Pith. sign in

REVIEW 3 minor 13 references

A Law of Iterated Expectation Primer for Causal Inference

T0 review · 0 major / 3 minor · reviewed 2026-06-26 · grok-4.3

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

Pith's one-line read The law of iterated expectation rewrites as the g-formula for causal effects in two equivalent forms under consistency, positivity and exchangeability.

desk verdict Clear teaching note on the g-formula via iterated expectations with solid examples, but no new results and not worth peer review. read the letter →

arxiv 2606.20078 v1 pith:EVMENONL submitted 2026-06-18 stat.OT

Ashley I. Naimi , Razieh Nabi , Lindsay J. Collin , Paul N. Zivich , Stephen R. Cole This is my paper · ORCID
classification stat.OT
keywords g-formulalawofiteratedexpectationcausalinferencestandardizationobservationaldataconfoundingpositivity
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

The paper demonstrates that a standard statistical identity becomes a tool for identifying causal effects from observational data when three assumptions hold. It presents the g-formula in a non-iterative form as one weighted average of conditional means and in an iterative form as nested expectations, showing the two versions are equivalent without parametric restrictions. Numerical examples with one binary confounder, mixed discrete and continuous confounders, and time-varying exposures illustrate how the identity applies in practice. A reader cares because the rewrite turns an abstract expectation rule into concrete formulas that recover causal quantities from observed data.

What carries the argument

The g-formula obtained by applying the law of iterated expectation to potential outcomes, expressed in either the single-average NICE version or the nested ICE version.

What would settle it

In data generated under consistency, positivity, and conditional exchangeability, the NICE and ICE versions of the g-formula produce numerically different values for the same causal contrast.

Watch

Extended reading notes

Core claim

Under the assumptions of causal consistency, positivity, and conditional exchangeability, the law of iterated expectation can be rewritten as a causal standardization formula (the g-formula) in two nonparametrically equivalent forms: a non-iterative conditional expectation (NICE) form involving a single weighted average of conditional outcome means, and an iterative conditional expectation (ICE) form involving nested expectations.

Load-bearing premise

Causal consistency, positivity, and conditional exchangeability hold in the data-generating process so the statistical identity identifies a causal effect.

Editorial extensions

If this is right

  • The g-formula identifies causal effects nonparametrically from observational data.
  • Both the single weighted average and the nested expectations versions yield identical results.
  • The rewrite applies to settings with time-fixed or time-varying exposures and mixed confounder types.
  • Simple numerical examples confirm the identity holds before moving to more complex data.

Reading between the lines

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

  • The two forms offer computational flexibility when implementing the g-formula in software.
  • The primer structure may reduce the barrier for applied researchers to use standardization methods.
  • The equivalence could guide development of estimators that switch between forms for efficiency.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The manuscript is an expository primer on the law of iterated expectation and its use in causal inference. It states that, under causal consistency, positivity, and conditional exchangeability, this law yields the g-formula in two algebraically equivalent nonparametric forms (NICE: a single weighted average of conditional outcome means; ICE: nested expectations) and illustrates both with three numerical examples of increasing complexity (binary confounder; mixed discrete/continuous confounders; time-varying with two time points).

Significance. The central claim is a direct, standard consequence of the law of total expectation under the listed identifying assumptions; the numerical examples supply concrete intuition for readers new to the material. The primer therefore has pedagogical value for teaching the g-formula, though it introduces no new theorems, parameter-free derivations, or empirical results.

minor comments (3)
  1. [Abstract] The abstract refers to 'integration notation' used to express the law of iterated expectation; the main text should include an explicit, self-contained definition of this notation early in the primer so that readers with limited statistical background can follow without external references.
  2. Section describing the three numerical examples should state the exact numerical values used for the conditional outcome means and the covariate distributions so that readers can reproduce the NICE and ICE calculations by hand.
  3. The manuscript should add a short concluding paragraph that explicitly contrasts the NICE and ICE forms with the more common 'plug-in' estimator language used in applied papers, to help readers translate the primer into practice.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of the manuscript's pedagogical value and for the recommendation of minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; standard identity restated with examples

full rationale

The manuscript is an expository primer that restates the standard nonparametric identification result for the g-formula via the law of iterated expectation under consistency, positivity, and conditional exchangeability. The two forms (NICE and ICE) are shown to be algebraically equivalent by the law of total expectation, with no new theorems, parameter fits, or empirical claims. No load-bearing self-citations, self-definitional steps, or reductions of predictions to inputs appear in the derivation chain; the content is self-contained against textbook derivations of the g-formula.

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

The paper relies entirely on standard probability identities and domain assumptions from causal inference; no free parameters, new entities, or ad-hoc inventions are introduced.

assumptions (2)
  • standard math Law of iterated expectation
    Foundational identity invoked to derive the g-formula forms.
  • domain assumption Causal consistency, positivity, and conditional exchangeability
    Required to interpret the statistical identity as a causal effect identifier.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Law of Iterated Expectation Primer for Causal Inference." pith.science (2026). https://pith.science/paper/EVMENONL

@misc{pith2026260620078,
  author       = {Pith},
  title        = {Pith review of: A Law of Iterated Expectation Primer for Causal Inference},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EVMENONL}},
  note         = {Machine review of arXiv:2606.20078}
}
read the original abstract

The g-formula is a foundational tool for identifying causal effects in observational data. This tool is based on the law of iterated expectation, a key mathematical identity in statistics. However, the notation with which the law of iterated expectation and the g-formula is expressed can be opaque to those with little background in statistics. We provide a primer introducing the law of iterated expectation, the integration notation used to express it, and its role for causal effect identification via the g-formula. Under the assumptions of causal consistency, positivity, and conditional exchangeability, the law of iterated expectation can be rewritten as a causal standardization formula (the g-formula) in two nonparametrically equivalent forms: a non-iterative conditional expectation (NICE) form involving a single weighted average of conditional outcome means, and an iterative conditional expectation (ICE) form involving nested expectations. We illustrate both forms using three progressively complex numerical examples: a time-fixed example with a single binary confounder, a time-fixed example with discrete and continuous confounders, and a time-varying example with two timepoints. We provide clarity on what the law of iterated expectation is, how it is related to the g-formula, and how to gain intuition of its mathematical formulations in actual data examples that can be generalized to a range of settings.

Figures

Figures reproduced from arXiv: 2606.20078 by the authors.

Figure 1
Figure 1. A directed acyclic graph (DAG) illustrating the causal struc￾ture among treatment A, outcome Y, and baseline covariate W. An arrow from one node to another indicates a direct causal effect. The path A Ð W Ñ Y represents confounding of the effect of A on Y by W. These observational data include 4,901 women with breast cancer assigned to tamoxifen use (A=1) or not (A=0), with an outcome measure of breast cancer recurr… view at source ↗
Figure 2
Figure 2. A DAG illustrating the causal struc￾ture for a time-varying confounding structure. The treatment and time-varying confounding variables A and Z are measured at two time￾points t P t0, 1u. The outcome Y is measured at the end of follow-up. This structure implies that: (a) the effect of A1 on Y is confounded by Z1, Z0, and A0; and (b) the effect of A0 on Y is confounded by Z0. Furthermore, part of the effect of A0 on … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references

  1. [1]

    A New Approach to Causal Inference in Mortality Studies with a Sustained Exposure Period–Application to Control of the Healthy Worker Survivor Effect.Mathe- matical Modelling

    Robins JM. A New Approach to Causal Inference in Mortality Studies with a Sustained Exposure Period–Application to Control of the Healthy Worker Survivor Effect.Mathe- matical Modelling. 1986;7:1393–1512

    1986

  2. [2]

    Springer

    Wasserman L.All of Statistics: A Concise Course in Statistical Inference. Springer. 2004

    2004

  3. [3]

    Cambridge University Press

    van der Vaart AW.Asymptotic Statistics. Cambridge University Press. 2000

    2000

  4. [4]

    Marginal Structural Models as a Tool for Standardization.Epi- demiol

    Sato T and Matsuyama Y. Marginal Structural Models as a Tool for Standardization.Epi- demiol. 2003;14:680–86

    2003

  5. [5]

    CRC Press

    Hernan MA and Robins JM.Causal Inference: What If. CRC Press. 2025

    2025

  6. [6]

    Defining and Identifying Average Treatment Effects.Amer- ican journal of epidemiology

    Naimi AI and Whitcomb BW. Defining and Identifying Average Treatment Effects.Amer- ican journal of epidemiology. 2023;192:685–687

    2023

  7. [7]

    Concerning the consistency assumption in causal inference.Epidemiol

    VanderWeele TJ. Concerning the consistency assumption in causal inference.Epidemiol. 2009;20:880–883

    2009

  8. [8]

    Parametric G-Formula Implementations for Causal Survival Analyses.Biometrics

    Wen L, Young JG, Robins JM, and Hernán MA. Parametric G-Formula Implementations for Causal Survival Analyses.Biometrics. 2021;77:740–753

    2021

Show all 13 references
  1. [9]

    An introduction to g methods.International journal of epidemiology

    Naimi AI, Cole SR, and Kennedy EH. An introduction to g methods.International journal of epidemiology. 2017;46:756–762

    2017

  2. [10]

    Using Longitudinal Tar- geted Maximum Likelihood Estimation in Complex Settings with Dynamic Interventions

    Schomaker M, Luque-Fernandez M, Leroy V , and Davies M. Using Longitudinal Tar- geted Maximum Likelihood Estimation in Complex Settings with Dynamic Interventions. Statistics in medicine. 2019;38:4888–4911

    2019

  3. [11]

    Analysis of occupational asbestos expo- sure and lung cancer mortality using the g formula.American journal of epidemiology

    Cole SR, Richardson DB, Chu H, and Naimi AI. Analysis of occupational asbestos expo- sure and lung cancer mortality using the g formula.American journal of epidemiology. 2013; 177:989–996. 20

    2013

  4. [12]

    2021; 174:595–601

    Naimi AI, Perkins NJ, Sjaarda LAet al.The Effect of Preconception-Initiated Low-Dose Aspirin on Human Chorionic Gonadotropin-Detected Pregnancy, Pregnancy Loss, and Live Birth : Per Protocol Analysis of a Randomized Trial.Annals of internal medicine. 2021; 174:595–601

    2021

  5. [13]

    nu”). Informally, this reference measure is a rule for assigning “sizes

    Díaz I. Machine learning in the estimation of causal effects: Targeted minimum loss-based estimation and double/debiased machine learning.Biostatistics. 2020;21:353–358. 21 Appendix: Integration, Probability, Statistics, and Measure Theory Readers encountering notation such as...

    2020

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.