Non-parametric decision trees and online HCI
Torben Sko, Henry Gardner, Michael A. Martin · 2013
This paper proposes that online HCI studies (such as web-surveys and remotely monitored usability tests) can benefit from statistical data analysis using modern statistical learning methods such as classification and regression trees (CARTs). Applying CARTs to the often large amount of data yielded by online studies can easily provide clarity concerning the most important effects underlying experimental data in situations where myriad possible factors are under consideration. The feedback provided by such an analysis can also provide valuable reflection on the experimental methodology. We discuss these matters with reference to a study of 1300 participants in a structured experiment concerned with head-interaction techniques for first-person-shooter games.