Pith. sign in

REVIEW 2 major objections 5 minor 58 references

Constructing g-computation estimators: two case studies in selection bias

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper shows that reading causal identification formulas as stacked estimating equations yields g-computation estimators that remove treatment-induced selection bias and bias from confounding-plus-selection settings with no joint…

desk verdict Useful, honestly-scoped estimators for two selection-bias settings, but the displayed g3 estimating equations contain a real typo that must be fixed (subtracts µ1 in the µ0 block), and the Case 2 exchangeability statement is imprecise. read the letter →

arxiv 2506.03347 v2 pith:FEZ2PYWM submitted 2025-06-03 stat.ME

Paul N Zivich , Haidong Lu This is my paper
classification stat.ME
keywords g-computationselectionbiasestimatingequationsM-estimationempiricalsandwichvariancetreatment-inducediteratedconditionalexpectationcausalinference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper shows that the standard g-computation algorithm can silently fail under two selection-bias structures and supplies two rebuilt estimators that succeed. In treatment-induced selection, conditioning on a post-treatment variable creates collider bias, so the paper replaces the usual unconditional pseudo-outcome mean with a mean taken within each treatment arm, $\mu_a = E[E(Y \mid A=a, X, S=1) \mid A=a]$. In the joint confounding-and-selection setting where no covariate set adjusts for both biases, the paper iterates expectations one level further, $\mu_a = E\{E[E(Y \mid A=a, Z, X, S=1) \mid A=a, Z]\}$. Both estimators are written as stacked estimating equations whose unbiasedness at the true parameter is proved, giving consistency and asymptotic normality with variance from the empirical sandwich estimator. Simulations confirm near-zero bias and nominal confidence-interval coverage where naive complete-case analysis and standard g-computation are badly biased.

What carries the argument

The machinery is the stacked estimating equation, a vector-valued estimating function $g(Z_i; \theta)$ whose root defines the estimator, built from score functions of the outcome model plus estimating functions for pseudo-outcome means or regressions and the causal contrast. M-estimation theory reduces the asymptotic proof to checking $E[g(Z; \theta)] = 0$ at the true $\theta$, and the empirical sandwich variance estimator then propagates uncertainty from every fitted model into the final confidence interval. The two case-specific moves are the treatment-arm indicators multiplying the pseudo-outcome mean equations in $g_2$, which restrict averaging to $A=a$, and the intermediate pseudo-outcome regression on $(A, Z)$ in $g_3$, which implements the iterated conditional expectation needed when no single covariate set closes both bias paths.

What would settle it

Run the Appendix 2.3 data-generating mechanism at $n = 10^6$ and compare $g_2$ with standard g-computation on the same data; the proof predicts $g_2$'s bias tends to zero while standard g-computation's bias remains near $-0.144$. If standard g-computation's bias also vanishes, the conditional-outer-expectation step is not needed, and if $g_2$'s bias does not vanish, the unbiasedness proof is wrong. The same check with $g_3$ versus the three standard adjustment sets in Appendix 3.3 settles the second case.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a causal identification formula can be read directly as an estimation recipe, provided each conditional expectation is represented by a correctly specified parametric model and each averaging step is encoded as an estimating function. For treatment-induced selection, the estimand must condition the outer average on treatment: $\mu_a = E[E(Y \mid A=a, X, S=1) \mid A=a]$. The proposed estimator $g_2$ stacks the complete-case score for $m(X; \beta) = E[Y \mid A, X, S=1]$, the conditional-mean equations $A_i(\hat{Y}_i^1 - \mu_1) = 0$ and $(1-A_i)(\hat{Y}_i^0 - \mu_0) = 0$, and $(\mu_1 - \mu_0) - \psi = 0$. For the confounding-plus-selection case with no joint adjustment set, $g_3$ adds an intermediate regression $m(V; \gamma_a) = E[\hat{Y}^a \mid A, Z]$ and averages the updated pseudo-outcomes $\tilde{Y}^a$. The paper proves each estimating equation has expectation zero at the true parameter under correct specification, so standard M-estimation theory yields $\sqrt{n}$ consistency and asymptotic normality; the empirical sandwich variance accounts for uncertainty in all fitted models.

Load-bearing premise

The whole construction relies on every fitted regression matching the true conditional expectation, namely the outcome model among completers and, in the second estimator, the intermediate pseudo-outcome regression, plus the causal assumptions that treatment is independent of potential outcomes given the adjustment variables and that selection into follow-up is independent of potential outcomes; misspecify any one of these and the estimator's consistency and coverage guarantees fall away.

Editorial extensions

If this is right

  • For the treatment-induced selection structure, standard g-computation that adjusts for the post-treatment variable $X$ is not just inefficient but heavily biased, while $g_2$ recovers the average causal effect with sandwich-variance confidence intervals at nominal coverage.
  • For the confounding-plus-selection structure with no joint adjustment set, no single covariate adjustment, $\{Z\}$, $\{X\}$, or $\{X, Z\}$, recovers the effect, while iterated g-computation $g_3$ does.
  • Expressing the estimator as a stack of estimating equations makes the uncertainty from every fitted model travel into the final variance through the empirical sandwich estimator, so bootstrap resampling is not required.
  • The proof template, derive the identification formula, write each conditional expectation as a parametric model, and check unbiasedness of each estimating equation at the truth, carries over to other causal structures with selection or missing data.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Read as a recipe, the same identification-formula-to-estimating-equations translation points toward multiply robust estimators: combine the g-computation stack with an inverse-probability-weighting stack, or derive the influence function from the identification formula and build an augmented estimator; the paper only sketches this direction.
  • A natural robustness experiment would replace the intermediate pseudo-outcome regression in $g_3$ with models stratified by treatment arm, or fully saturated interactions, and check whether bias and coverage improve when the outcome model is misspecified; the paper suggests stratification weakens modeling assumptions but does not test it.
  • Because the sandwich variance is valid only when all estimated nuisance functions are included in the stacked equations, feeding $g_2$ or $g_3$ with data-adaptive model or variable selection would require those selection steps to be part of the stack, or the reported confidence intervals could understate uncertainty.
  • One implication left implicit is that the translation can be inverted: given any identification formula expressed as nested conditional expectations, an epidemiologist can mechanically write down the corresponding estimating stack, making novel estimators checkable in a few lines of code.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper presents a didactic framework for translating causal identification results into g-computation estimators expressed as stacked estimating equations. Two case studies are treated: treatment-induced selection bias (Case 1, estimator g2) and joint confounding and selection bias without a common adjustment set (Case 2, estimator g3). For each case, the authors rederive the identifying formula, propose a corresponding M-estimator, prove unbiasedness of the estimating equations under correct model specification, and verify finite-sample performance by simulation. The simulations show near-zero bias, standard error ratios near one, and nominal coverage for the proposed estimators, while standard comparators are biased. The paper explicitly acknowledges reliance on correct parametric specification and provides reproducible code.

Significance. If the central claims hold, the paper provides a useful and accessible template for epidemiologists to construct g-computation estimators for selection-bias structures that are not amenable to a single adjustment set. The use of stacked estimating equations with the empirical sandwich variance estimator is practical and connects the estimators to existing software. The formal unbiasedness proofs in Appendices 1-3 are standard, the simulation results match the theory, and the code is publicly available. The main caveat is that consistency and nominal inference are conditional on correct specification of every parametric model, a limitation the authors explicitly acknowledge. Overall, the paper makes a solid methodological contribution to applied causal inference, though the displayed definition of g3 must be corrected before the stated results are internally consistent.

major comments (2)
  1. [Section 5.1 and Appendix A3.2] The displayed stacked estimating function g3 contains a load-bearing typographical error. In Section 5.1, the fourth block is written as \tilde Y^0_i - \mu1, and the same error is repeated in Appendix A3.2. At the true parameter, E[\tilde Y^0] = \mu0 by the identification result in Eq. (3) and the proof in A3.2, so the expectation of this block is \mu0 - \mu1, which is generally nonzero. Thus the displayed estimating equations are not unbiased at the true θ and directly contradict the paper's central claim that g3 is unbiased and hence consistent; the stacked system may even have no solution when \mu1 ≠ \mu0. The proof in A3.2 and the simulation in Table 2 evidently use \tilde Y^0_i - \mu0, so the error is in the displayed definition rather than the evaluated mathematics. Nevertheless, the definition must be corrected in both places before the manuscript can be accepted, and the authors should verify that the released code implements the corrected block.
  2. [Appendix A3.1] The identification assumptions for Case 2 are stated imprecisely. The exchangeability assumptions are written as "Y^a ⨿ A | Z" and "Y^a ⨿ A, Z, X"; the latter does not involve S and is not the assumption used in the proof. The proof requires a conditional selection exchangeability statement such as Y^a ⨿ S | A, Z, X, together with a positivity assumption such as Pr(S = 1 | A = a, Z, X) > 0. As written, the identification derivation in A3.1 starts from an ambiguous or incorrect assumption, making the proof not self-contained. This is a load-bearing issue because the identification result in Eq. (3) is the foundation for the estimator g3. The authors should restate the full set of exchangeability and positivity assumptions precisely.
minor comments (5)
  1. [Section 5.1] The phrase "how we an build" should read "how we can build." Also, Figure 3's caption describes the structure as "treatment-induced selection bias," but the case is confounding and selection bias lacking a superset; the caption should match the text.
  2. [Appendix A2.2 and A3.4] In the proof of unbiasedness for g2, the phrase "law of total exchangeability" appears to be a typo for "law of total expectation." Additionally, in Appendix 3 the heading "A2.4: Inverse Probability W eighting Estimator" should be "A3.4" and the word "Weighting" is misspelled with a space; the same typo appears in Appendix 2's heading.
  3. [Appendices A2.3 and A3.3] The data-generating mechanisms contain the typo "Beroulli" for "Bernoulli." Also in A2.3 the outcome model for S=1 is written as "Bernoulli(expit(0.5 + 0.75U - 1A))", which is clear but the coefficient notation could be simplified for readability.
  4. [Section 2 and Discussion] There are minor spelling issues such as "esitmation" in Section 2 and "ammenable" and "cusal" in the Discussion. These do not affect the science but should be corrected in a final proofreading pass.
  5. [Section 6] The Discussion acknowledges that the authors assumed the parametric outcome model was correctly specified, but the main text's consistency claims would be clearer if they explicitly stated that all parametric models, including the intermediate pseudo-outcome regression m(V; γ_a) in g3, must be correctly specified for the displayed unbiasedness and nominal coverage to hold.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the identification results are rederived in the appendices and the estimator consistency proofs rest on explicit model-specification assumptions rather than on fitted targets or load-bearing self-citation.

full rationale

The paper's derivation chain is self-contained. The motivating identification formulas (Equations 1, 2, and 3) are each re-proved in Appendices A1.1, A2.1, and A3.1 from stated exchangeability, positivity, and consistency assumptions, rather than being imported as black boxes. The consistency and asymptotic normality arguments in A1.2, A2.2, and A3.2 show unbiasedness of each stacked estimating function under correct specification of the parametric outcome and pseudo-outcome models; these proofs use the identification results plus the definition of the models. The final estimating function (µ1 - µ0) - ψ = 0 is a definitional link between the target parameter and the means, not a fitted quantity, and the paper explicitly flags the parametric specification requirement in the Discussion. Self-citations such as Zivich et al. [1] for the Case 2 identification structure are accompanied by in-paper proofs, so they are not load-bearing. The simulation studies compare estimators against a known true ψ generated from a data-generating mechanism, so there is no fitted input masquerading as a prediction. The displayed typo in Section 5.1 and A3.2, where the fourth block of g3 shows '~Y^0_i - µ1' instead of '~Y^0_i - µ0', is an internal inconsistency that should be corrected, but it is a notational/transcription error and not a circularity: the accompanying proof and the implemented simulation use the correct µ0 block. Accordingly, no step in the paper's derivation reduces to its own inputs by construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The estimators add no new physical entities or fitted universal constants. All free parameters are working regression coefficients. The main epistemic weight rests on causal identification assumptions and correct parametric specification, both stated in the paper. The Case 2 exchangeability assumption is imprecisely written in A3.1 but the proof requires the standard version.

free parameters (2)
  • Outcome model coefficients beta
    Estimated from complete cases in m(W; beta) = E[Y | A,Z,X,S=1]; key source of pseudo-outcomes, but they are standard regression parameters, not physical constants.
  • Intermediate pseudo-outcome model coefficients gamma_1, gamma_0
    Estimated by regressing Yhat^a on A and Z; needed to form Ytilde^a in Case 2; assumed correctly specified.
assumptions (4)
  • standard math M-estimation regularity conditions (finite-dimensional theta, unique root, smooth estimating functions) hold.
    Used to turn unbiased estimating equations into consistency and sqrt-n asymptotic normality; see Section 2 and Appendix 1.2.
  • domain assumption Causal consistency, exchangeability, and positivity hold as encoded in Figures 2 and 3.
    Required for the causal parameter mu_a to equal the observed-data functionals in Equations 1 through 3; proofs in Appendices 1 through 3.
  • domain assumption Correct specification of m(W; beta) and m(V; gamma_a) for the relevant conditional expectations.
    The unbiasedness proofs set model-predicted expectations equal to true conditional expectations (A1.2, A2.2, A3.2); the paper acknowledges this in the Discussion.
  • domain assumption For Case 2, conditional selection exchangeability Y^a independent of S given A, Z, X.
    Implied by Figure 3 and used in A3.1, though the appendix statement is garbled; without it the iterated expectation identity fails.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Constructing g-computation estimators: two case studies in selection bias." pith.science (2026). https://pith.science/paper/FEZ2PYWM

@misc{pith2026250603347,
  author       = {Pith},
  title        = {Pith review of: Constructing g-computation estimators: two case studies in selection bias},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FEZ2PYWM}},
  note         = {Machine review of arXiv:2506.03347}
}
read the original abstract

G-computation is a useful estimation method that can be adapted to address various biases in epidemiology. However, these adaptations may not be obvious for some complex causal structures. This challenge is an example of the much wider issue of translating a causal diagram into a novel estimation strategy. To highlight these challenges, we consider two recent cases from the selection bias literature: treatment-induced selection and co-occurrence of biases that lack a joint adjustment set. For each case study, we show how g-computation can be adapted, describe how to implement that adaptation, show some general statistical properties, and illustrate the estimator using simulation. To simplify both the theoretical study and practical application of our estimators, we express the proposed g-computation estimators as stacked estimating equations. These examples illustrate how epidemiologists can translate identification results into a g-computation estimator and study the theoretical and finite-sample properties of a novel estimator.

Figures

Figures reproduced from arXiv: 2506.03347 by the authors.

Figure 2
Figure 2. Single world intervention graph for treatment-induced selection bias [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Single world intervention graph for treatment-induced selection bias [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 51 canonical work pages

  1. [1]

    On the Use of Covariate Supersets for Identification Conditions,

    P. N. Zivich, B. E. Shook-Sa, J. K. Edwards, D. Westreich, and S. R. Cole, “On the Use of Covariate Supersets for Identification Conditions,” Epidemiology, vol. 33, pp. 559–562, mar 29 2022

    work page 2022

  2. [2]

    Toward a Clearer Definition of Selection Bias When Estimating Causal Effects,

    H. Lu, S. R. Cole, C. J. Howe, and D. Westreich, “Toward a Clearer Definition of Selection Bias When Estimating Causal Effects,” Epidemiology, vol. 33, p. 699, 9 2022

    work page 2022

  3. [3]

    Selection Bias Requires Selection: The Case of Collider Stratification Bias,

    H. Lu, G. S. Gonsalves, and D. Westreich, “Selection Bias Requires Selection: The Case of Collider Stratification Bias,” American Journal of Epidemiology , p. kwad213, nov 3 2023

    work page 2023

  4. [4]

    The Evolution of Selection Bias in the Recent Epidemiologic Literature—A Selective Overview,

    H. Lu, C. J. Howe, P. N. Zivich, G. S. Gonsalves, and D. Westreich, “The Evolution of Selection Bias in the Recent Epidemiologic Literature—A Selective Overview,” American Journal of Epidemiology , p. kwae282, aug 12 2024

    work page 2024

  5. [5]

    A Potential Outcomes Approach to Selection Bias,

    E. Kenah, “A Potential Outcomes Approach to Selection Bias,” Epidemiology (Cambridge, Mass.) , vol. 34, pp. 865–872, nov 1 2023. PMID: 37708480

    work page 2023

  6. [6]

    Simple graphical rules for assessing selection bias in general-population and selected-sample treatment effects,

    M. B. Mathur and I. Shpitser, “Simple graphical rules for assessing selection bias in general-population and selected-sample treatment effects,” American Journal of Epidemiology , p. kwae145, jun 20 2024

    work page 2024

  7. [7]

    Marginal structural models and causal inference in epidemiology,

    J. M. Robins, M. A. Hernan, and B. Brumback, “Marginal structural models and causal inference in epidemiology,” Epidemiology, vol. 11, pp. 550–60, 9 2000. edition: 2000/08/24

    work page 2000

  8. [8]

    Estimating marginal structural model parameters for time-fixed, binary actions with g-computation and estimating equations,

    P. N. Zivich and B. E. Shook-Sa, “Estimating marginal structural model parameters for time-fixed, binary actions with g-computation and estimating equations,” American Journal of Epidemiology , vol. 194, pp. 1464–1466, may 7 2025

    work page 2025

Show all 58 references
  1. [9]

    Reflection on modern methods: combining weights for con- founding and missing data,

    R. K. Ross, A. Breskin, T. L. Breger, and D. Westreich, “Reflection on modern methods: combining weights for con- founding and missing data,” International Journal of Epidemiology , p. dyab205, sep 18 2021

    2021

  2. [10]

    R. M. Daniel, Double Robustness , pp. 1–14. John Wiley & Sons, Ltd, 2018. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781118445112.stat08068 DOI: 10.1002/9781118445112.stat08068

    2018 doi

  3. [11]

    Implementation of G-computation on a simulated data set: demonstration of a causal inference technique,

    J. M. Snowden, S. Rose, and K. M. Mortimer, “Implementation of G-computation on a simulated data set: demonstration of a causal inference technique,” American journal of epidemiology , vol. 173, no. 7, pp. 731–738, 2011. edition: 03/16

    2011

  4. [12]

    G-computation, propensity score-based methods, and targeted maximum likelihood estimator for causal inference with different covariates sets: a comparative simulation study,

    A. Chatton, F. Le Borgne, C. Leyrat, F. Gillaizeau, C. Rousseau, L. Barbin, D. Laplaud, M. L´ eger, B. Giraudeau, and Y. Foucher, “G-computation, propensity score-based methods, and targeted maximum likelihood estimator for causal inference with different covariates sets: a co...

    2020

  5. [13]

    A Practical Example Demonstrating the Utility of Single-world Intervention Graphs,

    A. Breskin, S. R. Cole, and M. G. Hudgens, “A Practical Example Demonstrating the Utility of Single-world Intervention Graphs,” Epidemiology, vol. 29, pp. e20–e21, 5 2018. edition: 2018/01/11

    2018

  6. [14]

    The Calculus of M-Estimation,

    L. A. Stefanski and D. D. Boos, “The Calculus of M-Estimation,” The American Statistician , vol. 56, pp. 29–38, feb 1

  7. [15]

    D. D. Boos and L. A. Stefanski, M-Estimation (Estimating Equations) , pp. 297–337. Springer Texts in Statistics, New York, NY: Springer, 2013. DOI: 10.1007/978-1-4614-4818-1 7

    2013 doi

  8. [16]

    M-estimation for common epidemiological measures: intro- duction and applied examples,

    R. K. Ross, P. N. Zivich, J. S. A. Stringer, and S. R. Cole, “M-estimation for common epidemiological measures: intro- duction and applied examples,” International Journal of Epidemiology , vol. 53, p. dyae030, apr 1 2024

    2024

  9. [17]

    Estimating Equations, Theory of,

    A. F. Desmond, “Estimating Equations, Theory of,” in Wiley StatsRef: Statistics Reference Online , John Wi- ley & Sons, Ltd, 2014. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781118445112.stat01932 DOI: 10.1002/9781118445112.stat01932

    2014 doi

  10. [18]

    Estimating functions and the generalized method of moments,

    J. Jesus and R. E. Chandler, “Estimating functions and the generalized method of moments,” Interface Focus, vol. 1, pp. 871–885, dec 6 2011. publisher: Royal Society

    2011

  11. [19]

    Carroll, David Ruppert, Leonard A

    Raymond J. Carroll, David Ruppert, Leonard A. Stefanski, and Ciprian M. Crainiceanu, Measurement Error in Nonlinear Models : A Modern Perspective, Second Edition , vol. 2nd ed. Raymond J. Carroll ... [et al.] of Monographs on Statistics and Applied Probability. Boca Raton, FL:...

    2006

  12. [20]

    Bootstrap Methods: Another Look at the Jackknife,

    B. Efron, “Bootstrap Methods: Another Look at the Jackknife,” The Annals of Statistics , vol. 7, pp. 1–26, 1 1979. publisher: Institute of Mathematical Statistics

    1979

  13. [21]

    Sampling distributions and the bootstrap,

    A. Kulesa, M. Krzywinski, P. Blainey, and N. Altman, “Sampling distributions and the bootstrap,” Nature Methods, vol. 12, pp. 477–478, jun 1 2015. Bandiera abtest: a Cg type: Nature Research Journals number: 6 Primary atype: News publisher: Nature Publishing Group Subject term...

    2015

  14. [22]

    On Variance of the Treatment Effect in the Treated When Estimated by Inverse Probability Weighting,

    S. A. Reifeis and M. G. Hudgens, “On Variance of the Treatment Effect in the Treated When Estimated by Inverse Probability Weighting,” American Journal of Epidemiology , vol. 191, pp. 1092–1097, may 20 2022

    2022

  15. [23]

    Delicatessen: M-Estimation in Python,

    P. N. Zivich, M. Klose, S. R. Cole, J. K. Edwards, and B. E. Shook-Sa, “Delicatessen: M-Estimation in Python,” arXiv:2203.11300, mar 21 2022. arXiv: 2203.11300

    2022 arXiv

  16. [24]

    The Calculus of M-Estimation in R with geex,

    B. C. Saul and M. G. Hudgens, “The Calculus of M-Estimation in R with geex,” Journal of Statistical Software , vol. 92, pp. 1–15, feb 18 2020

    2020

  17. [25]

    Estimating causal effects from epidemiological data,

    M. A. Hern´ an and J. M. Robins, “Estimating causal effects from epidemiological data,” Journal of Epidemiology and Community Health, vol. 60, pp. 578–586, 7 2006. PMID: 16790829 PMCID: PMC2652882

    2006

  18. [26]

    The consistency statement in causal inference: a definition or an assumption?,

    S. R. Cole and C. E. Frangakis, “The consistency statement in causal inference: a definition or an assumption?,” Epi- demiology, vol. 20, no. 1, pp. 3–5, 2009

    2009

  19. [27]

    Positivity: Identifiability and Estimability,

    P. N. Zivich, S. R. Cole, and D. Westreich, “Positivity: Identifiability and Estimability,” arXiv:2207.05010, jul 11 2022

    2022 arXiv

  20. [28]

    Using simulation studies to evaluate statistical methods,

    T. P. Morris, I. R. White, and M. J. Crowther, “Using simulation studies to evaluate statistical methods,” Statistics in Medicine, vol. 38, no. 11, pp. 2074–2102, 2019. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/sim.8086

    2019 doi

  21. [29]

    Array programming with NumPy,

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. del R ´ ıo, M. Wiebe, P. Peterson, P. G´ erard-Marchant, K. Sheppar...

    2020

  22. [30]

    Scipy 1.0: fundamental algorithms for scientific computing in Python,

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, and J. Bright, “Scipy 1.0: fundamental algorithms for scientific computing in Python,” Nature Meth- ods, vol. 17, no. 3, pp. 261–272, 2020

    2020

  23. [31]

    Data Structures for Statistical Computing in Python,

    W. McKinney, “Data Structures for Statistical Computing in Python,” Proceedings of the 9th Python in Science Confer- ence, pp. 56–61, 2010. event-title: Proceedings of the 9th Python in Science Conference

    2010

  24. [32]

    Empirical Sandwich Variance Estimator for Iterated Conditional Expectation g-Computation,

    P. N. Zivich, R. K. Ross, B. E. Shook-Sa, S. R. Cole, and J. K. Edwards, “Empirical Sandwich Variance Estimator for Iterated Conditional Expectation g-Computation,” Statistics in Medicine , vol. 43, no. 29, pp. 5562–5572, 2024. eprint: https://onlinelibrary.wiley.com/doi/pdf/1...

    2024 doi

  25. [33]

    Leveraging external validation data: the challenges of transporting measurement error parameters,

    R. K. Ross, S. R. Cole, J. K. Edwards, P. N. Zivich, D. Westreich, J. L. Daniels, J. T. Price, and J. S. A. Stringer, “Leveraging external validation data: the challenges of transporting measurement error parameters,” Epidemiology, vol. 35, no. 2, pp. 197–207, 2024

    2024

  26. [34]

    Targeted Learning of the Mean Outcome under an Optimal Dynamic Treatment Rule,

    M. J. van der Laan and A. R. Luedtke, “Targeted Learning of the Mean Outcome under an Optimal Dynamic Treatment Rule,” Journal of causal inference , vol. 3, pp. 61–95, 3 2015. PMID: 26236571 PMCID: PMC4517487

    2015

  27. [35]

    Synthesis estimators for transportability with positivity violations by a continuous covariate,

    P. N. Zivich, J. K. Edwards, B. E. Shook-Sa, E. T. Lofgren, J. Lessler, and S. R. Cole, “Synthesis estimators for transportability with positivity violations by a continuous covariate,” Journal of the Royal Statistical Society Series A: Statistics in Society , p. qnae084, sep 2 2024

    2024

  28. [36]

    Econometric methods for fractional response variables with an application to 401(k) plan participation rates,

    L. E. Papke and J. M. Wooldridge, “Econometric methods for fractional response variables with an application to 401(k) plan participation rates,” Journal of Applied Econometrics , vol. 11, no. 6, pp. 619–632, 1996. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/%28SIC...

    1996

  29. [37]

    Quasi-Likelihood and Optimal Estimation,

    V. P. Godambe and C. C. Heyde, “Quasi-Likelihood and Optimal Estimation,” International Statistical Review , vol. 55, no. 3, pp. 231–244, 1987. publisher: [Wiley, International Statistical Institute (ISI)]

    1987

  30. [38]

    Revisiting representativeness,

    H. Lu, P. Zivich, J. Rudolph, Z. Liew, B. Mukherjee, and F. Li, “Revisiting representativeness,” International Journal of Epidemiology, vol. 54, no. 4, p. dyaf109, 2024

    2024

  31. [39]

    M. R. Kosorok, ed., Z-Estimators, pp. 251–262. Springer Series in Statistics, New York, NY: Springer, 2008. DOI: 10.1007/978-0-387-74978-5 13

    2008 doi

  32. [40]

    Z-estimation and stratified samples: application to survival models,

    N. E. Breslow, J. Hu, and J. A. Wellner, “Z-estimation and stratified samples: application to survival models,” Lifetime Data Analysis, vol. 21, pp. 493–516, oct 1 2015

    2015

  33. [41]

    Estimating equations for survival analysis with pooled logistic regression,

    P. N. Zivich, S. R. Cole, B. E. Shook-Sa, J. B. DeMonte, and J. K. Edwards, “Estimating equations for survival analysis with pooled logistic regression,” arXiv preprint arXiv:2504.13291 , 2025. 9

    2025 arXiv

  34. [42]

    The Robust Inference for the Cox Proportional Hazards Model,

    D. Y. Lin and L. J. Wei, “The Robust Inference for the Cox Proportional Hazards Model,” Journal of the American Statistical Association , vol. 84, no. 408, pp. 1074–1078, 1989. publisher: [American Statistical Association, Taylor & Francis, Ltd.]

    1989

  35. [43]

    Penalized Regressions: The Bridge versus the Lasso,

    W. J. Fu, “Penalized Regressions: The Bridge versus the Lasso,” Journal of Computational and Graphical Statistics , vol. 7, no. 3, pp. 397–416, 1998

    1998

  36. [44]

    Penalized Estimating Equations,

    W. J. Fu, “Penalized Estimating Equations,” Biometrics, vol. 59, no. 1, pp. 126–132, 2003

    2003

  37. [45]

    A unified class of penalties with the capability of producing a differentiable alternative to l1 norm penalty,

    H. Haselimashhadi, “A unified class of penalties with the capability of producing a differentiable alternative to l1 norm penalty,” Communications in Statistics - Theory and Methods , vol. 48, pp. 5530–5545, Nov. 2019

    2019

  38. [46]

    What can be estimated? Identifiability, estimability, causal inference and ill-posed inverse problems,

    O. J. Maclaren and R. Nicholson, “What can be estimated? Identifiability, estimability, causal inference and ill-posed inverse problems,” arXiv:1904.02826 [cs, math, stat] , jul 20 2020. arXiv: 1904.02826

    1904 arXiv

  39. [47]

    Nonparametric identification is not enough, but randomized controlled trials are,

    P. M. Aronow, J. M. Robins, T. Saarinen, F. S¨ avje, and J. Sekhon, “Nonparametric identification is not enough, but randomized controlled trials are,” arXiv:2108.11342 [stat] , sep 26 2021. arXiv: 2108.11342

    2021 arXiv

  40. [48]

    Targeted maximum likelihood estimation for causal inference in observational studies,

    M. S. Schuler and S. Rose, “Targeted maximum likelihood estimation for causal inference in observational studies,” American Journal of Epidemiology , vol. 185, no. 1, pp. 65–73, 2017

    2017

  41. [49]

    Doubly robust estimation of causal effects,

    M. J. Funk, D. Westreich, C. Wiesen, T. St¨ urmer, M. A. Brookhart, and M. Davidian, “Doubly robust estimation of causal effects,” American Journal of Epidemiology , vol. 173, no. 7, pp. 761–767, 2011. Edition: 03/08

    2011

  42. [50]

    Demystifying Statistical Learning Based on Efficient Influence Functions,

    O. Hines, O. Dukes, K. Diaz-Ordaz, and S. Vansteelandt, “Demystifying Statistical Learning Based on Efficient Influence Functions,” The American Statistician , vol. 76, pp. 292–304, Jan. 2022

    2022

  43. [51]

    Pulling back the curtain: the road from statistical estimand to machine-learning based estimator for epidemiologists (no wizard required),

    A. Renson, L. Montoya, D. E. Goin, I. D ´ ıaz, and R. K. Ross, “Pulling back the curtain: the road from statistical estimand to machine-learning based estimator for epidemiologists (no wizard required),” arXiv preprint arXiv.2502.05363 , 2025

    2025 arXiv

  44. [52]

    Five Facts About Influence Functions,

    S. R. Cole, A. Breskin, B. E. Shook-Sa, P. N. Zivich, M. G. Hudgens, and J. K. Edwards, “Five Facts About Influence Functions,” Epidemiology, vol. 36, pp. 467–472, July 2025

    2025

  45. [53]

    Double robust variance estimation with parametric working models,

    B. E. Shook-Sa, P. N. Zivich, C. Lee, K. Xue, R. K. Ross, J. K. Edwards, J. S. A. Stringer, and S. R. Cole, “Double robust variance estimation with parametric working models,” Biometrics, vol. 81, p. ujaf054, June 2025

    2025

  46. [54]

    Double/debiased machine learning for treatment and structural parameters,

    V. Chernozhukov, D. Chetverikov, M. Demirer, E. Duflo, C. Hansen, W. Newey, and J. Robins, “Double/debiased machine learning for treatment and structural parameters,” The Econometrics Journal , vol. 21, pp. C1–C68, Feb. 2018

    2018

  47. [55]

    Machine Learning for Causal Inference: On the Use of Cross-fit Estimators,

    P. N. Zivich and A. Breskin, “Machine Learning for Causal Inference: On the Use of Cross-fit Estimators,” Epidemiology, vol. 32, pp. 393–401, May 2021

    2021

  48. [56]

    Machine Learning and Causal Inference,

    P. N. Zivich, A. Breskin, and E. H. Kennedy, “Machine Learning and Causal Inference,” in Wiley StatsRef: Statistics Reference Online, pp. 1–8, John Wiley & Sons, Ltd, 2022

    2022

  49. [57]

    The use of plasmodes as a supplement to simulations: a simple example evaluating individual admixture estimation methodologies,

    L. K. Vaughan, J. Divers, M. A. Padilla, D. T. Redden, H. K. Tiwari, D. Pomp, and D. B. Allison, “The use of plasmodes as a supplement to simulations: a simple example evaluating individual admixture estimation methodologies,” Computational statistics & data analysis , vol. 53...

    2009

  50. [2002]

    publisher: Taylor & Francis eprint: https://doi.org/10.1198/000313002753631330

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.