Cognitive Hierarchy Theory¶
Predict strategic play by mixing reasoning-depth types whose positive levels best respond to beliefs over all strictly lower levels.
Core Idea¶
Cognitive hierarchy theory models strategic-game play as a population of different reasoning-depth types. An explicit nonstrategic step-0 rule anchors the recursion. A positive step \(k\) best responds to a belief that opponents occupy a normalized mixture of all strictly lower steps, never its own or higher steps. Population frequencies then weight the type-specific choices into an aggregate forecast. This relaxes equilibrium's mutually correct-belief assumption while preserving payoff-sensitive choice within each modeled type.[^ref-d76078be9c10]
Scope of Application¶
Camerer, Ho and Chong's 2004 one-shot-game implementation takes uniform step-0 play and a Poisson level-frequency distribution with mean \(\tau\). These are tractable model conventions, not universal requirements or a direct measure of an individual's cognitive ability. The authors' later generalized version changes anchor and lower-level weighting assumptions. Their original examples span beauty-contest number guessing and simultaneous market entry, but any new application must specify its own payoffs, anchor, frequencies and data.[ref-d76078be9c10][ref-ce1b9de53790]
Clarity¶
Standard predecessor-only level-\(k\) reasoning lets type \(k\) best respond to type \(k-1\) alone. CH instead responds to a mixture of types $0$ through \(k-1\). In a two-thirds-of-average beauty contest, the authors' uniform step-0 mean is 50 and step 1 guesses near 33; step 2 is not generically 22, because its belief mixes steps 0 and 1 with \(\tau\)-dependent weights. CH is also distinct from quantal-response equilibrium's noisy-choice relaxation and from a general hierarchy of cognitive skills.[ref-d76078be9c10][ref-ce1b9de53790]
Manages Complexity¶
The model compresses heterogeneous initial choices into an inspectable relation: game and payoffs → zero-step anchor → level frequencies and truncated beliefs → recursive best responses → aggregate play. The original 24 beauty-contest data sets yielded a median fitted \(\tau\) near 1.61, while the entry games showed aggregate behavior near demand alongside low-demand overentry and high-demand underentry. Neither finding is universal. A common one-parameter \(\tau\) strengthens cross-game predictive discipline but may fit an individual game worse than refitted frequencies.[^ref-d76078be9c10]
Abstract Reasoning¶
Read a CH claim by separating player type, belief about opponent types, best-response action and observed mixture. Ask how level 0 is justified and whether a parameter predicts held-out games or was fitted to the target observations. Changing the anchor can propagate through every higher step. The original paper proves CH approaches certain equilibria reachable by finitely many iterated weak-dominance deletions as \(\tau\) grows, but expressly states that it does not converge to Nash in all games. More reasoning is not a universal equilibrium theorem.[^ref-d76078be9c10]
Knowledge Transfer¶
The recursion transfers between unlike payoff games, not their fitted guesses or entry rates. Live Representation is the proposed strict parent: CH maps a strategic-game population into structured reasoning types, beliefs and predicted choices. Bounded Rationality is a broad neighbor, not automatically a strict parent under its full live signature. The strategic-game belief and payoff commitments keep CH domain-specific. The source-used abbreviation “CH” remains an unadjudicated alias pending collision review.[^ref-d76078be9c10]
[^ref-d76078be9c10]: Colin F. Camerer, Teck-Hua Ho and Juin-Kuan Chong, “A Cognitive Hierarchy Model of Games”, Quarterly Journal of Economics 119(3) (2004), 861–898, DOI 10.1162/0033553041502225; original published-paper text inspected via the repository-linked published PDF copy, especially pp.862–864, 867, 870–875 and 889–890. [^ref-ce1b9de53790]: Juin-Kuan Chong, Teck-Hua Ho and Colin F. Camerer, “A Generalized Cognitive Hierarchy Model of Games”, Games and Economic Behavior 99 (2016), 257–274, original author-repository abstract and publisher Introduction inspected for model-family and level-\(k\) distinctions.
Relationships to Other Abstractions¶
Current abstraction Cognitive Hierarchy Theory Domain-specific
Parents (1) — more general patterns this builds on
-
Cognitive Hierarchy Theory is a kind of Representation Prime
CH models strategic-game behavior through a structured mapping from players and payoffs to reasoning types, beliefs and predicted actions.
Hierarchy path (1) — routes to 1 parentless root
- Cognitive Hierarchy Theory → Representation → Abstraction
Neighborhood in Abstraction Space¶
Cognitive Hierarchy Theory sits in a crowded region of the domain-specific corpus (25th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.
Family — Game-Theoretic Models & Paradoxes (37 abstractions)
Nearest neighbors
- Guess ⅔ of the Average — 0.91
- Bayesian Nash Equilibrium — 0.90
- Game balance — 0.89
- Dominated Strategy — 0.89
- Max-dominated strategy — 0.89
Computed from structural-signature embeddings · 2026-10-08