For Complex Data, Let's Give Up on Interpretability

• Activity Number: 291
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Bertrand Clarke*+ and Jennifer Clarke and Camillo Valdes
• Companies: Univ. Miami and University of Miami and University of Miami
• Keywords: bacterial genomes ; prediction ; model uncertainty

Abstract

Subject matter researchers are often concerned with real physical problems and want statisticians to help them understand `what's really going on'. The problem is that for many (if not most) real problems there is huge model uncertainty or mis-specification. So, announcing a model or even a model class is misleading. Also, we usually get better predictions when we take model uncertainty into account. So, as a generality, we should try to predict well first and make any other inferences from a good predictor rather than trying to model a physical problem directly. This is the reverse of what we were taught and what subject matter researchers want. We illustrate these ideas with next generation sequencing data from bacterial genomes.