these social exchanges. It is these contingencies that actually bring about the changes in social behavior. The point is, to understand what happens to families during separation, one must actually observe the mothers' interactions with members of her primary group (e.g., friends, children, relatives). Repeated observations would provide the information necessary to understand how these processes unfold over time.
The formulation about mechanisms for maintenance also focused our attention on another set of questions that are crucial in understanding this process. What are the relations among maternal stress, irritability and sadness? Is stress a stimulus that controls both these maternal reactions? To what extent are maternal sadness and irritability elicited by stress and to what extent are they shaped by the reactions of key persons in the mother's social environment? Future studies that explore the process model outlined in the present report will have to address these questions.
The original study was designed to contribute to our understanding of the antisocial and depressed reactions of sons during the first years of separation. The data set functioned very well in explicating that problem. In designing the study, maternal depression was of sufficient interest to lead to the inclusion of multiple measures of depression (CES-D, daily diary, telephone contacts). At the time, it seemed that each of these represented different methods and could therefore function as separate indicators in defining the latent construct Depression. We took a similar perspective on the problem of defining the stress, support, and irritability constructs. It was only later that we learned that all of these measures are a variation on a theme of self-report and, indeed, they were not separate indicators but a single one.
The current study by Forgatch ( 1987) is an attempt to correct these problems. If she is successful, then it will be possible to use structural equation modeling to model many of the mechanisms discussed in the present report. In either case one thing is very clear, and that is that the study of maternal depression is characterized by some extraordinarily complex problems of measurement. What is startling about this is that in an area so heavily researched, these problems of method variance have been largely neglected or simply ignored.
The writers gratefully acknowledge the support provided by MH 38318 (Family Mental and Policy Research Branch) for the collection and analyses of the data. We wish to thank Karen Gardner and her staff for the care with which these data were collected and to Martie Skinner and her colleagues who carried out the statistical analyses.