The principal aims of this project are to study a number of statistical problems in the analysis of epidemiologic data. The problems studied are relevant to the analysis of case-control data and ecologic epidemiologic data. Since odds ratios form the nucleus of many epidemiologic analyses, this research addresses issues in model building with odds ratios, or more generally log-linear modeling. This research will contrast selected statistical estimation and testing procedures using both computer and theoretical techniques and the methodology will be applied to select cancer case-control data sets. The focus will be on problems associated with log-linear modeling in the face of numerous predictor and stratification variables. Such situations inevitably lead to problems in the selection of variables for analysis. Furthermore, with numerous variables the resulting cross-tabulations lead to sparse, perhaps incomplete multidimensional tables. These problems will be studied in this research. For ecologic epidemiological data, the issue of standardization of incidence rates arises. In this research various methods of rate standardization or regression adjustment will be compared. Emphasis will be on the problems which typically arise in the analysis of epidemiologic data. This research will lead to improved statistical methodology for the analysis of complex epidemiologic data. The issues studied have immediate implications for the analysis of epidemiologic data.
Showing the most recent 10 out of 14 publications
