pith. sign in

arxiv: 2605.03278 · v2 · submitted 2026-05-05 · 📊 stat.ME · cs.AI

Copula-Based Endogeneity Correction for Doubly Robust Estimation of Treatment Effect

Sahil Shikalgar , Md. Noor-E-Alam This is my paper

Pith reviewed 2026-05-08 18:59 UTC · model grok-4.3

classification 📊 stat.ME cs.AI
keywords doubly robust estimationendogeneity correctionGaussian copulatreatment effectscausal inferenceobservational dataNHANESblood pressure
0
0 comments X

The pith

Gaussian copulas correct for endogeneity in doubly robust treatment effect estimates without instruments.

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

The paper develops a way to make doubly robust estimates of treatment effects reliable even when variables like proxy measures are correlated with unmeasured factors. It incorporates Gaussian copulas to capture the dependence between these endogenous covariates and the error terms in the models for treatment assignment and outcomes. This correction keeps the key property that the estimator is consistent if either the treatment model or the outcome model is correctly specified. Simulations show the standard approach has large bias from endogeneity while the new one does not. An example using survey data on nutritional counseling and blood pressure shows how the correction can change conclusions to match other evidence.

Core claim

The authors introduce a copula-corrected doubly robust estimator for the treatment effect that models the joint distribution of endogenous covariates and regression errors via Gaussian copulas in both the treatment and outcome equations. This yields consistent estimates of the average treatment effect while retaining the double robustness property, requiring only that one of the two models be correctly specified. Monte Carlo simulations across various data-generating processes confirm that the naive estimator is biased but the corrected version recovers the true effect, and the NHANES application demonstrates that the corrected estimate of counseling's effect on blood pressure becomes insign

What carries the argument

The Gaussian copula that links the marginal distributions of the endogenous covariates and the error terms to allow consistent estimation in the doubly robust framework.

If this is right

  • The corrected estimator eliminates substantial bias observed in naive doubly robust estimation under endogeneity in simulations.
  • Application to NHANES data indicates that nutritional counseling has no statistically significant effect on blood pressure after correction, unlike the naive positive association.
  • The method provides a practical alternative for estimating treatment effects when instrumental variables are unavailable.
  • Double robustness is preserved, so the estimator remains consistent if either the treatment or outcome model is correctly specified despite endogeneity.

Where Pith is reading between the lines

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

  • This approach may be extended to other copula specifications if the dependence structure deviates from Gaussian.
  • Researchers could test the method in randomized trial data where endogeneity is artificially introduced to validate performance.
  • Connections to other parametric corrections for endogeneity in observational studies could be explored for robustness.
  • The change in NHANES conclusions suggests it could alter interpretations in similar healthcare studies with proxy variables.

Load-bearing premise

The Gaussian copula correctly represents the dependence between the endogenous covariates and the error terms, and this modeling choice does not break the double robustness when only one model is correct.

What would settle it

Generating data with a non-Gaussian dependence structure such as a t-copula, applying the Gaussian copula estimator, and checking whether it still produces unbiased estimates when one model is correctly specified would falsify the claim if bias persists.

Figures

Figures reproduced from arXiv: 2605.03278 by Md. Noor-E-Alam, Sahil Shikalgar.

Figure 1
Figure 1. Figure 1: Bias (%) by endogeneity level and misspecification in scenario 1. (−5.6% Scenario 1, −2.5% Scenario 2) compared to the naive estimator (−21.6% Sce￾nario 1, −27.6% Scenario 2). Under outcome model misspecification (Outcome Wrong), the CEDR estimator achieves −2.0% bias (Scenario 1) and −1.2% bias (Scenario 2), compared to −15.8% and −23.2% for the naive estimator. 3.2.4. Variance–Bias Tradeoff The CEDR esti… view at source ↗
Figure 2
Figure 2. Figure 2: Bias (%) by endogeneity level and misspecification in scenario 2. proxies for metabolic health. Both variables are endogenous, that CEDR is designed to address. 4.1. Data Description We use publicly available data from the NHANES pre-pandemic cycle (2017–March 2020), a nationally representative cross-sectional survey administered by the National Center for Health Statistics. NHANES collects demographic, di… view at source ↗
read the original abstract

Doubly Robust (DR) estimation of treatment effect relies on an untestable assumption that is the absence of unobserved confounding. This assumption is par- ticularly problematic in the context of healthcare research, where variables like pre- scription refill rates serve as proxies for unobserved behaviors such as medication adherence. These proxy variables are often endogenous, exhibiting correlation with the regression error term due to unmeasured confounding or measurement error. We propose a copula-corrected doubly robust estimator that addresses endogeneity in both the treatment and outcome models without requiring instrumental variables. Gaussian copulas model the joint distribution of endogenous covariates and the error term, enabling consistent estimation while preserving the doubly robust property that requires correct specification of either the treatment or outcome model, not both. Monte Carlo simulations demonstrate that naive DR estimation exhibits substantial bias under endogeneity, whereas our corrected estimator recovers unbiased treatment effects across different data-generating processes. We apply our method to examine the effect of nutritional counseling on blood pressure using the National Health and Nutrition Examination Survey (NHANES) data. Naive DR estimation suggests counseling is associated with increased blood pressure. After copula correction, this effect becomes statistically insignificant, consistent with literature showing modest effects of nutri- Counseling in reducing blood pressure. Our methodology provides researchers with a practical tool for obtaining treatment effects in the presence of endogeneity.

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 / 1 minor

Summary. The paper proposes a copula-corrected doubly robust estimator for average treatment effects that models endogeneity between covariates and error terms in both the treatment (propensity) and outcome equations via Gaussian copulas, without requiring instrumental variables. It asserts that this correction preserves the double-robustness property (consistency if at least one of the two models is correctly specified), demonstrates bias reduction relative to naive DR in Monte Carlo experiments, and applies the method to NHANES data on nutritional counseling and blood pressure, where the corrected estimate is statistically insignificant (unlike the naive DR result).

Significance. If the double-robustness property is rigorously preserved after incorporating the estimated copula dependence parameter, the approach would provide a practical, IV-free tool for bias correction in observational studies with endogenous proxies, which is common in healthcare and social-science applications. The reported Monte Carlo bias reduction and the change in substantive conclusion on the NHANES example illustrate potential utility, though the strength hinges on verification that the parametric copula adjustment does not introduce non-vanishing bias under partial model correctness.

major comments (2)
  1. [Abstract / estimator definition] Abstract and estimator construction: the central claim that the Gaussian-copula adjustment preserves double robustness is load-bearing but unsupported by any explicit derivation showing that the joint estimation of the copula dependence parameter leaves the estimator consistent when only the propensity score or only the outcome regression (including its copula link) is correctly specified. Because the copula supplies the explicit functional form of the dependence, misspecification of the copula family introduces a bias term whose cancellation under single-model correctness must be demonstrated algebraically rather than asserted.
  2. [Monte Carlo simulations] Monte Carlo section: the reported bias reduction is shown under data-generating processes that presumably match the Gaussian-copula assumption; to substantiate the DR claim, the simulations must include explicit cases in which the copula is misspecified while exactly one of the treatment or outcome models remains correct, and report whether the estimator remains consistent in those cases.
minor comments (1)
  1. [Abstract] Abstract, final paragraph: the phrase 'modest effects of nutri- Counseling' appears to be a line-break artifact and should be corrected to 'nutritional counseling'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address the major comments point by point below, acknowledging where the current version requires strengthening through additional derivations and simulations.

read point-by-point responses

  1. Referee: [Abstract / estimator definition] Abstract and estimator construction: the central claim that the Gaussian-copula adjustment preserves double robustness is load-bearing but unsupported by any explicit derivation showing that the joint estimation of the copula dependence parameter leaves the estimator consistent when only the propensity score or only the outcome regression (including its copula link) is correctly specified. Because the copula supplies the explicit functional form of the dependence, misspecification of the copula family introduces a bias term whose cancellation under single-model correctness must be demonstrated algebraically rather than asserted.

    Authors: We agree that the manuscript currently asserts preservation of the double-robustness property without an explicit algebraic demonstration of consistency when the copula dependence parameter is jointly estimated and only one of the two models is correctly specified. In the revised manuscript we will add a dedicated appendix section deriving the asymptotic consistency of the copula-corrected DR estimator under the stated partial correctness conditions, explicitly showing how any bias arising from copula-family misspecification cancels when either the propensity-score model or the outcome model (including its copula link) is correctly specified. revision: yes

  2. Referee: [Monte Carlo simulations] Monte Carlo section: the reported bias reduction is shown under data-generating processes that presumably match the Gaussian-copula assumption; to substantiate the DR claim, the simulations must include explicit cases in which the copula is misspecified while exactly one of the treatment or outcome models remains correct, and report whether the estimator remains consistent in those cases.

    Authors: We concur that the existing Monte Carlo design assumes the correct copula family and therefore does not yet fully test the double-robustness claim under copula misspecification. In the revision we will augment the simulation study with additional scenarios in which the fitted copula family differs from the data-generating copula (e.g., Clayton or Frank when data are generated under Gaussian dependence) while exactly one of the treatment or outcome models remains correctly specified, and we will report finite-sample bias and coverage results for these cases. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained under explicit modeling assumptions

full rationale

The paper derives a copula-augmented doubly robust estimator by specifying Gaussian copulas to capture dependence between endogenous covariates and error terms in the treatment and outcome models. The copula parameter is estimated from data as part of the joint model, and the treatment effect is then obtained via the corrected estimating equations. This construction does not reduce the final estimator to a tautological restatement of the fitted copula parameter or the raw data by construction; the target parameter remains a distinct functional of the observed outcomes, treatments, and covariates. No self-citations appear load-bearing in the abstract or described chain, no uniqueness theorems are invoked from prior author work, and no fitted input is relabeled as an independent prediction. The preservation of the doubly robust property is asserted under the maintained parametric assumptions rather than by definitional equivalence, leaving the estimator falsifiable against external benchmarks such as standard DR estimators or IV-based alternatives. The approach is therefore self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the standard doubly robust assumption plus a parametric copula model for dependence; no new entities are postulated.

free parameters (1)
  • copula dependence parameter
    Parameter(s) of the Gaussian copula that capture correlation between endogenous regressors and error terms; fitted from the data.
axioms (2)
  • domain assumption Either the treatment propensity model or the outcome regression model is correctly specified
    Core assumption of doubly robust estimation invoked to preserve consistency.
  • domain assumption The joint distribution of endogenous covariates and errors is adequately modeled by a Gaussian copula
    The modeling choice that enables the correction without instruments.

pith-pipeline@v0.9.0 · 5538 in / 1272 out tokens · 27226 ms · 2026-05-08T18:59:11.969548+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.

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

23 extracted references · 23 canonical work pages

  1. [1]

    Pharmacoepidemiology and drug safety , volume=

    Machine learning for improving high-dimensional proxy confounder adjustment in healthcare database studies: An overview of the current literature , author=. Pharmacoepidemiology and drug safety , volume=. 2022 , publisher=

    work page 2022

  2. [2]

    Evaluation & the Health Professions , volume=

    The perilous use of proxy variables , author=. Evaluation & the Health Professions , volume=. 2021 , publisher=

    work page 2021

  3. [3]

    Clinical Nutrition , volume=

    Efficacy of nutritional recommendations given by registered dietitians compared to other healthcare providers in reducing arterial blood pressure: systematic review and meta-analysis , author=. Clinical Nutrition , volume=. 2018 , publisher=

    work page 2018

  4. [4]

    Hypertension , volume=

    Dietary approaches to stop hypertension dietary intervention improves blood pressure and vascular health in youth with elevated blood pressure , author=. Hypertension , volume=. 2021 , publisher=

    work page 2021

  5. [5]

    Health and nutrition examination survey plan and operations, 1999-2010 , author=

    work page 1999

  6. [6]

    Journal of the Academy of Marketing Science , volume=

    Dealing with regression models’ endogeneity by means of an adjusted estimator for the Gaussian copula approach , author=. Journal of the Academy of Marketing Science , volume=. 2025 , publisher=

    work page 2025

  7. [7]

    , author=

    Using copulas to enable causal inference from nonexperimental data: Tutorial and simulation studies. , author=. Psychological Methods , volume=. 2023 , publisher=

    work page 2023

  8. [8]

    Journal of the American statistical Association , volume=

    Statistics and causal inference , author=. Journal of the American statistical Association , volume=. 1986 , publisher=

    work page 1986

  9. [9]

    , author=

    Estimating causal effects of treatments in randomized and nonrandomized studies. , author=. Journal of educational Psychology , volume=. 1974 , publisher=

    work page 1974

  10. [10]

    Biometrics , volume=

    Doubly robust estimation in missing data and causal inference models , author=. Biometrics , volume=. 2005 , publisher=

    work page 2005

  11. [11]

    Biometrika , volume=

    The central role of the propensity score in observational studies for causal effects , author=. Biometrika , volume=. 1983 , publisher=

    work page 1983

  12. [12]

    Journal of the American statistical Association , volume=

    Estimation of regression coefficients when some regressors are not always observed , author=. Journal of the American statistical Association , volume=. 1994 , publisher=

    work page 1994

  13. [13]

    The Econometrics Journal , volume=

    Modelling sample selection using Archimedean copulas , author=. The Econometrics Journal , volume=. 2003 , publisher=

    work page 2003

  14. [14]

    Marketing Science , volume=

    Handling endogenous regressors by joint estimation using copulas , author=. Marketing Science , volume=. 2012 , publisher=

    work page 2012

  15. [15]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Making sense of sensitivity: Extending omitted variable bias , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2020 , publisher=

    work page 2020

  16. [16]

    Epidemiology , volume=

    Bias formulas for sensitivity analysis of unmeasured confounding for general outcomes, treatments, and confounders , author=. Epidemiology , volume=. 2011 , publisher=

    work page 2011

  17. [17]

    Epidemiology , volume=

    Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable , author=. Epidemiology , volume=. 2006 , publisher=

    work page 2006

  18. [18]

    American Journal of Political Science , volume=

    Rain, rain, go away: 194 potential exclusion-restriction violations for studies using weather as an instrumental variable , author=. American Journal of Political Science , volume=. 2025 , publisher=

    work page 2025

  19. [19]

    2015 , publisher=

    Causal inference in statistics, social, and biomedical sciences , author=. 2015 , publisher=

    work page 2015

  20. [20]

    2007 , institution=

    An economic analysis of exclusion restrictions for instrumental variable estimation , author=. 2007 , institution=

    work page 2007

  21. [21]

    Management science , volume=

    Endogeneity in brand choice models , author=. Management science , volume=. 1999 , publisher=

    work page 1999

  22. [22]

    American journal of epidemiology , volume=

    Doubly robust estimation of causal effects , author=. American journal of epidemiology , volume=. 2011 , publisher=

    work page 2011

  23. [23]

    Journal of Economic perspectives , volume=

    Instrumental variables and the search for identification: From supply and demand to natural experiments , author=. Journal of Economic perspectives , volume=. 2001 , publisher=

    work page 2001