How should we measure exploration?

Recent studies used multi-armed bandit tasks to investigate individual differences in exploration strategies. However, the psychometric properties of these measures remain largely unexplored. We examined the test-retest reliability, convergent and divergent validity, and external validity of model-based estimates of exploration strategies. Our results revealed poor to moderate reliability, with minimal correlations for the same strategy across tasks. By refining the computational models, we identified two convergently valid latent factors representing value-guided and directed exploration. However, these factors showed no significant correlation with self-reported exploration tendencies and only weak associations with mood fluctuations. To improve future research, we suggest simplifying the computational models and using multiple tasks to more accurately model exploration strategies and mitigate spurious correlations arising from task-specific factors. A strong positive correlation with working memory and a lack of significant associations with questionnaire-based measures highlight both the potential and limitations of current behavioral paradigms in capturing exploration behavior.