Multivariate Linear Regression Analysis of Childhood Psychopathology using Multiple Informant Data

It is common in psychiatric epidemiologic studies of childhood psychopathology to have multiple informant reports of mental health outcomes. The key challenge in analyzing multiple informant data concerns how they should best be represented in statistical models. Here we propose multivariate linear regression as the preferred method when the multiple informant outcome data are continuous. This approach permits the informant-specific information about mental health outcomes to be included in a single regression analysis, at the same time adjusting for the correlation between informant responses. The advantages of using a multivariate model include the ability to: (1) test for informant differences in outcome and assess if the effect of a risk factor on the outcome varies by informant; (2) estimate separate effects for each informant where necessary, or common effects where appropriate; (3) estimate the correlation between informant reports; (4) appropriately handle missing data by including data from all subjects with at least one informant report. To illustrate the application of this approach an example from the Connecticut Child Study is presented, examining risk factors for``Internalizing'' behavior using parent and teacher informants.