The paper by Stahl and Wilson develops and tests a theory of human behavior in 3 × 3 symmetric games, proposing a family of five boundedly rational archetypes based on players' models of other players and their ability to identify optimal choices. The authors designed and conducted an experiment to identify these archetypes, including a rational expectations type, and estimate parameters defining these types. The experimental evidence rejects the rational expectations type but confirms the boundedly rational theory. The study contributes to a descriptive and prescriptive theory of games, providing insights into how players make decisions in strategic situations. The formal theory is developed using a three-parameter family of probabilistic choice functions, and the experimental design involves participants playing 12 symmetric 3 × 3 games without feedback. The statistical analysis, using bootstrapping, confirms the presence of boundedly rational types and excludes the rational expectations type from the sample. The results suggest that most participants exhibit one "model of other players" across all games, with a significant portion being level-2 types. The study also addresses the robustness of the findings and discusses the implications for understanding human behavior in games.The paper by Stahl and Wilson develops and tests a theory of human behavior in 3 × 3 symmetric games, proposing a family of five boundedly rational archetypes based on players' models of other players and their ability to identify optimal choices. The authors designed and conducted an experiment to identify these archetypes, including a rational expectations type, and estimate parameters defining these types. The experimental evidence rejects the rational expectations type but confirms the boundedly rational theory. The study contributes to a descriptive and prescriptive theory of games, providing insights into how players make decisions in strategic situations. The formal theory is developed using a three-parameter family of probabilistic choice functions, and the experimental design involves participants playing 12 symmetric 3 × 3 games without feedback. The statistical analysis, using bootstrapping, confirms the presence of boundedly rational types and excludes the rational expectations type from the sample. The results suggest that most participants exhibit one "model of other players" across all games, with a significant portion being level-2 types. The study also addresses the robustness of the findings and discusses the implications for understanding human behavior in games.