Parental Disruption and Adult Well-Being: A Cross Cohort Comparison

Although many studies examine the link between parental divorce and child well-being, some theories of the effects of divorce suggest that the negative associations should have declined over time. However, few studies have examined the extent to which the associations have remained stable over time. Using data from two British cohorts, we analyse both shorterand longer-term outcomes of children who experienced a parental divorce and the extent to which the associations have changed over time. Estimating similar models for both cohorts, we find little evidence of any change in the size of the relationship as divorce became more commonplace. INTRODUCTION Amongst British children born since the 1950s divorce has replaced death as the main cause of family disruption and rising divorce rates have led to increases in the proportions of children that have experienced the break up of their parents’ marriage. There is now a substantial body of research that has demonstrated for a range of nations that children whose parents divorce are more likely to be disadvantaged on a range of childhood, adolescent and adult outcomes (for reviews see Amato and Keith 1991a, 1992b; McLanahan and Sandefur 1994; Rogers and Pryor 1998; and Sigle-Rushton and McLanahan 2004). In the British context, statistically significant associations of poor outcomes with parental divorce have been reported using studies of individuals from three different birth cohorts born in 1946, 1958 and 1970 (see for example, Wadsworth and MacLean 1986; Kiernan 1992; Cherlin Kiernan and Chase-Lansdale 1995; Kiernan 1997). While results for individual cohorts are well-documented, very few studies have examined the extent to which the associations have remained stable over time. This is an important issue to explore because some well-cited theories of the effects of divorce suggest that the negative associations should have declined over time. For instance, as divorce becomes more commonplace, the selection hypothesis would suggest that the average child of a divorced family would come from less problematic families. As a result, the bias due to omitted variables (third variables that are associated with both parental divorce and subsequent disadvantage) should be lessened. In addition, the liberalisation of divorce law and a greater reliance on mediation are both likely to have reduced family conflict and stress. Family conflict and stress have both been strongly implicated as correlates of both childhood and adolescent problems (Amato and Keith 1991a; Morrison and Cherlin 1995). Taking a wider perspective, as alternative family structures have become more widely accepted, divorce has been accompanied by less stigma and any negative effects of community disapproval should have lessened over time. Finally, with increased family breakdown came increasing levels of information on the effects of divorce on children. Consequently, parents may have become more aware of the potential difficulties and tried to mitigate the effects of separation on their children. Motivated by these considerations, we seek to determine whether there is evidence for the hypothesis that parental divorce has become less strongly associated with poor child outcomes during childhood and adulthood as divorce has become more commonplace. To date there is mixed evidence on whether the impact of divorce on children has changed over time. An early review of the literature reported that divorce effect sizes had decreased when early and older studies were compared (Amato and Keith 1991), and more recent evidence suggests that the positive relationship between parental divorce and own divorce have both attenuated over time (Wolfinger 1999). But other recent studies find no changes over time in the associations of father absence with educational attainment and occupational status. Indeed, recent studies suggest that the effects of family structure on, for example, educational attainment have remained fairly constant in both the United States and the United Kingdom from the 1960s to the 1990s (Ely et al. 1999; Biblarz and Raftery 1999). In this study we make use of data from two British cohorts, born in 1958 and 1970 and followed up from birth into adulthood, to analyse the life experiences and chances of children who experienced a parental divorce and the extent to which the negative effects of parental divorce have remained stable or not over time. Although born only twelve years apart, the social contexts in which the two cohorts grew up were very different. Most pertinent for this study, the divorce rate climbed dramatically throughout the 1970s and the cohort born in 1970 were far more likely to see their parents divorce and to grow up in a social environment where divorce was more common and alternative family structures increasingly condoned. The prospective nature of the data allows us to examine whether the impact of divorce on behavioural and academic attainment during childhood has changed over time; as well as whether the size of the effects for adult well-being (at age 30 in the case of the 1970 cohort and age 33 in the case of the 1958) in terms of educational qualifications attained, being on welfare benefits and emotional well-being have changed. Additionally, these longitudinal data allow us to start with a sample of children whose parents were still together at the first childhood follow-up and control for characteristics that predated family disruption and thus control for some selection effects. METHODS Data This study uses data from the National Child Development Survey (NCDS) and the British Cohort Study (BCS), two nationally representative, longitudinal studies of birth cohorts in Great Britain. Both studies are similar in design; the NCDS study follows the lives of a cohort of children born in one week of March in 1958, and the BCS study follows another cohort born in one week of April in 1970. Both original samples provide information on over 17,000 births, and the data collected at the baseline provides detailed information on the provision of anteand post-natal services for the study of perinatal and infant mortality. Both studies have followed the cohorts over time and later waves, although not always interviewing children at the same ages, are similarly designed and include a broad range of socio-economic, demographic, health and attitudinal measures (Despotiduou and Shepherd 1998). The first two follow-up interviews took place when the NCDS children were aged seven and 11 and the BCS children were aged five and 10. Although the studies do not use exactly the same instruments, there is a good deal of overlap in the information provided. At both waves, detailed information on the child’s behaviour was obtained from the mother (and from teachers at age seven and 11 in the NCDS and at age 10 in the BCS), and information on the child’s academic ability was assessed using standardised tests. Parent and household information was also collected at these childhood waves. In addition, both cohorts were interviewed as adults; the fifth NCDS follow-up interview was conducted when cohort members were aged 33, and the fifth BCS follow-up interview was conducted in 2000 when the cohort members were aged 30. These data provide a unique opportunity to research two groups of children born twelve years apart and to explore changes in the association of family disruption with subsequent disadvantage over a period when family breakdown was becoming more commonplace. The length of time between interviews means that sample attrition, particularly at older ages, is inevitable. The age 30 BCS response rate -defined as the number of achieved interviews divided by the initial sample of cohort members -was 69.9% (Collins et al 1 A third childhood follow-up interview was conducted when both cohorts were aged 16. 2 NCDS cohort members were also interviewed at age 23 and 42, and BCS cohort members were interviewed via a short postal questionnaire at age 26. 3 By the age 30 interview, 1.6% of cohort members were classified as permanent or proxy refusals, 1.2% had emigrated so were not contacted for interview, and 0.6% had died. These individuals (along with a small number who were found to have a birthday outside of the 2001). When NCDS cohort members were interviewed at age 33, the response rate was similar. Out of an initial sample of 16,455 cohort members known and thought to be alive, 11,363 were successfully traced and interviewed. Comparisons show that the achieved samples do not differ a great deal from other survey samples of the British population although there is a slight under-representation of the most disadvantaged groups (Fogelman 1983; Shepherd 1997). Sample Selection For this analysis we restrict each sample to include only those children who were living with both biological parents at the time of their first follow-up interview (wave 1). For our examination of age 10 or 11 outcomes, we further restrict the sample to those whose parents provided information on their family structure at the second follow-up interview (wave 2). Similarly, when we examine adult outcomes, we restrict the sample to those individuals for whom we can determine whether or not there was a parental divorce or separation before the age of 17. The latter information was collected for both cohorts at the third follow-up interview, administered at age 16 for both samples. In addition, members of the BCS cohort were asked, at the age 30 interview, about any family disruptions that occurred during childhood. These additional questions are used when the age 16 information is not available. Members of the NCDS cohort were asked to provide only limited information on family disruptions at age 33, but at their sixth follow-up at age 42, they were asked to provide detailed information. In order to maximise our samples, when information on family structure at age 16 is missing, we use information collected at ages 33 and 42 to determine survey reference week) are not included as part of the initial sample in the calculation of the response rate (Collins et al 2001). whether or not a family disruption occurred. Because we are interested in parental dissolution and its association with indicators of disadvantage, we eliminate from the sample those individuals who experienced a parental death. The restrictions outlined above do not reduce the achieved samples appreciably. For the NCDS sample we have 14,707 children with information on their family structure at age 7, 13,520 (91.9%) of whom are living with both natural parents. Of these, 1,998 children are missing information on their family structure at the second follow-up at age 11. This leaves 11,522 children with information on their family structure. After removing from the sample those children who are living in foster care, who experienced a parental death and who are missing information on some of the control variables (we remove from the sample the small number of individuals who were missing information on their father’s occupational class and their parent’s housing tenure at age 7), the NCDS sample totals 11,225 children. For the first BCS follow-up at age five, there are 13,135 children with information on their family structure, 11,849 (90.2%) of whom are living with both natural parents. Of these, 1,499 children are missing information on their family structure at the second follow-up at age 10. Most of these 1,499 are missing because their families were not interviewed at wave 2. There are few missing cases as a result of non-response. This leaves 10,350 children with information on their family structure. After removing from the sample those children who were living in foster care, who experienced a parental death or who were missing information on some of the control variables, our BCS sample includes 10,162 children. Of the 13,520 cases in the NCDS sample that were identified as living with both natural parents at the first follow-up wave, 9,503 (about 70%) have information on family disruptions up to age 16, and are not eliminated as result of parental death or missing information at wave 1. For the BCS sample, the percentages are somewhat higher. Of the 11,849 individuals who were reported at age five to be living with both natural parents, 9,242 (about 78%) have information on parental separation or death up to the age of 16 and are not eliminated because of a parental death or because they lack information on some of the wave 1 control variables. Unfortunately, for both sub-samples, not all cases have complete information for all of the outcome variables that we consider. Rather than further restrict our samples, we opt to include as many valid observations as possible. This means that our sample sizes fluctuate and are not entirely comparable, but the gains from using as much valid information as possible most probably exceed the benefits of using entirely identical but smaller (and potentially less representative) samples. Restricting the samples to those children who were living with both natural parents at wave 1 allows us to obtain baseline information on child and family characteristics for a group of children from non-disrupted families, some of whom later experienced a family dissolution and some of whom did not. Because we are interested in both the short and long term associations, we examine both short-term, pre-adolescent outcomes (measured at age 11 for the NCDS sample and at age 10 for the BCS sample), and longer term adult outcomes. Using logistic regression, we examine the associations of parental disruption, controlling for a range of pre-disruption antecedents. Dependent variables Child well-being. In order to assess the well-being of children, we use measures of the child’s temperament and academic success at age 11 for the sample drawn from the NCDS cohort and at age 10 for the sample drawn from the BCS cohort. For temperament, we examine three measures each of which is constructed using parental responses to a series of questions concerning their children’s behaviour. A battery of questions, devised by Rutter and colleagues (1970), was asked at the first and second follow-ups of both studies. In most cases, parents were provided with a series of descriptions and asked to report whether each description certainly applied, somewhat applied, or did not apply to their child. Although both the wording and coding of the inventory was somewhat different for the BCS cohort at age 10, we have attempted to define our categories at age 10 in as meaningful and consistent a way as possible. Nonetheless, the differences between the two inventories are substantial enough that readers should interpret differences across samples with some caution. Following Hobcraft (1998), we group 11 items into three categories. We use parental assessments of how often the child fights with other children, is irritable, is destructive, and is disobedient to construct a measure of “aggression”. We use parental reports of the extent to which their child is a worrier, a loner, miserable or tearful, and afraid of new situations to construct an “anxiety” measure. Finally, characterisations of the child as being squirmy or fidgety, having twitches or mannerisms, and having difficulties concentrating are used to construct a “restlessness measure”. 4 At age 10, parents were given, for each description, a line with “certainly applies” at one end and “does not apply” on the other. They were then asked to “...make a vertical mark though the line...to indicate the extent to which the statement applies to your child’s behaviour”. Where the mark fell on the line was then coded into a scale from 199 with 99 being the most extreme agreement with the statement and 1 being the most extreme disagreement with the statement. We have divided the 1 99 scale into thirds corresponding to the certainly applies, somewhat applies and does not apply categories. Each item was coded on a scale of 0 to 2 with 0 meaning not applies, 1 meaning somewhat applies, and 2 meaning certainly applies. Within each group, the items were summed together to create three overall scores ranging from 0 8 for the aggression and anxiety scores and 0 6 for the restlessness score. We then classified each sum as low, moderate, high or missing. For aggression and anxiety, a sum total of 0 or 1 was coded as low, 2 or 3 was coded as moderate, and greater than or equal to 4 was coded as high. In the case of restlessness, a sum of zero was coded as low, 1 or 2 was coded as medium, and greater than 3 was coded as high. Our three temperament outcomes are constructed as indicator variables that equal one if the child scores high on aggression, anxiety or restlessness, respectively. The distributions of these outcomes are presented for both samples in Table 1. Although the distributions are similar across cohorts, the NCDS sample has a slightly higher percentage of cases with high scores for anxiety and restlessness. In both samples, 12% of the cases are coded as having high aggression scores and high anxiety score is the most common of the three outcomes (26% of the NCDS sample and 20% of the BCS sample have a high anxiety score). We use scores on standardised academic tests to measure academic achievement at the second follow-up interviews. For both cohorts, the outcome is constructed by combining a verbal test score and a quantitative test score, but because the students were not administered the same test, cohort differences should be interpreted with caution. Each test score was standardized to have a mean of zero and a variance of 1, and the two standardized test scores were then added together. Poor academic achievement is measured with an indicator variable that equals one for those children whose test scores place them in the bottom quartile of the distribution of scores (for the full sample of children with test scores at age 11 or 10, depending on the sample). Frequencies for this outcome are presented in Table 1. For both cohorts, our restricted sample is slightly less likely to have had low scores suggesting a small amount of positive selection on academic performance as a result of our sample restrictions. Longer-term adult outcomes. We measure socio-economic and psychological wellbeing in adulthood using three indicator variables, the distributions of which are presented in the lower half of Table 1. To measure poor academic success and labour market preparation, we use an indicator variable that is set equal to one for those cohort members who lack any academic or vocational qualifications at age 33 or 30 for the NCDS and the BCS samples, respectively. In Table 1, we see that, across cohorts, a lack of educational qualifications became less common over time (Bynner and Joshi 2002). In 1991, 12% of the NCDS sample reported having no qualifications, and in 2000, just 7% of the BCS sample reported having no academic or vocational qualifications. Economic well-being and instability is measured using an indicator that equals one if the cohort member reports being in receipt of non-universal (means tested or targeted) benefits when they were interviewed in the fifth follow-up survey. Benefit receipt is relatively common in both samples and the frequencies are similar across the two cohorts. About 16 percent of the NCDS sample reported receipt of non-universal benefits at age 33 and 14 percent of the BCS sample reported receipt of non-universal benefits at age 30. Mental health and well-being is measured using the Malaise Inventory, a 24-item battery of questions designed to identify those individuals at high risk of depression (Rutter et al 1970). The items cover a range of symptoms associated with depression, and, similar to previous work, we classify those individuals answering yes to at least seven of the 24 items as being at high risk of depression (Richman 1978; Rutter et al 1976). As documented elsewhere, high malaise has become more common in the younger cohort (Bynner et al 2002). Individuals in the BCS sample are much more likely to have a high malaise score at age 30 than were individuals in the NCDS sample at a similar age.