CONTACT: Allen Li, Associate Director of the Population Research Center at the RAND Corporation; firstname.lastname@example.org; 310.393.0411 x 6694
This paper summarizes the argument and findings of a longer and more technical paper that won the 2007 Graduate Student Paper Award in Social Demography from the Section on Population of the American Sociological Association.
Many research studies have shown that, on average, children of divorce have more behavior problems than children growing up in two-parent families. But the question for social scientists is whether the problems seen in the children of divorced parents were caused by the divorce, or whether something else caused BOTH the divorce and the children's problems.
Researchers wonder, in particular, whether some couples have personal characteristics and/or parenting patterns that increase the chance that their children will have behavior problems AND ALSO increase the chance that the couple will be unable to resolve marital issues. If this "something else" causes both divorce and behavior problems, then it is likely that that children would still have had problems even if their parents had somehow managed to remain married.
How do we look for that "something else"? We know that it is a mistake to compare children of divorced parents with children of continuously-married parents without taking into account differences between divorcing families and continuously married families PRIOR to the marital disruption. Parents who are more likely to divorce may also be more likely to be impoverished, to live in disadvantaged neighborhoods, to be less educated, to have been raised in divorced families themselves, or to have more children than average. These factors may impair a child's well-being whether the parents stay together or not, but also be more likely to produce a marital disruption.
To test the effect of pre-existing family characteristics versus the effect of divorce itself, prior studies have used statistical analysis to "control" for the differences we can see between divorced and continuously-married families prior to the disruption. This is done by taking into account the socioeconomic status of the parents, their race or ethnicity, and other "variables" that can be determined by having respondents fill out a paper or computerized questionnaire. Some studies also take into account prior differences in child well-being between the two types of families prior to the disruption. The old consensus is that taking these pre-existing factors into account helps explain some of the association between parental divorce and children's behavior problems - but not all. It reduces the average difference between the two groups but still leaves some average deficits for children of divorce, deficits that are not explained by controlling for these observed differences.
But what about the unique characteristics of each family that we do not as yet have the tools to measure? Things such as personality, parenting strategies, and detailed aspects of a person's biography all affect children, but researchers haven't been able to measure many of these constructs, far less to include them in large-scale studies. Therefore many studies end up comparing apples and oranges. The proper test of the impact of divorce on children is not to compare the children of divorced parents to the children of continuously-married families, and thus risk ignoring all the unobservable factors that may lead both to greater behavioral problems and to higher chances of divorce. It works better to compare the behavior problems of the same child before and after divorce. So, traditional methods often do not adequately estimate the impact on children of being in a family that is headed for divorce.
Several recent studies, including one of my own, which use more advanced and sophisticated research methods, present a powerful challenge to the old consensus that the average impact of divorce on children is negative. These studies are able to eliminate the impact of both "observable" and "unobservable" family differences that result in variations in child outcome, independent of divorce, and this provides a more accurate estimate of the "true" impact of divorce.
All these new studies have discovered the same thing: The average impact of divorce in society at large is to neither increase nor decrease the behavior problems of children. These studies suggest that divorce, in and of itself, is not the cause of the elevated behavior problems we see in children of divorce. They include Aughinbaugh, Pierret, and Rothstein (2005), Foster and Kalil (2007), and Li (2007).
While previous studies have compared the outcomes of children whose parents divorced to those of children whose parents remained together, I use a longitudinal study that measures changes in the behavior of children whose parents were not divorced at the beginning of the study but who divorced later. This allows me to investigate the counterfactual question, "What would have happened to the children's behavior if their parents had remained married?" For an example of how this method works, and why other methods tend to over-estimate the impact of divorce on children's behavior problems, see the Appendix at the end of this report.
The data I used included all children born to a national representative sample of American women born between 1958 and 1965. These same women had been surveyed repeatedly since 1979, and their children had been surveyed since 1988. Forty-seven percent of these mothers in my sample had been divorced by 2002. I used a 28-item checklist to measure behavior problems for children between 4 and 15 years of age. Mothers in each of the biennial survey filled out a questionnaire about whether their child engaged in behaviors such as cheating, deliberately breaking things, crying or arguing frequently and so forth. The mother of an average boy reported 8.7 items and the mother of an average girl reported 7.8 items that are often or sometimes true.
My study included a national sample of 6,332 children. It revealed that the estimated effect of a parental divorce on children's behavior problems is so small that fewer than half of the divorced mothers would observe a one-item increase in the 28-item BPI checklist of their child. This is not a statistically significant effect.
Why would I get this result when other carefully-constructed studies, which controlled for observed differences, found larger, and statistically significant effects of divorce? The kind of observed differences that show up in surveys may fail to catch subtle differences between families that eventually divorce and those who do not. For example, certain aspects of child temperament and behavior are associated with parental personality traits that may be hereditary. If a child has parents with difficult temperaments and divorce-prone personality traits, the child will likely exhibit greater behavior problems whether or not the parents divorce, but the child will also be exposed to a higher risk of parental divorce. Or take the fact that the resources parents are able and willing to provide for their children may vary dramatically across marriages and across divorces. If so, there may be "good" parents and "bad" parents, as well as "good" spouses and "bad" spouses. It is plausible that a "bad" spouse may well have been a "bad" parent prior to marital disruption (and may, thus, have been a factor in causing the disruption).
Disengaged or unloving parents are detrimental for children's emotional well-being and behavior. The lack of love on the part of one or both parents may increase the chance that the parents will divorce, but it may also create behavior problems in children whether or not their parents divorce. If so, we should not attribute the worse behavior of their children to the divorce itself, but to the impact of the unloving parent or parents. The point is that "bad" marriages are more likely to harm children's well-being than good ones AND more likely to lead to divorce, and a marriage can be "bad" in many unobserved ways.
I am not saying that divorce doesn't increase the behavior problems of some children, because I have focused only on the "average effect of divorce for the divorced." It is possible that the dissolution of some marriages decreases some children's behavior problems and the dissolution of others increases children's behavior problems, so that they cancel each other out, creating the zero effect that I found when I totaled the average effect of divorce. However, for this to be true, one must admit that while certain divorces harm children, others benefit them. My findings contradict the widely-accepted claim that MOST divorces increase children's behavior problems and that only a tiny minority of divorces do NOT.
It should be noted that my findings are only relevant to the kind of marriages where parents have qualities that make them likely to divorce. They should not be interpreted to imply that breaking up a randomly selected marriage in society would not lead to increased behavior problems for the children. But these findings do imply that to help children of divorce, social scientists and policy-makers should seek to understand and intervene in the processes both before and after a marriage comes apart, rather than seeking to simply prevent the divorce from occurring.Works Cited
Aughinbaugh, A., C.R. Pierret, and D.S. Rothstein. 2005. "The Impact of Family Structure Transitions on Youth Achievement: Evidence from the Children of the NLSY79." Demography 42:447-68
Foster, EM. and A. Kalil. 2007. "Living Arrangements and Children's Development in Low-Income White, Black, and Latino Families. Child Development 78:1657-74
Li, J-C. A. 2007. "The Kids are OK: Divorce and Children's Behavior Problems." RAND Labor and Population Working Paper, WR-489. Santa Monica, CA.
Appendix: How did I get my results?
Fixed-effects modeling, which is often used by economists, also helps us to understand the complex behavior of people in families.
To understand my method, consider two children of divorced parents, where the divorces occurred when they were both age 9. Suppose we measure their behavior problems once every two years, as I in fact did with a much larger sample. Here is a stylized example to illustrate conceptually how my method and the traditional method would yield different results in assessing the impact of divorce on children's behavior problems:
Age 4 6 8 10 12 14
Kid A 7 7 7 9 9 9 (parents divorced at age 9)
Kid B 8 8 8 8 8 8 (parents divorced at age 9)
For Kid A, the effect of divorce is a 2-item increase on the behavior problem index; for Kid B, the effect is 0.
Now consider two more children whose parents are continuously married.
Age 4 6 8 10 12 14
Kid C 6 6 6 6 6 6 (parents not divorced, up to age 14+)
Kid D 4 4 4 4 4 4 (parents not divorced, up to age 14+)
The traditional estimate of the effect of divorce would take the average behavior problems of Kids A and B (which is 8), and the average of Kids C and D (which is 5), and then calculate the difference between the two averages. The difference between the kids in the divorcing group and the kids in the non divorcing group would then be 3-item increase.
The new way of looking at this is called a fixed-effects estimate. Instead of comparing the divorced kids to the kids from married families, I compare them to themselves, before and after parental divorce. I do this by averaging the pre- and post-divorce differences for Kids A (9-7=2) and B (8-8=0). Hence, the fixed-effects estimate of the effect is a 1-item increase in behavior problems, a much lower estimate of the impact of divorce.
Or consider another child, whose behavior had begun to deteriorate before the divorce (again at age 9) and continued to do so afterwards:
Age 4 6 8 10 12 14
Kid E 6 7 8 9 10 11
Notice that between age 4 and 8, prior to the divorce, the child's behavior problem went up by one every two years. Had the parents avoided the divorce, we would have expected that the child's behavior problems continued to increase at the same rate. In other words, if the trajectory of a child's behavior problems stayed its predisruption course after the divorce, we should not claim that there is any impact of divorce because that is what would have happened had the parents remained married. This example also illustrates another subtle point that even the previous fixed-effects estimate may overstate the impact of divorce if the unobserved factors operate in a way that changes the level of child well-being but does not alter the trajectory in child well-being. Consider how we calculate the fixed-effects estimate by taking the difference between the pre- and post-divorce averages: The estimated effect will be (9+10+11)/3 - (6+7+8)/3 = 10-7 = 3 under the fixed-effects specification, whereas under the "random-trends" specification controlling for dynamic selection [ugh, technical], the estimated effect is 0.
Allen Li presented more of his findings at the 11th annual conference of the Council on Contemporary Families, in a panel entitled "Should They Stay or Should They Go?" The conference, "Family Issues in Contention," was held on April 25th and 26th at the University of Illinois at Chicago.
RESPONSES AND OTHER PERSPECTIVES
INTERVENTIONS CAN HELP COUPLES AND CHILDREN
Philip A. Cowan, Professor of Psychology, Emeritus, University of California, Berkeley
Li's paper remind us that the potential negative impact of divorce on children always has to be evaluated in light of the negative impact of distressed relationships on children of couples who stay together. Philip Cowan and Carolyn Cowan have completed three longitudinal studies that focus on couple relationships and children's development. One of the main findings of these studies is that in both married and cohabiting couples, high, unresolved conflict or relationships low in warmth and communication play a detrimental role in the development of children from ages 5-14. But for many couples, these negative patterns can be lessened. In our studies, preventive interventions for parents not showing major distress at the beginning have resulted in long-term benefits for mothers and fathers and also for the children.
A DIFFERENT SET OF FINDINGS WITH TWIN STUDIES
Robert E. Emery, Professor of Psychology, Director of the Center for Children, Families, and the Law, University of Virginia
Divorce does not occur at random, and Li is correct that when we take characteristics that increase the risk for divorce into account, this reduces differences in psychological adjustment when comparing children from married and divorced families. My colleagues and I have also been concerned about these "selection" effects, so we have studied thousands of twins (and their children) in several investigations to control not only for family and social background but also for genetics. Our twin design also reduces the presumed effects of divorce on children, but unlike Li's findings, differences between children from married and divorced families remain statistically significant. Of equal importance, in other research we have found that, while children from divorced families generally do not suffer from psychological *problems,* they do experience much more extensive psychological *pain,* including difficult memories and ongoing concerns about the fallout of divorce. For example, graduations and weddings can be turned into anxiety-ridden events for children whose parents are divorced and still do not get along. On an even more painful note, many young people whose parents have divorced also report doubting their fathers' love, a particularly painful experience.
WHAT HAPPENS AFTER DIVORCE ALSO COUNTS
Constance Ahrons, Professor of Sociology, Emerita, University of Southern California
This study adds to similar findings showing that, on average, children's behavior problems usually attributed to divorce actually pre-date the parental divorce. But the study should not be interpreted to say divorce has no effect on children as and after it occurs. Rather it shows that divorce must be understood in a broader context, by highlighting the systemic process of which the divorce is but one point in time. The findings from my Binuclear Family Study, a 20-year longitudinal study of 98 post-divorce families, clearly show that what mattered most to children was how their parents got along after the divorce. Interviews with 173 children 20 years after their parents' divorce revealed that when divorced parents were able to maintain a civil, low-conflict and sometimes cooperative relationship with one another, the children experienced no long-term problems associated with the divorce. However, when parents remained embroiled in conflict or totally disengaged from one another, their children continued to be distressed even 20 years later. We need to stop blaming "the divorce" for children's problems and start helping parents understand how their behavior before and after divorce impacts their children, and learn how they can reduce their children's stress.
USE CAUTION IN INTERPRETING THE ORIGINS OF PRE-DISRUPTION EFFECTS
Virginia Rutter, Professor of Sociology, Framingham State College
The fixed effects model Li uses suggests that predisruption effects are the only factor in why some small but significant percentage of children of divorce do worse than children in continuously married families--at least through age 14. But we don't know what those predisruption effects are. Li speculates that it is personality or hereditary factors. In looking at the detailed examination of what causes marital distress that leads to divorce, it makes more sense to me to consider how complex challenges that people face in their lives--challenges that stem from the environment--may help to account for those predisruption effects. A complex series of negative life events and stresses in one's biography may make both parenting--and getting along with one's partner--more challenging. I don't think we have cause to say there is a "divorce gene" or some people who are just "bad spouses" or "bad parents": a host of sociological and psychological research helps us piece together how all the complicated details of one's biography can lead to some sad situations in divorcing families, even for parents who are doing their best.
ASSESSING THE COSTS AND BENEFITS OF DIVORCE
Betsey Stevenson and Justin Wolfers, Professors of Business and Public Policy, The Wharton School, University of Pennsylvania
In assessing the impact of divorce, disentangling causation from correlation can be a difficult, and sometimes impossible, job. For instance, we know that divorced people are more likely to drink, use drugs, have lower income, and are less happy. But did divorce cause these bad things or did these bad things in fact cause the divorce? For example, when a man is laid-off, the couple is more likely to divorce in subsequent years. Other research has shown that the happiest people are the most likely to get married, while those divorcing are less happy than the average married person. Moreover, people who divorce are actually happier a few years after the divorce compared with their happiness in the years prior to the divorce.
Li's paper argues that much research on divorce overstates its personal costs for children. On April 15, a coalition of family values groups released a report that overstated the economic costs of divorce to taxpayers by adding up the costs, while ignoring the benefits. The report estimated that divorce and unwed motherhood costs the U.S. taxpayer $112 billion per year. But the research by Ananat and Michael (2008) that they rely on actually shows that divorce helps the financial situation of almost as many women as it hurts, and among those who gain, the gains are larger than the losses among those who lose. This report counts up the costs to the taxpayer from the women who lose income, but refuses to count even a single dollar of the rise in taxes or the decrease in poverty linked to those who gained income. To assess the consequences of divorce we need to consider the benefits as well as the costs. Our research has shown that making divorce easier through the implementation of unilateral divorce laws led to a 8-16 percent decline in female suicide, roughly a 30 percent decline in domestic violence for both men and women, and a 10 percent decline in females murdered by their partners.
WHY WE NEED LONGITUDINAL STUDIES
Jennifer Glass, Professor of Sociology, University of Iowa
Dr. Li's work confirms that divorce is not a randomly occurring event among married couples, but the consequence of hard-to-measure interactions and behaviors that can impact children both before and after their parents separate. The beauty of longitudinal evidence is that it can help illuminate the cause of children's behavior problems where classic experimental evidence is impossible to obtain (an experimenter can't randomly assign couples to divorce or stay together and then watch what happens with their children's subsequent behavior).
About CCF: The Council on Contemporary Families is a non-profit, non-partisan organization dedicated to providing the press and public with the latest research and best-practice findings about American families. Our members include demographers, economists, family therapists, historians, political scientists, psychologists, social workers, sociologists, as well as other family social scientists and practitioners. Founded in 1996 and based at the University of Illinois at Chicago, the Council's mission is to enhance the national understanding of how and why contemporary families are changing, what needs and challenges they face, and how these needs can best be met.
To learn more about other briefing papers and about our annual April conferences, including complimentary press passes for journalists, contact Stephanie Coontz, CCF's Director of Research and Public Education, at email@example.com.