Pith. sign in

REVIEW 2 major objections 6 minor 61 references

This tutorial establishes a systematic comparison of four mediation-analysis approaches for exposure mixtures, showing that the environmental risk score yields the lowest relative bias for global indirect effects while each method targets a

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 17:23 UTC pith:I3L4DNJY

load-bearing objection Useful tutorial with a simulation section that overstates what it shows; fix the benchmark and the linear-only DGP and it's publishable. the 2 major comments →

arxiv 2509.10916 v1 pith:I3L4DNJY submitted 2025-09-13 stat.ME stat.AP

A Tutorial on Conducting Mediation Analysis with Exposure Mixtures

Yiran Wang , Yi-Ting Lin , Sean McGrath , John D. Meeker , Sung Kyun Park , Joshua L. Warren , Bhramar Mukherjee This is my paper
classification stat.ME stat.AP
keywords causal mediation analysisexposure mixturesenvironmental healthBayesian kernel machine regressionenvironmental risk scoreprincipal component analysissimulation studydirect and indirect effects
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This tutorial paper aims to clarify what four popular approaches to mediation analysis with exposure mixtures actually estimate: single-exposure analysis (with and without co-exposure adjustment), principal-component-based mediation, environmental-risk-score-based mediation, and Bayesian kernel machine regression causal mediation analysis. Through simulations with linear, additive data, it argues that unadjusted single-exposure analysis is severely biased for the global indirect effect, that adjusted single-exposure analysis trades power for specificity, that principal-component analysis introduces substantial bias when components do not align with the mediational signal, and that the environmental risk score gives the lowest relative bias among summary methods. In a PROTECT birth-cohort example with 11 phthalates, LTE4 as mediator, and head circumference as outcome, all methods produce global indirect effects indistinguishable from zero, while MCNP and MEP consistently appear as exposure-specific indirect-effect signals. The paper's practical payoff is a decision framework tying method choice to the target estimand and the tolerable trade-off between bias, power, and interpretability.

Core claim

The paper's central claim is a comparative mapping of estimands, assumptions, and operating characteristics across four mediation-analysis strategies for exposure mixtures. It shows that under linear data-generating processes, the unadjusted single-exposure approach inflates the global natural indirect effect by over 345%, adjusted single-exposure analysis holds relative bias under 25% but detects true active exposures only weakly (TPR as low as 0.03) while controlling false positives below 0.02, environmental-risk-score mediation achieves the lowest relative bias (as low as 11% under strong mediation), and BKMR-CMA, though flexible, does not outperform summary methods in this linear setting

What carries the argument

The method comparison rests on three modeling devices: (1) product-method regression, where the indirect effect is the product of exposure-on-mediator and mediator-on-outcome coefficients, fitted as linear models without exposure–mediator interaction; (2) exposure summary scores—unsupervised principal components or a supervised environmental risk score built by penalized regression on a training split—that replace the mixture with a scalar or low-dimensional surrogate; and (3) Bayesian kernel machine regression, which represents the exposure–mediator and exposure–outcome relationships nonparametrically through a Gaussian kernel and samples counterfactual outcomes via MCMC to obtain posterior

Load-bearing premise

The performance comparison assumes the true exposure–mediator and mediator–outcome relationships are linear and additive with no exposure–mediator interactions, which is exactly the setting where the nonparametric flexibility of BKMR-CMA is not allowed to show its value.

What would settle it

Generate simulation data that include a nonlinear or interactive exposure–mediator or exposure–outcome relationship (for example, a quadratic exposure effect or an exposure–exposure interaction) and compare the relative bias of the global NIE estimated by BKMR-CMA versus ERS-MA; if BKMR-CMA's bias does not fall below ERS-MA's in that setting, the paper's implied conditional conclusion would need revision.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If a researcher's goal is a valid global indirect effect of a mixture, unadjusted single-exposure analysis should be avoided; it can inflate the global NIE by several-fold.
  • Adjusted single-exposure analysis targets exposure-specific effects, not the mixture effect, and requires multiple-testing control; its power is low when co-exposures are highly correlated.
  • PC-MA's bias is sensitive to the number of retained components, so the variance-explained criterion is not a reliable proxy for mediational relevance.
  • ERS-MA can provide a low-bias global summary but requires a training/analysis split that reduces sample size and yields effects for the score, not for individual exposures.
  • BKMR-CMA offers the only fully nonparametric route among the four and supports variable selection, but in the linear simulation it does not show superior bias or sensitivity.

Where Pith is reading between the lines

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

  • Because the simulation generates data from linear additive models, the comparison is silent on BKMR-CMA's core advantage; under nonlinear or interactive true effects, its relative performance could reverse, which would be a natural follow-up test.
  • The near-zero global indirect effects in PROTECT may reflect genuine absence of mediation through LTE4, but with n=175 the intervals are wide; a larger cohort or a different mediator could reveal nonzero total mediation that summary methods would capture.
  • The ERS-MA result is likely tied to the elastic net's ability to recover the sparse signal in the simulations; in mixtures with dense, weak effects, the score's advantage may shrink or disappear.
  • The tutorial's decision framework could be extended to binary outcomes and multiple mediators, where product-method equivalence breaks down and the methods' divergences would likely widen.

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

2 major / 6 minor

Summary. This tutorial reviews four approaches to mediation analysis with exposure mixtures: single-exposure mediation analysis (SE-MA) with and without adjustment for co-exposures, principal-component-based mediation analysis (PC-MA), environmental-risk-score-based mediation analysis (ERS-MA), and Bayesian kernel machine regression causal mediation analysis (BKMR-CMA). The paper formalizes the causal estimands and identification assumptions underlying each method, provides R implementation code and DAGs, and compares the methods through a simulation study with a linear, additive data-generating process under varying sample sizes and mediator strengths. The methods are illustrated on 11 phthalate exposures, LTE4 as a mediator, and head circumference Z-score in the PROTECT cohort. The paper explicitly states that it introduces no new methodology and is intended as a practice-oriented synthesis.

Significance. If the simulation comparison is properly anchored, this would be a useful contribution to the applied environmental-mixture literature. The tutorial's strengths include the explicit DAGs, assumption checklists, the decision framework in Figure 5, the summary table (Table 1), and reproducible R code for all four methods. The PROTECT application is instructive in showing method-dependent signals and near-zero global indirect effects. The paper also gives credit to the difficulty of defining causal effects for exposure summaries. However, the simulation benchmark has a load-bearing flaw in how 'true' NIEs are defined for PC-MA and ERS-MA, and the simulation design does not exercise the nonlinear/interactive settings where BKMR-CMA is most valuable.

major comments (2)
  1. [Section 3.2, Performance Metrics; Section 3.3.1, Relative Bias] The 'true' NIE for PC-MA and ERS-MA is defined by fitting the same model to a large (n=100,000) dataset generated from the same DGP. This makes the relative-bias metric a measure of finite-sample convergence to each method's own probability limit, not a measure of bias relative to the causal global NIE of the exposure mixture. Since the denominator is method-specific, cross-method comparisons such as 'ERS-MA … average relative bias as low as 18%' versus PC-MA's '103% to 121%' are not on a common scale; a badly misspecified method could appear to have low relative bias. Either compute an analytic causal global NIE for PC-MA and ERS-MA under the linear DGP (feasible for fixed PC loadings and fixed ERS coefficients), or relabel the metric as 'method-specific asymptotic consistency' and remove causal-bias language. This is load-bearing because the paper's central simulation claim is that ERS
  2. [Section 3.1, Simulation Settings and Data Generation; Section 3.3.2; Section 5] The DGP is linear and additive, with no exposure-mediator interactions and no nonlinear exposure-response functions; the paper states 'we assume no exposure-mediator interactions.' Because BKMR-CMA's main added value is its ability to capture nonlinear and interactive effects, this simulation cannot demonstrate that benefit. The conclusions in Sections 3.3.2 and 5 about BKMR-CMA's 'balanced trade-offs' and 'flexible' behavior are therefore conditional on a linear, additive world, and the comparison may understate BKMR-CMA's value in the settings for which it was designed. Add at least one scenario with nonlinearities or interactions, or clearly state that the simulation scope is limited to linear additive settings and temper the corresponding conclusions.
minor comments (6)
  1. [Section 3.3.1] The phrase 'highly strongly influenced' should be corrected to 'strongly influenced.'
  2. [Figure B3 caption] The caption says relative bias is 'the absolute value of the average deviation' but the text in Section 3.2 defines it as (NIE_hat - NIE_true)/NIE_true x 100%, which is signed. Please reconcile the definition.
  3. [Section 4 and Section 5] In Section 4, exposures with NIE p-values <= 0.2 are called 'potentially active,' but Section 5 refers to 'significant indirect effects' for MCNP and MEP. Given the threshold and the wide confidence intervals, 'suggestive' would be more accurate than 'significant.'
  4. [Tables A1/A2] The tables label a column 'P-value (FDR)' but the text says exposures are flagged using raw p-values <= 0.2; the FDR-adjusted values in the tables are all much larger. Please clarify which quantity drives the highlighting.
  5. [Section 2.5.1] The hierarchical variable selection code hard-codes 'groups = c(groups, 4)' for the mediator's own group. A brief comment explaining that the number '4' must match the number of exposure clusters plus one would prevent user error.
  6. [Section 2.2 and Section 3.2] For SE-MA, the paper should explicitly state how the 'global NIE' is formed from exposure-specific estimates (e.g., summing exposure-specific NIEs) so the simulation metric matches the conceptual discussion in Section 2.2, which emphasizes that SE-MA does not directly estimate a global mixture effect.

Circularity Check

1 steps flagged

The simulation defines NIE_true for PC-MA and ERS-MA as the large-sample estimate from the same method, so the headline relative-bias comparisons measure self-consistency rather than bias against the causal mixture effect.

specific steps
  1. self definitional [Section 3.2 (Performance Metrics), relied on in Section 3.3.1 and Figure B3]
    "Relative Bias = (NIE_hat − NIE_true)/NIE_true × 100% ... For PC-MA and ERS-MA, which involve dimension reduction or penalized regression steps, we estimate NIE_true empirically by fitting the respective models to a large reference dataset with 100,000 observations generated under the same data-generating mechanism. This large-sample estimate serves as a proxy for the population-level truth."

    For PC-MA and ERS-MA, NIE_true is defined as the probability limit of the very estimator whose bias is being assessed. The formula therefore reduces to (θ̂_n − θ̂_∞)/θ̂_∞, where θ̂_∞ is the same method applied to a 100,000-subject dataset. This measures finite-sample convergence of a method to its own asymptotic target, not closeness to the causal global NIE of the mixture. A method whose large-sample estimand is far from the causal effect—e.g., a PC that misses mediator-relevant signal or an ERS built from outcome regression—can appear nearly unbiased. Consequently, the comparative claims in Section 3.3.1, including 'ERS-MA outperforms both unadjusted SE-MA and PC-MA across all scenarios' and 'average relative bias is as low as 18%', are self-consistency results. They do not establish whi

full rationale

The paper is a tutorial and does not introduce new methodology; its mediation formulas are standard external results and the implementation relies on published R packages, so there is no load-bearing self-citation chain. The main circularity is confined to the simulation benchmark: for PC-MA and ERS-MA, the paper defines NIE_true as the large-sample fit of the same method, making the relative-bias metric a self-consistency check rather than an accuracy comparison against the causal estimand. This undermines the headline performance comparison, which is a central claimed contribution. The linear-additive, no-interaction simulation design is a limitation rather than a circularity, and the ERS-related self-citations are appropriate references to prior method development rather than circular support. Net: partial circularity in the evaluation of the central claim, but the tutorial's estimand discussion, code, and real-data illustration retain independent content.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The tutorial relies on standard causal mediation assumptions, and its simulation conclusions depend on the linear data generating process and hand-chosen effect sizes. No new entities are introduced.

free parameters (5)
  • Exposure-mediator effect sizes (alpha_x) = 0.3, 0.6, 0.9
    Chosen by hand in simulation (Section 3.1); determines indirect effect strength and TPR/FPR results.
  • Mediator-outcome effect (beta_m) = 0.4
    Chosen by hand in simulation; scales all indirect effects.
  • Within-group correlations = 0.1, 0.4, 0.8
    Chosen by hand for the three exposure groups; affects multicollinearity and relative method performance.
  • PIP thresholds = 0.1, 0.3, 0.5
    Used to declare exposures active in BKMR-CMA; directly affects TPR/FPR.
  • Minimum exposures in ERS = 3
    Iteratively adjust lambda1 to ensure at least 3 exposures retained; a modeling choice affecting ERS construction.
axioms (4)
  • domain assumption Consistency, positivity, SUTVA, sequential ignorability, and cross-world independence
    Standard causal identification assumptions stated in Section 2.1.3, required for natural direct and indirect effect estimation.
  • domain assumption No exposure-mediator interaction in linear models
    Assumed in SE-MA, PC-MA, ERS-MA (Section 2.1.4 and code); relaxable but not in the main analysis.
  • domain assumption Sufficiency of the retained PCs / ERS
    PC-MA requires Y,M independent of X given S,C (Section 2.3); ERS-MA requires the score to capture all relevant exposure information (Section 2.4).
  • ad hoc to paper Linearity and additivity of the simulated data generating process
    Section 3.1 uses linear models with no interactions, which is a simulation choice that affects the comparison.

pith-pipeline@v1.3.0-alltime-deepseek · 32231 in / 10693 out tokens · 103121 ms · 2026-08-04T17:23:19.419106+00:00 · methodology

0 comments
read the original abstract

Causal mediation analysis is a powerful tool in environmental health research, allowing researchers to uncover the pathways through which exposures influence health outcomes. While traditional mediation methods have been widely applied to individual exposures, real-world scenarios often involve complex mixtures. Such mixtures introduce unique methodological challenges, including multicollinearity, sparsity of active exposures, and potential nonlinear and interactive effects. This paper provides an overview of several commonly used approaches for mediation analysis under exposure mixture settings with clear strategies and code for implementation. The methods include: single exposure mediation analysis (SE-MA), principal component-based mediation analysis, environmental risk score-based mediation analysis, and Bayesian kernel machine regression causal mediation analysis. While SE-MA serves as a baseline that analyzes each exposure individually, the other methods are designed to address the correlation and complexity inherent in exposure mixtures. For each method, we aim to clarify the target estimand and the assumptions that each method is making to render a causal interpretation of the estimates obtained. We conduct a simulation study to systematically evaluate the operating characteristics of these four methods to estimate global indirect effects and to identify individual exposures contributing to the global mediation under varying sample sizes, effect sizes, and exposure-mediator-outcome structures. We also illustrate their real-world applicability by examining data from the PROTECT birth cohort, specifically analyzing the relationship between prenatal exposure to phthalate mixtures and neonatal head circumference Z-score, with leukotriene E4 as a mediator. This example offers practical guidance for conducting mediation analysis in complex environmental contexts.

Figures

Figures reproduced from arXiv: 2509.10916 by Bhramar Mukherjee, John D. Meeker, Joshua L. Warren, Sean McGrath, Sung Kyun Park, Yiran Wang, Yi-Ting Lin.

Figure 1
Figure 1. Figure 1: Directed acyclic graphs (DAGs) illustrating the mediation framework for exposure mix￾tures. The upper panel represents the general framework, while the lower panel applies this frame￾work to the PROTECT dataset. The (average) natural indirect effect (NIE) is defined as E[Y (x, M(x)) − Y (x, M(x ∗ ))], which captures the change in outcome when the mediator changes from the level under x ∗ to that under x, w… view at source ↗
Figure 2
Figure 2. Figure 2: Directed acyclic graphs comparing single exposure mediation analysis (SE-MA) with and without adjustment for co-exposures. Based on these two models, NDE, NIE, and TE of exposure Xj can be estimated as NDE( [ xj ,x∗ j |C) = (xj − x ∗ j )βˆ x, (1) NIE( d xj ,x∗ j |C) = (xj − x ∗ j )βˆmαˆx, (2) TE( c xj ,x∗ j |C) = (xj − x ∗ j )(βˆ x + βˆmαˆx), (3) where βˆ x, βˆm and ˆαx are the estimated coefficients from … view at source ↗
Figure 3
Figure 3. Figure 3: Directed acyclic graph illustrating the use of environmental risk score (ERS) in mediation analysis of exposure mixtures. In the training set (Y, X, C)Train, a supervised prediction model is fit to approximate E[Y | X,C], yielding an estimated function ˆfC . This function maps the exposures in the analysis set XA to a scalar ERS, computed as ˆfC (XA). When the model includes only main effects, this reduces… view at source ↗
Figure 4
Figure 4. Figure 4: Spearman correlation plot of the 11 phthalate exposures in the PROTECT dataset. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Practice-oriented decision framework for choosing mediation methods with correlated exposure mixtures. Estimands are defined with respect to the exposure representation adopted and should be interpreted accordingly. This paper compares four strategies for conducting mediation analysis with correlated exposure mixtures: single exposure mediation analysis (SE-MA), principal component-based mediation anal￾ysi… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

61 extracted references · 27 canonical work pages

  1. [1]

    VanderWeele

    Tyler J. VanderWeele. Mediation Analysis: A Practitioner’s Guide.Annual Review of Public Health, 37(1):17–32, 2016. doi: 10.1146/annurev-publhealth-032315-021402

    doi:10.1146/annurev-publhealth-032315-021402 2016

  2. [2]

    Peng, Christopher D

    Francesca Dominici, Roger D. Peng, Christopher D. Barr, and Michelle L. Bell. Protecting Human Health From Air Pollution: Shifting From a Single-pollutant to a Multipollutant Approach.Epidemiology, 21(2):187–194, 2010. doi: 10.1097/EDE.0b013e3181cc86e8

    work page doi:10.1097/ede.0b013e3181cc86e8 2010

  3. [3]

    Toccalino, Julia E

    Patricia L. Toccalino, Julia E. Norman, and Jonathon C. Scott. Chemical mixtures in untreated water from public-supply wells in the U.S. — Occurrence, composition, and potential toxicity. Science of The Total Environment, 431:262–270, 2012. doi: 10.1016/j.scitotenv.2012.05.044

    work page doi:10.1016/j.scitotenv.2012.05.044 2012

  4. [4]

    Associations of cumulative exposure to heavy metal mixtures with obesity and its comorbidities among U.S

    Xin Wang, Bhramar Mukherjee, and Sung Kyun Park. Associations of cumulative exposure to heavy metal mixtures with obesity and its comorbidities among U.S. adults in NHANES 2003–2014.Environment International, 121:683–694, 2018. doi: 10.1016/j.envint.2018.09.035

    work page doi:10.1016/j.envint.2018.09.035 2003

  5. [5]

    Ettinger, Joel Schwartz, H´ ector Lamadrid- Figueroa, Mauricio Hern´ andez-Avila, Chitra Amarasiriwardena, Howard Hu, David C

    Birgit Claus Henn, Lourdes Schnaas, Adrienne S. Ettinger, Joel Schwartz, H´ ector Lamadrid- Figueroa, Mauricio Hern´ andez-Avila, Chitra Amarasiriwardena, Howard Hu, David C. Bellinger, Robert O. Wright, and Martha Mar ´ ıa T´ ellez-Rojo. Associations of Early Childhood Manganese and Lead Coexposure with Neurodevelopment.Environmental Health Perspectives,...

    work page doi:10.1289/ehp.1003300 2012

  6. [6]

    Coull, and Robert O

    Birgit Claus Henn, Brent A. Coull, and Robert O. Wright. Chemical mixtures and chil- dren’s health.Current Opinion in Pediatrics, 26(2):223–229, 2014. doi: 10.1097/MOP. 0000000000000067

    work page doi:10.1097/mop 2014

  7. [7]

    Environmental exposure mixtures: questions and methods to address them.Current epidemiology reports, 5(2):160–165, 2018

    Ghassan B Hamra and Jessie P Buckley. Environmental exposure mixtures: questions and methods to address them.Current epidemiology reports, 5(2):160–165, 2018. doi: 10.1007/ s40471-018-0145-0

    2018

  8. [8]

    Application of an analytical framework for multivariate mediation analysis of environmental data.Nature communications, 11(1):5624, 2020

    Max T Aung, Yanyi Song, Kelly K Ferguson, David E Cantonwine, Lixia Zeng, Thomas F McElrath, Subramaniam Pennathur, John D Meeker, and Bhramar Mukherjee. Application of an analytical framework for multivariate mediation analysis of environmental data.Nature communications, 11(1):5624, 2020. doi: 10.1038/s41467-020-19335-2

    work page doi:10.1038/s41467-020-19335-2 2020

  9. [9]

    V. Zitko. Principal component analysis in the evaluation of environmental data.Marine Pollution Bulletin, 28(12):718–722, 1994. doi: 10.1016/0025-326X(94)90329-8

    work page doi:10.1016/0025-326x(94)90329-8 1994

  10. [10]

    Reid and K.L

    M.K. Reid and K.L. Spencer. Use of principal components analysis (PCA) on estuarine sedi- ment datasets: The effect of data pre-treatment.Environmental Pollution, 157(8-9):2275–2281,

  11. [11]

    Jianzhong Ma and Christopher I. Amos. Principal Components Analysis of Population Ad- mixture.PLoS ONE, 7(7):e40115, 2012. doi: 10.1371/journal.pone.0040115

    work page doi:10.1371/journal.pone.0040115 2012

  12. [12]

    Williams

    Herv´ e Abdi and Lynne J. Williams. Principal component analysis.WIREs Computational Statistics, 2(4):433–459, 2010. doi: 10.1002/wics.101

    doi:10.1002/wics.101 2010

  13. [13]

    Meeker, Siob´ an D

    Sung Kyun Park, Yebin Tao, John D. Meeker, Siob´ an D. Harlow, and Bhramar Mukherjee. Environmental Risk Score as a New Tool to Examine Multi-Pollutants in Epidemiologic Re- search: An Example from the NHANES Study Using Serum Lipid Levels.PLoS ONE, 9(6): e98632, 2014. doi: 10.1371/journal.pone.0098632. 33

    work page doi:10.1371/journal.pone.0098632 2014

  14. [14]

    Bobb, Linda Valeri, Birgit Claus Henn, David C

    Jennifer F. Bobb, Linda Valeri, Birgit Claus Henn, David C. Christiani, Robert O. Wright, Maitreyi Mazumdar, John J. Godleski, and Brent A. Coull. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures.Biostatistics, 16(3):493–508, 2015. doi: 10.1093/biostatistics/kxu058

    work page doi:10.1093/biostatistics/kxu058 2015

  15. [15]

    Devick, Jennifer F

    Katrina L. Devick, Jennifer F. Bobb, Maitreyi Mazumdar, Birgit Claus Henn, David C. Bellinger, David C. Christiani, Robert O. Wright, Paige L. Williams, Brent A. Coull, and Linda Valeri. Bayesian kernel machine regression-causal mediation analysis.Statistics in Medicine, 41(5):860–876, 2022. doi: 10.1002/sim.9255

    work page doi:10.1002/sim.9255 2022

  16. [16]

    Max T Aung, Youfei Yu, Kelly K Ferguson, David E Cantonwine, Lixia Zeng, Thomas F McElrath, Subramaniam Pennathur, Bhramar Mukherjee, and John D Meeker. Cross-sectional estimation of endogenous biomarker associations with prenatal phenols, phthalates, metals, and polycyclic aromatic hydrocarbons in single-pollutant and mixtures analysis approaches. Enviro...

    work page doi:10.1289/ehp7396 2021

  17. [17]

    Robins and Sander Greenland

    James M. Robins and Sander Greenland. Identifiability and Exchangeability for Direct and Indirect Effects:.Epidemiology, 3(2):143–155, 1992. doi: 10.1097/00001648-199203000-00013

    doi:10.1097/00001648-199203000-00013 1992

  18. [18]

    Direct and indirect effects

    Judea Pearl. Direct and indirect effects. InProceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, UAI’01, pages 411–420, San Francisco, CA, USA, 2001. The Association for Uncertainty in Artificial Intelligence, Morgan Kaufmann Publishers Inc. ISBN 1558608001. doi: 10.5555/2074022.2074073. URLhttps://dl.acm.org/doi/10.5555/ 2...

    arXiv 2001

  19. [19]

    Vanderweele and Stijn Vansteelandt

    Tyler J. Vanderweele and Stijn Vansteelandt. Conceptual issues concerning mediation, inter- ventions and composition.Statistics and Its Interface, 2(4):457–468, 2009. doi: 10.4310/SII. 2009.v2.n4.a7

    work page doi:10.4310/sii 2009

  20. [20]

    T. J. VanderWeele and S. Vansteelandt. Odds Ratios for Mediation Analysis for a Dichotomous Outcome.American Journal of Epidemiology, 172(12):1339–1348, 2010. doi: 10.1093/aje/ kwq332

    doi:10.1093/aje/ 2010

  21. [21]

    A general approach to causal mediation analysis

    Kosuke Imai, Luke Keele, and Dustin Tingley. A general approach to causal mediation analysis. Psychological Methods, 15(4):309–334, 2010. doi: 10.1037/a0020761

    doi:10.1037/a0020761 2010

  22. [22]

    VanderWeele

    Linda Valeri and Tyler J. VanderWeele. Mediation analysis allowing for exposure–mediator interactions and causal interpretation: Theoretical assumptions and implementation with SAS and SPSS macros.Psychological Methods, 18(2):137–150, 2013. doi: 10.1037/a0031034

    work page doi:10.1037/a0031034 2013

  23. [23]

    CRC Press, 1st edition, 2024

    Peng Ding.A first course in causal inference. CRC Press, 1st edition, 2024. doi: 10.1201/ 9781003484080

    2024

  24. [24]

    Applied mediation analyses: a review and tutorial.Epidemiology and health, 39:e2017035, 2017

    Theis Lange, Kim Wadt Hansen, Rikke Sørensen, and Søren Galatius. Applied mediation analyses: a review and tutorial.Epidemiology and health, 39:e2017035, 2017. doi: 10.4178/ epih.e2017035

    2017

  25. [25]

    Formal mode of statistical inference for causal effects.Journal of statistical planning and inference, 25(3):279–292, 1990

    Donald B Rubin. Formal mode of statistical inference for causal effects.Journal of statistical planning and inference, 25(3):279–292, 1990. ISSN 0378-3758. doi: 10.1016/0378-3758(90) 90077-8

    work page doi:10.1016/0378-3758(90 1990

  26. [26]

    VanderWeele.Explanation in Causal Inference: Methods for Mediation and Interac- tion

    Tyler J. VanderWeele.Explanation in Causal Inference: Methods for Mediation and Interac- tion. Oxford University Press, New York, NY, 2015. ISBN 978-0-19-932587-0. 34

    2015

  27. [27]

    Identification, Inference and Sensitivity Analysis for Causal Mediation Effects.Statistical Science, 25(1):57–71, 2010

    Kosuke Imai, Luke Keele, and Teppei Yamamoto. Identification, Inference and Sensitivity Analysis for Causal Mediation Effects.Statistical Science, 25(1):57–71, 2010. doi: 10.1214/ 10-STS321

    2010

  28. [28]

    Baron and David A

    Reuben M. Baron and David A. Kenny. The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations.Journal of Personality and Social Psychology, 51(6):1173–1182, 1986. doi: 10.1037/0022-3514.51.6.1173

    doi:10.1037/0022-3514.51.6.1173 1986

  29. [29]

    Multiple comparisons among means.Journal of the American statistical association, 56(293):52–64, 1961

    Olive Jean Dunn. Multiple comparisons among means.Journal of the American statistical association, 56(293):52–64, 1961. doi: 10.1080/01621459.1961.10482090

    arXiv 1961

  30. [30]

    A simple sequentially rejective multiple test procedure.Scandinavian journal of statistics, pages 65–70, 1979

    Sture Holm. A simple sequentially rejective multiple test procedure.Scandinavian journal of statistics, pages 65–70, 1979

    1979

  31. [31]

    Yoav Benjamini and Yosef Hochberg. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.Journal of the Royal Statistical Society Series B: Statistical Methodology, 57(1):289–300, 1995. doi: 10.1111/j.2517-6161.1995.tb02031.x

    arXiv 1995

  32. [32]

    Coull, Tyler J

    Baoyi Shi, Christine Choirat, Brent A. Coull, Tyler J. VanderWeele, and Linda Valeri. CMA- verse: A Suite of Functions for Reproducible Causal Mediation Analyses.Epidemiology, 32 (5):e20–e22, 2021. doi: 10.1097/EDE.0000000000001378

    work page doi:10.1097/ede.0000000000001378 2021

  33. [33]

    Mediation Analysis with Multiple Mediators

    Tyler VanderWeele and Stijn Vansteelandt. Mediation Analysis with Multiple Mediators. Epidemiologic Methods, 2(1), 2014. doi: 10.1515/em-2012-0010

    work page doi:10.1515/em-2012-0010 2014

  34. [34]

    Inverse odds ratio-weighted estimation for causal mediation anal- ysis.Statistics in medicine, 32(26):4567–4580, 2013

    Eric J Tchetgen Tchetgen. Inverse odds ratio-weighted estimation for causal mediation anal- ysis.Statistics in medicine, 32(26):4567–4580, 2013. doi: 10.1002/sim.5864

    work page doi:10.1002/sim.5864 2013

  35. [35]

    Imputation strategies for the estima- tion of natural direct and indirect effects.Epidemiologic Methods, 1(1):131–158, 2012

    Stijn Vansteelandt, Maarten Bekaert, and Theis Lange. Imputation strategies for the estima- tion of natural direct and indirect effects.Epidemiologic Methods, 1(1):131–158, 2012. doi: 10.1515/2161-962X.1014

    work page doi:10.1515/2161-962x.1014 2012

  36. [36]

    Mediation analysis with time varying exposures and mediators.Journal of the Royal Statistical Society Series B: Statistical Method- ology, 79(3):917–938, 2017

    Tyler J VanderWeele and Eric J Tchetgen Tchetgen. Mediation analysis with time varying exposures and mediators.Journal of the Royal Statistical Society Series B: Statistical Method- ology, 79(3):917–938, 2017. doi: 10.1111/rssb.12194

    work page doi:10.1111/rssb.12194 2017

  37. [37]

    James Robins. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect.Mathematical modelling, 7(9-12):1393–1512, 1986. doi: 10.1016/0270-0255(86)90088-6

    doi:10.1016/0270-0255(86)90088-6 1986

  38. [38]

    Gibson, Yanelli Nunez, Ahlam Abuawad, Ami R

    Elizabeth A. Gibson, Yanelli Nunez, Ahlam Abuawad, Ami R. Zota, Stefano Renzetti, Katrina L. Devick, Chris Gennings, Jeff Goldsmith, Brent A. Coull, and Marianthi-Anna Kioumourtzoglou. An overview of methods to address distinct research questions on envi- ronmental mixtures: An application to persistent organic pollutants and leukocyte telomere length.Env...

    work page doi:10.1186/s12940-019-0515-1 2019

  39. [39]

    Jolliffe and Jorge Cadima

    Ian T. Jolliffe and Jorge Cadima. Principal component analysis: A review and recent de- velopments.Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2065):20150202, 2016. doi: 10.1098/rsta.2015.0202

    arXiv 2065

  40. [40]

    R Foundation for Statistical Computing, Vienna, Austria, 2025

    R Core Team.R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2025. URLhttps://www.R-project.org/. 35

    2025

  41. [41]

    PhD thesis, University of Michigan, Ann Arbor, MI, 2023

    Jonathan Boss.Shrinkage Methods for High-Dimensional Regression and Mediation Models. PhD thesis, University of Michigan, Ann Arbor, MI, 2023

    2023

  42. [42]

    factoextra: Extract and visualize the results of multivariate data analyses.https://CRAN.R-project.org/package=factoextra, 2020

    Alboukadel Kassambara and Fabian Mundt. factoextra: Extract and visualize the results of multivariate data analyses.https://CRAN.R-project.org/package=factoextra, 2020. R package version 1.0.7

    2020

  43. [43]

    The application of electronic computers to factor analysis.Educational and psychological measurement, 20(1):141–151, 1960

    Henry F Kaiser. The application of electronic computers to factor analysis.Educational and psychological measurement, 20(1):141–151, 1960. doi: 10.1177/001316446002000116

    doi:10.1177/001316446002000116 1960

  44. [44]

    Sung Kyun Park, Zhangchen Zhao, and Bhramar Mukherjee. Construction of environmental risk score beyond standard linear models using machine learning methods: Application to metal mixtures, oxidative stress and cardiovascular disease in NHANES.Environmental Health, 16(1):102, 2017. doi: 10.1186/s12940-017-0310-9

    work page doi:10.1186/s12940-017-0310-9 2017

  45. [45]

    Chipman, Edward I

    Hugh A. Chipman, Edward I. George, and Robert E. McCulloch. BART: Bayesian addi- tive regression trees.The Annals of Applied Statistics, 4(1):266–298, 2010. doi: 10.1214/ 09-AOAS285. URLhttps://doi.org/10.1214/09-AOAS285

    doi:10.1214/09-aoas285 2010

  46. [46]

    Super learner.Statistical applica- tions in genetics and molecular biology, 6(1), 2007

    Mark J Van der Laan, Eric C Polley, and Alan E Hubbard. Super learner.Statistical applica- tions in genetics and molecular biology, 6(1), 2007. doi: 10.2202/1544-6115.1309

    arXiv 2007

  47. [47]

    Regularization and Variable Selection Via the Elastic Net.Journal of the Royal Statistical Society Series B: Statistical Methodology, 67(2):301–320, 2005

    Hui Zou and Trevor Hastie. Regularization and Variable Selection Via the Elastic Net.Journal of the Royal Statistical Society Series B: Statistical Methodology, 67(2):301–320, 2005. doi: 10.1111/j.1467-9868.2005.00503.x

    arXiv 2005

  48. [48]

    Yi Yang, Yuwen Gu, and Hui Zou. gcdnet: The (adaptive) lasso and elastic net penalized least squares, logistic regression, hybrid huberized support vector machines, squared hinge loss support vector machines and expectile regression using a fast generalized coordinate descent algorithm.https://CRAN.R-project.org/package=gcdnet, 2022. R package version 1.0.6

    2022

  49. [49]

    Bobb, Birgit Claus Henn, Linda Valeri, and Brent A

    Jennifer F. Bobb, Birgit Claus Henn, Linda Valeri, and Brent A. Coull. Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression.Environmental Health, 17(1):67, 2018. doi: 10.1186/s12940-018-0413-y

    work page doi:10.1186/s12940-018-0413-y 2018

  50. [50]

    Cambridge university press, 2000

    Nello Cristianini and John Shawe-Taylor.An introduction to support vector machines and other kernel-based learning methods. Cambridge university press, 2000. ISBN 9780511801389. doi: 10.1017/CBO9780511801389

    work page doi:10.1017/cbo9780511801389 2000

  51. [51]

    Semiparametric regression of multidimensional genetic pathway data: least-squares kernel machines and linear mixed models.Biometrics, 63 (4):1079–1088, 2007

    Dawei Liu, Xihong Lin, and Debashis Ghosh. Semiparametric regression of multidimensional genetic pathway data: least-squares kernel machines and linear mixed models.Biometrics, 63 (4):1079–1088, 2007. doi: 10.1111/j.1541-0420.2007.00799.x

    arXiv 2007

  52. [53]

    causalbkmr: Causal inference with bayesian ker- nel machine regression.https://github.com/zc2326/causalbkmr, 2025

    Zilan Chai, Katrina Devick, and Linda Valeri. causalbkmr: Causal inference with bayesian ker- nel machine regression.https://github.com/zc2326/causalbkmr, 2025. R package version 0.1.0, commit 4a33c0bbe76c7b5c55604064413b9be3e379e901

    2025

  53. [54]

    Wolff, Stephanie M

    Mary S. Wolff, Stephanie M. Engel, Gertrud S. Berkowitz, Xiaoyun Ye, Manori J. Silva, Chenbo Zhu, James Wetmur, and Antonia M. Calafat. Prenatal Phenol and Phthalate Expo- sures and Birth Outcomes.Environmental Health Perspectives, 116(8):1092–1097, 2008. doi: 10.1289/ehp.11007. 36

    work page doi:10.1289/ehp.11007 2008

  54. [55]

    Effect of envi- ronmental phthalate exposure on pregnancy duration and birth outcomes.International Journal of Occupational Medicine and Environmental Health, 29(4):683–697, 2016

    Kinga Pola´ nska, Danuta Ligocka, Wojciech Sobala, and Wojciech Hanke. Effect of envi- ronmental phthalate exposure on pregnancy duration and birth outcomes.International Journal of Occupational Medicine and Environmental Health, 29(4):683–697, 2016. doi: 10.13075/ijomeh.1896.00691

    arXiv 2016

  55. [56]

    Bloom, Edward L

    Michael S. Bloom, Edward L. Valachovic, Thoin F. Begum, John R. Kucklick, John W. Brock, Abby G. Wenzel, Rebecca J. Wineland, Lori Cruze, Elizabeth R. Unal, and Roger B. Newman. Association between gestational phthalate exposure and newborn head circumference; impacts by race and sex.Environmental Research, 195:110763, 2021. doi: 10.1016/j.envres.2021.110763

    arXiv 2021

  56. [57]

    Prenatal exposure to phthalate esters and PAHs and birth outcomes.Environment International, 36(7):699–704, 2010

    Yayoi Suzuki, Mayu Niwa, Jun Yoshinaga, Yoshifumi Mizumoto, Shigeko Serizawa, and Hiroaki Shiraishi. Prenatal exposure to phthalate esters and PAHs and birth outcomes.Environment International, 36(7):699–704, 2010. doi: 10.1016/j.envint.2010.05.003

    work page doi:10.1016/j.envint.2010.05.003 2010

  57. [58]

    Regularization paths for generalized linear models via coordinate descent.Journal of Statistical Software, 33(1):1–22, 2010

    Jerome Friedman, Trevor Hastie, and Robert Tibshirani. Regularization paths for generalized linear models via coordinate descent.Journal of Statistical Software, 33(1):1–22, 2010. doi: 10.18637/jss.v033.i01

    doi:10.18637/jss.v033.i01 2010

  58. [59]

    Jennifer F. Bobb. bkmr: Bayesian kernel machine regression.https://CRAN.R-project. org/package=bkmr, 2022. R package version 0.2.2

    2022

  59. [60]

    Sensitivity analysis in observational research: introducing the e-value.Annals of internal medicine, 167(4):268–274, 2017

    Tyler J VanderWeele and Peng Ding. Sensitivity analysis in observational research: introducing the e-value.Annals of internal medicine, 167(4):268–274, 2017. doi: 10.7326/M16-2607

    doi:10.7326/m16-2607 2017

  60. [61]

    dendextend: an r package for visualizing, adjusting, and comparing trees of hierarchical clustering.Bioinformatics, 2015

    Tal Galili. dendextend: an r package for visualizing, adjusting, and comparing trees of hierarchical clustering.Bioinformatics, 2015. doi: 10.1093/bioinformatics/btv428. URL https://doi.org/10.1093/bioinformatics/btv428. 37 A Supplementary Tables T able A1:Estimates for TE, NDE and NIE for each phthalate exposure using the single exposure mediation analys...

    doi:10.1093/bioinformatics/btv428 2015

  61. [2009]

    doi: 10.1016/j.envpol.2009.03.033

    work page doi:10.1016/j.envpol.2009.03.033 2009