When I stood up in public and said that behaviors learned at home do not transfer to outside-the-home contexts (Harris, 1999), a chorus of developmentalists stood up and said "Yes they do!" and they pointed to evidence from intervention studies.

The intervention studies the developmentalists had in mind are those in which parents are taught more effective methods for managing their obstreperous children -- methods that employ reinforcement for good behavior and time-outs for bad behavior, instead of yelling and hitting. In many of these studies, the interventionists actually succeed in changing the parents' behavior, at least temporarily. When this happens, the children's behavior at home often improves as well. But the developmentalists claimed that teaching parents better ways of managing their obstreperous children also improves the children's behavior in school. If this were the case, it would be a clear contradiction of the predictions generated by my theory.

But it is not the case. I came to that conclusion after a long, hard look at the published research. Rex Forehand, a leading researcher in the field of parent-training interventions, came to the same conclusion. In a review of twenty years of research in this field, Wierson and Forehand (1994) reported that training can improve the parents' behavior toward the child and the child's responsiveness to the parents. "However," Wierson and Forehand admitted, "research has been unable to show that the child's behavior is modified at school" (p. 148). Likewise, a school-based intervention can improve the child's behavior in school but will have no effect on troublesome behavior at home (Barkley et al., 2000; Grossman et al., 1997). Although some researchers claim to have shown transfer of the effects of an intervention from one context to another (e.g., Cowan & Cowan, 2002; Dishion & Andrews, 1995; Forgatch & DeGarmo, 1999), I have examined these studies and found them to have serious flaws (see Harris, 2000b; Harris, 2002).

The developmentalists believed that the evidence from intervention studies was the best evidence against my theory -- the best evidence that what children learn at home transfers to outside-the-home settings. Instead, intervention studies provide strong evidence in favor of my theory. Improving children's behavior at home, by modifying their parents' child-rearing style, does not improve their behavior at school. Likewise, improving behavior at school does not improve it at home. It is possible to improve children's behavior in both places, but to do that we need home-based interventions and school-based interventions.

Now we come to more subtle sources of bias. In my forthcoming paper in Developmental Psychology (Harris, 2000), I mentioned an intervention study by Forgatch and DeGarmo (1999). The intervention was designed to improve the mothers' child-rearing methods and it succeeded in doing so -- on average, the mothers in the intervention group used less "coercive parenting" and showed more "positive involvement" than those in the control group. The question is, did this favorable change in the atmosphere of the home produce an improvement in the child's behavior in school? My theory predicts that a home-based intervention of this kind will have no effects on the child's behavior in school, and Forgatch and DeGarmo's results were consistent with my prediction: they found no significant differences in school behavior between the control group and the intervention group.

Unwilling to take no for an answer, Forgatch and DeGarmo (1999) did a post-hoc data analysis -- a path analysis. The path analysis gave them the results they were looking for: it showed that those mothers who improved in their child-rearing methods were more likely to have children who behaved well in school.

It's easy to see why this kind of analysis would make sense to a developmentalist. The author of another intervention study patiently explained it to me in e-mail, in capital letters:

The problem is that the path analysis introduces a bias that Feinstein (1985), the epidemiologist, calls "a compliance-determined susceptibility bias." "A particularly tricky problem occurs," Feinstein says, "if compliance depends on certain personality traits that may also be associated with the outcome event" (p. 303). He gives a true example: a randomized trial of a drug designed to reduce blood lipids. The trial showed that the patients who complied well with the drug regimen had significantly lower fatality rates than patients who complied poorly. The catch was that the patients who received only a placebo but who complied well with the regimen also had significantly lower fatality rates!

People who comply with interventions are likely to differ in personality and intelligence from those who do not comply. Because these characteristics are heritable, parents who respond well to the demands of an intervention are more likely to have children who respond well to the demands of school. Assignment to the intervention or control group is supposed to be random. The path analyses used by Forgatch and DeGarmo (1999), and by Cowan and Cowan in a new intervention study that will be published in 2002, are equivalent to letting the participants decide for themselves which group they will be in.