Cultural transmission of move choice in chess

Egor Lappo; Marcus W. Feldman; Noah A. Rosenberg

arxiv: 2302.10375 · v3 · pith:7EH6LWADnew · submitted 2023-02-21 · ⚛️ physics.soc-ph · q-bio.PE

Cultural transmission of move choice in chess

Egor Lappo , Noah A. Rosenberg , Marcus W. Feldman This is my paper

Pith reviewed 2026-05-24 09:39 UTC · model grok-4.3

classification ⚛️ physics.soc-ph q-bio.PE

keywords cultural evolutionchessmove choicefrequency-dependent biasprestige biassuccess biasDirichlet-multinomialsocial learning

0 comments

The pith

A Dirichlet-multinomial model of elite chess games detects negative frequency-dependent bias in the transmission of certain moves.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors fit a population-level statistical model to decades of recorded chess games by leading players to quantify how cultural transmission biases shape move choice in specific positions. They report that negative frequency-dependent bias appears in the dynamics of some moves, while patterns for other moves remain compatible with prestige bias or success bias. These findings are interpreted as possible signatures of broader shifts in the chess community, such as the rise of computer engines and online broadcasts. If the model parameters capture transmission processes rather than artifacts, the work supplies a concrete case study of how different social learning rules coexist within one well-documented cultural domain.

Core claim

For moves made in specific positions, the relative effects of frequency-dependent bias, success bias, and prestige bias on move frequencies are evaluated using the Dirichlet-multinomial likelihood. Negative frequency-dependent bias plays a role in the dynamics of certain moves, and other moves are compatible with transmission under prestige bias or success bias. These apparent biases may reflect recent changes, namely the introduction of computer chess engines and online tournament broadcasts.

What carries the argument

The Dirichlet-multinomial likelihood model that estimates the relative strength of frequency-dependent, success, and prestige biases from observed move frequencies across games.

If this is right

Moves that are currently rare in a given position should increase in frequency over time under negative frequency-dependent transmission.
Moves whose frequencies track the choices of high-prestige or high-success players should show stable or rising adoption even when common.
The strength of each bias can be tracked across decades, allowing detection of shifts that coincide with technological introductions.
The same modeling approach can separate the contribution of different biases when multiple transmission rules operate simultaneously on the same cultural trait.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar statistical signatures of mixed biases might appear in other domains where agents repeatedly choose among a finite set of strategies with public performance records.
If computer assistance alters transmission biases, the model predicts measurable changes in move diversity after the widespread adoption of engines in any game or field with analogous choice structures.
The coexistence of negative frequency dependence and prestige effects could stabilize cultural diversity within a population while still allowing rapid uptake of successful innovations.

Load-bearing premise

The fitted parameters of the Dirichlet-multinomial model can be read as direct evidence of specific cultural transmission biases rather than as artifacts of unmodeled factors such as evolving chess theory or shifts in the player population.

What would settle it

Re-fitting the model after explicitly controlling for documented changes in opening theory or player demographics and finding that the frequency-dependent and prestige parameters lose statistical significance would undermine the bias interpretation.

read the original abstract

The study of cultural evolution benefits from detailed analysis of cultural transmission in specific human domains. Chess provides a platform for understanding the transmission of knowledge due to its active community of players, precise behaviors, and long-term records of high-quality data. In this paper, we perform an analysis of chess in the context of cultural evolution, describing multiple cultural factors that affect move choice. We then build a population-level statistical model of move choice in chess, based on the Dirichlet-multinomial likelihood, to analyze cultural transmission over decades of recorded games played by leading players. For moves made in specific positions, we evaluate the relative effects of frequency-dependent bias, success bias, and prestige bias on the dynamics of move frequencies. We observe that negative frequency-dependent bias plays a role in the dynamics of certain moves, and that other moves are compatible with transmission under prestige bias or success bias. These apparent biases may reflect recent changes, namely the introduction of computer chess engines and online tournament broadcasts. Our analysis of chess provides insights into broader questions concerning how social learning biases affect cultural evolution.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Applies Dirichlet-multinomial to chess moves to flag negative frequency dependence and prestige/success biases, but provides no fitting details or time controls so the bias claims rest on untested assumptions.

read the letter

The paper's main move is to take decades of elite chess games, treat move frequencies in given positions as data, and fit a Dirichlet-multinomial model that lets them compare frequency-dependent, success, and prestige biases. That is new as an application; the abstract does not claim the framework itself is original, but the domain is. They report that some moves show negative frequency dependence while others are compatible with prestige or success bias, and they note that computer engines and broadcasts could be driving recent shifts. That is the extent of the positive contribution: a clean, quantitative case study in one well-recorded cultural domain. Credit for keeping the model simple and for flagging the engine confound themselves. The soft spots are exactly where the stress-test note says. No description of data selection, no account of how the model was fit or validated, and no explicit conditioning on time or player cohort. Because the same frequencies are used both to estimate the bias parameters and to claim those parameters reflect transmission, any unmodeled temporal change in theory or player pool can produce the same pattern. The abstract itself lists those alternatives, so the interpretation is not yet secured. This is for people who already work on cultural transmission models and want another empirical example to think about. A reader who needs the methods to be reproducible or the confounds addressed will not get much yet. It is coherent enough on its own terms to deserve a referee who can check the fitting procedure and ask for time controls; I would not desk-reject on the abstract alone.

Referee Report

2 major / 1 minor

Summary. The paper claims to analyze cultural transmission in chess using a Dirichlet-multinomial model on move choice data from leading players over decades. It finds that negative frequency-dependent bias influences the dynamics of certain moves, while other moves are compatible with prestige bias or success bias, possibly due to the introduction of computer chess engines and online broadcasts.

Significance. This work provides a detailed empirical analysis of social learning biases in a well-recorded cultural domain. If the model successfully isolates transmission effects from confounders, it would strengthen evidence for specific biases like negative frequency dependence in cultural evolution, offering insights applicable to other fields. The long-term data and precise behaviors in chess are strengths for such studies.

major comments (2)

[Abstract] Abstract: the abstract acknowledges that apparent biases may reflect recent changes from computer engines and online broadcasts, yet the likelihood comparison does not test whether the biases persist after conditioning on time or player cohort; this leaves the attribution to transmission biases open to alternative explanations such as evolving chess theory.
[Model] Model description: bias effects are evaluated by fitting the Dirichlet-multinomial model directly to the move-frequency data used to test the biases, so reported compatibility with each bias is defined in terms of the fitted parameters themselves rather than an independent test.

minor comments (1)

[Abstract] Abstract: provides no details on data selection criteria, model fitting procedure, validation, or error assessment, which limits evaluation of support for the claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below.

read point-by-point responses

Referee: [Abstract] Abstract: the abstract acknowledges that apparent biases may reflect recent changes from computer engines and online broadcasts, yet the likelihood comparison does not test whether the biases persist after conditioning on time or player cohort; this leaves the attribution to transmission biases open to alternative explanations such as evolving chess theory.

Authors: We agree that conditioning on time or player cohort would help isolate transmission effects from secular changes in chess theory. Our Dirichlet-multinomial analysis is population-level and aggregates across the full time span; the abstract already flags the possible role of engines and broadcasts. In revision we will expand the discussion section to state this limitation explicitly and outline time-conditioned extensions as future work. revision: partial
Referee: [Model] Model description: bias effects are evaluated by fitting the Dirichlet-multinomial model directly to the move-frequency data used to test the biases, so reported compatibility with each bias is defined in terms of the fitted parameters themselves rather than an independent test.

Authors: The model is fitted to the observed frequencies and compatibility is assessed by comparing the estimated parameters (and associated likelihoods) against neutral and alternative bias specifications. This is the standard inferential procedure for Dirichlet-multinomial models of transmission; the data determine the parameter values, so the procedure is not circular. revision: no

Circularity Check

0 steps flagged

No significant circularity in statistical inference of transmission biases

full rationale

The paper fits a Dirichlet-multinomial model to chess move-frequency data to estimate parameters associated with frequency-dependent, success, and prestige biases, then reports which biases are compatible with the observed dynamics for specific moves. This is a standard empirical inference procedure, not a deductive derivation that reduces to its inputs by construction. The abstract explicitly notes that apparent biases may reflect unmodeled factors such as computer engines, showing the analysis does not treat the fitted parameters as uniquely forced. No self-citations, uniqueness theorems, or ansatzes are invoked as load-bearing steps in the provided text. The central claim remains an interpretation of model output against external data rather than a tautological renaming or self-definition.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the Dirichlet-multinomial likelihood captures cultural transmission and that fitted bias parameters can be causally interpreted; no independent evidence for these mappings is supplied in the abstract.

free parameters (1)

bias strength parameters
Parameters controlling the strength of frequency-dependent, success, and prestige biases are fitted to the observed move frequencies.

axioms (1)

domain assumption Move choice probabilities in specific positions follow a Dirichlet-multinomial distribution.
This distribution is adopted as the likelihood for the population-level model.

pith-pipeline@v0.9.0 · 5715 in / 1010 out tokens · 27902 ms · 2026-05-24T09:39:44.677180+00:00 · methodology

Cultural transmission of move choice in chess

Core claim

What carries the argument

If this is right

Where Pith is reading between the lines

Load-bearing premise

What would settle it

discussion (0)