Latent Class Analysis of Gameplay Metrics from Youth Playing Robot ChampionsTM: Relations of Class Membership to Persistence and Intensity

Abstract

Purpose: Game metrics assessing in-game actions (e.g., keystrokes, challenges, and time on task) have become a staple component to understanding the attraction youth have towards a video game. Data from automatic log files can be used to assess how players interact with various aspects of the game and whether they comprehend the game mechanics. Despite a rapid burgeoning in the use of game metrics, few studies have used this type of objective instrumentation to examine whether users display unique patterns of gameplay. To address this gap, this study examines a temporal slice of game metrics obtained from a robotics platform using person-centered strategies to empirically examine unique gameplay characteristics. We also validate gameplay strategies using objective measures of persistence and intensity of play.

Methodology: We used in-game metrics collected over a two-month window in 2024 obtained from 153,658 players who played 199,055 sessions of the Roblox sandbox game Robot Champions^TM. Latent class analysis (LCA), a model-based clustering approach, was used with seven game metric indicators to ascertain whether there are distinct patterns of gameplay.

Findings: LCA identified four unique gameplay styles comprised of Fully Engaged, Engaged in Training, Engaged in Building, and Engaged in Driving. Fully Engaged involved taking advantage of multiple aspects of the game whereas Engaged in Driving gameplay only involved driving the robot, neglecting other game features like queuing for or starting matches or benefiting from tutorials. Measures of persistence (number of games completed, steps completed, and duration of play) and a measure of intensity were used to characterize class membership. Between-class mean differences in persistence and intensity were also examined using multinomial logistic regression with post-hoc pairwise comparisons. Analysis of a small number of players who completed multiple games revealed that, over time, players engaged in multiple styles of gameplay.

Originality: This study is unique in that we used a person-centered clustering approach using in-game metrics as opposed to lower resolution and less reliable self-reports from players.

Impact: Findings are discussed in terms of ways game developers can utilize  log file data to learn more about the unique ways players engage in gameplay and what drives their actions during the game. This information can be utilized to improve robotics game design and configure learning mechanics. Ultimately, game metrics may provide fresh insight into player engagement and inform game designers which aspects of the game are fun and rewarding.