|When is the Relationship Between Facts a Causal One?|
By Philip Cowan, Professor of Psychology, Emeritus, University of California, Berkeley; email@example.com; and Carolyn Cowan, Professor of Psychology, Emeritus, University of California, Berkeley; firstname.lastname@example.org
Even when we can check the accuracy of facts, as Cherlin's paper urges us to do, the next step is to examine critically the way that people interpret the relationship of one fact to another. It is a fact, for example, that substantial numbers of children are growing up in single parent families. Or, more precisely, it is a fact that many children are growing up in households that do not contain two parents who are married to each other. (Some of these families have only one parent in the home; others may have two parents who are cohabiting.) It is also a fact that, in general, children and couples in non-married families are not faring as well as those in married families. Members of such families have less income and lower levels of physical and mental health, and the children have more emotional problems and behavior problems (Waite & Gallagher, 2000). It is tempting to conclude from these facts that living in a single parent household is the cause of these difficulties.
But when we look at all people who live in single parent households, we find a larger number of people with pre-existing financial, health, or emotional disadvantages than we find in the married-couple population. It may be these characteristics of the adults in the family, rather than single-parenthood per se, that make them less likely to get or stay married and more likely to raise children who exhibit behavior problems. If so, it would be inaccurate to say that divorce or unwed behavior causes child problems. In many cases, the problems seen in children raised by these individuals might still develop even if the parents were able - or were forced - to stay together.
The problem of overstating causal conclusions from correlational data is not the sole property of the political left or political right. Both sides are too quick to draw support from social science research when correlations support their cherished conclusions. Supporters of the political right tend to select studies that show correlations between divorce and negative outcomes for children. Supporters of the political left select studies that show correlations between poverty and negative outcomes for children. The point here is not that these claims are wrong, but that most studies do not provide evidence for the causal assertions and policy conclusions that are made on both sides of the political spectrum.
We need to make a slight digression here. It is always legitimate and possible to make policy arguments on moral or value grounds. That is, if one's values lead to the conclusion that cohabitation is a sin, and that having children compounds that sin, it is part of the bedrock of democracy that one can argue strongly on moral grounds that laws should be made to prevent cohabitation and foster marriage. What we're concerned with here are cases in which individuals claim that conclusions based on their values are "proven" by social science research.Basic problems in the interpretation of research facts
1. Causal facts always imply a direction of effects - the cause, A, comes before the effect, B. But statements based on statistical correlations can never tell us about the direction of effects. For example, it is a fact that there is a correlation between being married and having better-functioning lives and between non-marriage and financial or emotional difficulties. However, we do not know whether marriage produces the partners' better functioning or whether better-functioning partners get married. That is, selection effects guiding who gets married may influence the results.
2. An important corollary of point 1 is that when two social trends vary together, it is not possible to conclude that one causes the other. Increases in the proportion of mothers of young children in the workforce occurred in the mid-20th century around the same time that the divorce rate went up. On the basis of these two facts alone we cannot point to women working as as a cause of the increase in divorce. Why not? First, we don't know from these two statistics whether the divorces occurred more often in the families of women who went to work. And second, we do not know whether these two trends are associated with other factors that may plausibly have caused the increase in divorce.
3. Reasoning backwards about causality produces backward thinking. Most newspaper and magazine articles on family issues rely on research that starts with outcomes of interest right now and looks backwards for potential explanations, because that's what most research does. For example, we take two groups of couples, one in which there are high levels of domestic violence, and another in which there has never been any domestic violence. We look at their histories, and find that the couples with domestic violence are much more likely to have been abused by their parents than the harmonious couples. Does this demonstrate that early abuse is a cause of domestic violence? No.
What's missing from the picture is information from studies that follow families forward. These studies usually find that even if some of the abused children grow up to form violent relationships with a spouse, the majority of children who experience early abuse do not wind up in violent relationships. In this example, even if early abuse were found to be a cause of domestic violence, we might try to change each partner's understanding of the past through psychotherapy, but we cannot reverse the early abuse.
Other examples, though, seem to suggest that if we can identify the cause, a quick fix is possible. We know, for example, that there is a correlation between cohabitation and higher rates of domestic violence, but it would be dangerous to conclude that a causal relation exists and recommend that cohabitors should marry. Rather than their failure to marry producing domestic violence, it may have been their stormy relationship that led them not to marry in the first place. If this were the case, a policy that created incentives to marry could result in increased harm to both the couple and their children.
4. Correlations can result from a third variable that produces the association between them. It is a fact that children whose parents are divorced, or who live with a single parent who never married, tend to have more emotional, behavioral, and academic difficulties than children whose parents are married. It is possible, though, that some of the negative effects of divorce and single parenthood come from the fact that these households have lower income, and that the consequences of low income in terms of reduced resources are responsible, at least in part, for children's difficulties.
5. Many studies of families focus on status and not on process or relationship quality. Most studies of marriage and divorce, especially in Sociology and Social Welfare, attempt to link couple status (married, cohabiting, divorced, single) with child and family outcomes. For example, in Waite and Gallagher's "The Case for Marriage," almost all of the studies they cite contrast married couples with cohabiting couples or single adults. In each chapter, they present evidence that the strongest positive findings occur for happily married couples. But in the policy summary at the end, the authors revert to the argument that "married is better," ignoring the issue of quality altogether.
What the advocates for marriage ignore, or dismiss, are the hundreds of studies showing that high unresolved marital conflict erodes couple relationships and affects children negatively (Cowan & Cowan, 2002; Cummings & Davies, 1994; Emery, 1999; Gottman & Notarius, 2002). Unless we are talking about "good" marriages, getting couples married will not provide a solution either to social problems of poverty or to individual problems of child behavior.
The importance of systematic studies that include randomized clinical trials with control groups
First, we need to be able to determine that whoever is providing the information does not have an interest in consciously or unconsciously skewing the results. For example, if the intervention staff is providing the data, it is easy to see how they might be motivated, or self-deluded, to make higher ratings of the participants on the post-intervention assessments. When program evaluations include data from outside observers, as well as therapists and clients, the inclusion of multiple perspectives makes statements about improvement more credible.
Second, even with a study that contains the most objective, unbiased assessments of outcome, a control group is still needed. You can't just claim that the program is a success if the participants show positive changes. What if the average participant in a job-training program has a statistically significantly higher income a year after the program ends? How can we rule out the possibility that these results come from an economic boom in which most families have higher incomes a year later? That is, the fact of increased income does not support a causal interpretation about the impact of the intervention until we know what happens to a comparable untreated group.
Similarly, what if a group of children become less aggressive with their peers after their parents take a class on managing children's aggression? Again, we need to know whether children whose parents did not take such a class also decreased in aggression as they grew older or whether we can show that the declines in aggression are associated with parents' more effective parenting strategies.
We know that it is not always possible to do controlled experiments. To test the hypothesis that married parents provide a better environment for children's development, we cannot assign some single parents to the "get married" group and others to the "remain single" controls. In this case, there are responsible ways of gathering data to rule out alternative hypotheses so that we can come to a more informed decision about the impact of marriage on children's adaptation. One method is to measure a number of variables that could possibly influence A and B groups differently and "subtract them" from the outcome to see if any effect of the intervention remains. This method is only as powerful as the thoughtfulness of the investigator in thinking about what else, outside of the intervention could have created the obtained results.
A second, more powerful method is do a longitudinal study (e.g., of the same children before and after divorce) and determine whether, on the average, any change in the children can be identified from the period before their parents divorce to the period after. Often these studies use a fixed effect method. For a good example of such a study, which came to quite different findings about the impact of divorce than did earlier studies, see CCF Briefing Report, "The Impact of Divorce on Children's Behavior Problems," (Li, 2008).
Both the political left and the political right have jumped to conclusions in the debate about marriage, based on the erroneous assumption that correlations support causal inferences. From the right we hear: "Married families do better; let's get those single moms married or make it harder for couples to divorce." From the left we hear: "Unmarried mothers are poor, and poor families have difficulty; let's give them money and jobs."
What we need to remember is that explanations of how two facts are connected seem simple but are often exceedingly complex. Unpacking the causal connection requires very thoughtful systematic research, accompanied by interventions, if possible, that test hypotheses about the direction of effects. We are aware that this kind of rigorous exploration takes time, and that policy decisions must often be made in the absence of scientific proof that the proposed action will have the desired effects. What we want to convey to social service providers and policy makers is that causality is extremely difficult to nail down. Everyone must read press releases and summaries of social science with a critical eye. The kind of complexity hidden within a "simple" correlation cannot normally be communicated or understood in simple sound bites about cause and effect.
Cherlin, A. J., Chase-Lansdale, P. L., & McRae, C. (1998). Effects of parental divorce on mental health throughout the life course. American Sociological Review, 63(2), 239-249.
Cowan, P. A., & Cowan, C. P. (2002). Interventions as tests of family systems theories: Marital and family relationships in children's development and psychopathology. Development and Psychopatholology. Special Issue on Interventions as Tests of Theories., 14, 731-760.
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.
A 5-Part Teleseminar Series for Academics, Clinicians, and Anyone Else Who Wants to Learn How to Get the Media to Cover You and Your Topic!
CCF is a volunteer organization whose busy professionals contribute their time and expertise gratis. Click here to support our mission by making a donation to CCF. Your contributions will help to fund our annual conferences and publicize our research briefings.