Predictive Modeling of Change in Depressive Disorder and Counts of Depressive Symptoms in Urban Youths

Publisher Website
Google Scholar
Abstract
A longitudinal multivariate model examined data from a sample of 2,415 adolescents residing in 10 cities to see if both depressive symptoms and diagnosis of depressive disorder change over time in association with the same salient events. The model posits that risk variables (e.g., stressful circumstances and traumatic events), protective variables (e.g., social relationships), and a Risk × Protective interaction statistically predict change in depression, whether operationalized as a disorder or as a symptom count. Multiple regression, controlling for the univariate effects of sex and race, is used to analyze the model for depressive symptoms, whereas logistic regression (which, unlike multiple regression, is appropriate for a nominal dependent variable) is used to analyze the model for depressive disorder. The two forms of analyses show that risk variables and protector variables explain change in depression, both as a symptom count and as a disorder. However, the multiple regression, being more sensitive to variation, reveals a Risk × Protective interaction. The multiple regression yields estimates of the percentage of variance explained (the total model explains 43%; risk and protective variables explain 15%). In contrast, the logistic regression yields log odds ratios (5.1 for traumas, 1.5 for stressful events, and 0.9 for social relationships). Given the similarity in support for the model, how one chooses to operationalize depression depends on the use one intends to make of the measure and the types of results yielded by its corresponding analytic strategy.

This publication has 25 references indexed in Scilit: