This proposal will develop methodology for the analysis of multiple informants and/or multiple assessment data collected in surveys designed to measure mental health outcomes or risk factors in community and service-based samples. The use of multiple informants (parents, teachers, children, clinicians, peers) is generally regarded as the best approach to obtain information about children's mental health, social functioning, and service use. Multiple sources data are also common in others of psychiatric research (e.g. multiple psychiatric assessments). Some of the difficulties encountered in combining data from multiple sources include systematic differences between sources with regard to some outcomes, a relatively low level of informant agreement, and missing information for some sources. Beginning with work published in 1995, we have developed a general approach to the analysis of multiple source data focusing on categorical outcomes. Specifically, we have developed methods for incorporating multiple source data into a logistic regression framework where either the outcomes, or the predictors, may arise from multiple sources. These methods accommodate the possibility for missing responses in one or more sources The current proposal seeks to pursue this line of research. First, we will develop a general regression model for the case where multiple sources are used to determine risk factors and which builds on our previous work in this setting. This model will be conceptually similar to our initial regression model with outcomes measured by multiple sources. It will allow the use of all available data in parsimonious model fitting, the study of the effects of different sources, and the inclusion of subjects of partly missing data.. We will complete the existing methodology by showing how our work can be extended to accommodate other types of outcome data, including continuous and countered multiple source data. We will revisit our proposed methods for non-response to show how an alternative approach may be useful. We will develop new methods for handling the analysis of data obtained in two stage designs. Finally, we will develop extensions of the methods which can be used to analyze multiple source data obtained repeatedly over time in longitudinal studies. We will develop appropriate software so that these methodological developments can be readily used in psychiatric research. We will devise macros and procedures which can be used with existing, widely available packages. We will apply our methodology to five existing data sets; four data sets come from ongoing studies which involve collaborations with this group of investigators. The proposed methods will also be directly applicable to several recent NIMH initiatives.
Showing the most recent 10 out of 47 publications