Alternative models of the stress buffering hypothesis.

The interactive effects of life events and social support on a DSM-III diagnosis of major depressive episode and on number of depressive symptoms were examined. Data are from a stratified random sample of 3,732 community-dwelling adults. The paper focuses on differences between linear probability models and logistic regression models with regard to the definition, detection, and interpretation of interaction effects. Results indicate that conclusions about the interaction of life events and social support are model dependent. Using a linear probability model, significant event by support interactions were observed for both depressive symptoms and major depression. Using logistic regression, which estimates interactions in terms of odds ratios, no significant event by support interactions were observed. Discussion addresses the interpretive implications of modeling interaction in terms of probability differences versus odds ratios.