Author(s): YANG, Y., Institution: SHANGHAI JIAO TONG UNIVERSITY, Country: CHINA, Abstract-ID: 1757

INTRODUCTION: Leaderboards are persuasive tools in health and fitness promotions. Yet, the underlying mechanisms by which different leaderboard ranks influence health and fitness behavior remain unclear (James et al., 2025). Building upon leaderboard literature and social comparison theory, we aim to examine the non-linear effects of leaderboard rank on physical activity and test how upward and downward social comparison orientations shape these effects, thereby explaining plateaus and progress patterns in fitness behavior.

METHODS: We designed a two-wave online survey with a one-week lag to mitigate common method bias and examine non-linear effects of leaderboard design and social comparison on physical activity. We focused on WeRun, a fitness leaderboard feature of WeChat, and administered the study via Credamo. Participants’ daily step counts (wave 2 steps as dependent variable), leaderboard ranks, and received likes over seven days were recorded by the WeRun app and self-reported by participants. Social comparison was assessed using a validated four-dimensional scale (Esteves et al., 2021). We integrated generalized additive models (GAM) to model non-linear associations and LightGBM with SHAP to quantify feature importance, thereby exploring plateau and progression patterns (Hofman et al., 2021). In total, 1,267 healthy Chinese users (64.4% female; mean age 30.2) were included in the analysis.

RESULTS: The GAM revealed non-linear effects (adjusted R² = 45.8%). Leaderboard rank showed a U-shaped relationship (EDF = 2.67, p < .001), with step counts lowest at mid-ranks. Received likes exhibited a plateau beyond 3–4 likes (EDF = 1.85, p = .006). Upward contrast decreased activity only at very low levels (EDF = 2.24, p = .045); downward contrast increased activity only at very high levels (EDF = 2.02, p = .025). Thus, physical activity is optimally promoted when users do not contrast with upward targets but strongly contrast with downward targets. Furthermore, LightGBM combined with SHAP identified leaderboard rank and received likes as the most important predictors of step counts. Partial dependence plots further confirmed this U-shaped relationship.

CONCLUSION: We found non-linear, threshold-based patterns in the effects of leaderboard rank on physical activity. The findings may practically inform the design of fitness leaderboards to mitigate demotivation among mid-rank users and theoretically extend social comparison theory.

REFERENCES:
Esteves et al., (2021). Online social games: The effect of social comparison elements on continuance behaviour. Information & Management, 58(4), 103452.
Hofman et al., (2021). Integrating explanation and prediction in computational social science. Nature, 595(7866), 181-188.
James et al., (2025). Is fitness technology-facilitated social comparison the thief of well-being? The mediating role of social comparison on the relationships between passion and performance self-esteem. Information Systems Research.