pith. sign in

arxiv: 2605.16108 · v2 · pith:OZELVV6Qnew · submitted 2026-05-15 · 📊 stat.ME · q-bio.QM· stat.AP

Estimating Association Between Paired Outcomes in Clustered Data with Informative Subgroup Size

Owen Visser , Somnath Datta This is my paper

Pith reviewed 2026-05-22 10:12 UTC · model grok-4.3

classification 📊 stat.ME q-bio.QMstat.AP
keywords informative cluster sizeinformative subgroup sizepaired outcomesmarginal associationweighted estimating equationsclustered datawithin-cluster resampling
0
0 comments X

The pith

Three weighted estimating approaches adjust marginal association estimates for paired outcomes when cluster or subgroup size is informative.

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

The paper proposes three weighted estimating approaches for the marginal association between paired outcomes in clustered data. The weights come from within-cluster resampling arguments that extend inverse cluster-size and subgroup-size weighting to handle paired outcome categories. A reader would care because standard estimates can be distorted when the number of observed units or their distribution across outcome categories relates to the outcomes themselves. The work also modifies an existing test for informative subgroup size using Stouffer's method to lower computational cost. Simulations separate unit-level from cluster-level sources of association, and an application to NHANES oral-health data illustrates how different weighting choices affect estimates for periodontal and caries outcomes.

Core claim

We propose three weighted estimating approaches for marginal association between paired outcomes in clustered data. The weights are derived from within-cluster resampling arguments and extend inverse cluster-size and subgroup-size weighting to paired outcome categories. We also modify an existing ISS testing procedure by utilizing Stouffer's method to reduce computational burden.

What carries the argument

Weights derived from within-cluster resampling arguments that extend inverse cluster-size and subgroup-size weighting to paired outcome categories.

If this is right

  • Pair-based weighting reduces bias when association arises through unit-level dependence and subgroup composition is informative.
  • Typical inverse-cluster weighting remains more stable when the association is primarily carried by latent cluster-level structure.
  • The modified testing procedure using Stouffer's method lowers computational burden for detecting informative subgroup size.
  • In NHANES oral-health data, filled-surface outcomes show stronger evidence of informative subgroup size and greater sensitivity to pair-based weighting than decayed-surface outcomes.

Where Pith is reading between the lines

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

  • Analysts may need to compare results across weighting schemes to assess how sensitive conclusions are to the assumed source of association.
  • The resampling-weight approach could be adapted to settings with more than two outcomes per cluster or with time-to-event paired data.
  • In fields that routinely analyze clustered pairs, such as dentistry or ophthalmology, routine reporting of both unweighted and weighted estimates would clarify robustness to informative size.

Load-bearing premise

The within-cluster resampling arguments yield weights that correctly adjust marginal association estimates for informative subgroup size across different sources of association without introducing bias from the weighting itself.

What would settle it

A simulation in which the true source of association (unit-level dependence versus latent cluster-level structure) is known in advance and the weighted estimators are checked for bias under controlled informative subgroup sizes.

Figures

Figures reproduced from arXiv: 2605.16108 by Owen Visser, Somnath Datta.

Figure 1
Figure 1. Figure 1: Representative heatmap of absolute bias relative to the complete-data target, [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
read the original abstract

Informative cluster size (ICS) and informative subgroup size (ISS) can distort marginal association estimates when the number of observed units, or their distribution across outcome-defined categories, is related to the outcomes under study. This issue is especially relevant for paired outcomes, where the observed association can depend on cluster size, paired-category composition, and the process by which units become available for analysis. We propose three weighted estimating approaches for marginal association between paired outcomes in clustered data. The weights are derived from within-cluster resampling arguments and extend inverse cluster-size and subgroup-size weighting to paired outcome categories. We also modify an existing ISS testing procedure by utilizing Stouffer's method to reduce computational burden. To evaluate the methods, we develop a simulator for clustered paired outcomes that separates unit-level association, latent cluster-level association, and outcome-dependent retention. Simulations show that pair-based weighting can reduce bias when association arises through unit-level dependence and subgroup composition is informative, but can attenuate association carried by latent cluster-level structure. Typical inverse-cluster weighting remains more stable when the association is primarily cluster-level. Application to NHANES oral-health data shows small positive periodontal and caries associations overall, with filled-surface outcomes showing stronger ISS evidence and greater sensitivity to pair-based weighting than decayed-surface outcomes. These results indicate that marginal association under ICS and ISS should be interpreted in relation to the source of association, observed-unit structure, and assumptions used to choose the weighting scheme.

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.

Referee Report

2 major / 3 minor

Summary. The manuscript proposes three weighted estimating approaches for marginal association between paired outcomes in clustered data subject to informative cluster size (ICS) and informative subgroup size (ISS). Weights are derived from within-cluster resampling arguments and extend inverse cluster-size and subgroup-size weighting to paired outcome categories. A modified ISS testing procedure using Stouffer's method is also presented. Simulations separate unit-level association, latent cluster-level association, and outcome-dependent retention, showing pair-based weighting reduces bias under unit-level dependence with informative subgroup composition but attenuates association carried by latent cluster-level structure. Typical inverse-cluster weighting is more stable for cluster-level association. The methods are applied to NHANES oral-health data, showing small positive periodontal and caries associations with greater sensitivity to pair-based weighting for filled-surface outcomes.

Significance. If the performance claims hold, the work offers practical guidance for marginal association estimation in clustered paired data where ICS or ISS may distort estimates. The simulation design that isolates different sources of association is a clear strength, as it demonstrates when pair-based versus cluster-based weighting is preferable. The NHANES application illustrates real-world relevance in oral-health epidemiology. These contributions could help practitioners interpret marginal associations conditional on the suspected mechanism of association and observed-unit structure.

major comments (2)
  1. [§2] §2 (Weighted estimating approaches): The central claim that the within-cluster resampling weights correctly adjust marginal association estimates for ISS without introducing bias from the weighting itself is load-bearing. Simulations show attenuation under latent cluster-level association, which raises the question whether the target marginal parameter is preserved across mechanisms; an explicit derivation or statement of the estimand under each data-generating process would resolve whether the attenuation reflects bias or a change in the target.
  2. [Simulation study] Simulation study section: While the separation of unit-level versus cluster-level association sources is useful, the reported results do not include the true marginal association value under each simulation scenario. Without this benchmark, it is difficult to determine whether the observed attenuation under cluster-level structure indicates a problem with the weighting scheme or simply a correctly estimated but different marginal quantity.
minor comments (3)
  1. [Abstract] Abstract: The statement that pair-based weighting 'can attenuate association carried by latent cluster-level structure' is important but would benefit from a brief parenthetical note on the magnitude of attenuation observed in simulations.
  2. [Methods] Methods: Notation for the three weighting schemes could be introduced with a small illustrative table showing how inverse cluster-size and pair-based weights differ for a simple 2x2 paired outcome table.
  3. [Data application] Data application: Reporting the distribution of cluster sizes and the proportion of informative subgroups in the NHANES sample would help readers assess the practical relevance of the ISS findings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive assessment of our work. The suggestions to clarify the target estimands and to report true marginal association values in the simulations are helpful for strengthening the interpretation of our results. We address each major comment below and have revised the manuscript to incorporate these clarifications.

read point-by-point responses

  1. Referee: [§2] §2 (Weighted estimating approaches): The central claim that the within-cluster resampling weights correctly adjust marginal association estimates for ISS without introducing bias from the weighting itself is load-bearing. Simulations show attenuation under latent cluster-level association, which raises the question whether the target marginal parameter is preserved across mechanisms; an explicit derivation or statement of the estimand under each data-generating process would resolve whether the attenuation reflects bias or a change in the target.

    Authors: We agree that an explicit statement of the estimand under each mechanism strengthens the manuscript. In the revised version, we have added a dedicated paragraph in Section 2 deriving the target marginal association parameter for the unit-level and latent cluster-level data-generating processes. Under unit-level dependence with informative subgroup size, the pair-based weights target the marginal association conditional on the realized subgroup composition, which is the natural parameter when association operates at the unit level. Under latent cluster-level association, the population marginal incorporates between-cluster variation; pair-based weighting then correctly targets a within-subgroup marginal that is attenuated relative to the unconditional quantity. This distinction shows that the observed attenuation is a consequence of targeting a different but well-defined estimand rather than bias induced by the weighting procedure itself. revision: yes

  2. Referee: [Simulation study] Simulation study section: While the separation of unit-level versus cluster-level association sources is useful, the reported results do not include the true marginal association value under each simulation scenario. Without this benchmark, it is difficult to determine whether the observed attenuation under cluster-level structure indicates a problem with the weighting scheme or simply a correctly estimated but different marginal quantity.

    Authors: We appreciate this point and have now included the analytically derived true marginal association values for each simulation scenario in the revised tables and accompanying text. These benchmarks confirm that, when association is generated at the unit level with informative subgroup composition, pair-based weighting recovers the true marginal with substantially lower bias. When association is generated at the latent cluster level, pair-based weighting estimates a different but correctly specified marginal (the average within-subgroup association), while standard inverse-cluster weighting recovers the unconditional marginal. The added values therefore demonstrate that the attenuation is not a defect in the weighting scheme but follows directly from the change in target parameter across mechanisms. revision: yes

Circularity Check

0 steps flagged

No significant circularity; weighting derivation is independent of target parameter

full rationale

The paper derives its three weighted estimating approaches from within-cluster resampling arguments that are presented as logically prior to and independent of the specific marginal association parameter being estimated. Simulations explicitly separate unit-level dependence, latent cluster-level association, and outcome-dependent retention to evaluate performance, providing an external benchmark rather than a self-referential loop. No equation or claim reduces a prediction or uniqueness result to a fitted input or self-citation by construction; the differential behavior under different association sources is reported as an empirical finding, not a tautology. The central claim therefore remains self-contained against the stated assumptions and simulation design.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of resampling-derived weights for paired categories and the simulator's ability to isolate unit-level versus cluster-level association mechanisms.

axioms (1)
  • domain assumption Within-cluster resampling arguments extend to derive unbiased weights for paired outcome categories under informative subgroup size.
    Stated as the basis for the three weighted estimating approaches in the abstract.

pith-pipeline@v0.9.0 · 5792 in / 1200 out tokens · 45465 ms · 2026-05-22T10:12:18.170332+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

  1. [1]

    Informative cluster sizes for subcluster-level covariates and weighted generalized estimating equations.Biometrics, 67(3):843–851, 2011

    Ying Huang and Brian Leroux. Informative cluster sizes for subcluster-level covariates and weighted generalized estimating equations.Biometrics, 67(3):843–851, 2011

    work page 2011

  2. [2]

    Inference for marginal linear models for clustered longitudinal data with potentially informative cluster sizes.Statistical Methods in Medical Research, 20(4):347–367, 2011

    Ming Wang, Maiying Kong, and Somnath Datta. Inference for marginal linear models for clustered longitudinal data with potentially informative cluster sizes.Statistical Methods in Medical Research, 20(4):347–367, 2011

    work page 2011

  3. [3]

    Can the unit size predict outcomes? testing for informativeness in three-level designs.Statistics in Medicine, 44(6):e70041, 2025

    Samuel Anyaso-Samuel, Somnath Datta, Eva Roos, and Jaakko Nevalainen. Can the unit size predict outcomes? testing for informativeness in three-level designs.Statistics in Medicine, 44(6):e70041, 2025

    work page 2025

  4. [4]

    Within-cluster resampling

    Elaine B Hoffman, Pranab K Sen, and Clarice R Weinberg. Within-cluster resampling. Biometrika, 88(4):1121–1134, 2001

    work page 2001

  5. [5]

    Marginal analyses of clustered data when cluster size is informative.Biometrics, 59(1):36–42, 2003

    John M Williamson, Somnath Datta, and Glen A Satten. Marginal analyses of clustered data when cluster size is informative.Biometrics, 59(1):36–42, 2003

    work page 2003

  6. [6]

    A rank-sum test for clustered data when the number of subjects in a group within a cluster is informative.Biometrics, 72(2):432–440, 2016

    Sandipan Dutta and Somnath Datta. A rank-sum test for clustered data when the number of subjects in a group within a cluster is informative.Biometrics, 72(2):432–440, 2016

    work page 2016

  7. [7]

    Marginal association measures for clustered data.Statistics in medicine, 30(27):3181–3191, 2011

    Douglas J Lorenz, Somnath Datta, and Susan J Harkema. Marginal association measures for clustered data.Statistics in medicine, 30(27):3181–3191, 2011

    work page 2011

  8. [8]

    Inferring marginal association with paired and unpaired clustered data.Statistical methods in medical research, 27(6):1806– 1817, 2018

    Douglas J Lorenz, Steven Levy, and Somnath Datta. Inferring marginal association with paired and unpaired clustered data.Statistical methods in medical research, 27(6):1806– 1817, 2018. 26

    work page 2018

  9. [9]

    Methods for observed-cluster inference when cluster size is informative: a review and clarifications.Biometrics, 70(2): 449–456, 2014

    Shaun R Seaman, Menelaos Pavlou, and Andrew J Copas. Methods for observed-cluster inference when cluster size is informative: a review and clarifications.Biometrics, 70(2): 449–456, 2014

    work page 2014

  10. [10]

    Dental caries are positively associated with periodontal disease severity.Clinical Oral Investigations, 23(10):3811–3819, 2019

    Robert Durand, Arezou Roufegarinejad, Fatiha Chandad, Pierre H Rompr´ e, Ren´ e Voyer, Bryan S Michalowicz, and Elham Emami. Dental caries are positively associated with periodontal disease severity.Clinical Oral Investigations, 23(10):3811–3819, 2019

    work page 2019

  11. [11]

    Testing for marginal covariate effect when the subgroup size induced by the covariate is informative.Statistical Methods in Medical Research, 33(7):1264–1277, 2024

    Samuel Anyaso-Samuel and Somnath Datta. Testing for marginal covariate effect when the subgroup size induced by the covariate is informative.Statistical Methods in Medical Research, 33(7):1264–1277, 2024

    work page 2024

  12. [12]

    Samuel A Stouffer, Edward A Suchman, Leland C DeVinney, Shirley A Star, and Robin M Williams Jr.The american soldier: Adjustment during army life.(studies in social psy- chology in world war ii), vol. 1. Princeton Univ. Press, 1949. 27

    work page 1949