Recognition: unknown
Posterior Invariance of Multiplicative Contrasts under Margin Constraints in Contingency Tables
Pith reviewed 2026-05-07 17:40 UTC · model grok-4.3
The pith
The posterior of a generalized odds ratio is invariant to fixing a margin if and only if its coefficients sum to zero within the margin.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Under multinomial sampling or models with fixed partition sums, and assuming marginal and conditional parameters are independent a priori, the posterior distribution of a generalized odds ratio is invariant to fixing a margin if and only if the coefficients defining the contrast sum to zero within the margin.
What carries the argument
Generalized odds ratios as multiplicative contrasts of multinomial cell probabilities, together with the zero-sum condition on their coefficients within each margin that produces posterior invariance.
If this is right
- Inferences about qualifying generalized odds ratios can be conducted equivalently under fixed-margin or random-margin models.
- The result extends classical invariance properties of ordinary odds ratios to a wider family of multiplicative contrasts.
- When coefficients fail to sum to zero within the margin, fixing it necessarily alters the posterior.
- Analysts can check the coefficient sums in advance to decide whether margin constraints will affect their conclusions about associations.
Where Pith is reading between the lines
- The criterion could let researchers fix margins in large tables for computational simplicity whenever the zero-sum condition holds, without changing the target posteriors.
- It may guide choice of which contrasts to study when an experiment or study design controls certain margins by construction.
- Similar coefficient-based tests might be developed for invariance under other sampling schemes or prior structures not covered here.
Load-bearing premise
Marginal and conditional parameters are independent in the prior, together with unspecified mild technical conditions required for the if-and-only-if statement.
What would settle it
Finding a generalized odds ratio whose coefficients sum to zero within a margin yet whose posterior distribution changes when that margin is fixed, or the converse case where non-zero-sum coefficients still yield invariance, would disprove the claim under the stated prior independence.
read the original abstract
Measures of association in contingency tables, such as odds ratios and their generalizations, are often studied under different sampling schemes that either fix or leave random the margins of the table. While classical results show that certain odds ratios are unaffected by constraining the margins, it is less clear when this invariance holds more generally. This paper studies posterior inference for a broad class of multiplicative contrasts of multinomial cell probabilities, which we refer to as generalized odds ratios, and addresses exactly when fixing a margin alters inference about them. We consider Bayesian inference under multinomial sampling and under models in which partition sums of the table are fixed in advance, and assume that the marginal and conditional parameters are independent a priori. Under additional mild assumptions, we show that the posterior distribution of a generalized odds ratio is invariant to fixing a margin if and only if the coefficients defining the contrast sum to zero within the margin.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript derives an if-and-only-if theorem for the invariance of the posterior distribution of generalized odds ratios (multiplicative contrasts of multinomial cell probabilities) under margin constraints in contingency tables. Assuming prior independence between marginal and conditional parameters, the posterior remains unchanged when fixing a margin precisely when the coefficients defining the contrast sum to zero within the margin, for both multinomial sampling and fixed-margin models.
Significance. If the result holds, it offers a clean Bayesian generalization of classical invariance properties for odds ratios and similar association measures. The if-and-only-if characterization, together with the explicit prior independence assumption, provides a precise criterion for when different sampling schemes produce identical posteriors on multiplicative contrasts. This strengthens the theoretical basis for model choice in Bayesian analysis of categorical data and has potential value in applications such as epidemiology. The manuscript delivers a parameter-free mathematical characterization with no circularity or fitted quantities.
minor comments (2)
- The abstract refers to 'additional mild assumptions' without elaboration. While the body of the paper presumably states the regularity conditions needed for posterior factorization, a brief parenthetical summary of the key conditions in the abstract or theorem statement would improve self-contained readability.
- A short numerical illustration (e.g., a 2x2 table with explicit coefficient vectors that do and do not sum to zero within a margin) placed early in the introduction would help readers quickly grasp the necessity direction of the result.
Simulated Author's Rebuttal
We thank the referee for their review and recommendation of minor revision. The referee's summary correctly identifies the paper's if-and-only-if result on posterior invariance of generalized odds ratios under margin constraints, given prior independence of marginal and conditional parameters. No specific major comments or points of criticism appear in the report.
Circularity Check
No significant circularity; mathematical if-and-only-if characterization is self-contained
full rationale
The paper derives a theorem establishing that, under prior independence of marginal and conditional parameters plus mild regularity conditions, the posterior of a generalized odds ratio is invariant to margin constraints if and only if the contrast coefficients sum to zero within the margin. This is a direct consequence of the Bayesian model factorization and does not reduce any prediction or central claim to a fitted input, self-definition, or self-citation chain. No load-bearing step relies on renaming known results or smuggling ansatzes; the result is a characterization of when invariance holds, not a statistical prediction forced by data.
Axiom & Free-Parameter Ledger
axioms (2)
- domain assumption marginal and conditional parameters are independent a priori
- ad hoc to paper additional mild assumptions
Reference graph
Works this paper leans on
-
[1]
kY i=1 θαi i |τj # ≡E
Alan Agresti.An introduction to categorical data analysis. John Wiley & Sons, 2018. Yvonne M Bishop, Stephen E Fienberg, and Paul W Holland.Discrete multivariate analysis: Theory and practice. Springer Science & Business Media, 2007. Jerome Cornfield. A method of estimating comparative rates from clinical data. applica- tions to cancer of the lung, breast...
2018
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.