Multivariable model building for descriptive and predictive research questions

Dr. Georg Heinze, Medical University of Vienna, Austria,

Center for Medical Data Science, Institute of Clinical Biometrics

Despite there being a large body of literature on the topic of multivariable model building, this topic poses a lot of challenges to data analysts when facing real data. I will discuss some of the major challenges that are particularly relevant for descriptive and predictive research questions, but may also occur in causal research: variable selection, the issue of highly correlated predictors, nonstandard distributions of predictors, nonlinear functional forms of the outcome-predictor association and missing data. I will discuss each of the topics with the type of research question (descriptive, predictive, causal) in mind. Some proposals for combined variable and functional form selection will be reviewed and exemplified by applying them to real data. The relevance of addressing model instability by resampling, and of transparently reporting all steps of model building will be highlighted. 

Prerequisites: This seminar is for quantitative scientists. 

*This seminar is available for RECR Credit, 1.5 Hours, Attendance will be verified and a survey must be completed afterwards with well thought out responses to receive RECR Credit.