Family Instability and Child Well-Being

Abstract

Past research suggests that children who experience multiple transitions in family structure may face worse developmental outcomes than children raised in stable two-parent families and perhaps even children raised in stable, single-parent families. However, multiple transitions and negative child outcomes may be associated because of common causal factors such as parents’ antecedent behaviors and attributes. Using a nationally-representative, two-generation longitudinal survey that includes detailed information on children’s behavioral and cognitive development, family history, and mother’s attributes prior to the child’s birth, we examine these alternative hypotheses. Our results suggest that, for white children, the association between the number of family structure transitions and cognitive outcomes is largely explained by mother’s prior characteristics but that the association between the number of transitions and behavioral outcomes may be causal in part. We find no robust effects of number of transitions for black children.

The instability of family structure has become an increasingly salient part of children’s lives in the United States over the past half-century. During this period, as is well-known, divorce rates increased (Cherlin 1992), as did the prevalence of nonmarital cohabitation, which is less stable than marriage (Bumpass and Lu 2000). Moreover, cohabitation and marriage appear to be more unstable in the United States than in most other developed countries; and a relatively high percentage of American children experience transitions into single-parent families and stepfamilies (Andersson, 2002; Heuveline, Timberlake and Furstenberg 2003).

A growing body of literature suggests that children who experience multiple transitions in family structure may fare worse developmentally than children raised in stable two-parent families and perhaps even than children raised in stable, single-parent families. This body of research presents what we call the instability hypothesis, the prediction that children are affected by disruption and changes in family structure as much as (or even more than) by the type of family structures they experience. If this hypothesis were true, it would suggest that a significant reinterpretation of the effects of family structure on children’s well-being may be warranted. For example, it would imply that a child born to a single parent might be as well off, or perhaps even better off, if the parent did not cohabit or remarry.

However, most empirical tests of the instability hypothesis have neglected an alternative explanation. The association between multiple transitions and negative child outcomes does not necessarily imply that the former causes the latter. In fact, multiple transitions and negative child outcomes may be associated with each other through common causal factors reflected in the parents’ antecedent behaviors and attributes. We call this explanation the selection hypothesis. A test of the instability hypothesis versus the selection hypothesis would therefore be of interest to sociologists in several sub-disciplines: family sociology, with its emphasis on family and household structure; the sociology of children and youth, with its concern for children’s well-being; life course studies, which emphasize the long-term effects of early events on individuals’ later lives; the sociology of crime and deviance, which includes studies of the early antecedents of anti-social behavior; and the sociology of education, with its interest in the determinants of academic achievement.

We test the selection hypothesis against the instability hypothesis in a statistical analysis of nationally representative longitudinal data that includes detailed information on children’s behavioral and cognitive development, family history, and mother’s attributes prior to the child’s birth. We use the 1979 through 2000 waves of the National Longitudinal Survey of Youth (NLSY79) and its 2000 mother-child supplement, the Children of the NLSY (CNLSY), for the statistical analysis. Outcome variables measured in 2000 include indicators of children's cognitive achievement, internalizing and externalizing behavior problems, and delinquent behavior in early and middle childhood and early adolescence (ages 5 to 14). These outcomes are well-known indicators of children’s current development, as well as predictors of well-being in late adolescence and early adulthood. A number of studies have examined the effects of family structure transitions on similar outcomes (Astone and McLanahan 1991, Najman, Behrens, Andersen, Bor, O’Callaghan, and Williams 1997, Kurdek, Fine, and Sinclair 1995, Martinez and Forgatch 2002, Pong and Ju 2000, Cockett and Tripp 1994, Amato and Sobolewski 2001).Key sets of predictor variables will be (1) measures of the children's family structure histories, including the number of transitions; and (2) indicators of the children’s mothers' attributes. The main advantage of the prospective, two-generation design of the NLSY79 and CNLSY over previous studies is that we can examine children’s outcomes measured in 2000 while controlling for their mothers’ own cognitive test scores and self-reports of delinquency, substance abuse, and other characteristics obtained 15 to 20 years earlier from the mothers themselves, when most of them were adolescents. We believe this to be a substantial improvement on the standard list of retrospectively-obtained controls, such as respondent’s parents’ education and respondent’s family structure while growing up, etc., used in other studies of family instability.

If the associations between family structure transitions and child outcomes are substantially reduced by accounting for mother’s antecedent attributes, we will conclude that the analyses provide support for the selection hypothesis. Both hypotheses would receive support if the effect of transitions were reduced but still statistically significant in the presence of controls for mother’s attributes.

BACKGROUND

In the mid-to-late 1980s, a few researchers speculated that the more transitions in family structure a child experiences, the worse would be his or her adjustment (e.g., Furstenberg and Seltzer 1986; Brody, Neubaum, and Forehand 1988). Transitions may include parents’ separation, a cohabiting romantic partner’s move into, or out of, the home of a single parent, the remarriage of a single (non-cohabiting) parent, or the disruption of a remarriage. The underlying concept was that children and their parents, whether single or partnered, form a functioning family system and that repeated disruption of this system may be more distressing than its long-term continuation. According to this research, repeated movements of parents and their partners and spouses in and out of the child’s household could produce a series of short-term crises that could reduce a child’s capacity for normal development. Even if no crises ensue, multiple transitions could undermine a child’s sense of security and trust which could, in turn, affect the child’s emotional development. Stable single-parent households or stepfamilies, in contrast, do not require that children adjust repeatedly to the loss of co-resident parents and parent-figures or the introduction of cohabiting partners and stepparents.

Research by Wu and Colleagues

Strong evidence for the instability hypothesis was first presented by Wu and Martinson (1993). They examined the association between several measures of family disruption during a woman’s childhood and the likelihood that she would have a premarital birth. Using data from the National Survey of Families and Households, they found that the likelihood of a premarital birth was significantly related to the number of transitions a woman had experienced while growing up but not to the amount of time she had spent in a single-parent family or whether she was living with a single parent as an adolescent. They concluded, “Clearly our results challenge conventional wisdom by suggesting a stronger effect of family instability and a weaker effect of prolonged exposure to a mother-only family than previous research would suggest” (1993, p. 225). In a follow-up article, Wu (1996) replicated his finding, using the 1979 to 1989 waves of the NLSY79, which contains better measures of family income, with the caveat that early exposure to a single-parent family was also associated with premarital births for African Americans.

Wu and Thomson (2001) extended research on the instability hypothesis to consider the association between family structure and age at sexual initiation, and found important differences by race. Using data from the 1979 to 1987 waves of the NLSY79 to estimate proportional hazard models, they found that for white women, the number of family structure transitions a woman had experienced was associated with a higher risk of first intercourse at an early age. For black women, the number of transitions experienced had no predictive effect, but family structure in adolescence was strongly associated with age at sexual initiation. The authors concluded that family turbulence explains higher rates of sexual intercourse for white women, while for black women, family structure influences sexual activity through the availability of parental control.

Studies of Instability Effects on Cognition and Behavior

Work by Wu and his colleagues focused on sexual activity and early childbearing, outcomes most often observed in mid-to-late adolescence. Other studies focusing on behavioral and cognitive outcomes salient to early and middle childhood and early adolescence have produced findings that are also consistent with the instability hypothesis, although few are nationally representative and none employ a two-generation design with data collected on the first generation prior to the second generation’s birth. In an Australian sample, children who experienced at least one transition in family structure during early childhood were more likely to have elevated levels of behavior problems by age five, regardless of mother’s marital status at birth (Najman, Behrens, Andersen, Bor, O’Callaghan, and Williams 1997). Studies of children in elementary school found that children who experienced two or more transitions were more likely to show disruptive behavior at school, to have poorer emotional adjustment, and to have lower grades and achievement scores compared to children who experienced no transitions or one transition (Kurdek, Fine, and Sinclair 1995, Martinez and Forgatch 2002). And among middle school and high school students observed longitudinally, experiencing more than one transition in family structure was associated with a higher likelihood of dropping out of school (Pong and Ju 2000).

THEORETICAL MECHANISMS

Although the statement that multiple transitions disrupt family functioning in ways that affect children’s cognitive and behavioral outcomes is plausible, it does not explain the mechanisms through which instability, if indeed it has a causal effect, operates. Similarly, the claim that a selection effect functions through parents’ antecedent behaviors and attributes offers little guidance about the pathways through which it operates. While there is no consensus on the mechanisms behind these potential effects, one can draw some theoretical inferences from the research literature.

Instability Effects

The theoretical case for an instability effect rests on the proposition that each of the transitions of parents, partners, and stepparents into and out of the household requires adjustments that can be stressful, at least initially, for the biological parent and the children, and that the cumulative effect on children’s well-being can be substantial. The nature of the transition in terms of changes in household composition is less relevant than the stress associated with moving from one form to another. To illustrate, many studies have shown that a child’s transition from a two-parent family to a single-parent family is associated with lower school engagement, poorer cognitive achievement, and more behavior and emotional problems. One might think that a second transition that adds a stepparent or cohabiting partner would boost children’s well-being, since it brings another adult and, usually, more income into the household.

But most studies show that children whose parents have remarried do not have higher levels of well-being than children in single-parent families (Coleman, Ganong, and Fine 2000). Their levels of behavior problems, for example, are similar to children in single-parent families and higher than children in two-biological-parent families (Hetherington and Clingempeel 1992). While many explanations have been suggested, the most common, according to the leading review article, are variants of stress models (Coleman, Ganong, and Fine 2000). The addition of the stepparent is said to increase stress among children and adults, as families adjust to new routines, as the biological parent focuses attention on the new partnership, or as stepchildren come into conflict with the stepparent. This increased stress could cause children to have more emotional problems or to perform worse in school. Thus, it could potentially have effects on behavioral or cognitive outcomes.

Selection Effects

Theorizing about selection effects is more straightforward, although the research evidence is still modest. The basic proposition is that parents’ personality characteristics and cognitive abilities have two effects: First, they affect the characteristics of their children, through the environment in the home, through genetic transmission, or more likely a combination of both. Second, they affect the parents’ abilities to maintain stable intimate partnerships. Consequently, family instability and children’s well-being co-vary but are not causally related. As an example of research in this genre, consider a behavioral genetic study of the effects of divorce on children, in which O’Connor, Plomin, Caspi, and DeFries (2000) compared the adjustment of children in two-parent and divorced biological and adoptive families. They reasoned that differences in the experience of divorce between children in biological families, on the one hand, and adoptive families, on the other hand, would likely reflect what they called “passive genotype-environment correlations” – the tendency for parents to transmit genetic traits to their children and to provide environmental experiences consistent with those traits – in the biological-parent families but not the adoptive parent families. Using a matched sample of about 200 adoptive families and 200 biological families in Colorado, they found some evidence that children’s lower academic achievement, self-esteem, and competence in divorced families reflected genetic factors but that children’s higher psychopathology in divorced families appeared to be environmental in origin.¹

Another oft-cited study of the effects of multiple family structure transitions on children’s development found that for a sample of boys from one metropolitan area, mothers’ antecedent “antisocial behavior” mediated the association between transitions and the boys’ adjustment, academic performance, and delinquent behavior in sixth grade (Capaldi and Patterson 1991). But in a later study of the same group of boys, Capaldi, Crosby, and Stoolmiller (1996) found a positive, significant association between parental transitions and sexual initiation that was only partially mediated when parental antisocial behavior was included in the model. The earlier study suggests that diminished parenting skills and diminished relationship skills may co-vary, creating the appearance of a causal relationship between multiple transitions and some dimensions of children’s adjustment when, in fact, parental vulnerabilities cause both the transitions and the adjustment difficulties. Yet the study is based on a small sample of boys of limited diversity from one metropolitan area (Capaldi and Patterson 1987), and the evidence is not consistent across outcomes. Neither Wu’s research nor later research has tested the selection hypothesis against the instability hypothesis on multiple outcomes in a large, representative sample of children where mothers’ characteristics that may be related to selection are measured prior to the birth of her children.

Context-Specific Differences in How Mechanisms Operate

In addition to providing some insights about theories of instability and selection effects, the research literature suggests additional complexities that need to be addressed. Particularly relevant for our study are the possibility that the effects of instability and selection differ by race and the possibility that they differ for children’s behavioral development and for their cognitive development.

Racial Differences

The finding by Wu and Thomson (2001) of a racial difference in the apparent effects of instability on sexual initiation raises the issue of whether instability, to the extent that it may constitute a true causal effect, could have different consequences for black and white children across a broader range of outcomes. A number of studies suggest that the effects of other measures of family structure (e.g., time spent in a single-parent family or premarital cohabitation) may have different effects on both child and adult outcomes for whites and blacks (Dunifon and Kowaleski-Jones 2002; Phillips and Sweeney 2005).

With regard to instability there are at least two mechanisms that could theoretically produce a difference according to racial group (McLoyd, Cauce, Takeuchi, and Wilson 2000). The first is the greater role of extended kin in the upbringing of black children. The involvement of grandmothers and other kin could ease the emotional strain and the increased burden of care caused by the departure of a parent or parent-like figure. Studies suggest that black grandparents are more likely to take on a parent-like role toward their grandchildren than are white grandparents (Cherlin and Furstenberg 1992). Moreover, black women in general appear to provide more practical support to their kin in areas such as child care, transportation, and household work than do white women (Sarkisian and Gerstel 2004). Second, black children may live in family and neighborhood environments where a greater number of stressful events occur, so that the relative effect of the added stress of a partnership change would be lower than for a child living in less stressful circumstances. Consistent with either of these mechanisms, a meta-analysis of studies found that the association between parental divorce and an adult child’s educational attainment and non-marital childbearing was weaker among blacks than among whites (Amato and Keith 1991). Although these pathways are speculative, the findings to date led us to conduct our statistical analyses separately for black and white children.²

Domain-Specific Effects

Several studies suggest the possibility that different sets of parental and family attributes relate to children’s cognitive outcomes than to their behavioral and emotional outcomes. Martinez and Forgatch (2002) found, for example, that mothers’ academic skills encouragement was associated with children’s academic functioning, while general effective parenting was associated with children’s behavior and emotional adjustment. Carlson and Corcoran (2001) found that mother’s aptitude was uniquely associated with cognitive achievement scores, while mother’s psychological well-being was more strongly associated with children’s behavior problems. And Ram and Hou (2003) found that effective parenting and parents’ psychological well-being were strongly associated with children’s emotional and behavioral outcomes after moving into a one-parent or stepparent family, whereas changes in family income and parents’ employment status were associated with children’s math and reading comprehension but were unrelated to emotional-behavioral outcomes. These studies suggest that different instability and selection effects could operate in the domain of cognitive outcomes than in the domain of emotional and behavioral outcomes.

Moreover, some aspects of children’s well-being may be more responsive to changes in family composition than others. For instance, Sun and Li (2002) concluded that children’s psychological well-being is more elastic and responsive to parents’ marital disruption than is their academic achievement. The latter, they argued, is affected through a cumulative process over time (e.g, an early deficit in reading or math leads to a greater disadvantage over time) that cannot be quickly reversed by an improvement in family conflict. Because these studies suggest domain-specific effects, we do not expect that instability and selection will necessarily function the same way in children’s cognitive outcomes as in their behavioral outcomes.

RESEARCH DESIGN AND METHODS

Overview of Study Sample and Measures

In this article, we test the selection hypothesis using 1979 to 2000 data from the NLSY79 and 2000 data from its mother-child supplement, the CNLSY. Several of the articles we have reviewed used earlier waves of the NLSY and CNLSY, but no one has used the more recent waves to conduct analyses of the effects of mothers’ early attributes and their children’s experiences of family instability on their children’s development. The NLSY79 is a multipurpose survey of more than 12,600 men and women first interviewed in 1979, when they were between the ages of 14 to 21 (a few were 22) (Center for Human Resource Research [CHRR] 2001). Every year until 1994, and every other year since then, interviewers have collected detailed, longitudinal data on respondents’ education, work, and family experiences. Since 1982, the NLSY79 has also collected data on female respondents’ pregnancy, fertility, and child care experiences. In 1986, the study expanded to include the CNLSY, a biennial assessment of the children of NLSY79 mothers. The CNLSY assesses children’s cognitive achievement, temperament, motor and social development, behavior problems, and perceived competence, and the children’s home environment. All children of NLSY79 mothers between the ages of 0 to 14 are eligible for interviews at each wave (CHRR 2002).

The CNSLY includes three relevant components. For children 5 years or older, a computer-assisted personal interview (CAPI) obtains age-appropriate child-reported assessments of cognitive achievement, perceived competence, and schooling. For children 4 years or older, mothers complete a self-administered assessment of the home environment, child behavior problems, and child temperament. Finally, children 10 and older complete a self-administered questionnaire that obtains information about the child’s relationship with parents and peers, school performance, attitudes, delinquency, and sexual activity. The cumulative response rate for the NLSY79 in 2000 was 83.4 percent (U.S. Bureau of Labor Statistics 2002). The analysis file consists of one record per eligible child, with data from the mother attached to the record. The merged mother/child file for 2000 includes data on 3,392 children of 1,965 mothers. Because all of a woman’s children are eligible for the CNLSY, our sample includes multiple children from the same household. We use robust standard errors to take into account the non-independence of observations within households.

Limitations of the Study Sample

Mother’s age at birth

Mothers of the children in the CNLSY were 14 to 21 years old when they were selected for the NLSY79 sample in 1979. This restricted age range limits the representativeness of the children in the CNLSY. Children born to mothers shortly after the 1979 interview have mothers who were relatively young at the child’s birth, compared to the average childbearing age in the U.S. population. Conversely, children born to women only recently have mothers who were relatively old at the child’s birth. Thus, the oldest and youngest children in the 2000 CNLSY are not representative of all children their age nationally because they tend to have younger or older mothers, respectively. In a separate analysis, available on the ASR web site, we restricted our sub-sample to children age 8 to 14 in 2000, whose mothers’ age at birth is similar, on average, to mothers’ mean age at birth nationally for children of the same age. In this article, however, we have included children as young as age 5, except in our analysis delinquent behavior, which was only asked of 10 to 14 year-olds. The results are nearly identical. (Recall that cognitive outcomes were not measured for children under age five and that children over 14 are not suitable for inclusion because they were interviewed using a different questionnaire that produced non-comparable data. Therefore, our sample includes some children born to relatively old mothers, but we are unable to include older children born to the youngest mothers.)

Child’s place of residence

A second limitation is that children of the CNLSY are eligible for interview only when they coreside with their mothers. If a child is living in a separate household or other arrangement in a given interview year, he/she is not interviewed, but if the child returns to the mother’s household by the following interview year, he/she is picked up again. Therefore, the 2000 wave of the CNLSY includes only those children currently residing with their mothers. Our indicator of transitions will underestimate the effect of multiple transitions on behavioral and cognitive outcomes if transitions out of maternal co-residence, in fact, have an additional effect.

Coverage of the Latino population

A third limitation is that all children in the CNLSY were born to mothers living in the United States in 1979; consequently the Latino sample does not include children whose parents arrived in the United States as part of the large migration streams of recent decades. It is therefore not representative of the 2000 Latino population of the United States. Only seven percent of children in the weighted 2000 CNLSY sample are identified as Latino/Hispanic, significantly less than the 16.6 percent of children age 0 to 14 who were identified as Latino/Hispanic in the 2000 Current Population Survey (U.S. Bureau of the Census 2001). We have therefore excluded Latino children from the analyses presented in this article. However, we will refer to the results of a separate analysis for Mexican-origin mothers (the largest ethnic subgroup among Hispanics) available on the ASR web site.

Analysis Plan

We conduct a two-step analysis. In step 1, which we will refer to as our baseline model, we estimate models for child outcomes measured in 2000 that are similar to the models presented in existing studies of the instability hypothesis. These models are intended to replicate the finding that, controlling for a standard set of background variables, the number of transitions a child has experienced is significantly associated with lower scores on cognitive indicators and higher scores on behavior problem indicators. In step 2, which we will refer to as our full model, we add in variables that reflect characteristics of the mother and her environment before the child was born (and for some measures, before any of her children were born). If the associations between transitions and child outcomes are substantially reduced in step 2, we will conclude that the analyses provide support for the selection hypothesis. Both hypotheses would receive support if the effect of transitions were substantially reduced but still statistically significant. Our models cannot determine how much of the selection effect, if one is indicated, reflects environmental influences and how much reflects genetic influences. We now turn to a discussion of our measures and models.

Child Outcomes

We focus on children’s cognitive and behavioral outcomes. We have selected these outcomes because cognitive and behavioral attributes are important indicators of children’s current development, as well as established predictors of adult well-being. For example, lower scores on measures of cognitive achievement are associated with indicators of future socioeconomic disadvantage, including lower educational attainment and lower income (Jencks and Phillips 1997). Higher scores on measures of internalizing or externalizing behaviors are also associated with lower educational attainment (Miech et al. 1999; McLeod and Kaiser 2004), teen parenthood (Kessler et al. 1997), downward occupational mobility (Caspi, Elder and Bem 1987), and marital instability (Caspi, Elder, and Bem 1988; Kessler, Walters, and Forthofer 1998).

Cognitive achievement

Assessments of a child’s cognitive achievement are based on responses to the Peabody Individual Achievement Test (PIAT). The PIAT is a wide-ranging measure of academic achievement for children age five and over. It is among the most widely used brief assessments of academic achievement, and has demonstrably high test-retest reliability and concurrent validity (CHRR 2002). The CNLSY includes three subscales of the PIAT: Mathematics, Reading Recognition, and Reading Comprehension. For each subscale, normalized percentile and standard scores are derived, on an age-specific basis, from the child’s raw score. The norming procedures are essentially a two-step process with the percentile scores being derived from the raw scores and the standard scores from the percentile scores. The norming sample has a mean of 100 and a standard deviation of 15.

Behavioral outcomes

The Behavior Problems Index (BPI) was created to measure the frequency, range, and type of childhood behavior problems for children age four and over. Many items were derived from the Achenbach Child Behavior Checklist and other child behavior scales (see CHRR 2002). Assessments are based on mother reports of child behavior. We use the standardized BPI externalizing behavior score derived by summing scores from 20 items with three possible response categories (alpha reliability =.87 for white children and .90 for black children). Externalizing behavior is associated with peer problems, hyperactivity, and conduct disorder, and includes behaviors such as being argumentative, being disobedient at school or at home, and having trouble relating to others (CHRR 2002). The behavior scale is nationally normed, but normed “same-gender” scores are not available; that is, we cannot assess directly how girls compare to other girls, and boys to other boys. As with the measures of cognitive achievement, we use standard scores derived from the percentile scores.

Self-Reports for 10- to 14-year-olds

Using mothers’ reports of children’s behavior introduces the risk of projection. Mothers who engaged in problem behaviors themselves may be more likely to report that their children engaged in problem behavior, even if independent assessments indicate otherwise. Teacher reports or self-reports by children avoid the potential problem of projection, but such data are not available for the full sample. However, the CNLSY does include self-reports of behavior for children age 10 to 14. From these data, we construct a 9-item scale of children’s delinquent behavior, including whether in the last year the child has stayed out late; hurt someone; lied; stolen; damaged school property; gotten drunk; skipped out of school; required a parent to come to the school; or stayed out without permission. The scale has a possible range of 0 (committed none of the behaviors) to 27 (committed each of the 9 behaviors more than two times in the last year). We use this scale as the dependent variable in analyses including children age 10 to 14 who completed the child self-assessment survey.

Predictors of Child Outcomes

Control variables

In step 1 our objective is to replicate existing analyses of the association between instability and child outcomes. We therefore include a standard set of covariates to control for attributes that other studies have suggested are associated with child outcomes. These include gender, age, and number of adults in the household. We conduct separate analyses for white and black children.

Family structure and changes in family structure

In step 1 of our analysis, we will also include various measures of the child’s family history: (1) For current family structure, following Wu (1996), we identify three categories: two-parent families (whether married or cohabiting); mother-only families; and families with a biological mother and stepfather (or cohabiting partner). Recall that all children were residing with their mothers in 2000. (2) For time spent in a mother-only family, we use an indicator of whether the child spent at least 75 percent of the years between ages 0 and 4 in a mother-only household,³ a measure developed by Wu and Martinson (1993) to predict whether a woman experienced a premarital birth.⁴ Wu and Martinson used this measure as an indicator of whether a child was socialized from birth to perceive non-marital childbearing as normative. We use the measure to represent whether the child was at an elevated risk of poverty and low paternal involvement in early childhood. (3) We also develop a measure of child’s number of family transitions. We focus on changes in the child’s household composition due to changes in the mother’s marital status and cohabitation status.⁵ We assume that the following changes are accompanied by changes in household composition: from single to either cohabiting or married, from cohabiting to single, and from married to separated, divorced, or widowed. We do not consider the change from cohabiting with a partner to marrying that same partner as a family transition; nor do we consider the movement from separation to divorce as a family transition.⁶ We sum the number of transitions since the child’s birth that the mother has made. We also include a dummy variable indicating whether any transition transpired within the last two years, in order to separate the effect of the recency of a transition from the cumulative effect of multiple transitions. We will also refer to a supplementary analysis on the ASR web site in which we used two measures of the numbers of family transitions: the number of transitions in which an adult left the household (e.g., through marital separation) and the number of transitions in which an adult entered the household (e.g., through remarriage).

Mother’s characteristics

We include characteristics of the mother that reflect the social context in which she was raised and her behavior and attributes in adolescence and early adulthood. These indicators include the number of family transitions the woman experienced growing up, age at first sex, age at first birth, underage alcohol use, early drug use, delinquent behaviors, self-esteem, early family structure transitions in adulthood, cognitive achievement, and age at the birth of the child in the CNLSY. Most of the indicators were collected between 1979 and 1986, prior to the birth of the child in the study. The exceptions are the woman’s age at the child’s birth and the count of family transitions that the woman experienced in her own household growing up, which was collected retrospectively in 1988. Based on theoretical expectations and empirical research, we expect that the selected characteristics will be related to women’s later union formation, union stability, and family environment.

Family transitions

The measure of mother’s own family transitions represents instability in her childhood family structure, and potentially influences the number of family structure transitions she experiences in adulthood, as well as her parenting style. Women were asked to identify with whom they coresided each year from birth to age 18. From their responses, we develop 23 family types (following Wu 1996), which fall into four broad categories: two-parent family, mother-only family, a family with one biological parent and one stepparent, and all other arrangements. We sum the number of transitions between those four categories that the mother experienced as a child.

Age at first sexual intercourse

We expect that mother’s early initiation of sexual intercourse (before age 16) will be associated with children’s lower cognitive achievement scores and greater behavioral problems. Among girls, age at first intercourse is indicative of responses to home and school environments, and may represent attitudes or experiences that mothers transmit to their children (Schvaneveldt, Miller, Berry, and Lee 2001; Paul, Fitzjohn, Herbison, and Dickson 2000; Billy, Landale, Grady, and Zimmerle, 1998; Miller and Heaton 1991; Elliott and Morse 1989; South, Haynie, and Bose 2005; Whitbeck, Yoder, Hoyt, and Conger 1999).

Age at first birth

We include a dummy variable indicating whether the mother was less than 20 years old at her first birth. Births to young women, and especially to teenagers, are associated with long-term socioeconomic disadvantage, manifested among mothers as a higher risk of poverty and welfare dependence, and a lower likelihood of marriage. The relationship between teen childbearing and future disadvantage has long been considered to be causal (Hayes 1987), but some studies suggest that unmeasured family background characteristics may influence both the likelihood that a woman will have a teenage birth and that she will be socioeconomically disadvantaged in adulthood (Geronimus and Korenman 1992; Hoffman, Foster, and Furstenberg 1993; Geronimus, Korenman and Hillemeier 1994; Hoffman 1998). Evidence from these studies suggests that the effects of a teenage birth, while still significant, are attenuated by family background.

Completed years of education at age 21

Mother’s educational attainment represents her potential socioeconomic status in adulthood. We consider her completed education at age 21, the youngest age at which mothers gave birth to children included in the analytic sample⁷.

Drug and alcohol use

Previous research has shown that drug use is often co-morbid with antisocial personality and externalizing behavior (Knight, Broome, Cross, and Simpson 1998), and adult antisocial personality traits are associated with poor parenting skills, which in turn predict poor child outcomes (Capaldi and Patterson 1991). Antisocial personality and externalizing behavior also may be transmitted to children genetically. We measure alcohol and drug use separately. One item indicates whether the mother engaged in frequent underage alcohol consumption, defined as consuming at least one or two drinks a week before age 18. In the second item, we use an index of mother’s hard drug use prior to her first marriage or the birth of her first child, whichever came first, based on responses to the 1984 waves of the NLSY79. Our index sums six dichotomous measures, including whether the woman ever used cocaine, amphetamines, barbiturates, tranquilizers, psychedelic drugs, or heroin (alpha reliability = .75). The index is then divided by the number of items to which the mother provided a valid response, generating an indicator of mother’s average drug use with a range from 0 (no drug use) to 1 (used each drug at least once). For respondents with more than two missing values, a score is imputed using regression-based methods.⁸

Delinquent and illegal behavior

Our index sums the number of non-drug related behaviors in which the woman engaged. Prior research indicates that mother’s early delinquent behavior is associated with children’s behavior problems, net of other controls (Kandel, Rosenbaum, and Chen 1994). We use an index of mother’s delinquent and illegal behaviors, based on responses to the 1980 wave of the NLSY79, when respondents were between 15 and 22 years old. Respondents were asked whether they had engaged in a variety of behaviors in the past year, including shoplifting, other theft, assault, or vandalism. The respondent was asked how many times she engaged in each activity in the last year, and responses are grouped into seven categories ranging from never (0) to more than 50 times (6). As with the indicator of drug use, the illegal behaviors index is divided by the number of valid responses for each case. The alpha reliability score for the 15-item index is .69. We use regression-based methods to impute an illegal behavior score for respondents with more than three missing values on individual items (see endnote 8).

Cognitive ability/achievement

NLSY79 respondents participated in the independent administration of the Armed Forces Vocational Aptitude Battery (ASVAB) in 1980, and scores from the battery were used to construct an approximate and unofficial Armed Forces Qualification Test (AFQT) score. We use the 1989 revised version of the score, which is based on the scoring method currently used by the Department of Defense. The AFQT is designed as a general measure of trainability for applicants to the Armed Forces, and scores in the general adult population probably reflect both variation in environmental factors such as educational attainment and genetic factors such as cognitive ability (Jencks and Phillips 1997). Cognitive ability/achievement could be transmitted to children by either environmental or genetic means. For missing values, we use regression-based imputation methods to impute an AFQT score.⁹

Self-esteem

Women’s antecedent mental health may be associated both with union stability and with children’s psychological well-being. There is some evidence that lower self-esteem (Kim and McKenry 2002) and depression (Horwitz, White, and Howell-White1996) negatively predict women’s entry into marriage, and life satisfaction is positively associated with marriage (Mastekaasa 1992; but see Lamb, Lee, and DeMaris 2003) and negatively associated with divorce (Mastekaasa 1994). Exposure to mothers’ depression in infancy and early childhood is also associated with young children’s internalizing and externalizing problems (Essex, Klein, Cho, and Kraemer 2003; Essex, Klein, Miech, and Smider 2001). Accounting for mothers’ psychological well-being also partially nets out the potential influence of reporter bias on mother-reported assessments of children’s development.

The Rosenberg Self-Esteem Scale, a 10-item scale that describes an individual’s degree of approval or disapproval toward oneself, was administered to NLSY79 respondents in 1980 (CHRR 2001).¹⁰ We reverse-code specified items in the self-esteem scale so that a higher score designates higher self-esteem, and take the average of responses to the 10 items for each individual (reliability alpha=.83).¹¹

RESULTS

Descriptive analysis

Table 1 describes the subsample of children we have drawn from the CNLSY. Statistics may be generalized to children age 5 to 14 in 2000 with white or black mothers who were between the ages of 21 and 38 when the children were born.¹² The descriptive data are presented for all children in the first set of columns and for white and black children separately in the second and third sets of columns. On average, the subsample scores are slightly above the national norm of 100 on the indicators of externalizing behavior and cognitive achievement. That is, compared to the national average, the sample shows slightly higher levels of externalizing behavior, as well as slightly higher levels of cognitive achievement as measured in the three domains measured by the PIAT: mathematics, reading recognition, and reading comprehension. (Mean differences between white and black children that are statistically significant at the .05 level are starred.) Among children 10 to 14, self-reported delinquent behavior has an observed range from 0 to 24, with a mean just over 3. Black children report significantly more delinquent behavior than white children, but the average scores for both groups are highly skewed toward 0.

Table 1.

Means and Standard Deviations, Children Aged 5 to 14 (unweighted)

	All		White Children		Black Children
	Mean	S.D.	Mean	S.D.	Mean	S.D.		Min.	Max.
Cognitive and Socioemotional Development
Externalizing behavior	100.638	14.249	100.084	13.451	101.951	15.917	*	83	175
Self-reported delinquent behavior (age 10–14)	3.084	3.669	2.839	3.626	3.569	3.711	*	0	24
Math ability	103.379	14.611	106.532	13.822	96.234	13.823	*	65	135
Reading Recognition ability	106.085	15.573	108.750	14.709	100.091	15.808	*	65	135
Reading Comprehension ability	102.137	14.448	105.243	13.644	95.251	13.795	*	65	135
Early household structure
Child born to married mother	0.797	0.402	0.914	0.281	0.529	0.500	*
At least 75% of first 4 yrs in mother-only hh	0.197	0.398	0.083	0.277	0.458	0.499	*
All years spent in mother-only household	0.161	0.368	0.059	0.236	0.398	0.490	*
Current household structure
Married, biological parents	0.642	0.480	0.761	0.427	0.369	0.483	*
Mother-only household	0.264	0.441	0.157	0.364	0.509	0.500	*
Mother and stepfather in household	0.094	0.292	0.082	0.275	0.123	0.329	*
Instability in family structure
Number of transitions child experienced	0.700	1.171	0.581	1.117	0.973	1.245	*	0	9
Child experienced transition w/n last 2 yrs	0.147	0.354	0.137	0.344	0.169	0.375
Control variables
Male	0.503	0.500	0.507	0.500	0.493	0.500
Age in years	9.728	2.683	9.585	2.686	10.055	2.651	*	5	14
No. of other adults in hh besides mother	0.960	0.633	0.973	0.506	0.929	0.855		0	5
Mother's characteristics prior to the birth of the child
Mother's 1st birth at <=age 19	0.184	0.387	0.101	0.301	0.375	0.484	*
Mother's age at child's birth	28.505	3.248	28.795	3.155	27.838	3.363	*	21	38
Mother sexually active before age 16	0.118	0.323	0.084	0.278	0.197	0.398	*
AFQT score, percentile	44.662	28.709	55.167	25.911	20.476	18.366	*	1	99
Mother's years of education at age 21	12.592	1.762	12.829	1.659	12.047	1.870	*	1	16
Mother's transitions growing up	0.451	0.821	0.394	0.810	0.584	0.831	*	0	5
Mother's prior mar/cohab transitions	1.341	1.171	1.464	1.161	1.060	1.145	*	0	8
Illegal activities	0.170	0.273	0.162	0.246	0.189	0.326	*	0	3.79
Underage alcohol consumption	0.457	0.498	0.518	0.500	0.318	0.466	*
Hard drug use	0.063	0.161	0.081	0.179	0.022	0.099	*	0	1
Positive self-esteem	3.221	0.406	3.237	0.398	3.183	0.422	*	1.9	4
N	2097		1462		635
N, ages 10–14	989		657		332

^*Differences by race are statistically significant at the .05 level

Source: NLSY79/CNLSY 2000

Multivariate Statistical Models

We begin our multivariate analysis with the full sample of children age 5 to 14 years. Because the cognitive achievement and behavior problems dependent variables for the full sample can be considered continuous and normally distributed, we use ordinary least squares (OLS) regression with robust standard errors. Robust standard errors account for clustering and facilitate the assumption that observations within households (siblings) are not independent, but observations between households are.¹⁴

For the baseline analysis, we test four models, each with a different dependent variable. As predictors, we include the control variables and indicators of family structure described above. For each dependent variable, the proposed OLS model takes the form:

• y_i = α + β₁X_i + β₂Z_i + β₃T_i + ε_i

where y_i is the predicted score for individual i, α is the intercept term, β₁ and β₂ are vectors of regression coefficients associated with a particular set of variables, X_i is a vector of control variables for individual i, and Z_i is a vector of family structure variables. T_i is a measure of the number of family structure transitions a child has experienced, β₃ is the regression coefficient associated with Ti, and ε_i is a random error term. We tested various specifications of the measure of family structure transitions, T_i, because we anticipated that the cumulative effect of transitions might be nonlinear. These specifications included categorical measures of the number of transitions experienced, and a variety of transformations, including the square, the cube, and the square root of the number of transitions experienced. The continuous, linear specification had the strongest association with each outcome. The linear relationship between transitions in family structure and children’s behavior has been established, although not in a nationally representative study (Capaldi and Patterson 1991).¹⁵ For reasons of space we do not present the coefficients for the control variables X_i The full results are available on the ASR web site.

Table 2 presents results from the baseline models separately for white children and black children.¹⁶ Among whites, the number of family structure transitions a child has experienced is positively associated with his or her externalizing behavior score, and is negatively associated with his or her scores on tests of cognitive achievement. The association is statistically significant in predicting externalizing behavior and math and reading recognition scores. Each dependent variable has a mean of 100 and a standard deviation of 15, so, for example, the observed effect of a single transition on the externalizing behavior for a white child when other indicators of family structure and demographic characteristics are accounted for is 2.261, or about one-seventh of a standard deviation. Three transitions (2.261*3) would be associated with nearly a 6.7-point increase, more than one-third of a standard deviation.

Table 2.

Ordinary Least Squares Regressions Predicting Child Behavior and Cognitive Development, Ages 5–14

	White Children				Black Children
		Cognitive Ability				Cognitive Ability
	External. Behavior	Math	Reading Rec.	Reading Comp.	External. Behavior	Math	Reading Rec.	Reading Comp.
Socialization
at least 75% of 1st 4 yrs in mom-only hh	4.95 [1.838]**	−4.61 [1.780]**	−7.555 [2.103]**	−5.677 [1.840]**	4.018 [1.791]*	−3.075 [1.485]*	−1.824 [1.577]	−1.23 [1.604]
Current household structure
(omitted category=married biological parents)
Mother-only household	1.823 [2.251]	−2.158 [1.795]	−3.2 [2.136]	−4.749 [1.857]*	6.05 [1.826]**	−3.34 [1.618]*	−5.431 [1.839]**	−4.566 [1.751]**
Mother and stepfather in hh	−2.336 [2.547]	−0.46 [2.055]	−4.503 [2.512]	−6.651 [2.197]**	3.431 [2.479]	−3.221 [2.724]	−6.557 [3.237]*	−7.098 [2.737]**
Instability in family structure
# of any transitions child exp'd	2.261 [0.755]**	−1.127 [0.452]*	−1.157 [0.575]*	−0.793 [0.513]	0.103 [0.720]	0.685 [0.612]	1.243 [0.760]	0.076 [0.548]
Constant	93.091 [1.818]**	108.656 [1.307]**	114.181 [1.316]**	109.644 [1.250]**	87.884 [2.991]**	101.951 [1.611]**	106.969 [1.787]**	101.595 [1.797]**
N	1363	1337	1336	1175	575	590	594	530
R-squared	0.07	0.04	0.07	0.07	0.09	0.05	0.07	0.05

Robust standard errors in brackets

^*significant at 5%;

^**significant at 1%

Source: NLSY79/CNLSY 2000

Among black children, there is almost no association between the number of transitions a child has experienced and the child’s externalizing behavior score, when other factors are accounted for.¹⁷ The magnitude of the coefficient for the number of transitions in that model is close to 0. The association between the number of transitions experienced and a child’s cognitive scores are positive, or opposite the hypothesized direction, but the associations are not statistically significant.

The baseline models in table 2 show that the nature of family structure in early childhood and at the time of interview are also associated with child well-being. For white children, being raised in a mother-only household for at least 75 percent of a child’s first four years is strongly predictive of behavioral and cognitive achievement scores.¹⁸ Living in a mother-only household early on is associated with an increase of nearly one-third of a standard deviation in the externalizing behavior score (β=4.95) and a one-third to one-half standard deviation decrease in cognitive achievement indicators. For black children, current family structure is more predictive of child well-being than is early family structure: living in a mother-only household at interview is associated with a greater than one-third standard deviation increase in the externalizing behavior score and a comparable decrease in cognitive achievement scores.

In the full model, we add indicators of mother’s characteristics measured before the birth of the child. The OLS model takes the form:

• y_i = α + β₁X_i + β₂Z_i + β₃T_i + β₄V_i + ε_i

where β₄ is a vector of coefficients associated with V_i, a vector of mother’s characteristics measured prior to the birth of child i. For space reasons, we present the coefficients for V_i only on the ASR web site, but we will discuss key coefficient estimates in this article.

Table 3 shows the results of the full model.¹⁹ For white children, the association between the number of family structure transitions a child has experienced and the child’s externalizing behavior score diminishes somewhat when mother’s antecedent characteristics are included ((β=1.99), but remains significant at the p<.01 level. These results suggest that both selection and causation may operate to explain the relationship between family structure instability and externalizing behavior among white children. That is, the relationship may be driven partly by a common association with attributes of the mother that were present before her child was born, and partly by what may be a true effect of instability on behavior. Five characteristics of the mother have a statistically significant effect on her child’s externalizing behavior at the p=.10 level or lower: her early initiation of sexual activity (initiation before age 16), experiencing a teen birth, and the number of marital/cohabitation transitions experienced before the child was born are positively associated with her child’s behavior problems, and her completed years of education and her score on the self-esteem scale are negatively associated.

Table 3.

Ordinary Least Squares Regressions Predicting Child Behavior and Cognitive Development with Controls for Mother's Characteristics Prior to the Birth of the Child, Ages 5–14

	White Children				Black Children
		Cognitive Ability				Cognitive Ability
	External. Behavior	Math	Reading Rec.	Reading Comp.	External. Behavior	Math	Reading Rec.	Reading Comp.
Socialization
At least 75% of 1st 4 yrs in mother-only hh	3.054 [1.842]	−0.912 [1.569]	−4.415 [1.884]*	−2.999 [1.637]	3.782 [1.821]*	−1.039 [1.265]	−0.499 [1.444]	−0.224 [1.332]
Current household structure
(omitted category=married biological parents)
Mother-only household	1.85 [2.108]	0.089 [1.742]	−1.184 [2.035]	−2.688 [1.775]	5.671 [1.996] **	−0.725 [1.568]	−2.734 [1.725]	−2.649 [1.629]
Mother and stepfather in hh	−2.629 [2.355]	−0.187 [1.920]	−3.75 [2.338]	−5.635 [1.987]**	1.621 [2.395]	0.011 [2.322]	−2.358 [2.735]	−.371 [2.103]
Instability in family structure
Number of transitions child exp'd	1.99 [0.726]**	−0.209 [0.426]	−0.259 [0.560]	0.251 [0.482]	0.29 [0.714]	0.842 [0.529]	1.45 [0.724]*	0.383 [0.510]
Constant	111.324 [8.455]**	88.613 [5.554]**	80.486 [6.949]**	68.798 [6.223]**	94.18 [15.770]**	71.093 [8.238]**	69.899 [9.672]**	59.262 [8.404]**
N	1363	1337	1336	1175	575	590	594	530
R-squared	0.12	0.16	0.16	0.18	0.14	0.21	0.22	0.28

Robust standard errors in brackets

^*significant at 5%;

^**significant at 1%

Source: NLSY79/CNLSY 2000

As for cognitive outcomes, after controlling for the mother’s attributes, the association between the number of family structure transitions a child has experienced and the child’s cognitive achievement diminishes nearly to 0 among white children. Thus, the apparent association seems to be due almost entirely to selection. Mother’s years of education and her AFQT score have a statistically significant, positive effect on each dimension of cognitive achievement.²⁰

Among black children, the association between family structure transitions experienced and behavior problems remains non-significant after accounting for mother’s antecedent characteristics, except for a positive association (p<.05) that develops between transitions and the child’s reading recognition score. The effect of transitions on the externalizing behavior score, which was very low in the baseline model, increases in magnitude but remains relatively small and statistically insignificant.²¹

Turning to the measures of the effects of family structure during the child’s first four years and the effects of current family structure, we see that among both white and black children, the inclusion of the mother’s antecedent characteristics generally reduces the apparent effects of these measures on cognitive achievement, although some coefficients remain significant or marginally significant for whites. But the apparent effect of family structure on externalizing behavior remains robust: the early family structure measure remains significant for blacks and marginally significant for whites (p<.10), and current residence in a mother-only family remains significant for blacks. These findings suggest that other dimensions of family structure beyond instability may have an effect on children’s behavior problems.

Children age 10 to 14 years

We turn to an analysis of self-reported delinquent behavior for children 10 to 14 who completed the self-assessment in the CNLSY in 2000. In this case, the dependent variable is a count of self-reported delinquent behaviors, and the variable is skewed toward 0 and has overdispersion (the variance of the variable is greater than its mean). Therefore, we move away from an OLS regression and use a negative binomial regression.

The interpretation of the negative binomial regression is slightly different from an OLS regression. For any continuous covariate x_k in the model, a unit change in x_k changes the expected count, Y, by a factor of exp(β_k), holding all other variables constant. Where x_k is discrete, with, for example, values of 0 or 1 as in a dichotomous covariate, the expected count changes by

• ΔE(Y|X)/Δx_k = E(Y|X, x_k = U) − E(Y|X, x_k = L),

holding all other variables constant, where X represents the string of covariates in the model, U represents the upper value of covariate x_k, and L represents the lower value of x_k. Results are shown in table 4. We describe the results in terms of change in the expected count of the score in the dependent variable.

Table 4.

Negative Binomial Regression Predicting Self-Reported Delinquent Behavior, Ages 10–14

	White Children		Black Children
	Model 1	Model 2	Model 1	Model 2
Socialization
At least 75% of 1st 4 yrs in mom-only hh	0.344 [0.145]*	0.18 [0.153]	0.128 [0.118]	0.098 [0.125]
Current household structure
(omitted category=married biological parents)
Mother-only household	0.295 [0.231]	0.218 [0.213]	0.264 [0.177]	0.225 [0.187]
Mother and stepfather in hh	−0.256 [0.208]	−0.309 [0.205]	0.047 [0.226]	−0.031 [0.241]
Instability in family structure
Number of transitions child exp'd	0.13 [0.049]**	0.116 [0.046]*	0.027 [0.043]	0.024 [0.045]
Constant	−1.221 [0.412]**	−0.894 [1.084]	−0.08 [0.620]	0.911 [1.346]
N	657	657	332	332

Robust standard errors in brackets

^*significant at 5%;

^**significant at 1%

Source: NLSY79/CNLSY 2000

Among white children, the association between number of transitions and delinquent behavior is statistically significant at the .01 level before and after mothers’ characteristics are added. In the full model (model 2), each additional transition is associated with a 12 percent increase in self-reported delinquent behavior (β=.116, [exp(β) − 1]=.123). Moreover, the magnitude of the association is substantial: a white child who experienced three transitions would have a predicted 41 percent increase in his or her delinquent behavior score, whereas living in a mother-only household for at least 75 percent of the first four years would have a predicted 15 percent increase in delinquent behavior score.²² In addition, neither of the coefficients for early or current residence in a mother-only household is statistically significant. These results suggest that the association between transitions experienced and delinquent behavior in middle childhood and adolescence may be causal, at least in part. This pattern of results is consistent with Wu’s claim that instability is a stronger predictor of child outcomes than is current or past family structure (Wu and Martinson 1993; Wu 1996).²³ Among black children, however, we find virtually no effect of the number of transitions and nonsignificant associations with early experience in single-parent families and current family structure.

DISCUSSION

The instability hypothesis, which has gained currency in the sociological literature during the last decade, posits that children are affected by disruption and changes in family structure as much as (or even more than) by the type of family structures they experience. An alternative explanation of the association between changes in family structure and negative child outcomes is the selection hypothesis, which posits that the association is due to common causal factors reflected in the parents’ antecedent behaviors and attributes. We have used nationally representative data from the NLSY79/CNLSY, a twenty-one-year panel study of women and their children, to test these hypotheses on the behavioral and cognitive development of white and black children age 5 to 14 in 2000. We employed three child outcome measures: a cognitive achievement test, a mother-reported scale of externalizing behavior problems, and (for 10-to-14 year olds) a self-reported scale of delinquent behavior. To represent family structure instability, we counted the number of marital and cohabitation transitions a child had experienced. To represent mothers’ attributes prior to the birth of the children in the CNLSY, we considered a variety of self-reported behaviors and experiences as well as scores from psychometric assessments. Our objective was to determine whether the addition of mothers’ prior characteristics substantially reduces the apparent effects of children’s family instability on their cognitive and behavioral development.

The results of our analyses are summarized in Table 5. A positive sign indicates a positive effect of a particular family structure characteristic on the specified outcome in the full model, a negative sign indicates a negative effect, and the absence of a sign indicates that there was no statistically significant relationship. (For the set of cognitive outcomes, we indicate whether the family structure variable had a significant association with two out of three outcomes.) For white children, mother’s prior characteristics sharply attenuate the association between the number of family structure transitions and cognitive achievement, as measured by the three subscales of the PIAT: mathematics ability, reading recognition, and reading comprehension. These results provide support for the selection hypothesis. But mother’s prior characteristics only mildly attenuate the association of the number of family structure transitions with mother-reported externalizing behavior problems and self-reported delinquent behavior. These results suggest that a portion of the association between family structure instability and children’s behavior in middle childhood and early adolescence may well be causal.

Table 5.

Summary of Associations of Family Structure Characteristics with Indicators of Chlid Development

	Mother-only household in first 4 years		Current residence w/mother only		Current residence w/mother and stepfather		Number of transitions
	White children	Black children	White children	Black children	White children	Black children	White children	Black children
Externalizing behavior	+	+		+			+
Cognitive achievement	−
Self-reported delinquent behavior							+

⁺positive association at p<.05 or higher;

⁻negative association

Why might cognitive development be weakly associated with family structure instability when mothers’ antecedent characteristics are accounted for, while children’s behavior may be causally related to instability? Our interpretation is necessarily speculative, but we might conclude that the components of parent-child interactions that are associated with cognitive development are less influenced by instability than are those components that are associated with children’s behavior. In addition, attendance at school may provide a consistent influence on cognitive development that is not disrupted by family transitions unless the child changes school. Alternatively, children’s cognitive development may be more strongly influenced by fixed genetic factors that do not respond to environmental change, while behavior is shaped more strongly either by a child’s environment, or by interactions between a child’s genetic endowments and social environment.

We also conducted some supplementary analyses that are available on the ASR web site. We estimated separate coefficients for the number of transitions in which an adult partner entered the household and the number of transitions in which an adult partner left the household. Only the coefficient for the number of entering transitions was statistically significantly related to child outcomes (and only for whites): the more entrances, the worse the behavioral outcomes. While this result may seem counterintuitive, it is consistent with our theoretical argument that a change in the household disrupts a family system and that the introduction of a stepparent could therefore be disruptive to a single-parent-and-children household. Since this is a serendipitous finding, however, we urge caution in its interpretation until it is confirmed or disconfirmed by future studies. In addition, we estimated all of the models presented in this article for children with Mexican-origin mothers only. As was the case for the black sub-sample, the number of transitions was not significantly related to child outcomes; but unlike the black sub-sample, other family structure measures were not significant either. We cannot determine whether these results are due to the small number of Mexican-origin women in the survey (N=261 in our subsample) or due to a real lack of effects of family structure. Moreover, we would repeat our caution that the Mexican-origin subsample is not representative of the current Mexican-origin population in the United States.

Our findings expand upon some other studies. A well-known study of white boys in a single U.S. metropolitan area found that accounting for mother’s prior characteristics explained the association between family structure instability and children’s adjustment in sixth grade (Capaldi and Patterson 1991). That study measured children’s adjustment as a single construct reflecting academic failure, delinquent behavior, and depression. Our study offers a refinement by suggesting that selection mechanisms may explain the association between instability and children’s cognitive achievement, but that both selection and causation may operate to explain the association with children’s externalizing and delinquent behavior. In addition, our results are consistent with a behavioral-genetic study of children in biological and adoptive families that concluded that differences in academic achievement between children in divorced and two-parent families appear to have a substantial genetic component, whereas differences in psychopathology appear to be environmental in origin (O’Connor et al. 2000). Although we did not test a behavioral-genetic model, we did find that mothers’ AFQT score and educational attainment before the children were born accounted for much of the apparent effect of instability on children’s cognitive achievement.

We found differences by race in the association of family structure with cognitive and behavioral development. Before accounting for mother’s prior characteristics, we observed a consistent association between family instability and white children’s behavior problems and cognitive achievement, but we found no such association for black children. Although this racial difference has not generally been noted in the literature, it is consistent with Wu and Thomson’s (2001) analysis of racial differences in the effects of family instability on early age of first sexual intercourse. It is also consistent with research that suggests that the unique impact of family conflict may be attenuated for black children, who, in comparison to white children, more often have extended kin available for emotional support and who may experience conflict in the context of more stressful circumstances and events(McLoyd, Cauce, Takeuchi, and Wilson 2000).

Differences in the effect of instability by race also may result from our sample restrictions. We have considered only children born to women who were between 21 and 38 years old at the child’s birth. The age range we selected is somewhat skewed for black women, who tend to begin childbearing at younger ages compared to white women. Our finding that instability is a relatively unimportant dimension of family structure for black children’s behavioral and cognitive development should be considered in light of the fact that the black children in our sample may have been higher-order births, or they may have been lower-order births born to women who began having children relatively late compared to other black women.

For both white and black children, we found a persistent association between living in a mother-only household during the child’s first four years and mother-reported externalizing behavior problems, and for white children, reading recognition. The magnitude of the association with living in a mother-only household in early childhood exceeds the association with the number of transitions for most children: a white child would have to experience two family structure transitions in order for instability to have a comparable effect on externalizing behavior problems. Current family structure also is also associated with children’s behavioral and cognitive development. Residing in a mother-only household at interview (compared to living in a two-parent household) is associated with an increase of more than one-third of a standard deviation in externalizing behavior problems for black children, and living with a stepfather is associated with a similar reduction in white children’s reading comprehension scores. These results suggest that researchers should not assume that the effects of instability dominate those of long-term or current residence in a single-parent family. Still, we found that self-reported delinquent behavior was more strongly associated with the number of transitions a child had experienced than with past or current family structure.²⁴

Cohabitation, non-marital births, divorce, and remarriage have transformed family structure in the United States during the last half-century, and the emergence of new family forms has altered both the institutional utility and the symbolic significance of the family (Cherlin, 2004). A well-established body of research has considered how the transformation to a society in which children spend a substantial amount of time in single parent families and stepfamilies affects their development. Less attention, however, has been paid to instability and change per se. We find that, for whites, the number of transitions is negatively associated with children’s behavior, although not their cognitive achievement, controlling for numerous measures of selectivity. These results suggest, although they do not prove, that instability may have a causal effect on some children’s behavior. This association is not necessarily as large as the association between living in a single-parent family and child outcomes, but it is noticeable nevertheless. Each break-up, divorce, remarriage, or new cohabitation brings with it a period of adjustment as parents, partners, and children establish ways to relate and to define their self-identity in a new setting. In the past, sociologists have evaluated the effect of multiple changes on individual and family well-being in the context of the geographic, class, and occupational mobility that have historically defined American society. Goode (1963), for example, argued that nuclear families, rather than extended families, fit the needs of industrial society for mobile supplies of labor. In these contexts, mobility has been seen largely as beneficial. But the consideration of instability within families draws attention to another dimension of transitions in the lives of children and families with potentially harmful and far-reaching consequences.

Biographies

Paula Fomby is an associate research scientist in the Department of Sociology at Johns Hopkins University. She directs the survey component of the Three-City Study, a project to evaluate the well-being of children and families in the post-welfare reform era. She earned her Ph.D. in Sociology at the University of Wisconsin-Madison in 2001.

Andrew J. Cherlin is Benjamin H. Griswold III Professor of Sociology and Public Policy at Johns Hopkins University. In 1989–1990 he was Chair of the Family Section, and in 1995–1996 Chair of the Population Section, of the American Sociological Association. In 2003 he received the Distinguished Career Award from the Family Section of the American Sociological Association. In 1999 he was President of the Population Association of America. He has received a Merit Award from the National Institute of Child Health and Human Development to support his research on the effects of family structure on children; and in 2005–2006 he was a Guggenheim Foundation Fellow.