Background: CREATE Lab at NYU developed a suite of digital games intended to improve specific cognitive skills like working memory, inhibition, and task switching. However, some players seemed to focus on maintaining accuracy in the games, while others focus on speed, especially as levels get harder. The lab was interested in near real-time analytical techniques that could help spot different strategies players were using, and understand whether they had an impact on changes in cognitive skills.

Action: I identified key variables using LASSO regression, and used clustering techniques, especially Latent class mixture models, on the reaction time and accuracy data derived from gameplay data logs to detect and characterize different types of players. Participants’ performance in terms of reaction time, but not accuracy, was associated with a different standardized measure of inhibitory control.

Result: Results suggested new ways to examine in-game performance to adapt level difficulty for individual players in real time.