Experiments on individual strategy updating in iterated snowdrift game under random rematching

Area of Science:

• Behavioral Economics
• Evolutionary Game Theory
• Computational Social Science

Background:

• The iterated snowdrift game models social dilemmas where cooperation benefits all, but individual defection yields higher payoffs.
• Understanding human strategy updating in dynamic environments like random rematching is crucial for evolutionary biology.

Purpose of the Study:

• To investigate human strategy updating rules in iterated snowdrift games under random rematching.
• To model individual learning behaviors using experience-weighted attraction (EWA) learning theory.

Main Methods:

• Conducted two laboratory experiments involving iterated snowdrift games with four groups of subjects.
• Applied EWA learning theory to model individual-level strategy updating.
• Validated the model against experimental data, with 75% of groups passing validation.

Main Results:

• Observed substantial heterogeneity in strategy updating, with most subjects (63.9%) using reinforcement learning (RL) or similar rules.
• RL learners' performance suffered on average; sensitivity to recent experiences and consideration of forgone payoffs benefited players.
• Random rematching led to faster learning from recent experience and more diverse updating rules compared to fixed pairings.

Conclusions:

• Most subjects employ RL-like updating rules even with changing opponents, though not always leading to better outcomes.
• Human behavioral strategies in social dilemmas are complex and influenced by opponent dynamics and learning mechanisms.