, J., Feins, J. D., & Richardson, T. M. (2003). What have we learned about housing mobility and poverty deconcentration. In J. Goering, & J. D. Feins (Eds.), Choosing a better life?: Evaluating the moving to opportunity social experiment (pp. 3–36). Urban Institute Press.Harding, D. J. (2003). Counterfactual models of neighborhood effects: The effect of neighborhood poverty on dropping out and teenage pregnancy. American Journal of Sociology, 109(3), 676–719. https://doi.org/10.1086/379217Harding, D. J. (2009). Violence, older peers, and the socialization of adolescent boys in disadvantaged neighborhoods. American Sociological Review, 74(3), 445–464. https://doi.org/10.1177/000312240907400306Harding, D. J., Sanbonmatsu, L., Duncan, G. J., Gennetian, L. A., Katz, L. F., Kessler, R. C., Kling, J. R., Sciandra, M., & Ludwig, J. (2021, February). Evaluating contradictory experimental and non‐experimental estimates of neighborhood effects on economic outcomes for adults. Working Paper 28454, National Bureau of Economic Research. https://doi.org/10.3386/w28454Harris, K. M. (2013). The add health study: Design and accomplishments. (Vol. 1, pp. 1–22). Carolina Population Center, University of NorthCarolina at Chapel Hill. https://doi.org/10.17615/C6TW87Hedman, L., van Ham, M., & Manley, D. (2011). Neighbourhood choice and neighbourhood reproduction. Environment and Planning A: Economy and Space, 43(6), 1381–1399. https://doi.org/10.1068/a43453Hernán, M. A., & Robins, J. M. (2020). Causal inference: What if. CRC Press.Hinke, S., Leckie, G., & Nicoletti, C. (2019). The use of instrumental variables in peer effects models. Oxford Bulletin of Economics and Statistics, 81(5), 1179–1191. https://doi.org/10.1111/obes.12299Jalovaara, M., & Fasang, A. E. (2020). Family life courses, gender, and mid‐life earnings. European Sociological Review, 36(2), 159–178. https://doi.org/10.1093/esr/jcz057Jencks, C., & Mayer, S. E. (1990). The social consequences of growing up in a poor neighborhood. In L. E. Lynn, Jr., & M. G. H. McGeary (Eds.), Inner‐City Poverty in the United States (pp. 111–186). National Academy Press.Kain, J. F. (1968). Housing segregation, negro employment, and metropolitan decentralization. The Quarterly Journal of Economics, 82(2), 175. https://doi.org/10.2307/1885893Katz, L. F., Kling, J. R., & Liebman, J. B. (2001). Moving to opportunity in Boston: Early results of a randomized mobility experiment. The Quarterly Journal of Economics, 116(2), 607–654. https://doi.org/10.1162/00335530151144113Kaufman, L., & Rousseeuw, P. J. (Eds.). (1990). Finding groups in data. John Wiley & Sons Inc. https://doi.org/10.1002/9780470316801King, G., & Nielsen, R. (2019). Why propensity scores should not be used for matching. Political Analysis, 27(4), 435–454. https://doi.org/10.1017/pan.2019.11Kravdal, Ø. (2003). The problematic estimation of “imitation effects” in multilevel models. Demographic Research, 9, 25–40. https://doi.org/10.4054/DemRes.2003.9.2Kulu, H. (2008). Fertility and spatial mobility in the life course: Evidence from Austria. Environment and Planning A: Economy and Space, 40(3), 632–652. https://doi.org/10.1068/a3914Leventhal, T., & Brooks‐Gunn, J. (2000). The neighborhoods they live in: The effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin, 126(2), 309–337. https://doi.org/10.1037/0033-2909.126.2.309Li, M., Johnson, S. B., Newman, S., & Riley, A. W. (2019). Residential mobility and long‐term exposure to neighborhood poverty among children born in poor families: A U.S. longitudinal cohort study. Social Science & Medicine, 226, 69–76. https://doi.org/10.1016/j.socscimed.2019.02.042Ludwig, J., Duncan, G. J., & Hirschfield, P. (2001). Urban poverty and juvenile crime: Evidence from a randomized housing‐mobility experiment. The Quarterly Journal of Economics, 116(2), 655–679. https://doi.org/10.1162/00335530151144122Ludwig, J., Liebman, J. B., Kling, J. R., Duncan, G. J., Katz, L. F., Kessler, R. C., & Sanbonmatsu, L. (2008). What can we learn about neighborhood effects from the moving to opportunity experiment. American Journal of Sociology, 114(1), 144–188. https://doi.org/10.1086/588741Lundborg, P. (2006). Having the wrong friends? Peer effects in adolescent substance use. Journal of Health Economics, 25(2), 214–233. https://doi.org/10.1016/j.jhealeco.2005.02.001Macindoe, H., & Abbott, A. (2004). Sequence analysis and optimal matching techniques for social science data. In M. Hardy, & A. Bryman (Eds.), Handbook of data analysis (pp. 386–406). SAGE Publications Ltd. https://doi.org/10.4135/9781848608184.n17Malmberg, B., & Andersson, E. K. (2019). Adolescent neighbourhood context and transition to parenthood: A longitudinal study. Population, Space and Place, 25(5), e2228. https://doi.org/10.1002/psp.2228McLanahan, S., & Percheski, C. (2008). Family structure and the reproduction of inequalities. Annual Review of Sociology, 34, 257–276. https://doi.org/10.1146/annurev.soc.34.040507.134549Merton, R. K. (1988). The Matthew Effect in science, II: Cumulative advantage and the symbolism of intellectual property. Isis, 79(4), 606–623. https://doi.org/10.1086/354848Perelli‐Harris, B., & Gassen, N. S. (2012). How similar are cohabitation and marriage? Legal approaches to cohabitation across Western Europe. Population and Development Review, 38(3), 435–467. https://doi.org/10.1111/j.1728-4457.2012.00511.xPerelli‐Harris, B., & Lyons‐Amos, M. (2015). Changes in partnership patterns across the life course. Demographic Research, 33, 145–178. https://doi.org/10.4054/DemRes.2015.33.6Raley, R. K. (2000). Recent trends and differentials in marriage and cohabitation: The United States. In L. J. Waite (Ed.), The ties that bind: Perspectives on marriage and cohabitation (pp. 19–39). Aldine de Gruyter.Richter, L. M. (2006). Studying adolescence. Science, 312(5782), 1902–1905. https://doi.org/10.1126/science.1127489Rindfuss, R. R. (1991). The young adult years: Diversity, structural change, and fertility. Demography, 28(4), 493–512. https://doi.org/10.2307/2061419Rosenbaum, E., & Harris, L. E. (2001). Low‐income families in their new neighborhoods. Journal of Family Issues, 22(2), 183–210. https://doi.org/10.1177/019251301022002004Sampson, R. J., Morenoff, J. D., & Gannon‐Rowley, T. (2002). Assessing “neighborhood effects”: Social processes and new directions in research. Annual Review of Sociology, 28(1), 443–478. https://doi.org/10.1146/annurev.soc.28.110601.141114Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277(5328), 918–924. https://doi.org/10.1126/science.277.5328.918Sampson, R. J., & Sharkey, P. (2008). Neighborhood selection and the social reproduction of concentrated racial inequality. Demography, 45(1), 1–29. https://doi.org/10.1353/dem.2008.0012Sassler, M. S., & Miller, A. (2017). Cohabitation nation: Gender, class, and the remaking of relationships. University of California Press.Sastry, N., Ghosh‐Dastidar, B., Adams, J., & Pebley, A. R. (2006). The design of a multilevel survey of children, families, and communities: The Los Angeles family and neighborhood survey. Social Science Research, 35(4), 1000–1024. https://doi.org/10.1016/j.ssresearch.2005.08.002Schunck, R. (2016). Cluster size and aggregated level 2 variables in multilevel models: A cautionary note. Methods, Data, Analyses, 10(1), 120. https://doi.org/10.12758/mda.2016.005Smock, P. J. (2000). Cohabitation in the United States: An appraisal of research themes, findings, and implications. Annual Review of Sociology, 26(1), 1–20. https://doi.org/10.1146/annurev.soc.26.1.1Sobel, M. E. (2006). What do randomized studies of housing mobility demonstrate? Journal of the American Statistical Association, 101(476), 1398–1407. https://doi.org/10.1198/016214506000000636South, S. J. (2001). The geographic context of divorce: Do neighborhoods matter. Journal of Marriage and Family, 63(3), 755–766. https://doi.org/10.1111/j.1741-3737.2001.00755.xSouth, S. J., & Baumer, E. P. (2000). Deciphering community and race effects on adolescent premarital childbearing. Social Forces, 78(4), 1379–1407. https://doi.org/10.1093/sf/78.4.1379South, S. J., Crowder, K., & Chavez, E. (2005). Exiting and entering High‐Poverty neighborhoods: Latinos, blacks and Anglos compared. Social Forces, 84(2), 873–900. http://www.jstor.org/stable/3598483South, S. J., & Crowder, K. D. (1999). Neighborhood effects on family formation: Concentrated poverty and beyond. American Sociological Review, 64(1), 113. https://doi.org/10.2307/2657281South, S. J., & Crowder, K. D. (2000). The declining significance of neighborhoods? Marital transitions in community context. Social Forces, 78(3), 1067–1099. https://doi.org/10.1093/sf/78.3.1067South, S. J., & Crowder, K. D. (2010). Neighborhood poverty and nonmarital fertility: Spatial and temporal dimensions. Journal of Marriage and Family, 72(1), 89–104. https://doi.org/10.1111/j.1741-3737.2009.00685.xSouth, S. J., Huang, Y., Spring, A., & Crowder, K. (2016). Neighborhood attainment over the adult life course. American Sociological Review, 81(6), 1276–1304. https://doi.org/10.1177/0003122416673029Speare, A. (1974). Residential satisfaction as an intervening variable in residential mobility. Demography, 11(2), 173–188. https://doi.org/10.2307/2060556Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments. Econometrica, 65(3), 557. https://doi.org/10.2307/2171753Stock, J., & Yogo, M. (2005). Asymptotic distributions of instrumental variables statistics with many instruments. In D. W. K. Andrews (Ed.), Identification and inference for econometric models (pp. 109–120). Cambridge University Press. http://www.economics.harvard.edu/faculty/stock/files/AsymptoticDistrib_Stock%2BYogo.pdfStuder, M. (2013). WeightedCluster library manual: A practical guide to creating typologies of trajectories in the social sciences with R. LIVES Working Paper, 24, 1–32. https://doi.org/10.12682/lives.2296-1658.2013.24Studer, M., & Ritschard, G. (2016). What matters in differences between life trajectories: A comparative review of sequence dissimilarity measures. Journal of the Royal Statistical Society: Series A (Statistics in Society), 179(2), 481–511. https://doi.org/10.1111/rssa.12125Terza, J. V., Basu, A., & Rathouz, P. J. (2008). Two‐stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531–543. https://doi.org/10.1016/j.jhealeco.2007.09.009Timberlake, J. M. (2007). Racial and ethnic inequality in the duration of children's exposure to neighborhood poverty and affluence. Social Problems, 54(3), 319–342. https://doi.org/10.1525/sp.2007.54.3.319vanHam, M., Boschman, S., & Vogel, M. (2018). Incorporating neighborhood choice in a model of neighborhood effects on income. Demography, 55(3), 1069–1090. https://doi.org/10.1007/s13524-018-0672-9van Ham, M., & Manley, D. (2012). Neighbourhood effects research at a crossroads. Ten challenges for future research introduction. Environment and Planning A: Economy and Space, 44(12), 2787–2793. https://doi.org/10.1068/a45439Wilson, W. J. (1987). The truly disadvantaged : The inner city, the underclass, and public policy. University of Chicago Press. https://www.ncjrs.gov/App/Publications/abstract.aspx?ID=127579Wilson, W. J. (1996). When work disappears. Political Science Quarterly, 111(4), 567. https://doi.org/10.2307/2152085Wodtke, G. T. (2013). Duration and timing of exposure to neighborhood poverty and the risk of adolescent parenthood. Demography, 50(5), 1765–1788. https://doi.org/10.1007/s13524-013-0219-zWu, Z., & Schimmele, C. M. (2005). Repartnering after first union disruption. Journal of Marriage and Family, 67(1), 27–36. https://doi.org/10.1111/j.0022-2445.2005.00003.xZito, R. C. (2015). Family structure history and teenage cohabitation. Journal of Family Issues, 36(3), 299–325. https://doi.org/10.1177/0192513X13490933 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Population Space and Place Wiley

Neighbourhood effects on early adulthood family life courses: A trajectory‐based approach

Population Space and Place , Volume 29 (2) – Mar 1, 2023

Loading next page...
 
/lp/wiley/neighbourhood-effects-on-early-adulthood-family-life-courses-a-Y8N870sz8J

References (90)

Publisher
Wiley
Copyright
© 2023 John Wiley & Sons Ltd.
ISSN
1544-8444
eISSN
1544-8452
DOI
10.1002/psp.2609
Publisher site
See Article on Publisher Site

Abstract

INTRODUCTIONGrounded in Wilson's (1987) seminal treatise on The Truly Disadvantaged, neighbourhood effects on family events have received considerable attention. Researchers have focused on the impact of neighbourhoods on various family events, including teenage pregnancy (Baumer & South, 2001; Brooks‐Gunn et al., 1993), cohabitation (Zito, 2015), nonmarital childbearing (South & Crowder, 1999), sexual activity (Brewster, 1994), marriage (South & Crowder, 1999) and divorcee (South, 2001). Most of these studies have shown that neighbourhood characteristics are correlated with family events above and beyond individual and family attributes.Although the literature to date gains crucial insights into the role of neighbourhoods in family life courses, it has notable limitations. First, as implicated in Wilson's theory (1987, p. 56), neighbourhoods are crucial in keeping alive ‘that family stability is the norm, not the exception’. Moreover, the occurrence, timing and sequencing of family events are known to have crucial consequences for the later life course (Jalovaara & Fasang, 2020; Rindfuss, 1991). Focusing on isolated family events, therefore, provides an incomplete representation of the impact of neighbourhood effects on family life courses and deviations from conventional life‐course trajectories because we lose sight of longer trajectories following specific transitions. Childhood conditions and early life events and experiences can shape later life courses and (dis)advantage can accumulate over the life course (Merton, 1988). This is particularly true in American society, where family behaviour varies widely by social background characteristics (Carlson et al., 2004; Cherlin, 2010). For example, childbearing out of wedlock may be followed by single parenthood or normative age marriage may be complemented by stable and more traditional nuclear families. When individuals from advantaged and disadvantaged neighbourhoods are clustered into particular family formation patterns, traditionally applied approaches give an incomplete representation.Another important challenge in testing the effects of neighbourhood conditions on the life course of individuals is the identification problem (van Ham et al., 2018). Although previous literature has found that neighbourhood conditions are correlated with specific family events, it remains unclear to what extent these associations are due to the causal influence of neighbourhoods. The outcome of interest, in my case family life courses, and neighbourhood conditions might be correlated due to the differential selection of adolescents and their parents into specific neighbourhoods (Sampson et al., 2002). Individuals who are likely to experience specific events could also be nonrandomly allocated into neighbourhoods. Consequently, the observed correlations between neighbourhood characteristics and family trajectories could be driven by selection effects rather than the ‘actual’ influence of neighbourhoods.This study aims to fill these gaps in the literature on neighbourhood effects and family formation using four waves of data from the National Longitudinal Study of Adolescent Health (Add Health). Specifically, I examine typical family trajectories in early adulthood between the ages of 15 and 28 in the United States and relate these trajectories to adolescent neighbourhood conditions.I contribute to the literature in two ways. First, beyond focusing on isolated family formation events as is done in previous studies (e.g., South, 2001; South & Crowder, 1999; Zito, 2015), I consider family formation patterns as ‘process outcomes’ (Abbott, 2005). This strategy avoids the ‘short view on analytical scope’ (Elder, 1985, p. 31) and acknowledges the interdependence of multiple family events. Consequently, using sequence and cluster analysis, I test how neighbourhood conditions affect longitudinal family life courses by looking at single family events in the context of others. Second, I utilize an instrumental variable that is directly related to neighbourhood conditions but not to family life courses to improve the identification of causal neighbourhood effects on family trajectories.THEORETICAL BACKGROUNDUntil the mid‐2000s, studies on neighbourhood effects were predominantly conducted using the US context (e.g., Crane, 1991; Sampson et al., 1997; South & Crowder, 2000; Wilson, 1987), whereas there has been an expansion on studies focusing on the ramifications of neighbourhoods in the European context (e.g., Bråmå, 2006; Hedman et al., 2011; van Ham & Manley, 2012). These studies on neighbourhood effects suggest several mechanisms for how neighbourhoods influence the later life course of adolescents. Perhaps the most influential theoretical explanation comes from Wilson (1987, 1996), who proposes three causal mechanisms that link neighbourhood disadvantage to adverse outcomes. The first mechanism is the epidemic or contagion model, which emphasizes the influence of peers in a neighbourhood. Individuals change or adjust their behaviour according to the benchmarks or social norms set by others in a group to gain the approval of or avoid conflict with others in the group (Asch, 1955; Festinger, 1954; Festinger et al., 1950). Evidence, indeed, indicates that adaptation of behaviour is strongly correlated with the number of relevant others already exhibiting that behaviour in the neighbourhood (Crane, 1991) and that friends' and peers' family formation behaviour are associated with each other (Balbo & Barban, 2014).According to this model, the distinctive normative patterns in disadvantaged neighbourhoods fail to encourage conventional life courses (Brewster, 1994). In contrast, neighbourhoods characterized by concentrated poverty and concomitant social isolation may legitimize alternative subcultures in peer groups that promote nonnormative behaviour such as risky sexual activity, adolescent nonmarital childbearing and less attentive contraceptive behaviour (South & Baumer, 2000).The second mechanism, the collective socialization perspective, emphasizes the positive influence of successful adult role models who serve as an important social control function over adolescents and demonstrate the benefits and viability of stable family structures and the expectation of remaining married. Contrary, poor adults in disadvantaged neighbourhoods do not provide such successful role models increasing the risk of nonconventional family formation patterns such as childbearing outside of marriage (South & Crowder, 2010) and family instability (South, 2001). The relevance of role models in the family life course is also emphasized recurrently in the social interaction effects and family formation behaviour literature: individuals learn from other role models through observation, imitation and modelling, which affects their decision‐making process (Bernardi & Klaerner, 2014; Buyukkececi, 2021; Buyukkececi et al., 2020). The learning process can take different forms, ranging from small observations to intensive discussions with role models about the consequences of certain behaviours.The third is the institutional model, which points to the importance of local institutions, resources, services and other organizations. These include educational institutions, teachers and police officers available in a neighbourhood. Legitimate opportunities to attain adult status and financial security may be scarce in poor neighbourhoods (Brewster, 1994). Consequently, the lack of desirable jobs and the weakness of educational institutions may reduce the costs of nonnormative transitions in disadvantaged neighbourhoods (Jencks & Mayer, 1990). Not only the opportunities but also the perceived opportunities of adolescents may have an impact on family formation behaviour. Individuals with low educational and labour market aspirations and low school attachment might be more likely to deviate from normative life course trajectories, while adolescents from advantaged backgrounds might delay entry into adult family statuses because of high educational and career expectations. Furstenberg (2010) explains that the timing and sequencing of transitions in adulthood, namely cohabitation, marriage and childbearing, become normative for youth from advantaged backgrounds only after completing education and securing employment.In line with these mechanisms, social disorganization theory asserts that delinquency based on structural and cultural factors affects the social order within and across communities (Sampson et al., 1997). Respondents who are victims of crime or fear crime in their neighbourhoods are expected to be less connected to and involved in community life. Neighbourhoods with severe poverty and other disadvantages are often associated with various types of social disorder conditions that may increase a lack of social integration, feelings of mistrust and a sense of insecurity, which could subsequently weaken relationship quality and relationship formation behaviours.Along with statistical advances and burgeoning large‐scale data resources (e.g., Earls & Buka, 1997; Sastry et al., 2006) that have enabled the identification of neighbourhood contexts and their characteristics, empirical research has built on these theoretical innovations such as the epidemic, collective socialization and institutional models (Sampson et al., 2002) in the last decades and examined the relationship between neighbourhood effects and youth outcomes. A vast share of these studies has focused on the role of neighbourhood conditions and family formation behaviour. Most findings, indeed, have shown that neighbourhood disadvantage is positively associated with nonnormative family formation events such as teenage childbearing (Baumer & South, 2001; Brooks‐Gunn et al., 1993; Harding, 2003), nonmarital fertility (South & Crowder, 1999, 2010) and divorcee (South, 2001). Moreover, neighbourhood disadvantage hastens the transition to marriage among white individuals, whereas it has the opposite effect on black individuals (South & Crowder, 2000).These studies primarily considered stable marriages as normative behaviour, whereas childbearing out of wedlock and marital disruption were regarded as nonnormative behaviour. Yet, new ways of living together have emerged in the United States as well as in other western countries. Cohabitation has become widespread and normative (Raley, 2000). Although it used to be more common among people with less education and disadvantageous background, there is evidence that the cohabitation gap has narrowed in recent decades among all educational groups (Cherlin, 2010). Indeed, Zito (2015) found no significant association between neighbourhood‐level economic disadvantage and teenage cohabitation using the National Longitudinal Study of Adolescent Health (Add Health) data and focusing on cohorts born after the 1970s.Although previous literature provides a snapshot of the relationship between neighbourhood conditions and family formation, these studies are generally limited to isolated focal transitions such as out‐of‐wedlock births or divorcees. Focusing on specific aspects of family formation and traditionally applied methods such as event history models have their limitations when it comes to examining the full range of neighbourhood effects on family formation behaviour. It has been well acknowledged that the occurrence, timing and sequencing of these transitions and events are interrelated (Rindfuss, 1991; Wilson, 1987). Individual life courses unfold over time and need to be studied over longer periods (Elder et al., 2003). This is particularly relevant in adolescence and early adulthood, a ‘demographically dense phase’ (Rindfuss, 1991, p. 496) in which a multitude of events take place and decisions are made (Richter, 2006). These early life events and experiences, as well as childhood conditions, shape later life, and inequality can accumulate over the life course as postulated by the cumulative inequality theory (Ferraro & Shippee, 2009; Merton, 1988). In American society, it may be even more important to take a longitudinal life course perspective when testing for neighbourhood effects and to conceptualize family events as a ‘process outcome’ (Abbott, 2005) that unfold from adolescence through early adulthood because family formation patterns vary considerably across the social structure. Nonmarital childbearing, single parenthood and family instability, for instance, are more common in socioeconomically disadvantaged families (Carlson & Furstenberg, 2006; Carlson et al., 2004), resulting in more complex and fluid families (Cherlin, 2010). In contrast, individuals from advantaged backgrounds are more likely to postpone marriage and parenthood and have stable marriages (McLanahan & Percheski, 2008). Accordingly, the consequences of childhood neighbourhood conditions may extend beyond single family events to influence later life outcomes (Elder, 1985; Elder et al., 2003), and it is important to understand how childhood neighbourhood conditions are associated with the social pathways of individuals' family lives.This study considers single family events in the context of others from adolescence through early adulthood. I expect that the likelihood of family trajectories characterized by early and possibly out‐of‐wedlock births and less stable relationships increases for adolescents growing up in disadvantaged neighbourhoods. Contrary, I expect that family life courses outlined by late union formation and childbearing combined with stable marriages are more accessible in advantaged neighbourhoods.Methodological issuesAnother important gap of knowledge in the previous literature focusing on neighbourhood effects and family formation concerns neighbourhood selection, which presents noteworthy obstacles to drawing conclusions on the causal role of neighbourhood effects (Sampson et al., 2002; Sobel, 2006). Most individuals do not choose where to live randomly, and sorting into a neighbourhood is considerably structured (van Ham & Manley, 2012) due to various reasons such as affordable rents, discrimination and availability of housing (Desmond, 2016). Accordingly, individuals living in certain neighbourhoods might have (un)observed characteristics that are correlated with the outcome of interest above and beyond the neighbourhood effects. For instance, individuals might not be poor(er) because they live in a neighbourhood with concentrated poverty, but they might live in a poor neighbourhood because the rents are affordable. ‘Choices are restricted by household preferences, resources, and restrictions, but also by constraints imposed by the structure of the housing market. Likely, poor households do not choose to move to poverty neighbourhoods, but move there because they cannot afford to live anywhere else’ (Hedman et al., 2011, p. 1395).To overcome this bias and examine the consequences of neighbourhood conditions for family formation, previous literature recurrently included multiple controls to account for individual‐ and family‐level characteristics and disentangle neighbourhood effects (e.g., Baumer & South, 2001; South & Crowder, 1999, 2010). However, it is highly likely that this strategy does not fully account for the selection factors that lead to spurious correlations between neighbourhood conditions and family events. In another study, Harding (2003) examined how neighbourhood conditions are related to school dropout and teenage pregnancy by utilizing Propensity Score Matching (PSM) and counterfactual models to improve the identification of neighbourhood effects. Yet, King and Nielsen (2019) indicate that this strategy often accomplishes the opposite of the indicated goal and even increases imbalance, inefficiency and bias under certain conditions building on three main arguments: First, PSM aims to mimic randomized experiments, rather than randomized block experiments, and the latter provides much better precision and control against confounding. Second, PSM introduces the ‘propensity score paradox’ due to further trimming of units, which increases the imbalance beyond a certain point, unlike other matching methods. Third, the effect estimate of PSM is more sensitive to the model specification compared to the other matching methods. For these reasons, the authors conclude that this strategy should not be used for causal inference.Another strand of research has attempted to improve the identification of neighbourhood effects by using social experiments such as the Moving to Opportunity (MTO) programme, in which families in high‐poverty neighbourhoods were randomly offered the opportunity to move out of the neighbourhood and live in a low‐poverty area. These studies have focused predominantly on outcomes such as economic well‐being (Katz et al., 2001), crime (Ludwig et al., 2001), and employment (Rosenbaum & Harris, 2001) and have emphasized that the findings of neighbourhood effects in nonexperimental studies are challenged by the results of the MTO experiment. Studies examining the consequences of neighbourhood conditions for family formation using a similar strategy are scarce. To the best of my knowledge, there is only one study by Chetty et al. (2016) that followed up with MTO participants over the long term using IRS data and found that neighbourhood disadvantage was positively associated with single parenthood. In a more recent study, Harding et al. (2021) emphasized that observational and experimental approaches yield different results for MTO adults. To illustrate that, the authors compared experimental and nonexperimental estimates of MTO with a parallel analysis of Panel Study from Income Dynamics (PSID). The results showed similarities between the nonexperimental estimates of MTO and PSID, but differences from the experimental estimates of MTO suggest that the neighbourhood effects on economic outcomes may be due to selection bias, at least for economic outcomes.Although the nonexperimental estimates of MTO and PSID produced relatively similar findings, it should still be noted that an important drawback of MTO is the population to which the experimental estimates can be generalized (Ludwig et al., 2008). The selected group for the programme was defined as the families with children living in public housing projects in poor neighbourhoods in five major cities in the United States. These families had to meet several requirements, such as having paid their rental payments on time and not having a criminal record (Goering et al., 2003). Moreover, only about a quarter of eligible families applied for the programme and randomization occurred among this group. As a result, Ludwig et al. (2008) emphasize that MTO participants' data are meaningful only for this subset of the population, namely individuals living in disadvantaged neighbourhoods that were interested in moving, and sufficiently completed the application process.In the last decades, a growing body of literature has modelled the likelihood of a family moving to a particular type of neighbourhood based on a set of neighbourhood characteristics (Bråmå, 2006; Clark & Ledwith, 2007) to improve the identification of neighbourhood effects. More recently, van Ham et al. (2018) further developed this approach and proposed an empirical framework in which they modelled neighbourhood choice with a conditional logit model and estimated the probability that each individual would choose a particular neighbourhood from a set of 203 neighbourhoods. In the second step, they incorporated the correction components from the first step into a model of neighbourhood effects, which allowed relaxing the selection bias assumption. This strategy accounts for the unobserved characteristics of individuals that affect the likelihood of selecting particular neighbourhoods based on predetermined observable conditions. Similar to studies using the MTO programme, this line of research has investigated economic outcomes or mobility behaviours and suggested that neighbourhood effects on adult outcomes may be driven by selection bias.In summary, causal evidence on neighbourhood effects on family events is limited to the experimental study by Chetty et al. (2016) who focused on single parenthood events in five large US cities, as discussed above. In this study, I test for causal neighbourhood effects with a nonexperimental design by examining a nationally representative sample using Add Health data and looking at single family events in the context of others longitudinally rather than focusing on isolated events. To improve the causal identification of neighbourhood effects, I propose an instrumental variable approach based on an exclusion restriction that assures that the instrumental variable is directly linked to the outcome (i.e., family trajectories) only through the main predictor (i.e., neighbourhood disadvantage).DATAI used four waves of the National Longitudinal Study of Adolescent Health (Add Health), a school‐based, nationally representative longitudinal study of US adolescents who attended grades 7–12 in Wave I (1994–95). The sampling strategy was based on a systematic random sample of high schools and ‘feeder’ schools (see Harris, 2013). The primary sampling frame was all high schools in the United States with a 11th grade and an enrolment of at least 30 students. From these high schools, a random sample of 80 schools was selected. The sample was stratified by ethnicity, region, school type and size, and urbanicity. Together with feeder schools, the final sample included 134 schools, and the number of neighbourhoods represented in the high schools ranged from 3 to 88 in the original sample. The Add Health cohort was followed for four rounds of in‐home interviews (Wave I in 1994–95, Wave II in 1996, Wave III in 2001–02, and Wave IV in 2008–09). The initial sample consists of 20,745 youth who completed the in‐home interview. During the in‐home interview, 85% of the individuals' parents were interviewed, which provided detailed information about the parents (commonly the mother). Only after the collection of Wave IV, the Add Health cohorts become old enough—aging between 26 and 33 years—to provide comprehensive information on early adulthood family formation.For the analyses, Add Health allowed me to combine two types of information: (i) family formation events and trajectories and (ii) neighbourhood and other childhood background characteristics. Each wave provides detailed information on relationships and fertility histories. These include the timing of each pregnancy and live births, the type of relationship with romantic partners as well as the start and end dates of each relationship.Add Health includes individual‐level information as well as school‐ and neighbourhood‐level information with its multilevel data structure. It provides contextual information on structural neighbourhood characteristics at the county, census tract, and census block group summary levels by linking individuals' communities and neighbourhoods to the preexisting US census measures. These indicators include rates, proportions and population measures ranging from the share of families below the poverty level to the percentage of workers in managerial or professional occupations. Moreover, Add Health provides unique identifiers at the neighbourhood level that allow me to determine individuals living in the same neighbourhood. Combining this information with data on family formation trajectories and neighbourhood characteristics is exceptionally suitable to assess neighbourhood effects on family formation trajectories.Sample selectionI focused on family formation trajectories between the ages of 15 and 28. The choice of the upper age limit was determined mainly by data availability. For instance, approximately 16% and 19% of the original sample would be excluded from the analyses if the age range of the family sequences were extended to 29 and 30 years, respectively. However, additional analyses focusing on these age ranges yielded qualitatively similar results (available upon request). In Wave IV, 92% of individuals from the initial sample were located and approximately 80% of these respondents were re‐interviewed resulting in 15,701 individuals. A total of 3310 respondents younger than 28 in Wave IV were excluded from the analysis. Furthermore, individuals with missing information in the parental questionnaire (n = 2025, i.e., 16% of the sample) were excluded. The proportion of individuals with missing information in the parental was very similar to the baseline sample, supporting the overall representativeness of the sample. A total of 178 respondents (2%) with more than five missing family states in their family trajectories between ages 15 and 28 and 609 individuals (6%) with missing information on independent variables or weights were excluded from the sample. Lastly, I was not able to determine the instrumental variable used to test for causal neighbourhood effects for 539 individuals (6%). Consequently, they were excluded from the analyses resulting in 9040 respondents. Moreover, the final analysis sample comprised 132 high schools and 2316 neighbourhoods. An overview and description, as well as missing information on each variable after sample restrictions, are shown in the Supporting Information: Table A1.METHODSMy empirical analyses proceeded in three steps. First, I used sequence (Abbott, 1995) and cluster analyses to identify and compare typical family formation trajectories. Second, I estimated how sorting into a particular trajectory is related to neighbourhood characteristics by employing multinomial logit models. In the final step, I utilized an instrumental variable estimation to determine whether neighbourhood conditions are causally related to respondents' family trajectories.Sequence analysisI used biographical information on relationship and fertility histories from the four waves of Add Health to identify family trajectories from ages 15 to 28. Family sequences were coded based on the most common 8 states that combined cohabitation, marital status, and the number of children that were also meaningful substantively: 1. ‘Single with no child’; 2. ‘Single with children’; 3. ‘Cohabitating with no child’; 4. ‘Cohabitating with children’; 5. ‘Married with no child’; 6. ‘Married with 1 child’; 7. ‘Married with 2 children’; and 8. ‘Married with 3+ children’.1For individuals whose family state information was missing in a given year, I additionally included another state indicating that the family state was missing.‘Single’ states referred to those who were neither cohabiting nor married. Separated or divorced individuals were captured in the sequential order of the states because these individuals became single following a cohabitation or marriage episode. I distinguished between cohabitation and marriage because of the different selections between cohabitation and marriage. Although cohabitation is becoming more widespread in the United States regardless of social background characteristics, there remains a social divide in expectations and behaviours for forming unions, with the most advantaged more likely to marry and the least advantaged more likely to cohabitate(Cherlin, 2010; Gibson‐Davis et al., 2018; Sassler & Miller, 2017). Moreover, cohabiters comprise a highly heterogeneous group. They may regard cohabitation as a precursor or alternative to marriage or singledom (Perelli‐Harris & Gassen, 2012; Smock, 2000). Cohabiters have less commitment, greater freedoms and a higher risk of separation (Perelli‐Harris & Lyons‐Amos, 2015). Contrarily, marriage is still regarded as the ultimate commitment in most western countries (Perelli‐Harris & Gassen, 2012). Considering cohabitation and marriage as separate states in the analyses is further important because the qualitative meaning of marriage may differ between direct marriages and marriages starting with cohabitation (Wu & Schimmele, 2005).After constructing the family states, I used to sequence and cluster analyses to identify early adulthood family formation trajectories. Life‐course research uses sequence analysis to study processes of life course trajectories—in my case family trajectories from ages 15 to 28—that are an ordered collection of states over time. Sequence analysis is usually combined with cluster analysis to identify the most similar sequences and group them into typologies. A common method used for sequence comparison in social sciences is optimal matching (OM; Abbott, 1995), which measures the distance between two sequences by calculating the ‘cost’ of turning one sequence into another (Macindoe & Abbott, 2004). It proceeds with three transformation operations: (i) substitution of one state with another, and (ii) insertion, or (iii) deletion of a state. The cost of each operation is assigned by the researcher, and the distance between two sequences is defined by calculating the minimum cost of turning one sequence into another. In the sequence and cluster analyses, I used OM, a method that remains the most used measure in sequence analysis (Studer & Ritschard, 2016), and specified constant substitution costs of 2 and indel costs of 1 to ensure that both the timing and order of the family states contribute to the calculation of similarity between sequences (Aisenbrey & Fasang, 2010; Macindoe & Abbott, 2004).To identify the appropriate number of groupings to be extracted, I used ward clustering. These results were combined with portioning around medoids, which allowed me to obtain the most discriminant groups in my sample (Studer, 2013). Guided by the average silhouette width (ASW) cut‐off criteria, which is based on comparing average within‐cluster distances and average between‐cluster distances, I retained the six‐cluster solution as the optimal grouping quantitatively with an ASW value of 0.39. The validation strategy shows that observations closer to 1 are well‐clustered, whereas groups become more heterogenous with smaller silhouette widths, and values smaller than 0.25 are not well‐structured (Kaufman & Rousseeuw, 1990). Accordingly, the highest ASW value indicated that the observations are most similar within their profiles and most distinct from the other profiles. The six‐cluster grouping also provided the most meaningful solution substantively and was supported by other cluster cut‐off criteria as shown in Supporting Information: Figure A1. For the sequence and cluster analyses, I used the R packages TramineR, TraMineRExtras, and WeightedCluster (Gabadinho et al., 2011; Studer, 2013). Moreover, sample weights were utilized to determine the representative family life courses in the United States with cluster and sequence analyses.Multinomial logit modelsI estimated the probability of sorting into a particular family typology with multinomial logit models. In the analyses, longitudinal weights adjusted for clustering and stratification, as well as the design type, were used. The main predictor was neighbourhood disadvantage, which was determined based on census blocks. Following South and Crowder (1999) and Baumer and South (2001), I measured neighbourhood disadvantage by a standardized index constructed from six highly correlated neighbourhood characteristics: (a) percentage of households receiving public assistance, (b) percentage of families below poverty level, (c) percentage of workers not in managerial or professional occupations, (d) percentage of persons aged 25 and older without a college education, (e) percentage of families earning less than $50,000, and (f) percentage of unemployed men (Cronbach's α = 0.88). These data come from the 1990 US census included in the first wave of Add Health, which allowed me to measure neighbourhood disadvantage before age 152Only 49 individuals (i.e., 0.67% of the sample) were 15 years old in 1990., which satisfied the temporal ordering of neighbourhood characteristics and outcomes. Only 49 individuals (i.e., 0.67% of the sample) were 15 years old in 1990. Following the determination of typical early adulthood family life courses, the probability of sorting into a particular cluster was estimated with the following equation:1Pr(ai=j)=exp⁡(β0j+β1jNDIi+β2j′Xi′+β3jφi)1+∑j=15exp⁡(β0j+β1jNDIi+β2j′Xi′+β3jφi), $\text{Pr}({a}_{i}=j)=\frac{\text{exp}{\rm{}}({\beta }_{0j+}{\beta }_{1j}{\mathrm{NDI}}_{i}+{\beta }_{2j}^{^{\prime} }{X}_{i}^{^{\prime} }+{\beta }_{3j}{\varphi }_{i})}{1+\sum _{j=1}^{5}\text{exp}{\rm{}}({\beta }_{0j+}{\beta }_{1j}{\mathrm{NDI}}_{i}+{\beta }_{2j}^{^{\prime} }{X}_{i}^{^{\prime} }+{\beta }_{3j}{\varphi }_{i})},$where Pr⁡(ai=j) $\text{Pr}{\rm{}}({a}_{i}=j)$ was the probability of sorting into a specific cluster for individual i. NDIi ${\mathrm{NDI}}_{i}$ was the endogenous neighbourhood disadvantage index and β1j ${\beta }_{1j}$ was the main coefficient of interest. Xi denoted the set of controls shown in Supporting Information: Table A1 that might be related to both neighbourhood disadvantage and family formation behaviour. This set of variables included sex as neighbourhood effects on adolescent outcomes might be different for men and women (see Leventhal & Brooks‐Gunn, 2000; for a review). I further controlled for race and immigrant status, given that there are notable racial differences both in neighbourhood disadvantage and family formation behaviour (e.g., South & Crowder, 1999, 2010). I also included a set of parental background characteristics that were used in previous literature and are likely to be related to both the main predictor and the outcome. These variables were the number of siblings, parental marital status, education and the natural logarithm of the equivalized household income, and parents' total number of relationships up to Wave I.3As parental education partly accounted for parental SES and including parental income would introduce more missing values, parental household income was not included in the main models. Nevertheless, models were replicated by also considering this variable as a robustness check and the findings are reported in Supporting Information: Appendix. Lastly, φi ${\varphi }_{i}$ represented an unobserved individual‐level component, which was normally distributed with mean 0 and variance 1.As discussed earlier, the main challenge in estimating neighbourhood effects is selection bias as a result of selective sorting into neighbourhoods, indicating that the outcome of interest is not independent of selection into neighbourhoods (Brooks‐Gunn et al., 1997; Duncan et al., 1997). One approach to disentangle neighbourhood effects from selection effects is to utilize an instrumental variable approach. Lundborg (2006), for instance, applied a strategy closer in spirit to the identification strategy used in the present study. The author implemented various average classmate characteristics, such as the share of individuals living in a single parent household as instrumental variables. Similarly, Fletcher (2012) used the availability of alcohol in classmates' households to assess peer effects on alcohol consumption.To have compelling have a compelling instrumental variable, the instrument must be (i) correlated with the main predictor (i.e., neighbourhood disadvantage) and (ii) validly excluded from the outcome of interest (i.e., family life courses). I exploited information on neighbourhoods and parental questionnaires, to have a valid instrument satisfying these conditions. Using neighbourhood identifiers from Wave I, each teenager in the sample was randomly linked to another teenager living in the same neighbourhood. Second, I utilized an item from the matched individual's parental questionnaire that indicated whether the assigned teenager's caregiver lived in that neighbourhood because of its proximity to the workplace. The item was based on a question included in the parental questionnaire of Wave I: ‘You live here because this neighbourhood is close to a place where you (or your spouse or partner) work now’ (see item PA28B in the parental questionnaire codebook).4About 37% of the parents interviewed in Wave I mentioned that they live here because the neighbourhood is close to the workplace in the original sample, whereas this number was 39% for the matched adult neighbours included in the analyses as shown in Supporting Information: Table A1.The strategy builds on the spatial mismatch hypothesis first proposed by Kain (1968) who focused on where individuals live and where jobs are located. The hypothesis posits that low‐income individuals living in disadvantaged neighbourhoods are far from job opportunities, whereas individuals from advantaged backgrounds voluntarily choose to live in neighbourhoods that are closer to their jobs or near employment opportunities. Accordingly, my approach is based on the assumption that a neighbour's parent's preference to live in a neighbourhood closer to work is correlated with neighbourhood disadvantage, whereas it is not (directly) related to the family life courses of the focal person. Moreover, as argued by Hernán and Robins (2020), the instrument must be associated with the main predictor, but the causal direction of this association or other factors such as selection is not important as long as the instrument is not related to the outcome of interest and the error term in the main equation. Accordingly, the instrumental variable employed satisfies the exclusion restriction even when selective individuals are sorted into specific neighbourhoods, as a neighbour's parent's work‐related housing preferences are unrelated to an individual's family trajectory.One way this assumption can be violated is if an adult neighbour's occupational housing preferences are directly related to the focal person's family life trajectory. For example, family formation behaviour and housing preferences are important determinants of short‐distance residential moves (e.g., Clark & Huang, 2003; Feijten & Mulder, 2002), and as family size increases, the likelihood of moving long distances for a job decreases (Kulu, 2008). Moreover, recent evidence shows that flexible working schedules or remote work promote work‐life balance and have a positive impact on fertility (Billari et al., 2019). Consequently, an individual's work‐related housing preferences could be related to his or her own family formation behaviour, which in turn could affect the trajectory of the focal person's family life. Even if this is the case, such an effect would be indicative of neighbourhood effects. This is because it would be consistent with the second mechanism of neighbourhood effects, namely the collective socialization perspective, which emphasizes the influence of adult role models in the neighbourhood.Despite these merits of the analytical strategy, I performed two sensitivity analyses with two additional instruments and compared the results with the main models to further test the reliability of my instrumental variable approach. The first additional instrument used was expected to be related to neighbourhood disadvantage and have no direct effect on the outcome. Contrary, the second instrumental variable was expected to be directly related to both the main predictor and the outcome. The analytical strategy, as well as the findings, are explained in detail in the Supplemental Analyses section.According to the rule of thumb proposed by Staiger and Stock (1997) and Stock and Yogo (2005), the F‐statistic for the significance of the instrument in the first stage regression should be greater than 10 to have a compelling instrumental variable that satisfies the criterion validity. Analyses testing for weak instruments revealed that the F‐statistic was 23.34 as shown in Supporting Information: Table A2. Although the estimated R‐squared value (0.0026) value was low suggesting a weak relationship between the main predictor and the outcome, the F‐statistic was above conventional requirements of instrument relevance, and neighbourhood disadvantage was significantly lower when a neighbour was living in that household because of proximity to the workplace. After validating my instrument, I used the two‐stage residual inclusion (2SRI) estimation strategy developed by Terza et al. (2008) to address the endogeneity in the multinomial logit model. In the first stage, the endogenous variable was regressed on the instrumental variable and exogenous regressors. In the second stage, residuals from the first stage regressions as regressors were included in the multinomial logit regression. The first stage was specified with the following equation:2Ni=α+β1Zi+β2′Xi′+φi+εi, ${N}_{i}=\alpha +{\beta }_{1}{Z}_{i}+{\beta }_{2}^{^{\prime} }{X}_{i}^{^{\prime} }+{\varphi }_{i}+{\varepsilon }_{i},$where Zi ${Z}_{i}$ was a dummy indicating whether the matched neighbour's caregiver was living in the neighbourhood because it is close to the place of work. Xi denoted the set of variables included in Equation 1. φi ${\varphi }_{i}$ was an unobserved individual‐level component and εi ${\varepsilon }_{i}$ was the error term, which was normally distributed with mean 0 and variance σ. All models were estimated with the Generalized Structural Equation Modelling (gsem) command of Stata. Given the sampling frame of Add Health, an alternative strategy might be to use a 2‐level multilevel model in which respondents were nested into schools. However, multilevel models were not used in the main analyses for two reasons. First, in the analysis sample, 58 of 132 schools had fewer than 50 observations, and modelling with unbalanced and small group sizes (i.e., spareness) could suffer from several problems, including upwardly biased estimates of fixed‐effects coefficients and convergence problems (Clarke, 2008; Clarke & Wheaton, 2007). For instance, Schunck (2016) estimated that at extreme parsimony (i.e., only 5–10 observations per cluster), the average relative bias exceeded 93%, and the bias was substantial even at moderate cluster sizes (i.e., 20–40 observations per cluster). Second, estimated effects can differ greatly from actual effects when including an average measure on the right‐hand side of the model constructed by aggregation within the data (Kravdal, 2003), and can also lead to inconsistent IV estimates (Hinke et al., 2019). Nevertheless, non‐IV models were replicated using a 2‐level model that allowed for the grouping of individual outcomes within schools, and the results are reported in Supporting Information: Appendix.RESULTSEarly adulthood family life coursesFigure 1 shows the sequence index plots of the six‐cluster solution: 1. Single childless; 2. Single parent; 3. Cohabitating childless; 4. Cohabitating parent; 5. Married, later parenthood; and 6. Married, early parenthood. Each row represents an individual's family trajectories from age 15 to 28. Descriptive information for each cluster, including the average length of time spent in eight specific family states, is presented in Table 1. Moreover, Supporting Information: Figure A2 shows relative frequency plots sorted by silhouette values (Fasang & Liao, 2014). It illustrates 50 sequences of each cluster with the highest silhouette values that most strongly represent the main features of the profiles.1TableDescriptive information on clustersSingle childless (33.78%)Single parent (5.72%)Cohabitating childless (16.70%)Cohabitating parent (11.38%)Married, later parenthood (16.50%)Married, early parenthood (15.91%)Panel A: State durationSNC12.915.46.784.338.15.25S1C0.045.890.070.560.020.25C0.680.525.942.041.050.86C1C0.071.170.235.830.050.4M0.180.090.240.124.392.03M1C0.040.290.070.240.283.09M2C00.280.010.210.021.59M3C0.010.150.020.190.010.41MIS0.080.220.640.490.080.12Panel B: Main characteristicsWoman0.420.710.470.650.530.63White0.680.440.770.680.840.83Black0.230.540.170.270.110.12American Indian or Alaskan Native0.010.010.010.020.010.01Asian or Pacific Islander0.080.020.050.030.050.04Neighbourhood disadvantage−0.150.35−0.070.28−0.150.1Average silhouette width0.670.280.310.280.270.1Panel C: Health, education and economic indicators in Wave IVGeneral healtha2.252.602.372.572.172.33At least some college0.750.550.650.430.780.58Total HH incomeb8.326.658.397.069.158.3Personal earnings40562.6324904.4437442.4525295.1341575.3732327.61aHigher values indicate less healthy scores.bBased on 12 income categories.Source: National Longitudinal Study of Adolescent Health (Add Health).1FigureSix‐cluster solution. Source: National Longitudinal Study of Adolescent Health (Add Health).The first profile, Single childless (33.78%), was the most common family life course in the sample aged 15–28. Most respondents in this group remained single without forming any union or childbearing. Women were underrepresented and the cluster had the highest ASW (ASW = 0.67) indicating that the individuals' life courses included in this profile were very homogenous. This is to be expected considering that most respondents in this group did not experience any changes in their family states and remained single with no child during the observation period. The second family trajectory identified, namely Single parent (5.72%) also included single individuals, but respondents in this group differently experienced nonmarital fertility. Moreover, this was the least common pathway and more than 70% of the group members were women. On average, individuals remained single without children for 5.4 years and lived with a child but without a partner for more than 5 years between the ages of 15 and 28.The third cluster group was Cohabitating childless (16.70%). This was the second most common family life course and other pathways apart from the Single childless group, which included more men than women, but the sex differences in this group were negligible. After age 15, individuals in this cluster were likely to remain single and childless for about 7 years and to live with a partner for 6 years without having children. The fourth pathway, Cohabitating parent (11.38%), was characterized by early cohabitation and childbearing. After being single and childless around 4.5 years, individuals entered cohabitation toward the end of their teenage years and had children out of wedlock, on average, within 2 years of starting cohabitation. About two out of three respondents in this group comprised women.The last two pathways were characterized by marriage. The Married, later parenthood (16.50%) was the third most common pathway and sex differences were pronounced less in this group similar to the Cohabitating childless profile. On average, individuals started cohabiting in their mid‐20s after living single for 8 years and cohabitated for a year before marriage. Moreover, most respondents remained childless during the observation period, and about one‐fifth of the respondents in this profile had a child at the end of the age range studied (i.e., age 28). Similar to the other profiles characterized by childbearing, women were overrepresented in the last group, namely Married, early parenthood (15.91%). Overall, respondents remained single for 5.2 years and started living with a partner in their early 20s, and married within a year following the entry into cohabitation. They were also likely to have their first and second children within the second and fifth years after the transition to marriage, respectively. About one‐third of those who became parents early but did not have their second child until age 28 were allocated to this group. Moreover, this profile had the lowest ASW (ASW = 0.10), suggesting that the life courses of the respondents were more heterogeneous in this group.Turning to the average neighbourhood disadvantage scores for each cluster, I found that respondents who belonged to the Single parent profile lived in the most deprived neighbourhoods during their childhood, followed by the Cohabitating parent group. Contrary, childless pathways were associated with lower neighbourhood deprivation.Panel C in Table 1 provides further information on race, health, education, and income indicators measured in Wave IV to gain further insight into the clusters. These indicators include self‐reported health status, educational attainment, total household income and personal earnings. Overall, black respondents were overrepresented more in pathways characterized by childbearing out of wedlock, namely Single parent and Cohabitating parent, while white respondents were overrepresented in pathways characterized by marriage. Those whose life course trajectories were characterized by singlehood fared better than their counterparts who were parents similar to the patterns observed when comparing neighbourhood conditions. Respondents sorted into the Married, later parenthood pathway had, on average, better scores on all indicators than the other profiles followed by the Single childless pathway. In contrast, the Single parenthood profile had the lowest scores on all indicators followed by the Cohabitating parent trajectory. The only exception was that the proportion of respondents who at least had some college education was relatively higher in the former group than in the latter.Access to early adulthood family life coursesNext, I estimated how sorting into these six family trajectories was related to neighbourhood disadvantage using multinomial logit models first without and then with the 2SRI estimation method as shown in Table 2 with the Single childless pathway being the reference group. For the comparability and ease of interpretation of the models, I estimated and presented the marginal effects.2TableMultinomial logit models of neighbourhood effects on family trajectories (marginal effects with standard errors in parentheses)Single parentSingle parent (IV)Cohabitating childlessCohabitating childless (IV)Cohabitating parentCohabitating parent (IV)Married, later parenthoodMarried, later parenthood (IV)Married, early parenthoodMarried, early parenthood (IV)(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)Neighbourhood disadvantage0.014***0.054*−0.004−0.0000.029****0.022−0.015−0.0140.034***0.034(0.004)(0.028)(0.009)(0.042)(0.007)(0.040)(0.010)(0.038)(0.010)(0.051)Woman0.026***0.031**−0.016−0.0160.036***0.034***0.0050.0040.073****0.070****(0.008)(0.012)(0.014)(0.014)(0.011)(0.011)(0.015)(0.015)(0.012)(0.012)Race (Ref: White)Black0.083****0.069****−0.023−0.0280.0100.010−0.084****−0.084****−0.089****−0.089****(0.014)(0.015)(0.017)(0.026)(0.016)(0.027)(0.016)(0.021)(0.016)(0.026)American Indian or Alaskan Native0.0040.0030.1170.1150.0080.008−0.022−0.0210.0410.040(0.025)(0.038)(0.077)(0.074)(0.047)(0.045)(0.074)(0.074)(0.067)(0.065)Asian or Pacific Islander−0.016−0.0200.0180.0210.0210.022−0.056**−0.054**−0.023−0.020(0.012)(0.020)(0.047)(0.048)(0.029)(0.027)(0.022)(0.022)(0.036)(0.036)Immigrant−0.025−0.037−0.036−0.035−0.049**−0.046*0.0250.025−0.007−0.006(0.021)(0.027)(0.041)(0.042)(0.024)(0.024)(0.026)(0.026)(0.026)(0.025)Number of siblings0.0030.003−0.019****−0.019****−0.002−0.002−0.001−0.0020.015****0.015***(0.003)(0.004)(0.005)(0.005)(0.005)(0.005)(0.005)(0.005)(0.004)(0.004)Parental marital status (Ref: Single)Married−0.0000.006−0.021−0.019−0.006−0.0060.0240.0240.118****0.115****(0.010)(0.012)(0.030)(0.031)(0.018)(0.019)(0.035)(0.033)(0.021)(0.021)Widowed−0.012−0.009−0.021−0.018−0.0000.000−0.028−0.0260.109**0.107**(0.015)(0.018)(0.041)(0.041)(0.029)(0.028)(0.047)(0.045)(0.044)(0.044)Separated0.0070.0170.0400.0410.0160.014−0.010−0.0090.072****0.070***(0.011)(0.016)(0.041)(0.041)(0.024)(0.024)(0.036)(0.034)(0.021)(0.021)Parental education (Ref: Less than high school)High school0.0140.028−0.002−0.001−0.014−0.016−0.013−0.013−0.029−0.029(0.014)(0.022)(0.018)(0.023)(0.015)(0.015)(0.018)(0.022)(0.030)(0.034)Some college or trade school−0.0050.007−0.013−0.011−0.031**−0.0320.0140.014−0.004−0.003(0.014)(0.022)(0.025)(0.032)(0.015)(0.021)(0.022)(0.030)(0.029)(0.037)Collage graduate−0.0060.010−0.0010.002−0.081****−0.080****0.033*0.033−0.083**−0.080*(0.014)(0.023)(0.026)(0.033)(0.015)(0.023)(0.019)(0.033)(0.034)(0.042)Parent's number of relationships0.0050.0050.022*0.021*0.025****0.024****−0.022*−0.022*0.0120.012(0.005)(0.006)(0.012)(0.011)(0.007)(0.007)(0.012)(0.012)(0.011)(0.010)Equivalized parental HH income−0.014**−0.0050.027***0.028*−0.028****−0.028*0.024***0.024*−0.016−0.015(0.006)(0.008)(0.010)(0.016)(0.007)(0.014)(0.009)(0.014)(0.011)(0.021)N8763876387638763876387638763876387638763*p < 0.10;**p < 0.05;***p < 0.01;****p < 0.001.Source: National Longitudinal Study of Adolescent Health (Add Health).The hazard of sorting into the Single parent pathway in comparison to the Single childless profile increased in disadvantaged neighbourhoods. This is in line with the theoretical expectations suggesting that individuals from advantaged backgrounds postpone starting a family due to high career or educational expectations (e.g., Furstenberg, 2010). This trajectory was also less accessible for black respondents and women. When employing the instrumental variable strategy in column 2, the neighbourhood effects became less significant. Nevertheless, neighbourhood effects were significant at a 10% level suggesting that above and beyond the selection effects, growing up in a more disadvantaged neighbourhood increases the probability of experiencing family life courses characterized by single parenthood.The Cohabitating childless profile was insignificantly associated with neighbourhood disadvantage as shown in column 3. While sorting into this pathway was less likely for black respondents, the number of parental romantic relationships increased the likelihood of being in this cluster. Findings were not altered notably with the inclusion of the instrumental variable in column 4. The chances of being in the Cohabitating parent profile, however, increased with neighbourhood disadvantage. Moreover, this profile was less accessible to men and immigrants. While parents' educational level was negatively related to being in this group, the number of relationships parents had increased the likelihood of sorting into this group. Effects became insignificant with the instrumental variable strategy in column 6, suggesting that individuals likely to experience this pathway characterized by living with a partner and having children out of wedlock were also sorted into specific neighbourhoods.The probability of being in the Married, later parenthood pathway was negatively correlated with neighbourhood disadvantage in comparison to the Single childless trajectory. Similar to the Cohabitating childless profile, there were no significant sex differences in sorting into this pathway. White respondents and those with parents who have higher levels of education and/or less amount of relationships in the past were more likely to be in this family trajectory characterized by cohabiting for a year in the mid‐20s before marriage, albeit remaining childless at least until age 28. Similar to sorting the Cohabitating parent profile, neighbourhood effects became insignificant in the analysis with the instrumental variable. The hazard of sorting into the last pathway, namely Married, early parenthood, increased in disadvantaged neighbourhoods. Moreover, the likelihood of being in this pathway was lower for black individuals and men. Findings also indicated evidence of intergenerational transmission of family formation behaviour: while having more siblings increased the likelihood of sorting into this pathway, respondents with a single parent were less likely to experience this family trajectory. Having a highly educated parent was also negatively associated with being in this profile. Nevertheless, neighbourhood effects became insignificant by the inclusion of the instrumental variables as shown in column 10. Overall, findings indicate that neighbourhood disadvantage is associated with family trajectories in early adulthood. Estimated effects were also similar in magnitude with the IV approach, but the standard errors were larger in these models. Accordingly, most of these associations became insignificant in further analyses using the IV approach suggesting that the relationship between neighbourhood disadvantage and family trajectories was driven by other factors. Only, sorting into the Single parenthood pathway remained significant after the consideration of the instrumental variable strategy to address the endogeneity of neighbourhood effects.Supplemental analysesI performed several robustness checks to strengthen the confidence in my findings. To test the reliability of my instrumental variable strategy, I employed two sensitivity analyses. In the first analysis, I used an item referring to whether the matched adult from the same neighbourhood lived there because he was born there, using a similar strategy as in the main models. The residential satisfaction model explains that families evaluate their housing and neighbourhood conditions based on their life cycle needs, such as entering marriage and childbearing (Speare, 1974), and use their human capital to move to more desirable neighbourhoods. While high‐income families are more likely to move to higher‐income neighbourhoods, socioeconomically disadvantaged individuals as well as nonwhite groups, particularly African Americans, face structural barriers and are less likely to move into more affluent neighbourhoods (South et al., 2005), suggesting that an important share of residential stability is related to mobility constraints rather than residential satisfaction (Coulton et al., 2012). Moreover, disadvantages among individuals staying in the same neighbourhoods may increase when wealthier residents move away (Sampson & Sharkey, 2008) and evidence further indicates that the cumulative burden of neighbourhood poverty is primarily driven by persistent neighbourhood poverty from birth rather than residential mobility (Timberlake, 2007) and parents' tenure in poor neighbourhoods are positively related to exposure to poor neighbourhoods (Li et al., 2019). Accordingly, I expected an adult neighbour's status of being born in that neighbourhood to be related to neighbourhood deprivation, while not being directly related to the focal person's family history. Indeed, criterion validity (F‐statistic = 202.95) was met and the instrument was significantly and positively associated with neighbourhood advantage. As shown in Panel A of Supporting Information: Table A3 in the Appendix, the results were remarkably similar to the estimates of the main instrumental variables. While neighbourhood disadvantage was significantly and positively associated with sorting into the Single parenthood cluster, it had no significant effect on access to the other profiles. The estimated effects were very similar in magnitude to those of the main models.Second, I used an instrumental variable expected to be correlated with both the main predictor and the outcome. To do this, I used the structural educational opportunity, which refers to the proportion of 12th graders enroled in an academic or college preparatory programme as an IV. The costs of nonnormative transitions such as teenage childbearing might be lower when educational opportunities are low. At the same time, legitimate opportunities might be scarce in poor neighbourhoods (Brewster, 1994). Building on this, I expected structural opportunities to be directly related to neighbourhood disadvantage and family life trajectories, violating the assumptions of the IV strategy. In fact, using an IV, directly related outcome, the estimated effects were very similar in significance and magnitude to the main multinomial logit models using no IVs, as shown in Panel B of Supporting Information: Table A2 in the Appendix. Taken together, these two sensitivity analyses provide further support for my IV estimates.In addition to these sensitivity analyses, I conducted several robustness checks. First, I employed multiple imputations instead of listwise deletion of control variables with missing values (see Supporting Information: Table A4). Second, I focused on family trajectories from ages 15 to 26 as shown in Supporting Information: Table A5. I was able to include two additional cohorts and the sample size increased to 11,584 with this strategy. Third, I focused on family trajectories from ages 15 to 30 and employed a longer time horizon to assess the sensitivity of the main findings (see Supporting Information: Table A6). Consequently, 1257 respondents (14%) were excluded from these analyses. The sequence and cluster analyses revealed a six‐cluster solution as the optimal solution in these specifications, and the determined family life courses were qualitatively similar to the main results. After identifying the clusters, I estimated the probability of sorting into these clusters employing 2SRI estimation methods.Fourth, I replicated the models by excluding respondents who weakly represented the main characteristics of the clusters in which they were located using the information on silhouette values, which range from −1 to 1, and denote the distance from other respondents included in the same profile. Low and negative silhouette scores indicate poorly classified individuals who do not reflect or weakly reflect the main characteristics of the cluster (Kaufman & Rousseeuw, 1990). Accordingly, I ran the main regression models by excluding (i) negative silhouette values (see Supporting Information: Table A7) and (ii) silhouette values smaller than 0.15 (see Supporting Information: Table A8). As shown in Supporting Information: Tables A5–A8, the estimated neighbourhood effects were qualitatively robust to all but one of these specifications. Only in the models in which I examined family trajectories from ages 15 to 26 (see Supporting Information: Table A5) were the neighbourhood effects not significantly associated with the risk of transitioning to the Single parenthood profile, in contrast to the main models. Yet, the p‐value was substantially close to the 0.10 threshold (p = 0.103).Sixth, I controlled for intergenerational closure and social cohesion. Previous literature points to two important channels associated with the second (i.e., collective socialization) and third mechanisms (i.e., institutional model) of neighbourhood effects that explain the relationship between neighbourhood disadvantage and nonnormative outcomes, namely, intergenerational closure (Coleman, 1988) and social cohesion (Sampson et al., 1997). Accordingly, as an additional exercise, I tested whether the observed effects in the main models were sensitive to accounting for these factors as shown in Supporting Information: Table A9. The measures were based on the construction of Harding (2009). Intergenerational closure comprised three items converted to a 5‐point scale: 1. if the respondent saw a neighbour's child getting into trouble, would he/she tell the neighbour; 2. if a neighbour saw the respondent's child getting into trouble, would the neighbour tell the respondent; and 3. the number of parents of the adolescent's friends the parent has talked to in the past 4 weeks. The social cohesion scale referred to what extent neighbourhood residents know and look out for each other. It is based on three true/false measures: 1. ‘You know most people in the neighbourhood’; 2. ‘In the past month, you have stopped on the street to talk with someone who lives in your neighbourhood’; and 3. ‘People in this neighbourhood look out for each other’. The inclusion of these items did not alter the main findings and both intergenerational closure and social cohesion were not significantly associated with accessing early adulthood family life courses.Seventh, I replicated the main multinomial logit models without the IV strategy, controlling for ‘whether the respondent's parents lived in that neighbourhood because they were born there’, to test whether this could explain the differences between the models without IV and IV as shown in Supporting Information: Table A10. Indeed, this item was negatively related to the likelihood of sorting into the Single parenthood trajectory, whereas no significant effects were found for access to the other profiles. Nonetheless, the estimated neighbourhood effects remained noticeably the same with the main non‐IV models. In addition, as Add Health used a school‐based design and the primary sampling frame was derived from high schools, I replicated the non‐IV models first, with multilevel models in which students were nested within high schools, and second, using robust standard errors. The results were also robust to these specifications and are reported in Supporting Information: Tables A11 and A12, respectively.Finally, there is the possibility that the timing of childbearing is not well accounted for and that there are differences in the timing of first birth within clusters. As another sensitivity analysis, I compared the trajectory approach to a traditional approach (e.g., Malmberg & Andersson, 2019) that is expected to be more sensitive to the timing of events using survival models. All multiprocess survival models were fitted using the ‘cmp’ command in Stata (Bartus & Roodman, 2014). These models focused on the relationship between neighbourhood disadvantage and timing of first birth. As shown in Supporting Information: Table A13, I used six different outcome measures in these analyses: (i) transition to parenthood referring to all types of births regardless of the type of union; (ii) transition to childbearing within marriage; (iii) transition to childbearing outside of marriage; (iv) transition to single parenthood; (v) cohabitating childbearing; and (vi) teenage childbearing. All models without the IV strategy showed that the likelihood of experiencing any of these six events examined was shorter in disadvantaged neighbourhoods. While estimated effects were strongest in transition to teenage childbearing followed by childbearing in cohabitation and single parenthood, neighbourhood disadvantage was associated weakest with childbearing in marriage. Nevertheless, the insignificant neighbourhood effects obtained with the IV estimation strategy suggest that these associations are due to selection rather than causal effects. Overall, this suggests that in addition to traditional approaches that focus on single events, examining single family formation events in the context of others offers important insights into understanding the consequences of neighbourhood conditions in childhood for the later life courses. For example, estimates from IV found that neighbourhood disadvantage, while not significantly associated with the transition to single parenthood, increased the risk of experiencing a life course characterized by years of single parenthood in which one has children but lives without a partner.CONCLUSIONAdopting a trajectory‐based approach and considering family life courses as a ‘process outcome’ (Abbott, 2005), this study identified typical early adulthood family trajectories in the United States using Add Health data. Subsequently, these typical family profiles were linked to childhood neighbourhood conditions. To disentangle neighbourhood effects from selection effects that lead to spurious correlations between neighbourhood conditions and family life courses, I utilized an instrumental variable that was directly related to neighbourhood conditions but not to individuals' family life courses.Sequence and cluster analyses revealed six typical family life trajectories between the ages of 15 and 28, characterized primarily by partnership formation and parenthood status. Childless life course trajectories without or with subsequent union formation, including marriage and cohabitation, exhibited lower neighbourhood deprivation on average than the pathways with parenthood. Multinomial logit models showed that sorting into the Married, later parenthood pathway was significantly less likely in comparison to being in the Single childless pathway in deprived neighbourhoods. Contrary, being in the pathways shaped by parenthood was more likely in neighbourhoods with higher levels of deprivation. This is in line with theoretical considerations (Furstenberg, 2010) that suggest that the timing and sequencing of family formation become normative for those from advantaged backgrounds only after certain transitions, such as education and job security, have been completed.Importantly, further analyses using the instrumental variables strategy to test for causal neighbourhood effects found that sorting into the Single parenthood and Married, early parenthood trajectories significantly increased with neighbourhood deprivation. In contrast, significant neighbourhood effects in being in the Cohabitating parents and Married, later parenthood pathways disappeared with the instrumental variable approach, suggesting that individuals traversing these family life courses were sorted into specific neighbourhoods. Taken together, findings indicate that while some neighbourhood effects in nonexperimental studies may be driven by selection bias rather than actual neighbourhood conditions as Harding et al. (2021) suggest, not all effects are necessarily due to the selection. Indeed, findings highlight the importance of considering single events in the context of others and family formation trajectories longitudinally over the life course to understand the ramifications of neighbourhood conditions. For instance, I identified two family life courses that included childbearing out of wedlock. Yet, respondents sorted into the Cohabitating parent profile lived in selective neighbourhoods that had unobserved characteristics that correlated with their subsequent family life, whereas neighbourhood disadvantage increased the odds of being in the Single parenthood pathway, as shown by the instrumental variable approach. This is important given that respondents from the latter group also had the lowest average health and income scores. As a result, disadvantaged neighbourhoods may trigger family life trajectories associated with undesirable outcomes.In addition to identifying causal consequences, it is also important for descriptive research to take a holistic perspective when examining the relationship between neighbourhood conditions and family events. For example, the multinomial logit models showed that neighbourhood poverty was not particularly associated with pathways characterized by childless cohabitation, similar to Zito (2015). Yet, the risk of sorting into the cohabitating pathways with childbearing was significantly associated with neighbourhood disadvantage. Similarly, neighbourhood effects had different consequences for the two identified marriage paths. While early marriage combined with multiple children was more common in disadvantaged neighbourhoods, late marriage and living childless until age 28 were less likely in advantaged neighbourhoods.I conclude with limitations and suggestions for future studies. First, I used the neighbourhood information included in Wave I to measure neighbourhood conditions. A drawback of this strategy is that individuals may move to a different neighbourhood and the duration of exposure to the neighbourhood conditions (Wodtke, 2013) as well as neighbourhood attainment (South et al., 2016) might be influential on individuals' life courses. Yet, information on the duration of exposure to the neighbourhoods identified in Wave I was not available. Moreover, the inclusion of neighbourhood information from other waves could be misleading due to the temporal ordering of the family sequences and determinants utilized because the information from other waves is measured after the beginning of the family trajectories. For these reasons, I used neighbourhood information from Wave I to capture neighbourhood effects.Second, there is a large increase in the standard errors for the neighbourhood effects estimates with the IV approach, suggesting that the instrument is weak. Although the F‐tests and additional sensitivity analyses with different IVs further support my IV estimation strategy, note that the insignificant effects found in the main IV model could be due to a weak IV and these results should be interpreted with caution. Third, one way to explain the observed effects using the instrumental variable approach is that a neighbour's parent is in contact with the focal person because they live in the same neighbourhood. Accordingly, the adult person's work‐related housing preferences could be related to other unobserved characteristics that are likely to influence the focal person's family formation patterns. In this case, the outcome and the instrumental variable employed would be directly associated. Yet, such an effect would still demonstrate the presence of neighbourhood effects, as this is one mechanism, namely collective socialization that explains how neighbourhoods can influence adolescents' family formation trajectories.Fourth, my analyses focused on family trajectories in early adulthood. Thus, I was unable to track the complete fertility and relationship formation patterns of individuals. Future studies could examine the role of neighbourhoods in later adulthood. Likewise, it would be interesting to assess the relevance of neighbourhood conditions in work‐life courses. Lastly, focusing on other single family formation events such as the timing of marriage, or divorcee with a similar instrumental variable strategy may shed light on whether the observed neighbourhood effects found in the previous studies were, indeed, causally linked to the family formation events.ACKNOWLEDGEMENTSThis project is financially supported by the Norface Joint Research Programme on Dynamics of Inequality Across the Life‐course, which is co‐funded by the European Union's Horizon 2020 research and innovation programme under grant agreement No. 724363. I further acknowledge funding from the European Research Council (ERC) under grant agreement no. 848861 (KINMATRIX). Open Access funding enabled and organized by Projekt DEAL.REFERENCESAbbott, A. (1995). Sequence analysis: New methods for old ideas. Annual Review of Sociology, 21(1), 93–113. https://doi.org/10.1146/annurev.so.21.080195.000521Abbott, A. (2005). The idea of outcome in US sociology. In G. Steinmetz (Ed.), The politics of method in the human sciences (pp. 393–426). Duke University Press.Aisenbrey, S., & Fasang, A. E. (2010). New life for old ideas: The “second wave” of sequence analysis bringing the “course” back into the life course. Sociological Methods & Research, 38(3), 420–462. https://doi.org/10.1177/0049124109357532Asch, S. E. (1955). Opinions and social pressure. Scientific American, 193(5), 31–35. https://doi.org/10.1038/scientificamerican1155-31Balbo, N., & Barban, N. (2014). Does fertility behavior spread among friends. American Sociological Review, 79(3), 412–431. https://doi.org/10.1177/0003122414531596Bartus, T., & Roodman, D. (2014). Estimation of multiprocess survival models with cmp. The Stata Journal: Promoting Communications on Statistics and Stata, 14(4), 756–777. https://doi.org/10.1177/1536867X1401400404Baumer, E. P., & South, S. J. (2001). Community effects on youth sexual activity. Journal of Marriage and Family, 63(2), 540–554. https://doi.org/10.1111/j.1741-3737.2001.00540.xBernardi, L., & Klaerner, A. (2014). Social networks and fertility. Demographic Research, 30, 641–670. https://doi.org/10.4054/DemRes.2014.30.22Billari, F. C., Giuntella, O., & Stella, L. (2019). Does broadband Internet affect fertility. Population Studies, 73(3), 297–316. https://doi.org/10.1080/00324728.2019.1584327Bråmå, Å. (2006). “White flight”? The production and reproduction of immigrant concentration areas in Swedish cities, 1990–2000. Urban Studies, 43(7), 1127–1146. https://doi.org/10.1080/00420980500406736Brewster, K. L. (1994). Race differences in sexual activity among adolescent women: The role of neighborhood characteristics. American Sociological Review, 59(3), 408. https://doi.org/10.2307/2095941Brooks‐Gunn, J., Duncan, G. J., & Aber, J. L. (Eds.). (1997). Neighborhood poverty (Vol. 1). Russell Sage Foundation. http://www.jstor.org/stable/10.7758/9781610440844Brooks‐Gunn, J., Duncan, G. J., Klebanov, P. K., & Sealand, N. (1993). Do neighborhoods influence child and adolescent development? American Journal of Sociology, 99(2), 353–395. https://doi.org/10.1086/230268Buyukkececi, Z. (2021). Does re‐partnering behavior spread among former spouses? European Journal of Population, 37(4–5), 799–824. https://doi.org/10.1007/s10680-021-09589-xBuyukkececi, Z., Leopold, T., van Gaalen, R., & Engelhardt, H. (2020). Family, firms, and fertility: A study of social interaction effects. Demography, 57, 243–266. https://doi.org/10.1007/s13524-019-00841-yCarlson, M. J., & Furstenberg, F. F. (2006). The prevalence and correlates of multipartnered fertility among urban U.S. parents. Journal of Marriage and Family, 68(3), 718–732. https://doi.org/10.1111/j.1741-3737.2006.00285.xCarlson, M. J., McLanahan, S., & England, P. (2004). Union formation in fragile families. Demography, 41(2), 237–261. https://doi.org/10.1353/dem.2004.0012Cherlin, A. J. (2010). Demographic trends in the United States: A review of research in the 2000s. Journal of Marriage and Family, 72(3), 403–419. https://doi.org/10.1111/j.1741-3737.2010.00710.xChetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the moving to opportunity experiment. American Economic Review, 106(4), 855–902. https://doi.org/10.1257/aer.20150572Clark, W. A. V., & Huang, Y. (2003). The life course and residential mobility in British housing markets. Environment and Planning A: Economy and Space, 35(2), 323–339. https://doi.org/10.1068/a3542Clark, W. A. V., & Ledwith, V. (2007). How much does income matter in neighborhood choice. Population Research and Policy Review, 26(2), 145–161. https://doi.org/10.1007/s11113-007-9026-9Clarke, P. (2008). When can group level clustering be ignored? Multilevel models versus single‐level models with sparse data. Journal of Epidemiology & Community Health, 62(8), 752–758. https://doi.org/10.1136/jech.2007.060798Clarke, P., & Wheaton, B. (2007). Addressing data sparseness in contextual population research. Sociological Methods & Research, 35(3), 311–351. https://doi.org/10.1177/0049124106292362Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, 95–120. https://doi.org/10.1086/228943Coulton, C., Theodos, B., & Turner, M. A. (2012). Residential mobility and neighborhood change: Real neighborhoods under the microscope. Cityscape (Washington, D.C.), 14(3), 55–89. http://www.jstor.org/stable/41958940Crane, J. (1991). The epidemic theory of ghettos and neighborhood effects on dropping out and teenage childbearing. American Journal of Sociology, 96(5), 1226–1259. https://doi.org/10.1086/229654Desmond, M. (2016). Evicted: Poverty and profit in the American city. Broadway Books.Duncan, G. J., Connell, J. P., & Klebanov, P. K. (1997). Conceptual and methodological issues in estimating causal effects of neighborhoods and family conditions on individual development. In J. Brooks‐Gunn, G. J. Duncan, & L. J. Aber (Eds.), Neighborhood poverty, volume 1: Context and consequences for children (pp. 219–250). Russell Sage Foundation.Earls, F., & Buka, S. L. (1997). Project on human development in Chicago neighborhoods. National Institute of Justice.Elder, G. H. (1985). Life course dynamics: Trajectories and transitions, 1968–1980. Cornell University Press.Elder, G. H., Johnson, M. K., & Crosnoe, R. (2003). The emergence and development of life course theory. In J. T. Mortimer, & M. J. Shanahan (Eds.), Handbooks of sociology and social research (pp. 3–19). Springer. https://doi.org/10.1007/978-0-306-48247-2_1Fasang, A. E., & Liao, T. F. (2014). Visualizing sequences in the social sciences. Sociological Methods & Research, 43(4), 643–676. https://doi.org/10.1177/0049124113506563Feijten, P., & Mulder, C. H. (2002). The timing of household events and housing events in the Netherlands: A longitudinal perspective. Housing Studies, 17(5), 773–792. https://doi.org/10.1080/0267303022000009808Ferraro, K. F., & Shippee, T. P. (2009). Aging and cumulative inequality: How does inequality get under the skin. The Gerontologist, 49(3), 333–343. https://doi.org/10.1093/geront/gnp034Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2), 117–140. https://doi.org/10.1177/001872675400700202Festinger, L., Schachter, S., & Kurt, B. (1950). Social pressures in informal groups: A study of human factors in housing. Harper.Fletcher, J. M. (2012). Peer influences on adolescent alcohol consumption: evidence using an instrumental variables/fixed effect approach. Journal of Population Economics, 25(4), 1265–1286. https://doi.org/10.1007/s00148-011-0365-9Furstenberg, F. F. (2010). On a new schedule: Transitions to adulthood and family change. The Future of Children, 20(1), 67–87. http://www.jstor.org/stable/27795060Gabadinho, A., Ritschard, G., Müller, N. S., & Studer, M. (2011). Analyzing and visualizing state sequences in R with TraMineR. Journal of Statistical Software, 40(4), 1–37. https://doi.org/10.18637/jss.v040.i04Gibson‐Davis, C., Gassman‐Pines, A., & Lehrman, R. (2018). “His” and “Hers”: Meeting the economic bar to marriage. Demography, 55(6), 2321–2343. https://doi.org/10.1007/s13524-018-0726-zGoering, J., Feins, J. D., & Richardson, T. M. (2003). What have we learned about housing mobility and poverty deconcentration. In J. Goering, & J. D. Feins (Eds.), Choosing a better life?: Evaluating the moving to opportunity social experiment (pp. 3–36). Urban Institute Press.Harding, D. J. (2003). Counterfactual models of neighborhood effects: The effect of neighborhood poverty on dropping out and teenage pregnancy. American Journal of Sociology, 109(3), 676–719. https://doi.org/10.1086/379217Harding, D. J. (2009). Violence, older peers, and the socialization of adolescent boys in disadvantaged neighborhoods. American Sociological Review, 74(3), 445–464. https://doi.org/10.1177/000312240907400306Harding, D. J., Sanbonmatsu, L., Duncan, G. J., Gennetian, L. A., Katz, L. F., Kessler, R. C., Kling, J. R., Sciandra, M., & Ludwig, J. (2021, February). Evaluating contradictory experimental and non‐experimental estimates of neighborhood effects on economic outcomes for adults. Working Paper 28454, National Bureau of Economic Research. https://doi.org/10.3386/w28454Harris, K. M. (2013). The add health study: Design and accomplishments. (Vol. 1, pp. 1–22). Carolina Population Center, University of NorthCarolina at Chapel Hill. https://doi.org/10.17615/C6TW87Hedman, L., van Ham, M., & Manley, D. (2011). Neighbourhood choice and neighbourhood reproduction. Environment and Planning A: Economy and Space, 43(6), 1381–1399. https://doi.org/10.1068/a43453Hernán, M. A., & Robins, J. M. (2020). Causal inference: What if. CRC Press.Hinke, S., Leckie, G., & Nicoletti, C. (2019). The use of instrumental variables in peer effects models. Oxford Bulletin of Economics and Statistics, 81(5), 1179–1191. https://doi.org/10.1111/obes.12299Jalovaara, M., & Fasang, A. E. (2020). Family life courses, gender, and mid‐life earnings. European Sociological Review, 36(2), 159–178. https://doi.org/10.1093/esr/jcz057Jencks, C., & Mayer, S. E. (1990). The social consequences of growing up in a poor neighborhood. In L. E. Lynn, Jr., & M. G. H. McGeary (Eds.), Inner‐City Poverty in the United States (pp. 111–186). National Academy Press.Kain, J. F. (1968). Housing segregation, negro employment, and metropolitan decentralization. The Quarterly Journal of Economics, 82(2), 175. https://doi.org/10.2307/1885893Katz, L. F., Kling, J. R., & Liebman, J. B. (2001). Moving to opportunity in Boston: Early results of a randomized mobility experiment. The Quarterly Journal of Economics, 116(2), 607–654. https://doi.org/10.1162/00335530151144113Kaufman, L., & Rousseeuw, P. J. (Eds.). (1990). Finding groups in data. John Wiley & Sons Inc. https://doi.org/10.1002/9780470316801King, G., & Nielsen, R. (2019). Why propensity scores should not be used for matching. Political Analysis, 27(4), 435–454. https://doi.org/10.1017/pan.2019.11Kravdal, Ø. (2003). The problematic estimation of “imitation effects” in multilevel models. Demographic Research, 9, 25–40. https://doi.org/10.4054/DemRes.2003.9.2Kulu, H. (2008). Fertility and spatial mobility in the life course: Evidence from Austria. Environment and Planning A: Economy and Space, 40(3), 632–652. https://doi.org/10.1068/a3914Leventhal, T., & Brooks‐Gunn, J. (2000). The neighborhoods they live in: The effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin, 126(2), 309–337. https://doi.org/10.1037/0033-2909.126.2.309Li, M., Johnson, S. B., Newman, S., & Riley, A. W. (2019). Residential mobility and long‐term exposure to neighborhood poverty among children born in poor families: A U.S. longitudinal cohort study. Social Science & Medicine, 226, 69–76. https://doi.org/10.1016/j.socscimed.2019.02.042Ludwig, J., Duncan, G. J., & Hirschfield, P. (2001). Urban poverty and juvenile crime: Evidence from a randomized housing‐mobility experiment. The Quarterly Journal of Economics, 116(2), 655–679. https://doi.org/10.1162/00335530151144122Ludwig, J., Liebman, J. B., Kling, J. R., Duncan, G. J., Katz, L. F., Kessler, R. C., & Sanbonmatsu, L. (2008). What can we learn about neighborhood effects from the moving to opportunity experiment. American Journal of Sociology, 114(1), 144–188. https://doi.org/10.1086/588741Lundborg, P. (2006). Having the wrong friends? Peer effects in adolescent substance use. Journal of Health Economics, 25(2), 214–233. https://doi.org/10.1016/j.jhealeco.2005.02.001Macindoe, H., & Abbott, A. (2004). Sequence analysis and optimal matching techniques for social science data. In M. Hardy, & A. Bryman (Eds.), Handbook of data analysis (pp. 386–406). SAGE Publications Ltd. https://doi.org/10.4135/9781848608184.n17Malmberg, B., & Andersson, E. K. (2019). Adolescent neighbourhood context and transition to parenthood: A longitudinal study. Population, Space and Place, 25(5), e2228. https://doi.org/10.1002/psp.2228McLanahan, S., & Percheski, C. (2008). Family structure and the reproduction of inequalities. Annual Review of Sociology, 34, 257–276. https://doi.org/10.1146/annurev.soc.34.040507.134549Merton, R. K. (1988). The Matthew Effect in science, II: Cumulative advantage and the symbolism of intellectual property. Isis, 79(4), 606–623. https://doi.org/10.1086/354848Perelli‐Harris, B., & Gassen, N. S. (2012). How similar are cohabitation and marriage? Legal approaches to cohabitation across Western Europe. Population and Development Review, 38(3), 435–467. https://doi.org/10.1111/j.1728-4457.2012.00511.xPerelli‐Harris, B., & Lyons‐Amos, M. (2015). Changes in partnership patterns across the life course. Demographic Research, 33, 145–178. https://doi.org/10.4054/DemRes.2015.33.6Raley, R. K. (2000). Recent trends and differentials in marriage and cohabitation: The United States. In L. J. Waite (Ed.), The ties that bind: Perspectives on marriage and cohabitation (pp. 19–39). Aldine de Gruyter.Richter, L. M. (2006). Studying adolescence. Science, 312(5782), 1902–1905. https://doi.org/10.1126/science.1127489Rindfuss, R. R. (1991). The young adult years: Diversity, structural change, and fertility. Demography, 28(4), 493–512. https://doi.org/10.2307/2061419Rosenbaum, E., & Harris, L. E. (2001). Low‐income families in their new neighborhoods. Journal of Family Issues, 22(2), 183–210. https://doi.org/10.1177/019251301022002004Sampson, R. J., Morenoff, J. D., & Gannon‐Rowley, T. (2002). Assessing “neighborhood effects”: Social processes and new directions in research. Annual Review of Sociology, 28(1), 443–478. https://doi.org/10.1146/annurev.soc.28.110601.141114Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277(5328), 918–924. https://doi.org/10.1126/science.277.5328.918Sampson, R. J., & Sharkey, P. (2008). Neighborhood selection and the social reproduction of concentrated racial inequality. Demography, 45(1), 1–29. https://doi.org/10.1353/dem.2008.0012Sassler, M. S., & Miller, A. (2017). Cohabitation nation: Gender, class, and the remaking of relationships. University of California Press.Sastry, N., Ghosh‐Dastidar, B., Adams, J., & Pebley, A. R. (2006). The design of a multilevel survey of children, families, and communities: The Los Angeles family and neighborhood survey. Social Science Research, 35(4), 1000–1024. https://doi.org/10.1016/j.ssresearch.2005.08.002Schunck, R. (2016). Cluster size and aggregated level 2 variables in multilevel models: A cautionary note. Methods, Data, Analyses, 10(1), 120. https://doi.org/10.12758/mda.2016.005Smock, P. J. (2000). Cohabitation in the United States: An appraisal of research themes, findings, and implications. Annual Review of Sociology, 26(1), 1–20. https://doi.org/10.1146/annurev.soc.26.1.1Sobel, M. E. (2006). What do randomized studies of housing mobility demonstrate? Journal of the American Statistical Association, 101(476), 1398–1407. https://doi.org/10.1198/016214506000000636South, S. J. (2001). The geographic context of divorce: Do neighborhoods matter. Journal of Marriage and Family, 63(3), 755–766. https://doi.org/10.1111/j.1741-3737.2001.00755.xSouth, S. J., & Baumer, E. P. (2000). Deciphering community and race effects on adolescent premarital childbearing. Social Forces, 78(4), 1379–1407. https://doi.org/10.1093/sf/78.4.1379South, S. J., Crowder, K., & Chavez, E. (2005). Exiting and entering High‐Poverty neighborhoods: Latinos, blacks and Anglos compared. Social Forces, 84(2), 873–900. http://www.jstor.org/stable/3598483South, S. J., & Crowder, K. D. (1999). Neighborhood effects on family formation: Concentrated poverty and beyond. American Sociological Review, 64(1), 113. https://doi.org/10.2307/2657281South, S. J., & Crowder, K. D. (2000). The declining significance of neighborhoods? Marital transitions in community context. Social Forces, 78(3), 1067–1099. https://doi.org/10.1093/sf/78.3.1067South, S. J., & Crowder, K. D. (2010). Neighborhood poverty and nonmarital fertility: Spatial and temporal dimensions. Journal of Marriage and Family, 72(1), 89–104. https://doi.org/10.1111/j.1741-3737.2009.00685.xSouth, S. J., Huang, Y., Spring, A., & Crowder, K. (2016). Neighborhood attainment over the adult life course. American Sociological Review, 81(6), 1276–1304. https://doi.org/10.1177/0003122416673029Speare, A. (1974). Residential satisfaction as an intervening variable in residential mobility. Demography, 11(2), 173–188. https://doi.org/10.2307/2060556Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments. Econometrica, 65(3), 557. https://doi.org/10.2307/2171753Stock, J., & Yogo, M. (2005). Asymptotic distributions of instrumental variables statistics with many instruments. In D. W. K. Andrews (Ed.), Identification and inference for econometric models (pp. 109–120). Cambridge University Press. http://www.economics.harvard.edu/faculty/stock/files/AsymptoticDistrib_Stock%2BYogo.pdfStuder, M. (2013). WeightedCluster library manual: A practical guide to creating typologies of trajectories in the social sciences with R. LIVES Working Paper, 24, 1–32. https://doi.org/10.12682/lives.2296-1658.2013.24Studer, M., & Ritschard, G. (2016). What matters in differences between life trajectories: A comparative review of sequence dissimilarity measures. Journal of the Royal Statistical Society: Series A (Statistics in Society), 179(2), 481–511. https://doi.org/10.1111/rssa.12125Terza, J. V., Basu, A., & Rathouz, P. J. (2008). Two‐stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531–543. https://doi.org/10.1016/j.jhealeco.2007.09.009Timberlake, J. M. (2007). Racial and ethnic inequality in the duration of children's exposure to neighborhood poverty and affluence. Social Problems, 54(3), 319–342. https://doi.org/10.1525/sp.2007.54.3.319vanHam, M., Boschman, S., & Vogel, M. (2018). Incorporating neighborhood choice in a model of neighborhood effects on income. Demography, 55(3), 1069–1090. https://doi.org/10.1007/s13524-018-0672-9van Ham, M., & Manley, D. (2012). Neighbourhood effects research at a crossroads. Ten challenges for future research introduction. Environment and Planning A: Economy and Space, 44(12), 2787–2793. https://doi.org/10.1068/a45439Wilson, W. J. (1987). The truly disadvantaged : The inner city, the underclass, and public policy. University of Chicago Press. https://www.ncjrs.gov/App/Publications/abstract.aspx?ID=127579Wilson, W. J. (1996). When work disappears. Political Science Quarterly, 111(4), 567. https://doi.org/10.2307/2152085Wodtke, G. T. (2013). Duration and timing of exposure to neighborhood poverty and the risk of adolescent parenthood. Demography, 50(5), 1765–1788. https://doi.org/10.1007/s13524-013-0219-zWu, Z., & Schimmele, C. M. (2005). Repartnering after first union disruption. Journal of Marriage and Family, 67(1), 27–36. https://doi.org/10.1111/j.0022-2445.2005.00003.xZito, R. C. (2015). Family structure history and teenage cohabitation. Journal of Family Issues, 36(3), 299–325. https://doi.org/10.1177/0192513X13490933

Journal

Population Space and PlaceWiley

Published: Mar 1, 2023

Keywords: family formation; instrumental variable; neighbourhood effects; sequence analysis

There are no references for this article.