The Bayesian Origin of the Probability Weighting Function in Human Representation of Probabilities
Pith reviewed 2026-05-18 09:29 UTC · model grok-4.3
The pith
The inverse-S probability weighting function arises because humans encode probabilities with higher precision near 0 and 1 and decode the noisy signals by minimizing Bayes risk.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Probabilities are represented by noisy internal signals that are decoded by Bayes-risk minimization. For bounded stimuli the resulting distortion decomposes into boundary regression, likelihood repulsion, and prior attraction. Consequently the inverse-S weighting pattern observed in behavior implies a stable U-shaped allocation of encoding precision with greater sensitivity near 0 and 1. This encoding is recovered directly from data and remains U-shaped even when the prior is altered by bimodal stimulus distributions.
What carries the argument
Bayesian encoding-decoding model whose distortion for bounded stimuli decomposes into boundary regression, likelihood repulsion, and prior attraction.
If this is right
- Greater encoding precision near 0 and 1 is recovered from judgment of relative frequency, lottery pricing, and risky choice tasks.
- The Bayesian model outperforms deterministic weighting functions, bounded log-odds models, uniform-encoding Bayesian accounts, and matched efficient-coding models on held-out data.
- When stimulus statistics are made bimodal the recovered prior shifts to match the new distribution while the U-shaped encoding remains unchanged.
Where Pith is reading between the lines
- The account predicts that any task using bounded probability-like quantities should exhibit similar U-shaped sensitivity once the prior is accounted for.
- If encoding precision allocation is metabolically or computationally costly, the U-shape may reflect an efficient compromise between boundary accuracy and interior stability.
Load-bearing premise
Decoding of noisy internal probability signals proceeds by minimizing Bayes risk.
What would settle it
An experiment that recovers a non-U-shaped encoding function from choice or judgment data while the inverse-S weighting pattern is still present would falsify the account.
read the original abstract
Humans systematically misrepresent probability in a stereotyped inverse-S pattern. It has been documented for decades, but its origin remains unexplained. We propose a Bayesian encoding-decoding account in which probabilities are represented by noisy internal signals and decoded by Bayes-risk minimization. For bounded probability stimuli, we show that distortion decomposes into boundary regression, likelihood repulsion, and prior attraction, yielding a key prediction: the classic inverse-S-shaped weighting pattern implies a U-shaped allocation of encoding precision with greater sensitivity near 0 and 1. Across judgment of relative frequency, lottery pricing, and risky choice, this U-shape is recovered from data without imposing any functional form on the encoding, and our framework outperforms deterministic weighting functions, bounded log-odds models, uniform-encoding Bayesian accounts, and matched efficient-coding models on held-out data. In a new dot probability estimation experiment with bimodal stimulus statistics, the recovered prior tracks the new distribution while the recovered encoding remains U-shaped. Together, these results identify the inverse-S-shaped probability weighting function as the joint product of a stable U-shaped encoding and a flexible prior, integrated by optimal Bayesian decoding.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a Bayesian encoding-decoding framework to explain the origin of the classic inverse-S-shaped probability weighting function. Probabilities are represented by noisy internal signals decoded via Bayes-risk minimization; for stimuli bounded on [0,1] the resulting distortion decomposes into boundary regression, likelihood repulsion, and prior attraction. This yields the central prediction that the observed inverse-S pattern implies a U-shaped allocation of encoding precision (higher near 0 and 1). The authors recover this U-shape nonparametrically from judgment, pricing, and choice data, show superior held-out predictive performance relative to deterministic weighting functions and alternative Bayesian models, and confirm that the recovered encoding remains U-shaped while the prior adapts in a new bimodal dot-estimation experiment.
Significance. If the central claim holds, the work supplies a mechanistic, process-level account of a long-standing descriptive regularity in decision science and cognitive neuroscience. Notable strengths are the non-parametric recovery of the encoding function, quantitative outperformance on held-out data across tasks, and the targeted new experiment that tests prior flexibility while holding the encoding prediction fixed. These elements move the literature from purely descriptive weighting curves toward falsifiable encoding-decoding models grounded in optimal inference.
major comments (2)
- [Theoretical derivation of distortion decomposition] The implication that the inverse-S weighting pattern entails U-shaped encoding precision is derived under the specific assumption that decoding follows Bayes-risk minimization (posterior mean under squared-error loss or equivalent). The manuscript does not examine whether the same observed weighting function could arise from a non-U-shaped encoding under alternative decoders such as MAP estimation or simple heuristics; this assumption is load-bearing for the theoretical claim that the U-shape is implied by the data rather than imposed by the model.
- [Results on non-parametric recovery across tasks] The non-parametric recovery of the encoding precision function is central to the empirical support, yet the abstract and main results provide no details on the procedure (regularization, basis functions, or uncertainty quantification). Without these, it is difficult to assess how strongly the recovered U-shape is constrained by the data versus the Bayesian decoding assumption itself.
minor comments (2)
- [Abstract] The abstract states that the framework 'outperforms' alternatives on held-out data but does not report the specific cross-validation scheme or effect sizes; adding these would strengthen the model-comparison claim.
- [Model presentation] Notation for the encoding precision function and the three distortion components should be introduced with a single equation or diagram early in the theoretical section to improve readability.
Simulated Author's Rebuttal
We thank the referee for their constructive and insightful comments. We address each major comment point by point below and outline the revisions we will make to strengthen the manuscript.
read point-by-point responses
-
Referee: [Theoretical derivation of distortion decomposition] The implication that the inverse-S weighting pattern entails U-shaped encoding precision is derived under the specific assumption that decoding follows Bayes-risk minimization (posterior mean under squared-error loss or equivalent). The manuscript does not examine whether the same observed weighting function could arise from a non-U-shaped encoding under alternative decoders such as MAP estimation or simple heuristics; this assumption is load-bearing for the theoretical claim that the U-shape is implied by the data rather than imposed by the model.
Authors: We agree that the decomposition and the implication of U-shaped encoding are derived specifically under Bayes-risk minimization (posterior mean). This decoder was selected because it is the Bayes-optimal estimator under squared-error loss and directly corresponds to the expected-loss minimization that arises in the pricing and choice tasks we analyze. We did not explore MAP or heuristic decoders in the original submission. In the revision we will add a dedicated subsection discussing whether an inverse-S weighting function can be generated by non-U-shaped encodings under MAP estimation or simple heuristics, and we will explicitly qualify that the claim of an implied U-shape holds under optimal Bayesian decoding with squared-error loss. revision: partial
-
Referee: [Results on non-parametric recovery across tasks] The non-parametric recovery of the encoding precision function is central to the empirical support, yet the abstract and main results provide no details on the procedure (regularization, basis functions, or uncertainty quantification). Without these, it is difficult to assess how strongly the recovered U-shape is constrained by the data versus the Bayesian decoding assumption itself.
Authors: We thank the referee for noting this gap in presentation. The non-parametric recovery employs cubic-spline basis functions with knots placed at quantiles of the stimulus range, L2 regularization whose strength is chosen by cross-validation, and uncertainty quantified by bootstrap resampling of the data. Full specification of the basis, regularization procedure, and bootstrap confidence bands appears in the Methods and Supplementary Information. In the revision we will add a concise description of these choices to the main-text Results section and ensure the abstract references the supplementary methods, thereby making the constraints on the recovered U-shape transparent. revision: yes
Circularity Check
No significant circularity; derivation follows from model equations and is tested on held-out and new data
full rationale
The paper first derives mathematically that, under noisy internal signals decoded by Bayes-risk minimization, distortion for bounded [0,1] stimuli decomposes into boundary regression, likelihood repulsion, and prior attraction. This decomposition produces the implication that observed inverse-S weighting corresponds to U-shaped encoding precision. The U-shape is then recovered nonparametrically by fitting the model to existing datasets (judgment of relative frequency, lottery pricing, risky choice) without imposing a functional form on the encoding function, and the framework is shown to outperform alternatives on held-out data. A new dot-probability experiment with bimodal stimulus statistics recovers a prior that tracks the altered distribution while the encoding remains U-shaped. No step reduces by construction to the target inverse-S pattern, no load-bearing self-citation chain is used, and the central implication is a direct consequence of the stated model equations rather than a tautology or fitted renaming. The analysis is therefore self-contained against external benchmarks.
Axiom & Free-Parameter Ledger
axioms (2)
- domain assumption Probabilities are represented internally by noisy signals
- domain assumption Decoding minimizes Bayes risk
Lean theorems connected to this paper
-
IndisputableMonolith/Cost/FunctionalEquation.leanwashburn_uniqueness_aczel (J-cost uniqueness) echoes?
echoesECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.
Bias(p) = A1,σ(p)·sign(0.5−p)√J(p) + (1−A2,σ(p))·d/dp(1/J(p)) + (1−A3,σ(p))·(1/J(p))·d/dp(log Pprior(p)) + O(σ⁴)
-
IndisputableMonolith/Foundation/AlphaCoordinateFixation.leancostAlphaLog_fourth_deriv_at_zero unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
log-odds encoding F(p) := log((p+β)/(1−p+β)) yields U-shaped J(p) = (1/σ)(1/(p+β)+1/(1−p+β))
What do these tags mean?
- matches
- The paper's claim is directly supported by a theorem in the formal canon.
- supports
- The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
- extends
- The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
- uses
- The paper appears to rely on the theorem as machinery.
- contradicts
- The paper's claim conflicts with a theorem or certificate in the canon.
- unclear
- Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.