This proposal involves a collaborative effort between prevention Scientists at five funded NIMH sites and statistical methodologists at four sites to develop new statistical methodology and research designs for preventive trials. Building on an R01 which developed new statistical methods for preventive trials at one NIMH Center at Johns Hopkins, this proposal is a competing continuation which broadens the methodologic basis of prevention science in mental health. The following three areas have been selected because of their critical importance for designing and analyzing preventive trials to demonstrate reduction in psychopathology. 1. Selection bias and attrition are the two major """"""""missing data"""""""" problems in preventive field trials. Because of the frequent nonrepresentativeness of participants in an intervention compared to the target population, factors influencing selection bias can have severe implications in interpretation of the results of a trial. We will investigate designs and analytical strategies which reduce bias in selection or on the inferences that are drawn. Two trials at Arizona State University (ASU) and the Institute of Social Research, University of Michigan (UM) will be used to evaluate these new procedures. The effects of differential attrition on inferences about the prevention effects are also important; data from Johns Hopkins (JH) will be used to evaluate statistical methods to deal with follow-up loss. 2. We will develop a set of principles for developing the most efficient designs for preventive trials, including what subjects or subgroups should be given which type of interventions and which measures should be collected on each subject. These principles can then be directly applied to develop more efficient programs in prevention. 3. We will also develop a set of statistical methods which address heretofore unanswered questions in existing preventive trials. These include new methods to evaluate an intervention's effect on development, better evaluation of the roles of mediators and modifiers, and systematic identification of variation in impact of an intervention by subgroups. These new methods will allow more appropriate modelling of the data than existing methods such as LISREL and other methods which are based on normality assumptions and linear relationships among variables.
Showing the most recent 10 out of 120 publications
