Family structure and the economic wellbeing of children in youth and adulthood.

Social Science Research 43 (2014) 30–44

Contents lists available at ScienceDirect

Social Science Research journal homepage: www.elsevier.com/locate/ssresearch

Family structure and the economic wellbeing of children in youth and adulthood q Leonard M. Lopoo a,⇑, Thomas DeLeire b,c a Department of Public Administration and International Affairs, Center for Policy Research, Maxwell School, Syracuse University, 426 Eggers Hall, Syracuse, NY 13244, United States b Georgetown Public Policy Institute, Georgetown University, United States c NBER, United States

a r t i c l e

i n f o

Article history: Received 30 November 2012 Revised 12 June 2013 Accepted 19 August 2013 Available online 30 August 2013 Keywords: Family structure Economic wellbeing

a b s t r a c t An extensive literature on the relationship between family structure and children’s outcomes consistently shows that living with a single parent is associated with negative outcomes. Few US studies, however, examine how a child’s family structure affects outcomes for the child once he/she reaches adulthood. We directly examine, using the Panel Study of Income Dynamics, whether family structure during childhood is related to the child’s economic wellbeing both during childhood as well as during adulthood. We find that living with a single parent is associated with the level of family resources available during childhood. This finding persists even when we remove time invariant factors within families. We also show that family structure is related to the child’s education, marital status, and adult family income. Once we control for the child’s demography and economic wellbeing in childhood, however, the associations into adulthood become trivial in size and statistically insignificant, suggesting that the relationship between family structure and children’s long-term, economic outcomes is due in large part to the relationship between family structure and economic wellbeing in childhood. Ó 2013 Elsevier Inc. All rights reserved.

1. Introduction The importance of family structure as it relates to outcomes for children is one of the most frequently investigated topics in the field of social demography. Descriptively, many studies show that children who grow up with married biological parents fare better in nearly all domains, including economically, cognitively, and emotionally, than those with parents who separate or divorce during their childhood or who live with a parent who was never married (Amato, 2000; Bjorkland et al., 2007; Gruber, 2004; McLanahan and Sandefur, 1994; Nock, 2000; Ribar, 2004). These differences persist even after controlling for a variety of demographic and socioeconomic factors (McLanahan and Sandefur, 1994; Ribar, 2004). Most research to date examines the relationship between family structure and outcomes measured during a child’s youth and adolescence, such as educational attainment (e.g., Astone and McLanahan, 1991, 1994; Biblarz and Raftery, 1999; McLanahan and Sandefur, 1994; Thompson et al., 1994), cognitive test scores (Gennetian, 2005) and psychological distress and smoking (Ermisch and Francesconi, 2001). Many of these outcomes are chosen because they are considered indicators of

q We are grateful for useful comments made by participants at the 2012 Institute for Research on Poverty Summer Workshop at the University of Wisconsin–Madison. We thank Noah Natzke for excellent research assistance. ⇑ Corresponding author. Fax: +1 315 443 1081. E-mail address: [email protected] (L.M. Lopoo).

0049-089X/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ssresearch.2013.08.004

L.M. Lopoo, T. DeLeire / Social Science Research 43 (2014) 30–44

31

the child’s success as an adult. Few US studies directly examine the relationship between family structure during a child’s youth and outcomes for that child once he/she reaches adulthood (Fronstin et al., 2001). Those that do frequently make use of retrospective data (e.g., Biblarz and Gottainer, 2000), have missing information on important potential mechanisms such as family income (e.g., Lang and Zagorsky, 2001), or focus on educational outcomes alone (e.g., Case et al., 2001; Francesconi et al., 2010; Ginther and Pollak, 2004). This paper adds to the extant literature on family structure and children’s outcomes in the US by addressing some of the potential weaknesses in the prior literature. First, we examine the links between family structure during a child’s youth and the child’s economic wellbeing both in childhood and in adulthood. To do so, we use a data source, the Panel Study of Income Dynamics (PSID), that allows us to tie the resources available to a child while growing up to the economic success of the same child once he/she becomes an adult. These data were not collected retrospectively, but, instead, were gathered over the course of the more than 40 years the PSID has been in existence. Second, we have information on the child’s educational attainment, marital status, and total family income in adulthood, factors that have never been considered together within the same study. We use them to identify the mechanisms that explain the relationship between family structure and adult economic wellbeing. Third, we have information on the maternal family structure for the first 18 years of a child’s life. This is important for two reasons. It allows us to examine a much wider array of family structures – including divorced mothers who remarry and divorced mothers who do not remarry – and to determine if there is heterogeneity in the estimated associations across these diverse structures. Using a wider window also means we are much more likely to classify children as living with a single parent than we would using a single year of data. Much of the previous literature on family structure and children’s outcomes was based on the living arrangements of the child during a particular year of the child’s life. For example, in their classic study, McLanahan and Sandefur (1994) define the family structure of a child based on the mother’s marital status in the child’s 16th year. Wolfe et al. (1996) demonstrate that choosing a single point in time to capture family circumstances will produce misleading results. Our study replicates the literature showing that single parenthood reduces the economic resources available to children while they are growing up. Economic resources, compared to parental behaviors, have been shown to be an important mechanism explaining the relationship between single parenthood and educational outcomes for children (Thompson et al., 1994). Our study suggests that the reductions in economic wellbeing associated with family structure that a child experiences in his/her youth have lasting effects across the life-course affecting educational attainment, marriage, and ultimately one’s total family income in adulthood. In fact, our results show that once we control for the child’s demography and economic wellbeing in childhood, the association between family structure in the child’s youth and the child’s economic wellbeing in adulthood disappear suggesting that the consequences of family structure on long-term outcomes of children are largely the result of economic experiences in childhood. Family structure is not randomly assigned; therefore, it is extraordinarily difficult to establish causal estimates of the impact of single parenthood. Associations may lack causal interpretations due to bias generated by omitted variables (Bjorkland et al., 2007; Bjorklund and Sundstrom, 2006; Francesconi et al., 2010; Ginther and Pollak, 2004; Ribar, 2004). It has been well established that single parenthood is associated with a number of characteristics that might affect family income. If one fails to account for these omitted variables, then estimates of the effect of family structure will also include the indirect effect of family structure that operates through these omitted variables. These problems rightly leave the literature on family structure and children’s outcomes vulnerable to criticisms related to internal validity. Without a controlled experiment that randomly assigns family structure, establishing causal estimates is exceedingly difficult. Our results, like those in the previous literature, are subject to this limitation. We use a difference model to remove any bias created by time-invariant omitted variables correlated with family structure and family income measured when the child was young. Nevertheless, the results for the child’s economic wellbeing in his/her youth are still vulnerable to bias generated by omitted time-varying factors, such as maternal mental and physical health. Furthermore, all of our results on adult outcomes are susceptible to this criticism. Therefore, our results might be described most accurately as ‘‘descriptive regressions;’’ a term Ginther and Pollak (2004) use in their work on this topic. While the estimation challenges that have only been partially addressed in the literature are important to document so as to improve our current understanding of the issue, it is equally important to note that policymakers in the United States continue to place great emphasis on marriage, particularly for the low-income population. One can see this in the Temporary Assistance for Needy Families (TANF) program where the goals include promoting healthy marriages and discouraging out-of-wedlock pregnancies. Arizona, Arkansas, and Louisiana have implemented laws that create a covenant marriage option for couples. Covenant marriages require counseling before a couple can divorce and a no-fault declaration is not an option for dissolution. Thus, while it can be difficult to generate causal estimates of the impact of family structure on children’s outcomes, it remains an important social policy topic and is, therefore, worthy of more scholarly attention.

2. Theory Growing up in a single-parent family can reduce the economic wellbeing of children through a number of channels. Many of the long-term economic effects of family structure operate by reducing the child’s human capital accumulation, which would reduce the child’s productivity as an adult (Becker, 1993). Perhaps the most straightforward channel involves the potential reduction in resources provided by one parent compared with two. Married parents have higher average family

32


incomes than single parents (Duncan and Hoffman, 1985; Fronstin et al., 2001; Ribar, 2004). All else equal, these differences would allow for greater investments in human capital, such as schooling, educational technology, and extracurricular lessons, and should ultimately translate into a higher standard of living for the child (Becker, 1991; Becker and Tomes, 1986). Furthermore, married parents have more time (between the two of them) to spend with their children than single parents, all else equal (Fronstin et al., 2001; Thompson et al., 1994). Economic models of specialization suggest that two parents can capitalize on their comparative advantages and create greater productivity than single parents (Becker, 1991). Two parents co-residing also experience economies of scale. The couple can pool their resources and share the cost of housing, appliances, and many other expenditures that do not need to be duplicated for a family (McLanahan and Sandefur, 1994; Ribar, 2004). Furthermore, should a couple split, women and children tend to receive less than their proportionate share of the couple’s pooled resources (Duncan and Hoffman, 1985; Weitzman, 1985; Smock, 1993). In other words, a mother and a child usually have to live on less than two-thirds of the pooled resources available to a family of three (McLanahan and Sandefur, 1994). This reduction in resources may translate into lower levels of parental investment in the child’s human capital as the custodial parent must split her time between the labor force and home production (Fronstin et al., 2001). Divorce, separation, or the death of a parent can produce inordinate amounts of stress that might harm the child’s economic wellbeing (Amato, 2000; Francesconi et al., 2010; Fronstin et al., 2001). Because a divorce is culmination of a process of marital disintegration that can transpire over many years and produces stress throughout that period, it can affect the emotional and physical development of children and, ultimately, their educational attainment (Amato, 2000; McLanahan and Sandefur, 1994). Parents/fathers who do not live in the same household also do not develop the same trust levels and commitment to their children (McLanahan and Sandefur, 1994). Children who reside in single-parent households may also assume adult roles, such as providing childcare for younger siblings, which could slow human capital accumulation (Weiss, 1979). At the same time, assuming these adult roles might foster responsibility and teach skills that increase the individual’s value in the labor market. The stresses created by single motherhood, whether it occurs through marital dissolution or never having been married, can also strain the mother child relationship. This might lead to inconsistent parenting and may disrupt the natural stages children progress through in their development (McLanahan, 1988; McLanahan and Bumpass, 1988). Of course, if children are in a difficult situation at home, for instance, one where there is physical or emotional abuse, removing the child from such a circumstance might prove beneficial (Amato, 2000; Fronstin et al., 2001; Gruber, 2004). Even in households in which abuse is not present, continual marital turmoil can be harmful to the child’s emotional wellbeing and create economic effects.1

3. Literature review While the empirical research on family structure and its effects on children’s outcomes is quite large, our study is one of only a handful of US studies that investigate the relationship between family structure and children’s outcomes once they reach adulthood. International studies are relatively more common. Most of this research, whether international or US-oriented, focuses on educational outcomes, and nearly all of this research, whether testing single parenthood collectively or breaking single parenthood into more homogeneous categories, shows that living with married biological parents is advantageous relative to growing up with a single parent. Over time, researchers have attempted to remove the omitted variable bias in estimates of the importance of family structure by using propensity score matching techniques, family fixed effects models, and instrumental variable models. These econometric methods tend to reduce the magnitude of the associations, and often, but not always, the point estimates in these more sophisticated studies are statistically insignificant. Internationally, researchers have studied Germany, Sweden, and the United Kingdom. Fronstin et al. (2001) use the British National Child Development Survey and find that parental divorce or death is associated with a reduction in the labor force participation of males and in the wages of females. Parental divorce also appears to moderate educational attainment for the British cohort. Ermisch and Francesconi (2001) use sibling fixed effects models to determine if single parenthood is related to educational attainment, economic inactivity, childbearing at a young age, and smoking in the 1991–1995 waves of the British Household Panel Survey. They report that single parenthood is negatively related to educational attainment and positively related to smoking. Furthermore, single parenthood during the child’s youth appears to be particularly harmful. Francesconi et al. (2010) use a variety of methods to remove potential omitted variable bias, including family fixed effects, propensity score matching, and two additional quasi-experimental methods for families in Germany. They estimate a negative relationship between dissolution between biological parents and educational attainment that diminishes in size and significance once they remove some of the unobserved heterogeneity. Bjorklund and Sundstrom (2006) report the same result using a large sample of siblings in Sweden. Gruber (2004) is one of a few studies that investigates the long-term economic impacts of family structure using data from the United States, asking if changes in states’ unilateral divorce laws are related to the economic wellbeing of children as adults. He finds that children exposed to unilateral divorce regimes while growing up have lower levels of education and family income as adults. Lang and Zagorsky (2001) ask if maternal and paternal coresident time is related to the economic 1 It is also true that the reduction in resources created when a parent leaves the child’s household is moderated by the reduction in their consumption (Fronstin et al., 2001).


33

wellbeing of children. They show that coresidence with biological parents is largely unrelated to education, marital status, and economic wellbeing. Other researchers have used a parental death to estimate heterogeneous effects of single parenthood in the US. Biblarz and Gottainer (2000) find that divorce compared to a paternal death leads to lower education levels, lower occupational prestige and happiness, while Lang and Zagorsky (2001) report no difference in outcomes comparing death to divorce. According to Ginther and Pollak (2004), growing up with both biological parents is positively correlated with educational outcomes. Once one adds information on family background characteristics, such as family income, generating ‘‘descriptive regressions,’’ then the estimated size of the relationship declines in magnitude and is often insignificant. Other researchers use family fixed effects and find a significant relationship between family structure and educational outcomes (Case et al., 2001; Evenhouse and Reilly, 2004). In a study similar to this one, Bjorkland et al. (2007) use both the National Longitudinal Survey of Youth, 1979 cohort (NLSY) and the Panel Study of Income Dynamics (PSID) to estimate the relationship between family structure and educational attainment and earnings for American families and a 20% random sample of individuals born in Sweden in 1964 and 1965, matching them with their siblings. Their findings suggest that the children of single parent families have lower educational attainment and earnings. However, once one removes within family variation, the results are no longer statistically significant. Again, while there are differences in model choices, measures of family structure, and context that create variation in the findings, the preponderance of the evidence to date suggests a positive relationship between growing up with married biological parents and the economic wellbeing of a child. Whether that relationship is causal is more ambiguous since once omitted variable bias is reduced, the coefficient estimates diminish in size and frequently become statistically insignificant.

4. Methods 4.1. Data The PSID serves as the data source for this analysis. The PSID is a longitudinal data set originally representative of households in the United States in 1968, the year the survey started. From 1968 through 1997, respondents were asked demographic and employment questions annually, and since 1997, respondents have been interviewed biannually. One of the unique features of the PSID is that children of the original 4802 sample members were retained in the sample after they left their parents’ households.2 This attribute allows one to follow a child over time thereby comparing the child’s economic circumstances in childhood to his/her economic wellbeing as an adult. In order to track the marital histories of the mothers of the children in our sample, we use the Marital History Supplement in the PSID. Our sample is comprised of children born between 1967 and 1978 to the original sample members, and we employ information reported on them between their birth and the year of their 17th birthday and again from 2000 to 2008, when they were adults.3

4.2. Measures 4.2.1. Family structure We classify a child’s family structure based on the marital status of the child’s mother beginning in the year of the child’s birth through the year of the child’s 17th birthday. Thus, we follow Ermisch and Francesconi (2001) and implicitly assume that children reside with their mothers during their childhood. Mothers are classified as (1) continuously married, (2) never married, (3) married after the birth of the child, or (4) married at the birth of the child but experiencing a subsequent marital dissolution. To be continuously married, the mother has to be married at the time of the child’s birth and not experience a change in marital status until at least the year of the child’s 18th birthday. To be never married, the mother has to be unmarried at the time of the child’s birth and not experience a change in marital status until at least the year of the child’s 18th birthday. Those in the married after the birth of the child category were unmarried at the time of the child’s birth, but subsequently married before the year of the child’s 17th birthday (and may have subsequently experienced a marital dissolution). Those in the dissolution category were married at the time of the child’s birth but subsequently experienced dissolution (separated, divorced, or became a widow).4 Among the dissolution group, we further examine two subcategories of mothers: (5) those who had a marital dissolution and then remarried before the year of the child’s 18th birthday and (6) those who had a dissolution and never remarried before the child was 18. 2 The PSID is composed of two independent samples. The first, called the Survey Research Center or SRC, sample is composed of a randomly drawn national sample, and the second, called the Survey of Economic Opportunities or SEO, is a group of low-income families. We use both subsamples in our analyses, and weight responses by the individual weights provided. See Hill (1992) for more details on the PSID sample design. 3 As detailed below, we used a number of selection criteria to create our analytic sample, which altered its composition slightly. Please see the data appendix for more details. 4 The PSID reports separated, divorced, and widowed as a single category.

34

L.M. Lopoo, T. DeLeire / Social Science Research 43 (2014) 30–44 Table 1 Descriptive statistics for mothers and children. Parent’s generation

Child’s generation

Family income Family income adjusted for family size Marital status (proportion) Educational attainment (in years)

69,208 (45,577) 34,027 (22,173)

81,548 (67,048) 50,896 (42,439) 0.662 13.87 (2.16)

Age Always married Never married Married after birth Marital dissolution Remarried Never remarried Female African American White Other Sample size

25.47 (5.39) 0.681 0.022 0.057 0.239 0.130 0.110

12.61 (2.10)

31.82 (3.38)

0.491 0.130 0.843 0.027 1406

Note: Authors’ calculations from the PSID. Incomes are inflated to 2008 dollars using the CPI-U-RS and CPI-U-X1. Standard deviations reported in parentheses. All statistics weighted by PSID individual weight.

Our measure of family structure has limitations, mainly, in that it does not track the physical presence of biological fathers, other relatives, and other non-relatives in the household. For example, the ‘‘never married’’ category could include single mothers and cohabiting mothers. Addressing this limitation is not easy using the retrospective data in the PSID.5 Table 1 provides information on our analytic sample. Sixty-eight percent of the children had mothers who were continuously married, 2.2% were never married, and nearly 6% married after the birth of their child. Of the mothers in the analytic sample, almost 24% were married when their child was born but experienced dissolution before the child’s 18th birthday. This last set of mothers can be separated into those that remarry (13% of total sample) and those that never remarry (11% of total sample). 4.2.2. Family income For each child, we calculate the child’s total family income at two points in time. First, we create a measure of the child’s ‘‘permanent income when a child,’’ which we define as the average of the total family income in each year between the child’s birth and year of the child’s 17th birthday. Total family income includes taxable income, such as earnings, interest, and dividends, and cash transfers, such as Social Security and welfare from all family members. All income figures are inflated to 2008 dollars using the CPI-U-X1 and the CPI-U-RS and are based on the mother’s family.6 We measure permanent income in childhood using the average of family income over the first 17 years of the child’s life because the research literature has shown that annual measures of income have considerable year-to-year variation due to the transitory component of income (Corak, 2001; Mayer and Lopoo, 2008; Solon, 1992).7 To make sure that we have the majority of income measures for the mother over the child’s lifetime, we divided the income history into the child’s first 5 years (birth through age 5), from age six through age 11, and from age 12 through age 17. We created the following decision rule for selection into the analytic sample: the mother’s family income measure must not be missing for more than 2 years during each 6-year interval of the child’s youth. This decision rule guarantees that the total mean family income over the child’s first 17 years is based on multiple observations and that the income measures utilized are taken throughout the child’s first 17 years. We estimate the child’s permanent income as an adult as the average of the total family income for the child between 2000 and 2008, a period when the child has reached adulthood (and potentially is investing in his/her children). We require

5 Of the 1406 cases in our analytic sample, 887 of the biological mothers were always married to the biological father, 152 had a marital dissolution and never remarried, and 165 of the mothers divorced and remarried. When these mothers remarried, they did not remarry the child’s biological father. Thus, for 86% of the sample (n = 1,204), we have complete information for the biological father. Clearly, we also know that the 77 never married did not marry the biological father, but we cannot tell if these women cohabited with the biological father, at least before 1983. Of the 125 mothers who married after the birth of the child, we know that 41 married the biological father because the unique identification number for the father is recorded in the PSID. The remaining 84 cases have missing information for the biological father’s identification number. Because we do not have the father’s identification number for the other 84 cases, we do not know if the mother was eventually married to the biological father. Thus, one should keep in mind that we do not know the exact relationship between the child, the biological mother, and the biological father for the never married or married after birth family structure categories. 6 Economists frequently use a natural logarithmic transformation of income measures in their empirical analyses. This transformation has the advantage of reducing the influence of outliers, and it allows one to interpret slopes in percentage terms rather than in levels. We report the coefficient estimates in levels because we believe this provides a straightforward interpretation of the mean differences in family structure types. As a sensitivity check, we reran our models logging the outcome. The results were unchanged. 7 Blau (1999) shows that the relationship between family income and children’s outcomes is stronger when one uses an estimate of permanent income rather than a single year.


35

that the child have 3 years of income during that period before retaining the child in the analytic sample. Because the children in our sample are of different ages over this period, we flexibly control for child’s age in all analyses using this measure.8 In addition to reporting total family income, we also use a family income measure adjusted for the size of the family. We follow the Congressional Budget Office (CBO, 2009) methodology and divide the total family income for the child by the square root of the family size for each year. We then take the mean of family income adjusted for family size. The square root of family size is used to allow for the economies of scale present in a family. Table 1 shows that the mean annual income for these children during their first 17 years was $69,208 and $32,027 when adjusted for family size. Once the children became adults, the mean annual income was $81,548 and the adjusted mean income was $50,896. In some analysis, we also measure mean total family income over years before and over years following a marital dissolution and, among always-married mothers, mean total family income measured from the year of the child’s birth to the year of the child’s eighth birthday and the mean total family income over the next 9 years. In addition to family income, we use the child’s educational attainment and marital status in some of our models. These variables serve as both outcomes and as potential mechanisms that explain the relationship between family structure and the child’s total family income in adulthood. In models when the educational attainment measure is the outcome, we use a linear measure of years of education. This allows for a straightforward interpretation of the child’s education level. The mean years of education completed is 13.87. We also use categorical measures when we include the child’s educational attainment as a control. As explained earlier, there is a literature that links family structure to educational attainment, and we use our data to weigh in on this question. Researchers have also shown that biological parents’ family structure during a child’s youth is related to the child’s family structure as an adult (Amato, 1996; Kiernan and Cherlin, 1999; McLanahan, 1988). Because we want to know about the child’s marital status when income was measured and because income was measured over a 9-year interval (2000–2008), we ask what proportion of time the child was married from 2001 to 2009.9 The mean proportion for marital status for the child as an adult is 0.662. 4.3. Empirical models 4.3.1. Family income in childhood We begin our analyses by estimating the effect of family structure on permanent income during childhood. In our analysis of the relationship between family structure and permanent income during childhood, we first estimate the following simple descriptive model10:

Y i ¼ a0 þ a1 Singlei þ ei ;

ð1Þ

where Yi is the permanent income during childhood for child i, and Singlei is an indicator for whether the mother of child i is a single parent (never married, married after birth of the child, or marital dissolution). In this model, the constant, a0, is the average permanent income for children who resided with parents who were always married, and the coefficient, a1, is the difference in the permanent income for those who resided with a single mother at some point during their first 17 years. We also report estimates from a more elaborate empirical specification that includes more detailed measures of family structure:

Y i ¼ b0 þ b1 NMi þ b2 MAi þ b3 DRi þ b4 DN i þ X i H þ ei

ð2Þ

where NMi is an indicator for whether the mother of child i was never married, MAi is an indicator for whether the mother of child i was married after birth of the child, DRi is an indicator for whether the mother of child i experienced a marital dissolution and subsequently remarried, DNi is an indicator for whether the mother of child i experienced a marital dissolution and did not remarry, and Xi are demographic characteristics including measures of the sex and race of the child, the age of the mother at the time of her first child’s birth11 and at the time income is measured, indicators of the mother’s education (less than high school, high school, some college, college [excluded], and a linear and quadratic measure of the child’s parity). The results for Eqs. (1) and (2) are potentially biased due to the same selection issues that plague virtually all of the family structure literature. One might also reasonably ask if the mothers whose marriages dissolve have different levels of permanent income because of the dissolution and the loss of a coresident adult in the labor force, or if those with low-levels of income are more likely to experience dissolution. The permanent income measure and the mother’s family structure are 8 Haider and Solon (2006) demonstrate that a current year’s income is closer to an individual’s permanent income as the individual ages. In other words, one has a more accurate indication of an individual’s permanent income measuring their income at age 35 than, say, age 25. For the youngest respondents in the PSID, we have income measures as young as 22. As a sensitivity check, we ran separate models using mean income values calculated for individuals only after they had reached age 28. To keep the sample sizes comparable, we required two years in the sample rather than three. The results using this more restricted data are qualitatively the same as those reported below. 9 Each PSID interview collected data on total family income from the previous year. For example, 2000 total family income was collected in 2001. To match the data for marital status to the income information and maintain our sample size, we use the family files of the PSID. 10 Ginther and Pollak (2004) refer to this simple model that reports mean differences as a ‘‘stylized fact.’’ 11 Research consistently shows that young maternal age is positively associated with low socioeconomic status (Geronimus, 1987; Geronimus and Korenman 1992).

36


measured during the same interval making it impossible to determine the causal direction. Furthermore, as explained earlier, it is well understood that single parents are quite different from parents who are always married in many ways other than their family structure. To remove some of the biases present in this first set of results, we make several modifications in subsequent analyses. First, we use a more homogeneous group of families than those in the complete analytic sample – we retain only the mothers who were married at the time of the birth of the child. We also use a difference model to ask if the mean total family income before the dissolution occurred is different from the mean total family income after the dissolution. Naturally, one should expect the total family income to increase as the workers in a family gain greater experience and their earnings increase. To provide a baseline comparison, i.e., to define what is a reasonable change in income over time, we measure changes in income among always-married mothers. To create this counterfactual change among always-married mothers, we compare the mean income from the year of the child’s birth to the year of the child’s eighth birthday to the mean income in the next 9 years. We chose year nine as the dividing point since the median year that dissolution occurred (among those who experienced it) was the year of the ninth birthday. We also remove the influence of time invariant factors by estimating a difference model. Using the set of mothers who were married in the year their child was born, we estimate the change in the mean income as a function of the family structure of the child. Consider equation:

Y it ¼ b1 AM it þ b2 DRit þ b2 DN it þ ei þ lit ;

ð3Þ

where t = 1 in the period before dissolution and t = 2, post-dissolution. The variable AMit is an indicator variable that, among children whose mothers were always married, is equal to zero in the first 9 years of observation (birth to age 8) and one for the next 9 years, DRit is an indicator that, among mothers who experienced a marital dissolution and subsequently remarried, is equal to zero in the period prior to the dissolution and is equal to one in the post-period, and DNit is measured similarly for mothers whose marriage dissolved but who did not remarry. Each observation has a family specific fixed effect, e, which we allow to be correlated with family structure. If one subtracts the income level in the post-period from pre-period, one generates the following difference model:

DY i ¼ b1 DAM i þ b2 DDRi þ b3 DDNi þ Dui ;

ð4Þ

where DYi represents the difference in mean income before and after the dissolution for child i, and DR and DN are indicators for whether the mother had a marital dissolution and remarried and had a marital dissolution and did not remarry. The benefit of this difference model is that time-invariant unobserved characteristics of the family, such as the mother’s socioeconomic status, preferences for a large family, marriage, or work, (i.e., e) are differenced out. In other words, mother-specific factors that are constant within the interval, whether observed or not, are not the explanation for any differences observed. Time-varying factors that are not included in the model continue to potentially bias our estimates, however. 4.3.2. Education and family income in adulthood In our second set of analytic models, (equivalent to Eqs. (1) and (2)) we estimate the relationship between family structure in the child’s first 17 years and three different outcomes: the child’s educational attainment, the child’s marital status, and the child’s permanent income as an adult. In addition to the specification described in Eq. (2), we add a linear and quadratic term for the child’s age when the child’s education and income was measured. For all analyses, we report results using both the permanent family income and results using permanent family income adjusted for the size of the family. All of the models are also estimated using the individual weights provided by the PSID. In models with controls, we demean the control variables to give the constant an interpretation that is relevant for our purposes. We report Huber White standard errors and because multiple children from the same family are present in the analytic sample, we adjust the standard errors to allow for dependence in the error term within the same families. To gain some sense of the potential endogeneity bias present in our results, we first report descriptive statistics for each family structure type. Table 2 provides descriptive statistics broken down by family structure category. The means reported show that there are large differences in observable characteristics by family structure category. For each control variable we report the result from a hypothesis test with the null hypothesis that the coefficient reported for each family structure category is equal to the value reported for the always married category. For example, the proportion of children that are female in the dissolution and remarried and the dissolution and never remarried groups is different statistically from the proportion of female children for always married mothers. In the PSID, the children in the continuously married families are also much more likely to be white compared to the never married families, the married after birth families, and the never remarried families. The continuously married also appear to have much higher levels of education. Compared to every other category, the always married mothers have much lower rates of dropping out of high school. The always married mothers are also much more likely to have completed college than the mothers who never married and those who married after giving birth. Always married mothers are also older when they had their first child on average and older when they had the child under observation in our study than the mothers in every other family structure category. For the last two variables, child’s age when income measured and the child’s parity, we only observe a statistically significant difference for the married after birth group. Collectively, these data show that there are noticeable differences in observables, and one might reasonably conclude that there are likely differences in unobservables that might create bias in our family structure estimates. Readers should keep this in mind when interpreting results.

37

L.M. Lopoo, T. DeLeire / Social Science Research 43 (2014) 30–44 Table 2 Descriptive statistics by family structure.

Female White African American Other race Mother high school dropout Mother completed high school Mother completed some college Mother completed college Mother’s age when had first child Mother’s age when she had this child Child’s age (when income measured) Child’s place in mother’s birth order Sample size Notes: p < 0.05;

Always married

Never married

Married after birth

Dissolution and remarried

Dissolution and never remarried

0.461 (0.499) 0.903 (0.296) 0.073 (0.260) 0.024 (0.152) 0.128 (0.334) 0.512 (0.500) 0.206 (0.405) 0.155 (0.362) 22.43 (3.71) 26.18 (5.15) 31.96 (3.33) 2.25 (1.68) 877

0.452 (0.501) 0.195 (0.399) 0.805 (0.399) 0 0.661 (0.476) 0.296 (0.460) 0.010 (0.100) 0.032 (0.178) 19.66 (4.30) 23.59 (6.68) 31.22 (3.26) 2.57 (2.59) 77

0.520 (0.502) 0.465 (0.501) 0.450 (0.499) 0.085 (0.280) 0.407 (0.493) 0.491 (0.502) 0.100 (0.301) 0.002 (0.048) 18.96 (2.99) 20.31 (4.10) 30.93 (3.67) 1.53 (0.960) 125

0.594 (0.493) 0.904 (0.295) 0.064 (0.246) 0.032 (0.176) 0.200 (0.401) 0.399 (0.491) 0.276 (0.448) 0.125 (0.332) 20.65 (3.04) 24.17 (4.60) 31.69 (3.52) 2.12 (1.25) 165

0.548 (0.499) 0.726 (0.448) 0.255 (0.437) 0.020 (0.139) 0.201 (0.402) 0.386 (0.488) 0.276 (0.449) 0.137 (0.345) 21.71 (4.33) 25.71 (6.15) 31.70 (3.25) 2.18 (1.48) 152

p < 0.01 for hypothesis test with null that mean is equal to value for always married.

5. Results In this section, we first show results from models estimating the relationship between family structure on the resources available to children during childhood. Second, we provide estimates from models of family structure on the adult outcomes of the children. 5.1. Family structure and economic resources during childhood Table 3 displays results for the association between the family structure and permanent income during childhood. Model 1 suggests that children whose mothers were not continuously married (single mother) have incomes that are about $25,000 less than the children of mothers who were continuously married. Given that always-married families earned about $77,200 annually, single mothers have total family incomes that are about two-thirds the size of always-married couples’ incomes. Model 2 breaks the family structure up into never married, married after birth, dissolution and remarried, and dissolution and never remarried (the omitted group is always married). This disaggregation shows that, among ‘‘single mother families,’’ there is substantial variation in economic wellbeing. Compared to those who were always married, those who were never married had family incomes that were about $57,000 less. Those who were not married at the birth of the child, but later married, had family incomes that were about $38,000 less. One key finding is that there are differences between those whose mother divorced and remarried and those with mothers who divorced and did not remarry in terms of the child’s economic wellbeing in childhood. Mothers who remarry have family incomes that are about $12,000 less that the always married, while those who do not remarry have family income levels that are more than $27,200 less. This difference between these two groups is statistically significant. Table 3 OLS models of family structure and permanent income during child’s youth. Model 1

Single mother

Model 2

Family income

Adjusted family income

25,052 (3177)

10,232 (1560)

Never married Married after birth Marital dissolution and remarried Marital dissolution and never remarried Constant Demographics Mother’s education Sample size

77,193 (2313) No No

37,289 (1129)

Model 3

Family income


56,926 (3380) 38,152 (3862) 11,946 (4420) 27,243 (3837) 77,193 (2316) No No

26,224 (1711) 17,424 (2086) 3718 (2135) 10,936 (1932) 37,289 (1130)

Family income

37,685 (4553) 18,389 (3476) 6650 (4132) 238888 (3289) 74,578 (2114) Yes No 1406

Model 4 Adjusted family income

Family income


15,988 (2502) 7943 (1696) 848 (1991) 9235 (1593) 35,960 (1013)

27,046 (4044) 13,476 (3105) 8351 (3790) 23,827 (3316) 74,275 (2029) Yes Yes

10,856 (2192) 5567 (1499) 1688 (1798) 9217 (1602) 35,816 (976)

Notes: p < 0.05; p < 0.01. Robust standard errors clustered at family level. Models 3 and 4 include indicator variables for the child’s sex and the child’s race; linear and quadratic terms for the mother’s mean age when income was measured and the child’s parity; and measures of the mother’s age when she had her first child and her age when she had the child under study. Mother’s education includes a set of indicators for less than high school, high school, some college. College or more is omitted category.

38


In Model 3, we control for demographic factors, and we see noticeable differences in the coefficients once these demographic variables are added. This suggests the presence of unobserved heterogeneity and the importance of adding controls to the models to remove potential omitted variable bias. In Model 3, one still observes statistically significant differences for never married mothers, mothers who marry after the birth and married mothers who break up and never remarry. The children of mothers who remarry have permanent income levels around $6,700 less than those who are always married, but this difference is not statistically significant, suggesting that remarriage may protect children from experiencing potential parental investment levels that are noticeably different from always-married families. In Model 4, in which we also control for mother’s education level, the family income levels for children of never married mothers, the mothers who marry after birth, and the married mothers who break up and never remarry are different from the children of mothers who were always married. The economic wellbeing of the children of mothers who experienced dissolution and remarry is again statistically significantly different from the wellbeing of the children of always married mothers. Compared to the other estimated differences, however, children who grow up in remarried household experience parental income levels that are much higher than other family structure types. We find nearly identical results when we make an adjustment for the size of the family (see Table 3). In Table 4, we report results where we restrict our sample to mothers who were married in the year their child was born. In this set of results, we compare the family incomes of the child before a marital dissolution to family income after the dissolution, artificially breaking-up the continuously married couples to provide a baseline for a reasonable change in family incomes over time. This model has the additional benefit of removing time-invariant factors for the children that may be biasing the family structure estimates reported in Table 3. The first column reports results for total family income and the second column reports results for family-size adjusted total family income. Married couples earn about $25,100 more annually in the last 9 years of the child’s formative years than during the first 9. Those who remarried earned about $14,100 more after the dissolution. Both estimates are statistically significantly different from zero, and the difference between the coefficient estimates is statistically significant. Thus, children who experience dissolution and have mothers who remarry are worse off after the dissolution relative to continuously married families. Those who never remarried earned about $2,600 less after the dissolution than before, and this result is not statistically significant. When we compare this change to the always married and remarried coefficients, the coefficient for the never married families is statistically different. In other words, children who grow up in continuously married families have the highest permanent income available relative to other family structures, and those who remarry after dissolution are better off than those who never remarry. Interestingly, not only do the children of mothers who experience dissolution and do not remarry do worse relative to continuously married families and remarried families, they are worse off in absolute terms after the dissolution (although the reduction in income is not statistically significant). When we adjust family income for family size, we find a slightly different story. Those who remarry actually do better than those who are continuously married and those who never remarry in the later part of the child’s formative years. This could be because mothers who remarry are more selective when choosing their mates, because both partners are more inclined to work, or because mothers without many children are more likely to remarry. The children who grow up in families with dissolution and without remarriage do worse following the dissolution than the continuously married families, but once one accounts for family size, the difference is not statistically significant. While the results in Table 4 are free of any bias from time-invariant factors for the children in our sample, time varying unobserved factors will still be problematic. For example if dissolution occurs simultaneously with the loss of a job, because the job loss will affect family income, one cannot be sure if it is the dissolution or the job loss that is source of the change in family income.

5.2. Family structure and the economic wellbeing of the child as an adult Next, we ask if the effects of family structure persist into adulthood. In Table 5, we estimate the relationship between family structure and the child’s educational attainment, measured once the child is an adult. In Model 1, the children of Table 4 Difference model of family income before and after marital dissolution, sample of mothers married when child born. Change in family income following dissolution

Always married Dissolution and remarried Dissolution and never remarried Sample size

Family income


25,112 (2219) 14,124# (4180) 2589##%% (4256) 1189

11,487 (1078) 22,245## (2601) 7745%% (2531) 1189

Notes: p < 0.05; p < 0.01 for H0: parameter = 0. p < 0.05; ## p < 0.01 for H0: estimate different from Always married. % p < 0.05;%%p < 0.01 for H0: estimate different from dissolution and remarried. We assigned a ‘‘dissolution’’ date at 8 years to always married mothers for comparison. Robust standard errors clustered at family level; five cases lost for those who divorced in the child’s 17th year and four cases were lost for mothers who divorced in the child’s 16th year and income was not available in the child’s 17th year. #

39


single mothers earn 0.74 fewer years of education than the children of always married parents, who earned just over 14 years on average. Model 2 shows that, as was the case for parental income, aggregating all of the single parents masks differences that exist across different family structures. For instance, the children of mothers who were never married had 1.7 fewer years of education than the children of mothers who were always married, while the children of the mothers that married after the birth of the child had about 1.2 fewer years of education. The children of parents who experienced dissolution and remarried and the children of dissolution without remarriage had about 0.54 fewer years, and both differences are statistically significant. In Model 3, we include controls for demographic factors, and the differences for those who never married, those who married after birth, and those who remarried after dissolution are reduced noticeably. The difference for never married mothers declines to about three-quarters of the previous estimated mean difference, the difference for children of mothers who married after the birth declines to less than half of the previous mean difference, and the difference for the remarried mothers drops over 40%. The difference for the dissolution and never remarrying group does not change appreciably and remains statistically significant. In Model 4, we control for the mother’s education level, and all of the differences decline again, with the exception for the difference for the dissolution groups. The children of dissolution without remarriage is the only family structure type that has a statistically significant difference compared to the children of continuously married mothers. In Model 5, we include a measure of the child’s permanent family income to test if the remaining educational difference for this family structure group can be explained by the permanent family income differences. The coefficient estimate for never remarried falls about 30% and is no longer statistically significant, suggesting that the educational differences observed are due, at least in part, to the income differences observed during childhood. In Table 6, we run the same models using the proportion of time between 2001 and 2009 that the (adult) child was married as the outcome. Model 1 shows that the children of continuously married parents are married around 71% of the time between 2001 and 2009. The proportion of time children of single parents were married is about 15.6 percentage points lower. When we disaggregate by family structure type, we find that the proportion of time children of never married mothers are married is about 35 percentage points lower (or about 50% less) than the children of continuously married parents. The proportion of time that children with mothers who married after the birth were married is about 20.4 percentage points lowTable 5 OLS models of family structure and child’s educational attainment. Model 1 Single mother Never married Married after birth Marital dissolution and remarried Marital dissolution and never remarried Constant Demographics Mother’s education Mother’s permanent income Sample size

Model 2

Model 3

Model 4

Model 5

1.693 (0.366) 1.211 (0.320) 0.516 (0.219) 0.560 (0.285) 14.11 (0.095) No No No

1.311 (0.503) 0.616 (0.314) 0.230 (0.211) 0.547 (0.253) 14.02 (0.087) Yes No No 1406

0.630 (0.509) 0.303 (0.295) 0.333 (0.189) 0.556 (0.224) 14.01 (0.082) Yes Yes No

0.424 0.204 0.270 0.374 13.97 Yes Yes Yes

0.738 (0.170)

14.11 (0.094) No No No

(0.498) (0.292) (0.184) (0.228) (0.080)

Notes: p < 0.05; p < 0.01; Robust standard errors clustered at family level. Models 3–5 include indicator variables for the child’s sex and the child’s race; linear and quadratic terms for the mother’s mean age when income was measured and the child’s parity; and measures of the mother’s age when she had her first child and her age when she had the child under study. Mother’s education includes a set of indicators for less than high school, high school, some college. College or more is omitted category.

Table 6 OLS models of family structure and child’s marital status. Model 1 Single mother Never married Married after birth Marital dissolution and remarried Marital dissolution and never remarried Constant Demographics Mother’s education Mother’s permanent income Child’s education Sample size

Model 2

Model 3

Model 4

0.358 (0.077) 0.204 (0.046) 0.110 (0.041) 0.145 (0.050) 0.712 (0.015) No No No No

0.162 (0.076) 0.153 (0.080) 0.083 (0.056) 0.080 (0.056) 0.105 (0.040) 0.098 (0.040) 0.093 (0.042) 0.087 (0.042) 0.694 (0.015) 0.692 (0.015) Yes Yes No Yes No No No No 1406

Model 5

Model 6

0.146 (0.080) 0.077 (0.056) 0.096 (0.041) 0.081 (0.042) 0.691 (0.015) Yes Yes Yes No

0.140 (0.077) 0.074 (0.056) 0.094 (0.041) 0.075 (0.043) 0.690 (0.015) Yes Yes Yes Yes

0.156 (0.030)

0.712 (0.015) No No No No

Notes: p < 0.05; p < 0.01; Robust standard errors clustered at family level. Models 3–5 include indicator variables for the child’s sex and the child’s race; linear and quadratic terms for the mother’s mean age when income was measured and the child’s parity; and measures of the mother’s age when she had her first child and her age when she had the child under study. Mother’s education includes a set of indicators for less than high school, high school, some college. College or more is omitted category.

40

Table 7 OLS models of family structure and child’s family income during adulthood. Model 1 Family income


24,088 (4493)

14,522 (2868)

Never married Married after birth Marital dissolution and remarried Marital dissolution and never remarried Mother’s permanent income (demeaned) Child: LTHS

Model 3

Model 4

Model 5

Model 6

Model 7

Family income


Family income


Family income


Family income


Family income


Family income


52,953 (5294) 35,656 (7912) 16,206 (5512) 21,506 (6967)

32,322 (3334) 22,939 (4964) 9719 (3602) 12,188 (4527)

30,993 (7522) 17,260 (8748) 9667 (5282) 15,398 (7152)

19,103 (4911) 11,591 (5454) 4551 (3440) 8477 (4675)

20,509 (7559) 12,547 (8553) 10,953 (5235) 15,357 (7268)

12,518 (4819) 8568 (5349) 5619 (3337) 8621 (4710)

7853 (8470) 6411 (8377) 7077 (4938) 4115 (7820) 0.472 (0.186)

5156 (5255) 4998 (5260) 3365 (3101) 2081 (5029) 0.274 (0.108)

3928 (9283) 4527 (7984) 4125 (4980) 705 (7325) 0.406 (0.185) 50,253 (6282) 40,470 (5629) 22,406 (6402)

2640 (5815) 3737 (4933) 1209 (3085) 254 (4649) 0.228 (0.105) 34,727 (3951) 28,719 (3381) 17,475 (3808)

89,226 (3138) No No

55,525 (1955)

86,167 (2902) Yes No

53,504 (1770)

85,825 (2867) Yes Yes

53,339 (1750)

83,459 (2490) Yes Yes

51,962 (1562)

82,968 (2497) Yes Yes

51,298 (1431)

2624 (8013) 1062 (7576) 278 (4516) 2797 (7057) 0.397 (0.189) 46,000 (6167) 38,851 (5352) 22,294 (6170) 46,637 (4579) 81,208 (2176) Yes Yes

1455 (5554) 3111 (4899) 412 (3031) 887 (4656) 0.226 (0.106) 33,958 (3989) 28,426 (3377) 17,454 (3801) 8436 (3320) 51,063 (1437)

Child: HS Child: SC Child married Constant Demo. Mother’s education Sample size

89,226 (3134) No No

55,525 (1952)

1406

Notes: p < 0.05; p < 0.01; Robust standard errors clustered at family level. Models 3–5 include indicator variables for the child’s sex and the child’s race; linear and quadratic terms for the mother’s mean age when income was measured and the child’s parity; and measures of the mother’s age when she had her first child and her age when she had the child under study. Mother’s education includes a set of indicators for less than high school, high school, some college. College or more is omitted category. The child’s education includes the same education categories. The child married variable is the proportion of observations between 2001 and 2009 the child was married.


Single mother

Model 2


41

er, the proportion of time children with parents who had a dissolution and remarried is about 11 percentage points lower, and the proportion of time children with parents who had a dissolution and did not remarry is about 14.5 percentage points lower. All of the estimates are statistically significant. Models 3 and 4 control for the demographics of the families and the mother’s education level. Interestingly, all of the coefficient estimates decline in magnitude in these models, and the differences for the dissolution categories are the only statistically significant differences that remain. In Model 5, we control for permanent family income differences, and the only statistically significant coefficient estimate is for the dissolution and remarried group. The point estimate, 0.096, is similar to the mean difference reported in Model 2. Even controlling for the human capital accumulation for the child does not explain the difference for the children of remarriage. The point-estimate is largely unaffected in Model 6 and remains statistically significant. All other point-estimates are negative suggesting a reduction in the likelihood of marriage, but the remarried group is the only category that remains statistically significant. In Table 7, we report results for the permanent income of the child as an adult as well as estimates for family income adjusted for family size. We focus on the results for family income since the two sets are similar. We also build our descriptive regressions eventually incorporating potential mechanisms, education and marriage, that could explain the relationship between the family structure a child experienced while growing up and the child’s economic wellbeing as an adult. Model 1 reveals that the children of mothers who were never married have incomes that are about $24,000 less than the children of continuously married parents, a difference of about 27%. In Model 2, we disaggregate the family structure measures and find that the children of never married parents have total family incomes in adulthood that are about $53,000 less than the children of continuously married mothers, while the children of those who marry after birth earn $36,000 less. As was the case for income during childhood, there are interesting differences among the children of dissolution that differ by the remarriage status of the child’s mother. The children of mothers who remarried earn roughly $16,200 less than the children of always married parents, while the children of mothers who do not remarry earn $21,500 less. Models 3 and 4 include controls for demographic factors and maternal education, and these controls explain a large portion of the differences observed. The difference for never-married mothers falls (in magnitude) from -$53,000 to around -$20,500. The difference for those married after the birth of the child falls (in magnitude) from -$36,000 to -$12,500 and is no longer statistically significant. The difference for the dissolution and remarried group falls (in magnitude) from -$16,200 to -$10,900. The never remarried group declines (in magnitude) from -$21,500 to -$15,400. In Model 5, we control for the child’s permanent income growing up. Our results show that the child’s economic wellbeing in childhood is statistically significantly related to the child’s economic wellbeing in adulthood. Furthermore, once permanent income in childhood is controlled none of the family structure factors remain statistically significant. The estimated difference declines to $-7,900 for the never married group, $-6400 for the married after birth group, -$7,100 for the remarried group, and -$4,100 for the never remarried group. Models 6 and 7 add the child’s education and marital status to the Model 5 specification. The point estimates continue to decline in magnitude with these mechanisms added to the models. In Model 7, the point estimate for the never married mothers is $2,624 compared to the mean difference reported in Model 2 of -$52,900. The coefficient estimate for the married after birth group is -$1,062 compared to the -$35,700 reported in Model 2. For the remarried group, the mean difference with the controls in Model 7 is $278. The difference reported in Model 2 was -$16,200. Finally, the coefficient estimate in Model 7 for the never remarried group is $2,800 compared to -$21,500 in Model 2. None of the point estimate for family structure in Model 7 are statistically significant. Interestingly, the permanent income measure from childhood and subsequent educational attainment and marriage continue to be significantly related to the child’s economic wellbeing adulthood, which suggests that the family structure association that persist into adulthood can largely be explained by the economic wellbeing of the child when growing up and the effects that that income can have on the child’s educational attainment and family structure as an adult.12 6. Discussion and conclusion Until recently, data have not been available to investigate the long-term consequences of single parenthood for children once they reach adulthood, and research that connects the family structure of a child’s parents while growing up to his/her economic wellbeing in adulthood in the US is limited. We use the PSID to determine whether the marital status of mothers during the child’s formative years affects the child’s economic well-being both in childhood and adulthood. Our results suggest the economic wellbeing of children who experience a maternal marital dissolution in their childhood reduces the economic resources available to the child’s family. If one accounts for family size, then remarriage appears to be particularly protective. All children have greater resources available to them later in their childhood, but the children of mothers who have a dissolution and remarry have more resources available per capita than the children of continuously married or the children of never remarrying mothers. Future research might investigate if this improvement is merely another form of selection in which mothers with few children are more likely to remarry giving them greater economic resources per capita. This finding is interesting in the context of the broader literature. Duncan and Hoffman (1985) showed that marital dissolution is more harmful to women than men and that remarriage can protect women. We find that this is particularly true for women with children and when one looks at their smoothed income rather than just looking at individual years. Of 12 Appendix B reports similar results when we use a measure of family structure similar to Bjorkland et al. (2007), proportion of time spent in a single mother household and time spent in a remarried household. The omitted group is time spent in a married household.

42


course, one cannot be sure this difference among those who experience dissolution is a remarriage effect. Mothers who experience dissolution and remarry may be different from mothers who do not remarry in ways that we have not controlled. If this remarriage difference is causal, the finding is of value to the broader literature on remarriage. Children who grow up with mothers who divorce and remarry do not fare as well on a variety of outcomes, including educational attainment and engaging in risky behaviors, on average, as children who grow up with continuously married parents (Astone and McLanahan, 1991; Coleman et al., 2000; Hoffman and Johnson, 1998; Ribar, 2004; Teachman et al., 1996). Our results suggest that these losses are probably not due to a reduction in parental resources: remarried families seem to have at least as much income as continuously married families, perhaps even more. One potential explanation for the poorer outcomes for the children of remarriage is the instability of a marital dissolution, where children may have to move into a different neighborhood with different peer networks and schools. Furthermore, the children may suffer from the stress of the break-up. Ribar (2004) also suggests that step-parents may not place the same value on their step-children, and invest in them less. Our findings also suggest that some of the differences persist into adulthood. The children of single parents have lower levels of education and spend less time married than those who have continuously married parents. Once we control for the child’s economic wellbeing during childhood, however, these difference are largely explained with the exception of divorce and remarried families and marriage. With respect to their income as an adult, the findings for the children of single mothers suggest that as adults, these children earn about a quarter less annually than the children of continuously married parents. Our descriptive regressions show that once we control for the child’s demography and mother’s education a large portion of the difference is explained. If we incorporate the child’s permanent income growing up as well as the child’s educational attainment and marital status, these differences no longer exist. Ultimately, then we find no persistent effects of family structure over time, particularly once one controls for the resources available to the child when young. Appendix A. Data appendix Fitzgerald et al. (1998) report that the PSID has lost approximately 50% of it cases due to attrition between 1968 and 1989. The lost cases come disproportionately from the lower tail of the socioeconomic distribution, and includes those with the greatest mobility, weak attachment to the labor force, and marital instability. Despite this attrition, at least as of 1989, results from the PSID do not appear to be affected. The 1406 cases in our analytic PSID sample met some fairly rigid data requirements. We required that the child had a linked identification number to his/her mother; the mother’s marital history was complete in each year for the child’s first 18 observed years; the mother had a reported education value; the child’s mother needed at least 3 years of total family income reported during each of the following periods of the child’s life: ages zero through 5, ages 6 through 11, and ages 12 through 17; the mother’s age when she had her first child was reported; and the child had a reported level of education as an adult. Given these selection criteria, the 1406 children used in the analytic sample is much smaller than the full sample of children born between 1967 and 1978 in the PSID. Below, we describe where cases were lost in the sample construction. To get some sense of the characteristics of the children lost in each step, we report the child’s sex, race, and family income as an adult. While there were many children born between 1967 and 1978 in the PSID, many were lost during the panel due to attrition while others were in subsamples that were discontinued. The sample weights should account for sample attrition and the discontinued subsamples. Our initial PSID sample had 1823 respondents with a weight and an income measure in adulthood.

Original sample Less mothers without a linked identification no Less mothers without a marital history Less mothers without an education measure Less children without a complete income history Less mothers without age at first birth Less children without an education value

Child is Child is female African American

Child is Child is white other race

Child’s family income in adulthood

Sample size

0.503 0.503 0.503 0.500 0.496

0.139 0.140 0.140 0.132 0.129

0.835 0.832 0.832 0.839 0.843

0.027 0.028 0.028 0.029 0.028

80,765 81,143 81,234 81,901 82,180

(65,639) (65,939) (66,167) (66,836) (66,885)

1823 1674 1651 1576 1444

0.496 0.491

0.129 0.130

0.843 0.843

0.028 0.027

82,197 (66,923) 81,548 (67,048)

1442 1406

Results suggest that the cases that were lost for this analysis were not random. The individuals in the analytic sample are more likely to be male (by around 1 percentage point), about a percentage point less likely to be African American, and about a percentage point more likely to be white than the original sample. Furthermore, the analytic sample has a slightly higher income than the original sample suggesting that some lower socioeconomic status individuals were lost due to our selection criteria. Readers should keep this in mind when interpreting results.

43


Appendix B Another tactic one might take to examine the relationship between family structure and economic wellbeing in adulthood is to examine the length of time children spent in different family structures, similar to Bjorkland et al. (2007). Below, we provide the results from models using the proportion of time single and the proportion of time remarried (the proportion married is the omitted group). Interestingly, the models reveal the primary finding that we report in this paper: once one takes into account the family’s permanent income, even controlling for the mother’s education level, the family structure measures have no relationship with the child’s income in adulthood (see Table B1).

Table B1 OLS estimates of proportion of childhood (ages 0–17) mother single and remarried.

Proportion single Proportion remarried Constant Demography Mother’s education Mother’s permanent income Child’s education Child’s marital status

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

46,572 (6071) 16,024 (10,998) 87,731 (2944) No No No No No

29,182 (5973) 7738 (10,674) 85,287 (2699) Yes No No No No

24,397 (5954) 10,968 (10,523) 84,938 (2651) Yes Yes No No No

7860 (8114) 9654 (10,282) 83,000 (2281) Yes Yes Yes No No

1824 (7761) 5823 (10,812) 82,095 (2124) Yes Yes Yes Yes No

4657 (7218) 2531 (10,040) 80,875 (1984) Yes Yes Yes Yes Yes

References Amato, Paul R., 1996. Explaining the intergenerational transmission of divorce. Journal of Marriage and the Family 58, 628–640. Amato, Paul R., 2000. The consequences of divorce for adults and children. Journal of Marriage and Family 62 (4), 1269–1287. Astone, Nan Marie, McLanahan, Sara S., 1991. Family structure, parental practices, and high school completion. American Sociological Review 56 (3), 309– 320. Astone, Nan Marie, McLanahan, Sara S., 1994. Family structure, residential mobility, and school dropout: a research note. Demography 31 (4), 575–584. Becker, Gary S., 1991. A Treatise on the Family, Enlarged Edition. Harvard University Press, Cambridge, MA. Becker, Gary S., 1993. Human capital: a theoretical and empirical analysis, with special reference to education, third ed. University of Chicago Press, Chicago. Becker, Gary S., Tomes, Nigel, 1986. Human capital and the rise of families. Journal of Labour Economics 4, 21–239. Biblarz, Timothy J., Gottainer, Greg, 2000. Family structure and children’s success: a comparison of widowed and divorced single-mother families. Journal of Marriage and the Family 62, 533–548. Biblarz, Timothy J., Raftery, Adrian E., 1999. Family structure, educational attainment, and socioeconomic success: rethinking the ‘pathology of matriarchy’. American Journal of Sociology 105 (2), 321–365. Bjorkland, Anders, Ginther, Donna K., Sundstrom, Marianne, 2007. Family structure and child outcomes in the USA and Sweden. Journal of Population Economics 20 (1), 183–201. Bjorklund, Anders, Sundstrom, Marianne, 2006. Parental separation and children’s educational attainment: a siblings analysis on Swedish register data. Economica 73 (292), 605–6245. Blau, David M., 1999. The effect of income on child development. The Review of Economics and Statistics 81 (2), 261–276. Case, Anne, Lin, I-Fen, McLanahan, Sara, 2001. Educational attainment of siblings in stepfamilies. Evolution and Human Behavior 22, 269–289. Coleman, Marilyn, Ganong, Lawrence, Fine, Mark, 2000. Reinvestigating remarriage: another decade of progress. Journal of Marriage and the Family 62, 1288–1307. Congressional Budget Office (CBO), 2009. Historical Effective Federal Tax Rates: 1979 to 2006. Congressional Budget Office, Washington, DC. Corak, Miles, 2001. Death and divorce: the long-term consequences of parental loss on adolescents. Journal of Labor Economics 19 (3), 682–715. Duncan, Greg J., Hoffman, Saul D., 1985. A reconsideration of the economic consequences of marital dissolution. Demography 22 (4), 485–497. Ermisch, John F., Francesconi, Marco, 2001. Family structure and child’s achievements. Journal of Population Economics 14 (2), 249–270. Evenhouse, Eirik, Reilly, Siobhan, 2004. A sibling study of stepchild wellbeing. Journal of Human Resources 39 (1), 248–276. Fitzgerald, John, Gottschalk, Peter, Moffitt, Robert, 1998. An analysis of sample attrition in panel data: the Michigan panel study of income dynamics. Journal of Human Resources 33 (2), 251–299. Francesconi, Marco, Jenkins, Stephen P., Siedler, Thomas, 2010. Childhood family structure and schooling outcomes: evidence for Germany. Journal of Population Economics 23, 1201–1231. Fronstin, Paul, Greenberg, David H., Robins, Philip K., 2001. Parental disruption and the labour market performance of children when they reach adulthood. Journal of Population Economics 14 (1), 137–172. Gennetian, Lisa, 2005. One or two parents? half or step siblings? The effect of family composition on young children’s achievement. Journal of Population Economics 18 (3), 415–436. Geronimus, Arline T., 1987. On teenage childbearing and neonatal mortality in the United States. Population and Development Review 13 (2), 245–279. Geronimus, Arline T., Korenman, Sanders, 1992. The socioeconomic consequences of teen childbearing reconsidered. The Quarterly Journal of Economics 107, 1187–1214. Ginther, Donna K., Pollak, Robert A., 2004. Family structure and children’s educational outcomes: blended families, stylized facts, and descriptive regressions. Demography 41 (4), 671–696. Gruber, Jonathan, 2004. Is making divorce easier bad for children? The long-run implications of unilateral divorce. Journal of Labor Economics 22 (4), 799– 833. Haider, Steven, Solon, Gary, 2006. Life-cycle variation in the association between current and lifetime earnings. American Economic Review 96 (4), 1308– 1320. Hill, Martha S., 1992. The Panel Study of Income Dynamics: A User’s Guide. Sage Publications, Inc., Newbury Park, CA. Kiernan, Kathleen E., Cherlin, Andrew J., 1999. Parental divorce and partnership dissolution in adulthood: evidence from a British Cohort. Population Studies 53, 39–48. Lang, Kevin, Zagorsky, Jay L., 2001. Does growing up with a parent absent really hurt? Journal of Human Resources 36 (2), 253–273.

44


Mayer, Susan E., Lopoo, Leonard M., 2008. Government spending and intergenerational mobility. Journal of Public Economics 92, 139–158. McLanahan, Sara S., 1988. Family structure and dependency: early transitions to female household headship. Demography 25 (1), 1–16. McLanahan, Sara, Bumpass, Larry, 1988. Intergenerational consequences of family disruption. American Journal of Sociology 94 (1), 130–152. McLanahan, Sara, Sandefur, Gary, 1994. Growing Up with a Single Parent: What Hurts, What Helps. Harvard University Press, Cambridge, MA. Nock, Steven L., 2000. Marriage as a public issue. The Future of Children 15 (2), 13–32. Ribar, David C., 2004. What Do Social Scientists Know about the Benefits of Marriage? A Review of Quantitative Methodologies. IZA Discussion Paper 998. Smock, Pamela J., 1993. The economic costs of marital disruption for young women over the past two decades. Demography 30 (3), 353–371. Solon, Gary, 1992. Intergenerational income mobility in the United States. American Economic Review 82 (3), 393–408. Teachman, Jay D., Paasch, Kathleen, Carver, Karen, 1996. Social capital and dropping out of school early. Journal of Marriage and the Family 58 (3), 773–783. Thompson, Elizabeth, Hanson, Thomas L., McLanahan, Sara S., 1994. Family structure and child well-being: economic resources vs. parental behaviors. Social Forces 73 (1), 221–242. Weiss, Robert, 1979. Growing up a little faster: the experience of growing up in a single parent household. Journal of Social Issues 35 (4), 97–111. Weitzman, Lenore J., 1985. The Divorce Revolution: The Unexpected Social and Economic Consequences for Women and Children in America. Free Press, New York. Wolfe, Barbara, Haveman, Robert, Ginther, Donna, An, Chong Bum, 1996. The ‘window problem’ in studies of children’s attainments: a methodological exploration. Journal of the American Statistical Association 91 (435), 970–982.

Lessons to the World: Sudanese Youth and the Wellbeing of Children.

Family structure as a predictor of screen time among youth.

Childhood Family Income and Violent Victimization During Youth and Young Adulthood: Trends in Hospital Care During 1988-2007 in Finland.

The role of positive youth development practices in building resilience and enhancing wellbeing for at-risk youth.

The wealth, health and wellbeing of Ireland's older people before and during the economic crisis.

Genetic testing of children for diseases that have onset in adulthood: the limits of family interests.

Commitment to the wellbeing of children worldwide.

Mental health and wellbeing in children and adolescents.

Criteria for the Transition to Adulthood, Developmental Features of Emerging Adulthood, and Views of the Future Among Greek Studying Youth.

Childhood neighbourhoods as third places: Developing durable skills and preferences that enhance wellbeing in adulthood.

Family Structure, Family Stability, and Outcomes of Five-Year-Old Children.

The relationship between psychosocial features of emerging adulthood and substance use change motivation in youth.

The Millennium child with autism: early childhood trajectories for health, education and economic wellbeing.

Parental caregiving of children with cancer and family impact, economic burden: nursing perspectives.

Restavèk children in context: Wellbeing compared to other Haitian children.

CHANGES IN NONMARITAL COHABITATION AND THE FAMILY STRUCTURE EXPERIENCES OF CHILDREN ACROSS 17 COUNTRIES.

Providing alcohol to underage youth: the view from young adulthood.

Social Class, Family Formation, and Delinquency in Early Adulthood.

Family Instability and Pathways to Adulthood in Urban South Africa.

Three-year change in the wellbeing of orphaned and separated children in institutional and family-based care settings in five low- and middle-income countries.

Youth unemployment and functional somatic symptoms in adulthood: results from the Northern Swedish cohort.

Screening for Depression in Children and Youth.

'Transitions are Scary for our Kids, and They're Scary for us': Family Member and Youth Perspectives on the Challenges of Transitioning to Adulthood with Autism.

Suicidal behavior of children and youth.