What can statistics education offer to data science?

Abstract

Data science relies heavily on statistical ideas, though it seems more concerned with prediction than statistics, which is more focused on modeling the data production process. This article will argue that the data scientist will do well to pay more attention to the likely disconnect between the chosen statistical model and the process it tries to emulate. Three learning goals are proposed and illustrated with elementary examples to help students grasp the idea. The disconnect is relevant to the replication crisis, yet is inadequately discussed in statistical communities. The lessons here are applicable to the education of statisticians.