pith. sign in

arxiv: 2605.19772 · v1 · pith:HKRIT2RKnew · submitted 2026-05-19 · 📊 stat.ME

Assessing covariate-adjusted risk differences in small-sample clinical trials

Martin Schnuerch , Alex Ocampo , Klaus K\"ahler Holst , Christian Stock This is my paper

Pith reviewed 2026-05-20 02:21 UTC · model grok-4.3

classification 📊 stat.ME
keywords risk differencecovariate adjustmentsmall sampleg-computationtype I errorclinical trialsMantel-Haenszelexact tests
0
0 comments X

The pith

Much of the Type-I error inflation in g-computation for covariate-adjusted risk differences stems from misalignment between the marginal estimand and the variance estimator rather than small sample size alone.

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

The paper compares methods for estimating marginal risk differences in randomized trials with binary outcomes and prognostic categorical covariates when the total sample size is 150 or smaller. It tests exact unconditional procedures, Mantel-Haenszel estimators, and several g-computation approaches that standardize over the covariates to produce a population-averaged effect. Standard Wald-type standard errors after g-computation produce inflated false-positive rates in the smallest samples, while robust or penalized variance estimators restore error control at some cost in power. Classical methods that do not rely on covariate adjustment keep Type-I error near the nominal level but forgo the precision gains that proper adjustment can deliver. The authors conclude that the observed inflation largely reflects a mismatch between what is being estimated and how its uncertainty is quantified, and they offer concrete guidance on aligning these pieces for reliable inference.

Core claim

In small-sample randomized clinical trials with binary endpoints and categorical baseline covariates, g-computation estimators for marginal risk differences show inflated Type-I error under standard Wald inference, yet this inflation is largely corrected by robust or penalized variance estimators; classical Mantel-Haenszel and exact unconditional tests maintain error control without adjustment but sacrifice efficiency.

What carries the argument

g-computation (standardization) for marginal risk differences, paired with alternative variance estimators, contrasted against Mantel-Haenszel and Suissa-Shuster exact tests.

If this is right

  • Robust or penalized variance estimation should be used with g-computation to keep Type-I error near nominal levels in samples of 150 or less.
  • Covariate adjustment can still improve precision once variance estimation is aligned with the target estimand.
  • Mantel-Haenszel and exact unconditional tests remain reliable fallbacks that avoid inflation but forgo efficiency gains from adjustment.
  • Trial protocols and analysis plans should explicitly match the chosen estimand, variance method, and inferential target.

Where Pith is reading between the lines

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

  • The same variance-mismatch problem may appear in observational studies that use standardization for risk differences with modest sample sizes.
  • Regulatory recommendations favoring marginal estimands could usefully specify variance practices that preserve error control in small trials.
  • Extending the simulations to continuous covariates or to time-to-event outcomes would test whether the misalignment pattern generalizes beyond binary endpoints.

Load-bearing premise

The simulation scenarios with N at most 150, prognostic categorical covariates, and the chosen data-generating processes accurately reflect the behavior of real small-sample randomized trials with binary endpoints.

What would settle it

A simulation or re-analysis of real trial data in which Type-I error remains substantially inflated even after switching to a variance estimator that matches the marginal estimand would indicate that sample size itself is the dominant driver.

Figures

Figures reproduced from arXiv: 2605.19772 by Alex Ocampo, Christian Stock, Klaus K\"ahler Holst, Martin Schnuerch.

Figure 1
Figure 1. Figure 1: Proportion of simulation runs with separation issues. Separation did not occur in [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Type I error rates (top row) and statistical power (middle and bottom row) for the [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Estimation bias, i.e. the difference between the average estimated and the true risk [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Root mean squared error for the investigated statistical methods across simulation [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Coverage of two-sided 95% confidence intervals for all considered statistical methods. [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
read the original abstract

Binary endpoints are common in clinical trials and conditional odds ratios have traditionally been used to assess treatment effects. However, the interpretation of odds ratios is difficult, they are non-collapsible and rely on strong assumptions in order to be a relevant overall summary measure for the trial. As an alternative, risk differences have gained increasing prominence as a more interpretable, clinically meaningful and assumption-lean measure of treatment effects. This shift has also been motivated by new regulatory guidance, which emphasizes the relevance of marginal estimands and encourages covariate adjustment. Yet, covariate-adjusted inference for risk differences, particularly in smaller samples, has methodological subtleties and lacks well-established best practices. We conduct a simulation study comparing methods for estimating and testing risk differences in small-sample ($N \leq 150$) randomized clinical trials with prognostic categorical baseline covariates, focusing on exact unconditional tests, Mantel-Haenszel methods, and $g$-computation (standardization) approaches. We find that several $g$-computation approaches exhibit inflated Type-I error in very small samples when standard Wald-type inference is applied, whereas robust or penalized variants improve error control at the expense of power. Classical methods such as the Mantel-Haenszel and Suissa-Shuster tests remain robust but may forgo efficiency gains from covariate adjustment. Overall, our results indicate that much of the observed Type-I error inflation reflects misalignment between estimand and variance estimation rather than small sample size alone. Based on these results, we provide practical recommendations to guide method selection that align the estimand, variance estimation, and inferential target.

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 / 2 minor

Summary. The manuscript reports a simulation study comparing methods for estimating and testing covariate-adjusted risk differences in small-sample (N ≤ 150) randomized trials with binary endpoints and prognostic categorical baseline covariates. Methods examined include exact unconditional tests, Mantel-Haenszel procedures, and g-computation (standardization) estimators using standard Wald, robust, and penalized variance approaches. The central claim is that Type I error inflation observed with standard g-computation largely reflects misalignment between the marginal risk difference estimand and the variance estimator rather than small sample size alone, with robust/penalized variants improving error control at the cost of power while classical methods remain robust but less efficient.

Significance. If the central attribution holds, the results would provide actionable guidance for method selection in small clinical trials emphasizing alignment of estimand, variance estimation, and inference target. This is relevant given regulatory focus on marginal estimands. The simulation design directly probes the claimed source of inflation, a methodological strength, though the evidential support for the primary claim is moderate without additional diagnostics.

major comments (2)
  1. [Simulation study] Simulation study section: the manuscript does not report a direct comparison of the g-computation Wald-type standard errors to the Monte Carlo (empirical) standard deviation of the estimator across replications. Without this calibration check, it remains unclear whether the standard variance estimators systematically underestimate sampling variability for the marginal estimand or whether the error-control gains from robust/penalized variants arise from general conservatism; this comparison is load-bearing for the claim that misalignment (rather than small N alone) primarily drives the observed Type I error inflation.
  2. [Methods] Methods and data-generating process: insufficient detail is provided on the exact parameter values, covariate distributions, outcome model coefficients, and number of distinct simulation scenarios. These specifics are necessary to evaluate whether the chosen N ≤ 150 settings with categorical prognostic factors accurately represent the behavior of real small-sample RCTs and thereby support generalizability of the misalignment conclusion.
minor comments (2)
  1. [Abstract] The abstract would benefit from explicitly stating the number of Monte Carlo replications and the precise criteria used to declare Type I error inflation or acceptable control.
  2. [Introduction] Notation for the marginal versus conditional risk difference could be introduced earlier and used consistently to avoid ambiguity when discussing estimand-variance alignment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below and have revised the manuscript to strengthen the simulation diagnostics and methodological transparency.

read point-by-point responses

  1. Referee: Simulation study section: the manuscript does not report a direct comparison of the g-computation Wald-type standard errors to the Monte Carlo (empirical) standard deviation of the estimator across replications. Without this calibration check, it remains unclear whether the standard variance estimators systematically underestimate sampling variability for the marginal estimand or whether the error-control gains from robust/penalized variants arise from general conservatism; this comparison is load-bearing for the claim that misalignment (rather than small N alone) primarily drives the observed Type I error inflation.

    Authors: We agree that a direct calibration of estimated standard errors against empirical Monte Carlo standard deviations would provide stronger support for attributing Type I error inflation to estimand-variance misalignment. In the revised manuscript we have added this comparison (new Table 3 and accompanying text in the Results section). For each g-computation variant we now report the average model-based SE and the empirical SD of the point estimates over 10,000 replications. The results show that the standard Wald SE systematically underestimates the true variability of the marginal risk difference for N ≤ 50, while robust and penalized variants yield ratios closer to unity, confirming that the inflation is driven primarily by the mismatch rather than small-sample bias alone. revision: yes

  2. Referee: Methods and data-generating process: insufficient detail is provided on the exact parameter values, covariate distributions, outcome model coefficients, and number of distinct simulation scenarios. These specifics are necessary to evaluate whether the chosen N ≤ 150 settings with categorical prognostic factors accurately represent the behavior of real small-sample RCTs and thereby support generalizability of the misalignment conclusion.

    Authors: We thank the referee for highlighting the need for greater reproducibility. The revised Methods section now includes the full logistic regression coefficients for the outcome model, the exact multinomial probabilities for each categorical covariate, the target marginal risk differences under the null and alternatives, and an explicit enumeration of all 48 simulation scenarios (combinations of N, number of covariates, and prevalence levels). These details have also been summarized in a new supplementary table to allow readers to assess how well the design reflects typical small-sample RCT settings. revision: yes

Circularity Check

0 steps flagged

No circularity in simulation-based method comparison

full rationale

The paper is a simulation study evaluating finite-sample performance of estimators and tests for covariate-adjusted risk differences under specified data-generating processes with N ≤ 150. All reported Type-I error rates, power, and recommendations are direct empirical outputs from the Monte Carlo experiments rather than mathematical derivations, fitted parameters relabeled as predictions, or claims justified solely by self-citation chains. The central claim that misalignment between estimand and variance estimator (rather than sample size alone) drives inflation is an interpretation of the simulation results and does not reduce to any self-definitional or fitted-input construction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The paper's conclusions rest on the validity of the simulation design choices and standard statistical assumptions for binary data and randomization; no new entities are postulated.

free parameters (2)
  • simulation sample sizes (N ≤ 150)
    Chosen by authors to represent small-sample regime; affects power and error rate comparisons.
  • number and distribution of categorical covariates
    Prognostic baseline factors selected to mimic clinical trial settings.
axioms (2)
  • domain assumption Randomized treatment assignment ensures exchangeability conditional on observed covariates.
    Invoked when interpreting covariate-adjusted estimates as causal risk differences.
  • domain assumption Binary outcome model is correctly specified for g-computation standardization.
    Required for the standardization approach to target the marginal risk difference.

pith-pipeline@v0.9.0 · 5819 in / 1458 out tokens · 33441 ms · 2026-05-20T02:21:50.033654+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

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.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    Down with odds ratios!,

    D. L. Sackett, J. J. Deeks, and D. G. Altman, “Down with odds ratios!,”Evid Based Med, vol. 1, pp. 164–166, Oct. 1996

    work page 1996

  2. [2]

    Standardized risk and description of results from multivariable modeling of a binary response,

    B.-R. Kim, R. L. Carter, P. V. Rao, M. Ariet, and M. B. Resnick, “Standardized risk and description of results from multivariable modeling of a binary response,”Biom J, vol. 48, no. 1, pp. 54–66, 2006

    work page 2006

  3. [3]

    The Odds Ratio and Logistic Regression

    R. G. Newcombe,Confidence Intervals for Proportions and Related Measures of Effect Size. Chapman & Hall/ CRC, Biostatistics Series, 2013. Section 11, “The Odds Ratio and Logistic Regression”, pp. 223–254. 25

    work page 2013

  4. [4]

    Making apples from oranges: Comparing noncol- lapsible effect estimators and their standard errors after adjustment for different covariate sets,

    R. Daniel, J. Zhang, and D. Farewell, “Making apples from oranges: Comparing noncol- lapsible effect estimators and their standard errors after adjustment for different covariate sets,”Biom J, vol. 63, no. 3, pp. 528–557, 2021

    work page 2021

  5. [5]

    Adjusting for Covariates in Randomized Clin- ical Trials for Drugs and Biological Products. Guidance for Industry

    US Food and Drug Administration (FDA), “Adjusting for Covariates in Randomized Clin- ical Trials for Drugs and Biological Products. Guidance for Industry.” Docket number, FDA-2019-D-0934, May 2023. URL:https://www.fda.gov/media/148910/download

    work page 2019

  6. [6]

    Covariate adjustment in randomized controlled trials: General concepts and practical considerations,

    K. van Lancker, F. Bretz, and O. Dukes, “Covariate adjustment in randomized controlled trials: General concepts and practical considerations,”Clin Trials, vol. 21, no. 4, pp. 399– 411, 2024

    work page 2024

  7. [7]

    Conditional and Unconditional Treatment Effects in Randomized Clinical Trials: Estimands, Estimation, and Interpretation,

    J. Wei, J. Xu, B. Bornkamp, R. Lin, H. Tian, D. Xi, X. Zhang, Z. Zhao, and S. Roychoud- hury, “Conditional and Unconditional Treatment Effects in Randomized Clinical Trials: Estimands, Estimation, and Interpretation,”Stat Biopharm Res, vol. 16, no. 3, pp. 371– 381, 2024

    work page 2024

  8. [8]

    ICH E9 (R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials to the Guideline on Statistical Principles for Clinical Trials

    International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use, “ICH E9 (R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials to the Guideline on Statistical Principles for Clinical Trials.”https://www.ich. org/page/efficacy-guidelines, Nov. 2019. Final version

    work page 2019

  9. [9]

    Statistical aspects of the analysis of data from retrospective studies of disease,

    N. Mantel and W. Haenszel, “Statistical aspects of the analysis of data from retrospective studies of disease,”J Natl Cancer Inst, vol. 22, no. 4, pp. 719–748, 1959

    work page 1959

  10. [10]

    Minimum expected cell size requirements for the Mantel- Haenszel one-degree-of-freedom chi-square test and a related rapid procedure,

    N. Mantel and J. L. Fleiss, “Minimum expected cell size requirements for the Mantel- Haenszel one-degree-of-freedom chi-square test and a related rapid procedure,”Am J Epi- demiol, vol. 112, no. 1, pp. 129–134, 1980

    work page 1980

  11. [11]

    Mantel-Haenszel and Related Methods for Multiple 2×2 Tables

    A. Agresti,Categorical Data Analysis. John Wiley & Sons, 3rd ed., 2013. Section 6.4, “Mantel-Haenszel and Related Methods for Multiple 2×2 Tables”, pp. 225–232

    work page 2013

  12. [12]

    A graphical approach to the identification and estimation of causal pa- rameters in mortality studies with sustained exposure periods,

    J. M. Robins, “A graphical approach to the identification and estimation of causal pa- rameters in mortality studies with sustained exposure periods,”J Chronic Dis, vol. 40, no. Suppl. 2, pp. 139S–161S, 1987

    work page 1987

  13. [13]

    Covariate adjustment and estimation of difference in proportions in randomized clinical trials,

    J. Liu and D. Xi, “Covariate adjustment and estimation of difference in proportions in randomized clinical trials,”Pharm Stat, vol. 23, no. 6, pp. 884–905, 2024

    work page 2024

  14. [14]

    An introduction to the augmented inverse propensity weighted estimator,

    A. N. Glynn and K. M. Quinn, “An introduction to the augmented inverse propensity weighted estimator,”Polit Anal, vol. 18, no. 1, pp. 36–56, 2010

    work page 2010

  15. [15]

    Variance reduction in randomised trials by inverse probability weighting using the propensity score,

    E. J. Williamson, A. Forbes, and I. R. White, “Variance reduction in randomised trials by inverse probability weighting using the propensity score,”Stat Med, vol. 33, pp. 721–737, Feb 2014

    work page 2014

  16. [16]

    Design of randomized clinical trials with a binary endpoint: Conditional versus unconditional analyses of a two-by-two table,

    E. L. Korn and B. Freidlin, “Design of randomized clinical trials with a binary endpoint: Conditional versus unconditional analyses of a two-by-two table,”Stat Med, vol. 43, no. 16, pp. 3109–3123, 2024

    work page 2024

  17. [17]

    Exact Unconditional Sample Sizes for the 2×2 Binomial Trial,

    S. Suissa and J. J. Shuster, “Exact Unconditional Sample Sizes for the 2×2 Binomial Trial,”J R Stat Soc Ser A, vol. 148, no. 4, p. 317, 1985

    work page 1985

  18. [18]

    Covariate adjustment in randomized clinical trials: From general theory to practical insights,

    M. S. Bannick, Y. Yi, and T. Ye, “Covariate adjustment in randomized clinical trials: From general theory to practical insights,”Clin Trials, May 2026

    work page 2026

  19. [19]

    Why estimands are needed to define treatment effects in clinical trials,

    O. N. Keene, H. Lynggaard, S. Englert, V. Lanius, and D. Wright, “Why estimands are needed to define treatment effects in clinical trials,”BMC Medicine, vol. 21, no. 1, p. 276, 2023. 26

    work page 2023

  20. [20]

    Marginal and conditional summary measures: Transportability and compatibility across studies

    A. Remiro-Az´ ocar, D. M. Phillippo, N. J. Welton, S. Dias, A. E. Ades, A. Heath, and G. Baio, “Marginal and conditional summary measures: Transportability and compatibility across studies.” Wiley StatsRef: Statistics Reference Online, 2025

    work page 2025

  21. [21]

    M. A. Hern´ an and J. M. Robins,Causal Inference: What If. Boca Raton: Chapman & Hall/CRC, 2024

    work page 2024

  22. [22]

    Estimating the variance of covariate- adjusted estimators of average treatment effects in clinical trials with binary endpoints,

    D. Magirr, C. Wang, A. Przybylski, and M. Baillie, “Estimating the variance of covariate- adjusted estimators of average treatment effects in clinical trials with binary endpoints,” Pharm Stat, vol. 24, no. 4, p. e70021, 2025

    work page 2025

  23. [23]

    The logic of inductive inference,

    R. A. Fisher, “The logic of inductive inference,”Journal of the Royal Statistical Society, vol. 98, no. 1, pp. 39–82, 1935

    work page 1935

  24. [24]

    Clarifying the role of the Mantel- Haenszel risk difference estimator in randomized clinical trials,

    X. Qiu, Y. Qian, J. Yi, J. Wang, Y. Du, Y. Yi, and T. Ye, “Clarifying the role of the Mantel- Haenszel risk difference estimator in randomized clinical trials,”Biometrics, vol. 81, no. 4, p. ujaf142, 2025

    work page 2025

  25. [25]

    A Note on the Mantel-Haenszel Estimators When the Com- mon Effect Assumptions Are Violated,

    H. Noma and K. Nagashima, “A Note on the Mantel-Haenszel Estimators When the Com- mon Effect Assumptions Are Violated,”Epidemiol Methods, vol. 5, no. 1, pp. 19–35, 2016

    work page 2016

  26. [26]

    On the Variance Estimator for the Mantel- Haenszel Risk Difference,

    T. Sato, S. Greenland, and J. M. Robins, “On the Variance Estimator for the Mantel- Haenszel Risk Difference,”Biometrics, vol. 45, no. 4, pp. 1323–1324, 1989

    work page 1989

  27. [27]

    A new approach to causal inference in mortality studies with a sustained exposure period – application to control of the healthy worker survivor effect,

    J. Robins, “A new approach to causal inference in mortality studies with a sustained exposure period – application to control of the healthy worker survivor effect,”Math Model, vol. 7, no. 9-12, pp. 1393–1512, 1986

    work page 1986

  28. [28]

    Covariate adjustment for two-sample treatment comparisons in randomized clinical trials: a principled yet flexible approach,

    A. A. Tsiatis, M. Davidian, M. Zhang, and X. Lu, “Covariate adjustment for two-sample treatment comparisons in randomized clinical trials: a principled yet flexible approach,” Stat Med, vol. 27, no. 23, pp. 4658–4677, 2008

    work page 2008

  29. [29]

    Covariate adjustment in randomized trials with binary outcomes: targeted maximum likelihood estimation,

    K. L. Moore and M. J. van der Laan, “Covariate adjustment in randomized trials with binary outcomes: targeted maximum likelihood estimation,”Stat Med, vol. 28, no. 1, pp. 39–64, 2009

    work page 2009

  30. [30]

    Robust variance estimation for covariate-adjusted unconditional treatment effect in randomized clinical trials with binary outcomes,

    T. Ye, M. Bannick, Y. Yi, and J. Shao, “Robust variance estimation for covariate-adjusted unconditional treatment effect in randomized clinical trials with binary outcomes,”Stat Theory Relat Fields, vol. 7, no. 2, pp. 159–163, 2023

    work page 2023

  31. [31]

    A robust score test in G-computation for covariate adjustment in randomized clinical trials leveraging different variance estimators via influence functions,

    X. Zhang, H. Chu, L. Liu, and S. Roychoudhury, “A robust score test in G-computation for covariate adjustment in randomized clinical trials leveraging different variance estimators via influence functions,”Stat Med, vol. 44, no. 7, p. e70080, 2025

    work page 2025

  32. [32]

    Standardized estimates from categorical regression models,

    M. M. Joffe and S. Greenland, “Standardized estimates from categorical regression models,” Stat Med, vol. 14, pp. 2131–2141, Oct. 1995

    work page 1995

  33. [33]

    Covariate-Adjusted Dif- ference in Proportions from Clinical Trials Using Logistic Regression and Weighted Risk Differences,

    M. Ge, L. K. Durham, R. D. Meyer, W. Xie, and N. Thomas, “Covariate-Adjusted Dif- ference in Proportions from Clinical Trials Using Logistic Regression and Weighted Risk Differences,”Drug Inf J, vol. 45, no. 4, pp. 481–493, 2011

    work page 2011

  34. [34]

    A general form of covariate adjustment in clinical trials under covariate-adaptive randomization,

    M. S. Bannick, J. Shao, J. Liu, Y. Du, Y. Yi, and T. Ye, “A general form of covariate adjustment in clinical trials under covariate-adaptive randomization,”Biometrika, vol. 112, no. 3, p. asaf029, 2025

    work page 2025

  35. [35]

    Improving precision by adjust- ing for prognostic baseline variables in randomized trials with binary outcomes, without regression model assumptions,

    J. A. Steingrimsson, D. F. Hanley, and M. Rosenblum, “Improving precision by adjust- ing for prognostic baseline variables in randomized trials with binary outcomes, without regression model assumptions,”Contemp Clin Trials, vol. 54, pp. 18–24, 2017. 27

    work page 2017

  36. [36]

    A solution to the problem of separation in logistic regression,

    G. Heinze and M. Schemper, “A solution to the problem of separation in logistic regression,” Stat Med, vol. 21, no. 16, pp. 2409–2419, 2002

    work page 2002

  37. [37]

    Firth’s logistic regression with rare events: accurate effect estimates and predictions?,

    R. Puhr, G. Heinze, M. Nold, L. Lusa, and A. Geroldinger, “Firth’s logistic regression with rare events: accurate effect estimates and predictions?,”Stat Med, vol. 36, no. 14, pp. 2302–2317, 2017

    work page 2017

  38. [38]

    On Generalized Score Tests,

    D. D. Boos, “On Generalized Score Tests,”Am Stat, vol. 46, no. 4, pp. 327–333, 1992

    work page 1992

  39. [39]

    Bias reduction of maximum likelihood estimates,

    D. Firth, “Bias reduction of maximum likelihood estimates,”Biometrika, vol. 80, no. 1, pp. 27–38, 1993

    work page 1993

  40. [40]

    The iterative bisection procedure: a useful tool for determining parameter values in data-generating processes in Monte Carlo simulations,

    P. C. Austin, “The iterative bisection procedure: a useful tool for determining parameter values in data-generating processes in Monte Carlo simulations,”BMC Med Res Methodol, vol. 23, no. 1, p. 45, 2023

    work page 2023

  41. [41]

    Handling missingness, failures, and non- convergence in simulation studies: A review of current practices and recommendations,

    S. Pawel, F. Bartoˇ s, B. S. Siepe, and A. Lohmann, “Handling missingness, failures, and non- convergence in simulation studies: A review of current practices and recommendations,” Am Stat, vol. 80, no. 1, pp. 31–48, 2026

    work page 2026

  42. [42]

    Agnostic notes on regression adjustments to experimental data: Reexamining Freedman’s critique,

    W. Lin, “Agnostic notes on regression adjustments to experimental data: Reexamining Freedman’s critique,”Ann Appl Stat, vol. 7, no. 1, pp. 295–318, 2013

    work page 2013

  43. [43]

    Leveraging prognostic baseline variables to gain preci- sion in randomized trials,

    E. Colantuoni and M. Rosenblum, “Leveraging prognostic baseline variables to gain preci- sion in randomized trials,”Stat Med, vol. 34, no. 18, pp. 2602–2617, 2015

    work page 2015

  44. [44]

    A com- parison of covariate adjustment approaches under model misspecification in individually randomized trials,

    M. S. Tackney, T. Morris, I. White, C. Leyrat, K. Diaz-Ordaz, and E. Williamson, “A com- parison of covariate adjustment approaches under model misspecification in individually randomized trials,”Trials, vol. 24, no. 1, p. 14, 2023

    work page 2023

  45. [45]

    Randomization Does Not Justify Logistic Regression,

    D. A. Freedman, “Randomization Does Not Justify Logistic Regression,”Stat Sci, vol. 23, no. 2, pp. 237–249, 2008

    work page 2008

  46. [46]

    Model-Robust Inference for Clinical Trials that Improve Precision by Stratified Random- ization and Covariate Adjustment,

    B. Wang, R. Susukida, R. Mojtabai, Masoumeh, Amin-Esmaeili, and M. Rosenblum, “Model-Robust Inference for Clinical Trials that Improve Precision by Stratified Random- ization and Covariate Adjustment,”J Am Stat Assoc, vol. 118, no. 542, pp. 1152–1163, 2023

    work page 2023

  47. [47]

    Small-sample adjustments for Wald-type tests using sand- wich estimators,

    M. P. Fay and B. I. Graubard, “Small-sample adjustments for Wald-type tests using sand- wich estimators,”Biometrics, vol. 57, no. 4, pp. 1198–1206, 2001

    work page 2001

  48. [48]

    Simple, efficient estimators of treatment effects in randomized trials using generalized linear models to leverage baseline variables,

    M. Rosenblum and M. J. van der Laan, “Simple, efficient estimators of treatment effects in randomized trials using generalized linear models to leverage baseline variables,”Int J Biostat, vol. 6, no. 1, p. Article 13, 2010

    work page 2010

  49. [49]

    Planning a method for covariate adjustment in individually randomised trials: a practical guide,

    T. P. Morris, A. S. Walker, E. J. Williamson, and I. R. White, “Planning a method for covariate adjustment in individually randomised trials: a practical guide,”Trials, vol. 23, p. 328, Apr 2022

    work page 2022

  50. [50]

    A new and improved confidence interval for the Mantel-Haenszel risk difference,

    B. Klingenberg, “A new and improved confidence interval for the Mantel-Haenszel risk difference,”Stat Med, vol. 33, no. 17, pp. 2968–2983, 2014

    work page 2014

  51. [51]

    margins: Marginal effects for model objects

    T. J. Leeper, “margins: Marginal effects for model objects.” R package, version 0.3.28,

    work page

  52. [52]

    URL:https://cran.r-project.org/package=margins

    work page

  53. [53]

    beeca: Exact unconditional inference for binary endpoints

    A. Przybylski, M. Baillie, C. Wang, and D. Magirr, “beeca: Exact unconditional inference for binary endpoints.” R package, version 0.2.0, 2024. URL:https://CRAN.R-project. org/package=beeca. 28

    work page 2024

  54. [54]

    Various versatile variances: An object-oriented im- plementation of clustered covariances in r,

    A. Zeileis, S. K¨ oll, and N. Graham, “Various versatile variances: An object-oriented im- plementation of clustered covariances in r,”J Stat Softw, vol. 95, no. 1, pp. 1–36, 2020

    work page 2020

  55. [55]

    Robincar2: Robust inference for covariate adjustment in randomized clinical trials

    L. Li, M. Bannick, D. Xi, T. Ye, and Y. Yi, “Robincar2: Robust inference for covariate adjustment in randomized clinical trials.” R package, version 0.1.1, 2025. URL:https: //cran.r-project.org/package=RobinCar2

    work page 2025

  56. [56]

    logistf: Firth’s bias-reduced logistic re- gression

    G. Heinze, M. Ploner, L. Jiricka, and G. Steiner, “logistf: Firth’s bias-reduced logistic re- gression.” R package, version 1.26.1, 2025. URL:https://cran.r-project.org/package= logistf. A Appendix A.1 Software implementation This appendix summarizes practical aspects of software implementation for the statistical meth- ods described and discussed in the ...

    work page 2025