Heterogeneity in Family Life Course Patterns and Intra-Cohort Wealth Disparities in Late Working Age

Considering soaring wealth inequalities in older age, this research addresses the relationship between family life courses and widening wealth differences between individuals as they age. We holistically examine how childbearing and marital histories are associated with personal wealth at ages 50–59 for Western Germans born between 1943 and 1967. We propose that deviations from culturally and institutionally-supported family patterns, or the stratified access to them, associate with differential wealth accumulation over time and can explain wealth inequalities at older ages. Using longitudinal data from the German Socio-Economic Panel Study (SOEP, v34, waves 2002–2017), we first identified typical family trajectory patterns between ages 16 and 50 with multichannel sequence analysis and cluster analysis. We then modelled personal wealth ranks at ages 50–59 as a function of family patterns. Results showed that deviations from the standard family pattern (i.e. stable marriage with, on average, two children) were mostly associated with lower wealth ranks at older age, controlling for childhood characteristics that partly predict selection into family patterns and baseline wealth. We found higher wealth penalties for greater deviation and lower penalties for moderate deviation from the standard family pattern. Addressing entire family trajectories, our research extended and nuanced our knowledge of the role of earlier family behaviour for later economic wellbeing. By using personal-level rather than household-level wealth data, we were able to identify substantial gender differences in the study associations. Our research also recognised the importance of combining marital and childbearing histories to assess wealth inequalities. Supplementary Information The online version contains supplementary material available at 10.1007/s10680-021-09601-4.

property and business assets. As a result, only 41 percent of GDR housing property was privately owned by the end of the 1980s compared to 91 percent in the FGR. Due to the lack of investment options in the GDR, most private wealth was therefore accumulated in savings accounts. At reunification, expropriated property was returned to their rightful owners, GDR Mark converted favourable to Deutsche Mark and large economic incentives were put in place to aid former GDR residents in their wealth accumulation (Hauser et al., 1996). Nevertheless, even decades after reunification wealth disparities between eastern and western Germany remain substantial particularly amongst the older population (Grabka & Westermeier, 2014).

S.2 Additional description of data and method S.2.1 SOEP retrospective family histories data
The SOEP provides retrospective marital and fertility information within the datasets BIOMARSY 1 and BIOBIRTH. Retrospective information is collected using a biographical questionnaire, which is administered once within one of the first years after panel entry. An exception is men's fertility history data, which have only been collected for men who entered the SOEP in 2000 or later. Retrospective datasets are updated annually using information provided within the personal questionnaire regarding the current family status and family events that may have occurred since 1st January of the previous year. Detailed information on retrospective data is available in Goebel (2017).

S.2.2 Additional information on personal-level SOEP wealth data
Whereas other panel studies commonly measure wealth at the household level and one household member provides information on the financial standing of the entire household, within the SOEP, wealth information is measured at the individual level. This means that each household member over 16 years of age is questioned about their personal and potentially shared assets and liabilities. The SOEP is thus currently the only household panel study that provides comprehensive personal-level wealth measures over four waves.
Wealth data collection follows several steps. First, a filter question (yes/no) is asked to assess whether the respondent personally holds a certain type of assets or liability. Second, if respondents answer in the affirmative, they are asked to provide the total value. Third, a second filter question (yes/no) is posed to assess whether those assets and liabilities are held jointly. This is only done for wealth components that can theoretically be owned jointly (e.g. housing equity). Fourth, if respondents affirm joint ownership, they are asked about their personal share in percentage points.
Using the total metric value of the wealth component and personal share, the SOEP team calculates the value of personally owned assets and liabilities. Based on all household members' personal wealth, the SOEP team further aggregates personal-level wealth to the householdlevel, so that SOEP users are provided with both personal-level and household-level wealth measurements (Grabka & Westermeier, 2015).
As previous research has almost exclusively relied on household-level wealth data in the analysis of wealth at older ages, we re-run our analyses using total per capita net wealth. To generate this measure, we use household-level wealth data, which in the SOEP is personal-level wealth aggregated to the household. We divide household-level wealth by the number of adults living in the household to obtain the per capita measure. Results of this supplementary analysis are provided in Figure S.1. and S.2. in this supplementary material. Although the general directions of the association of interest are in line with our main results, due to the nature of the measure and the neglect of within-couple wealth differences, gender differences are substantially reduced for the per capita measure.

S.2.3 Confounders and additional measures used within the descriptive and multivariate analyses
A range of baseline confounders are included as control variables in the regression analyses, as they partially predict both selection into certain family pathways and base-level wealth. These include: a dummy for migration background to indicate whether respondents or their parents had immigrated to Germany; a categorical measure of the number of siblings (none (ref.), 1 sibling, 2 siblings, 3 or more siblings); a continuous measure of parents' occupational status defined by the Standard International Occupational Prestige Scale (SIOPS); and a categorical measure of parents' highest education level (low (ref.), intermediate, high). Additionally, the regression models control for respondents' age as a continuous measure to capture maturation effects and account for age related wealth differences within our sample; respondents' birth cohorts (1943-1950 (ref.), 1951-1958, 1959-1967) to consider cohort effects; and marital status changes between ages 50 and 59 (depending on age at last observation) by including three dummy variables that capture the entry into marriage, or marital dissolution either through separation and divorce or through widowhood.
While the present paper does not aim to explain the specific mechanisms of wealth accumulation associated with different family trajectories, we partially address the resource accumulation potential of major family trajectories within our descriptive analyses. For this, we use the following human capital trajectory measures separately for men and women: respondents' highest level of education (low, intermediate, high), number of years of employment, number of unemployment episodes, and the mode of the Standard International Occupational Prestige Scale (SIOPS) score.

S.2.4 Justifications for the use of a multi-channel sequence analysis
Sequence analysis applications examining interdependencies across (multiple) life domains have typically relied on two strategies: combining states across sequences of two or more domains, or performing domain-specific sequence analysis and averaging the resulting pairwise dissimilarities. Instead, we chose MCSA because it enables us to consider all possible interactions across domain-specific states (i.e. using a full set of combined states was not feasible given sample limitations) and properly acknowledging relevant cross-domain interdependence (often not accomplished when averaging domain-specific distances).
Key requirements for MCSA applications are that the study domains should be interdependent, and the domain-based dissimilarities should be associated with MCSA dissimilarities. Deploying the approach proposed by Piccarreta (2017)

S.2.5 Assessment of cluster quality of 11-cluster solution
The overall average silhouette width (ASW) of the 11-cluster solution is 0.34, which indicates that the homogeneity of the clusters is moderate. It is worth noting that the two remarriage clusters (i.e. remarriage with low fertility and remarriage with high complexity) display ASW below .20 suggesting low within-cluster homogeneity. This was expected since these clusters combine sets of complex sequences that vary on the timing and duration of state episodes, but there are relevant common sequencings, including union dissolution, repartnering and multiple fertility episodes. See sequence-specific and cluster-specific ASW as well as cluster-specific sample distributions in Figure S.4. and Table S.2., respectively.     (2002,2007,2012,2017); non-imputed, unweighted. Notes: Data are from Socio-Economic Panel Survey v34 (2002,2007,2012,2017); imputed, unweighted. * p<.05, ** p<.01, *** p<.001