How Feedback Frequency and Environmental Structure Impact Learning in Strategic Interactions

Empirical research on learning in games has largely focused on simple, well-structured environments, devoting comparatively little attention to how environmental structure and feedback frequency jointly shape learning dynamics. This leaves open important questions about how agents learn when tasks vary in difficulty or when feedback is delayed or absent. We address this gap with an online experiment using a 2 (environment) × 4 (feedback frequency) factorial design. 800 participants played 30 dominance-solvable normal-form games against a computer opponent. The environment was either curriculum-ordered, with games grouped into blocks of increasing difficulty, or randomized. Feedback about the computer's actions was provided always, never, every five or ten rounds. Results show that under full feedback, a curriculum-ordered environment enhances performance. When no feedback is provided, overall performance converges across environments, yet feedback-free learning emerges only in the structured setting, where repeated exposure to similarly difficult games supports within-block improvement. Analyses of feedback frequency reveal a strong interaction with environmental structure: in curriculum-ordered environments, full feedback outperforms delayed feedback, contradicting the idea that unsupervised periods foster self-teaching. Conversely, in random environments, full feedback yields the worst performance, while delayed feedback mitigates confusion generated by noisy cues. These findings indicate that structured environments generally support learning, either by boosting performance or by avoiding detrimental effects. Unstructured environments, instead, can hinder learning under full feedback. Optimal feedback provision is thus environment-dependent: full feedback is preferable in structured settings, whereas delayed feedback is more effective when task structure is weak.