Co-Principal Investigator: Raudenbush, Stephen

The purpose of this grant is to produce a general statistical framework and an associated set of statistical guidelines and tools for studying such impact variability. The project will address three major design issues: 1) A treatment effect may be moderated by covariates that are measured with error; 2) The covariates and outcome, whether continuous or discrete, may be only partially observed; and 3) Compliance to treatment assignment may be imperfect. The project will also extend these issues to RCTs in which a treatment effect may be random and moderated by covariates.

The research team will develop the statistical analysis by estimating the parameters of the theoretical model via maximum likelihood from incomplete data using adaptive Gauss-Hermite Quadrature. They will extend this approach to multi-site trials in which each can be regarded as possessing a unique set of principal strata. First, the software will be programmed in C for initial development and testing. Then, the team will render a more user-friendly interface through R and Shiny web applications. The research team will also provide a user's guide and  results from the analysis of data from multisite trials, and software documentation. The team will prepare research manuscripts, present at conferences, and teach at in workshops for education researchers.