Model 1. When the child outcome was a continuous variable, linear mixed-effects models were fitted to the data with child outcome as the dependent variable, maternal treatment and time(study week)as independent variables,and an interaction term representing treatment by time. In addition, age and gender of child and study site were included as covariates. Other child and maternal baseline characteristics that were significantly different between treatments were also tested but were dropped from the model if they did not contribute to an appreciable difference in the results. A statistically significant interaction term indicated a differential treatment effect. Cor-relations between related measures over time, as well as non- independence of observations among siblings, were handled by including nested random effects in the model (i.e.,within- subject observations nested within family) (32). When child outcomes were binary variables (child diagnoses) or count variables(child symptoms),logistic random-effects regression models (for binary measures) and Poisson random-effects regression (for count measures) were used to determine differential effects of maternal treatment on these outcomes(33). Repeated measure over time and nonindependence of siblings as well as potential confounding variables were handled as described for continuous outcomes.