This project involves new statistical methodology as well as an application of that methodology. The application is a meta-analysis of rat and mice experiments in the literature on the effect of dietary fat on the promotion of mammary tumors. We model the tumor promoting effect of different types of fatty acids (saturated, monounsaturated, n-3 polyunsaturated, and n-6 polyunsaturated fats) through a conditional logistic model. The new statistical methodology uses the sandwich estimator of variance in the conditional logistic model. Using this new methodology, we correctly model the asymptotic variance of the dietary effects regardless of whether those effects are the same for each experiment (i.e. fixed effects) or whether those effects vary from experiment to experiment (i.e. random effects). One paper on the application has been published and a second one is scheduled to appear. A paper on the methodology is also scheduled to appear.
