A Typology of Players in the Game Physics Playground

This paper examines whether players of the educational physics game Physics Playground fall into distinct playstyle groups, motivated by the inconsistent learning and engagement outcomes that educators have observed when using games. Building on Bartle's player taxonomy and later survey-based typologies by Yee and Kahn et al., the authors take a different approach: they apply latent class analysis to in-game log data from 138 players, drawn from more than six million individual actions, rather than relying on self-reported questionnaires. The variables modeled included badges earned, machines drawn, freeform drawings, erasures, level starts and restarts, and total events logged.

The analysis identified two-class and three-class solutions that were both defensible. The two-class model separated achievers, who earned significantly more gold and silver badges and played more levels, from all other players. The three-class model, which the authors argue maps most closely onto Bartle's framework, distinguished achievers (about 17 percent of players), explorers or "tinkerers" (about 51 percent), who built and revised complex machines while largely ignoring optimal badge-earning solutions, and disengaged players (about 32 percent), who showed low activity across nearly every measured feature. The authors interpret the disengaged group as plausibly containing the socializer and killer styles that a single-player environment cannot accommodate.

The study is explicitly limited by its modest sample. The authors note that 138 players can detect only large effects, and that the BIC criterion suggested up to ten classes, many of which captured small outlier groups rather than stable subpopulations. The model fit statistics for solutions beyond three classes were weak, and the authors acknowledge that the three-class solution was selected partly on interpretability. They call for larger samples with thousands of players, multi-context data that include both single-player and multiplayer settings, and the incorporation of real-time affect measures to better understand how different player types experience the game.