StatPlayground: A Sandbox for Learning Practical Statistics

Abstract

Inferential statistics is a frequent task in several fields such as HCI, Psychology, and Medicine. Research shows that inferential statistics is often used incorrectly because the underlying statistical concepts are misunderstood. From interviews with students in an HCI lab, we find that, in addition to theoretical knowledge of statistics, novice analysts require statistical know-how, i.e., practical knowledge of how various data characteristics are inter-related and how they influence significance test selection and statistics, to analyze data. However, current learning resources such as books and online searches are not adequate to help learn statistical know-how. As a possible solution, we present StatPlayground, an interactive web app that supports exploratory learning of statistical know-how. StatPlayground does this by allowing users to modify data via direct-manipulation of visualizations, to see how those changes affect other data characteristics such as the shape of the distribution and variance of the data, as well as the resulting significance test and statistics such as effect size and p-value. StatPlayground can be combined with traditional teaching methods and can help prepare students for real-world analysis. Our evaluation of StatPlayground with graduate students shows the potential of StatPlayground to help learn statistical know-how and design implications for simulation tools for learning statistics.