Bridging the Gap Between Tools for Learning and for Doing Statistics

My dissertation focused on the tools we use to do and teach statistics. For a general overview of the problems I am thinking about, see my page on the future of statistical programming, or read my full dissertation

Committee:

Abstract: Computers have changed the way we think about data and data analysis. While statistical programming tools have attempted to keep pace with these developments, there is room for improvement in interactivity, randomization methods, and reproducible research.

In addition, as in many disciplines, statistical novices tend to have a reduced experience of the true practice. This dissertation discusses the gap between tools for teaching statistics (like TinkerPlots, Fathom, and web applets) and tools for doing statistics (like SAS, SPSS, or R). We posit this gap should not exist, and provide some suggestions for bridging it. Methods for this bridge can be curricular or technological, although this work focuses primarily on the technological.

We provide a list of attributes for a modern data analysis tool to support novices through the entire learning-to-doing trajectory. These attributes include easy entry for novice users, data as a first-order persistent object, support for a cycle of exploratory and confirmatory analysis, flexible plot creation, support for randomization throughout, interactivity at every level, inherent visual documentation, simple support for narrative, publishing, and reproducibility, and flexibility to build extensions.

While this ideal data analysis tool is still a few years in the future, we describe several projects attempting to close the gap between tools for learning and tools for doing statistics. One is curricular, a high school level Introduction to Data Science curriculum. Others are technological, like the experimental LivelyR interface for interactive analysis, the MobilizeSimple R package, and Shiny microworlds.

Much of this work was inspired by Biehler (1997), which describes the attributes necessary for a software package for learning statistics. Biehler’s vision has largely come to light, but advancements in computing and ‘data science’ have moved the goalposts, and there is more to learning statistics than before. This work envisions a tool not only encompassing these advancements, but providing an extensible system capable of growing with the times.