pith. sign in

arxiv: 2606.26638 · v1 · pith:DBPTK7RYnew · submitted 2026-06-25 · 📊 stat.ME

Multivariable Mendelian randomization with weak instruments: a comparison of Bayesian and frequentist methods

Andrew J. Grant , Ashish Patel , Stephen Burgess This is my paper

Pith reviewed 2026-06-26 03:34 UTC · model grok-4.3

classification 📊 stat.ME
keywords Mendelian randomizationmultivariableweak instrumentsBayesian methodsfrequentist methodscausal inferencesimulation study
0
0 comments X

The pith

Bayesian MVMR-Pony reduces bias and improves coverage compared to frequentist methods in multivariable Mendelian randomization with weak instruments.

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

The paper compares a Bayesian method called MVMR-Pony against frequentist approaches for addressing weak instruments in multivariable Mendelian randomization. Weak instruments occur when genetic variants associate strongly with one exposure but only weakly with it after accounting for the other exposures in the model. Simulation studies examine scenarios driven by correlated genetic effects, measurement error, and mediation, finding that MVMR-Pony yields lower bias, higher coverage, controlled type I error rates, and greater power. A sympathetic reader would care because reliable causal estimates become harder to obtain when analyzing multiple exposures at once, and the Bayesian approach offers one way to mitigate that problem.

Core claim

In simulation studies, the MVMR-Pony Bayesian method outperforms frequentist approaches with respect to bias, coverage, type I error rates, and power across settings where weak instrument bias arises due to correlated genetic effects, measurement error, and mediation.

What carries the argument

MVMR-Pony, a Bayesian framework for multivariable Mendelian randomization that mitigates weak instrument bias.

If this is right

  • In settings with correlated genetic effects, MVMR-Pony provides more accurate causal effect estimates than frequentist alternatives.
  • When measurement error is present in the exposures, the Bayesian method produces less biased results.
  • In mediation scenarios, MVMR-Pony maintains better control of type I error while retaining higher power.
  • The approach supports valid inference even when instruments are only weakly associated with an exposure conditional on the others.

Where Pith is reading between the lines

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

  • Re-analysis of published multivariable MR studies that used frequentist methods on datasets with multiple correlated exposures could test whether effect estimates change substantially.
  • Future work could examine whether the performance advantage persists when the number of exposures grows beyond the two- or three-exposure cases typical in current simulations.
  • Software implementations that allow routine use of the Bayesian method alongside standard frequentist tools would let analysts compare results directly on the same data.

Load-bearing premise

The simulation studies accurately represent the conditions under which weak instrument bias occurs in real multivariable Mendelian randomization analyses with multiple exposures.

What would settle it

A simulation or real-data analysis in which frequentist methods achieve lower bias or higher coverage than MVMR-Pony under the same weak-instrument conditions generated by correlated effects, measurement error, or mediation.

Figures

Figures reproduced from arXiv: 2606.26638 by Andrew J. Grant, Ashish Patel, Stephen Burgess.

Figure 1
Figure 1. Figure 1: Simulation results for the scenario with increasingly correlated genetic effects, ρ, showing the mean estimate, coverage, and rejection rate (that is, power) using the IVW, GMM, GRAPPLE, srivw, and MVMR-Pony methods. The dashed lines indicate the true causal effect (for mean estimate) or the nominal level (for coverage). Also shown are the F statistics and conditional F statistics for each value of ρ. Note… view at source ↗
Figure 2
Figure 2. Figure 2: Simulation results for the scenario with increasing measurement error on exposures, ν, showing the mean estimate, coverage, and rejection rate (that is, power) using the IVW, GMM, GRAPPLE, srivw, and MVMR-Pony methods. The dashed lines indicate the true causal effect (for mean estimate) or the nominal level (for coverage). Also shown are the F statistics and conditional F statistics for each value of ν [P… view at source ↗
Figure 3
Figure 3. Figure 3: Simulation results for the scenario with increasing level of mediation, α, showing the mean estimate, coverage, and rejection rate (that is, power) using the IVW, GMM, GRAPPLE, srivw, and MVMR-Pony methods. The dashed lines indicate the true causal effect (for mean estimate) or the nominal level (for coverage). Also shown are the F statistics and conditional F statistics for each value of α. Note that some… view at source ↗
Figure 4
Figure 4. Figure 4: Log odds ratio for Alzheimer’s disease point estimate and 95% confidence interval (or credible interval) per unit increase in eGFR and UACR, using (A) non-pleiotropy-robust methods, and (B) pleiotropy-robust methods [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
read the original abstract

Weak instruments are a well known limitation for valid causal inference in Mendelian randomization studies. In the single exposure setting, weak instrument bias can be mitigated by selecting genetic instruments which are strongly associated with the exposure according to p-value and/or F-statistic thresholds. However, in the multi-exposure setting, genetic instruments may be strongly associated with an exposure but weakly associated with it conditional on all other exposures in the analysis. It is therefore more difficult to guarantee conditionally strong instruments in multivariable Mendelian randomization. Weak instrument bias can be mitigated using modelling approaches, however there are fewer methods for doing this in the multivariable case compared with the single exposure case. In this paper, we consider a method for mitigating weak instrument bias in multivariable Mendelian randomization using a Bayesian framework: MVMR-Pony. We compare this method with existing frequentist methods. We show using simulation studies that the MVMR-Pony method outperforms the frequentist approaches with respect to bias, coverage, type I error rates, and power, across settings where weak instrument bias arises due to correlated genetic effects, measurement error, and mediation.

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

0 major / 3 minor

Summary. The manuscript introduces MVMR-Pony, a Bayesian method for multivariable Mendelian randomization (MVMR) intended to mitigate weak instrument bias arising from correlated genetic effects, measurement error, and mediation. It compares this approach to existing frequentist methods and claims, based on simulation studies, that MVMR-Pony outperforms them with respect to bias, coverage, type I error rates, and power across the simulated settings.

Significance. The topic is relevant given the prevalence of weak instruments in MVMR applications. A simulation-based comparison of Bayesian and frequentist approaches under multiple bias mechanisms is an appropriate evidentiary standard for a methodological paper of this type. If the simulations are representative, the results could inform method choice in practice; the paper appropriately confines its claims to the simulated conditions rather than asserting general superiority.

minor comments (3)
  1. The simulation studies section would benefit from a table explicitly listing all parameter values (instrument strengths, correlation coefficients, sample sizes, and number of replicates) to facilitate reproducibility and assessment of coverage of realistic scenarios.
  2. Consider adding a brief real-data illustration, even if secondary, to show how MVMR-Pony behaves on actual genetic data with weak instruments.
  3. Clarify in the methods whether the Bayesian priors in MVMR-Pony are chosen in a data-dependent way or are fully pre-specified, as this affects the interpretation of the performance metrics.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript, recognition of its relevance, and recommendation for minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents MVMR-Pony as a new Bayesian approach for MVMR with weak instruments and evaluates it exclusively via simulation studies that generate data under three explicit mechanisms (correlated genetic effects, measurement error, mediation). Performance metrics (bias, coverage, type I error, power) are computed against these independently generated datasets rather than being derived from or fitted to the method's own parameters. No equations, uniqueness theorems, or ansatzes are shown to reduce by construction to the paper's inputs or to prior self-citations. The evidentiary basis is external simulation benchmarks, which is the standard and non-circular approach for this class of methodological work.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper relies on standard Mendelian randomization assumptions (relevance, exchangeability, exclusion restriction) and on the design of its simulation studies; no free parameters or invented entities are described in the abstract.

axioms (1)
  • domain assumption Standard Mendelian randomization assumptions hold (no pleiotropy, no population stratification, etc.).
    Implicit in all MR studies and required for the simulation scenarios to be meaningful.

pith-pipeline@v0.9.1-grok · 5727 in / 1229 out tokens · 36576 ms · 2026-06-26T03:34:03.281105+00:00 · methodology

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

51 extracted references · 4 canonical work pages

  1. [1]

    Mendelian randomization: Using genes as instruments for making causal infer ences in epidemiology

    Lawlor, DA, Harbord, RM, Sterne, JAC, Timpson, N, and Davey S mith, G. Mendelian randomization: Using genes as instruments for making causal infer ences in epidemiology. Stat Med. 2008;27(8):1133–1163

    2008

  2. [2]

    An introduction to instrumental variables for epid emiologists

    Greenland, S. An introduction to instrumental variables for epid emiologists. Int J Epidemiol. 2000;29(4):722–729

    2000

  3. [3]

    Instruments for causal inference : an epidemiologists dream? Epidemiology

    Hernan, MA and Robins, JM. Instruments for causal inference : an epidemiologists dream? Epidemiology. 2006;17(4):360–372

    2006

  4. [4]

    Mendelian ran domization analysis with multiple genetic variants using summarized data

    Burgess, S, Butterworth, A, and Thompson, SG. Mendelian ran domization analysis with multiple genetic variants using summarized data. Genet Epidemiol . 2013;37(7):658–665

    2013

  5. [5]

    Multivariable Mendelian randomizat ion: the use of pleiotropic genetic variants to estimate causal effects

    Burgess, S and Thompson, SG. Multivariable Mendelian randomizat ion: the use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181(4):251–260

    2015

  6. [6]

    High-thro ughput multivariable Mendelian randomization analysis prioritizes apolipoprotein B as key lipid risk factor for coronary artery disease

    Zuber, V, Gill, D, Ala-Korpela, M, Langenberg, C, et al. High-thro ughput multivariable Mendelian randomization analysis prioritizes apolipoprotein B as key lipid risk factor for coronary artery disease. Int J Epidemiol . 2021;50(3):893–901

    2021

  7. [7]

    Mendelian randomisa- tion for mediation analysis: current methods and challenges for imple mentation

    Carter, AR, Sanderson, E, Hammerton, G, Richmond, RC, et al. Mendelian randomisa- tion for mediation analysis: current methods and challenges for imple mentation. Eur J Epidemiol. 2021;36:465–478

    2021

  8. [8]

    Estimating causal effects on a disease pr ogression trait using bivariate Mendlian randomisation

    Cai, S and Dudbridge, F. Estimating causal effects on a disease pr ogression trait using bivariate Mendlian randomisation. Genet Epidemiol . 2025;49:e22600

    2025

  9. [9]

    Instrumental variables regression w ith weak instruments

    Staiger, D and Stock, JH. Instrumental variables regression w ith weak instruments. Econo- metrica. 1997;65(3):557–586. 15

    1997

  10. [10]

    Testing and correctin g for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization

    Sanderson, E, Spiller, W, and Bowden, J. Testing and correctin g for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization. Stat Med . 2021;40(25): 5434–5452

    2021

  11. [11]

    Mendelianrandomization v0.9 .0: updates to an R package for performing Mendelian randomization analyses using sum marized data [version 1; peer review: 2 approved]

    Patel, A, Ye, T, Xue, H, Lin, Z, et al. Mendelianrandomization v0.9 .0: updates to an R package for performing Mendelian randomization analyses using sum marized data [version 1; peer review: 2 approved]. Wellcome Open Res . 2023;8(449)

    2023

  12. [12]

    Efficient Design for Mendelian Random ization Studies: Sub- sample and 2-Sample Instrumental Variable Estimators

    Pierce, BL and Burgess, S. Efficient Design for Mendelian Random ization Studies: Sub- sample and 2-Sample Instrumental Variable Estimators. Am J Epidemiol 07 . 2013;178(7): 1177–1184

    2013

  13. [13]

    Mendelia n randomization

    Sanderson, E, Glymour, MM, Holmes, MV, Kang, H, et al. Mendelia n randomization. Nat Rev Methods Primers . 2022;2(1):6

    2022

  14. [14]

    Bias in causal estimates from me ndelian randomization studies with weak instruments

    Burgess, S and Thompson, SG. Bias in causal estimates from me ndelian randomization studies with weak instruments. Stat Med. 2011;30(11):1312–1323

    2011

  15. [15]

    A n examination of multivari- able Mendelian randomization in the single-sample and two-sample summ ary data settings

    Sanderson, E, Davey Smith, G, Windmeijer, F, and Bowden, J. A n examination of multivari- able Mendelian randomization in the single-sample and two-sample summ ary data settings. Int J Epidemiol . 2019;48(3):713–727

    2019

  16. [16]

    Bias in multivariable Mendelian randomization studies due to measurement error on exposures

    Zhu, J, Burgess, S, and Grant, AJ. Bias in multivariable Mendelian randomization studies due to measurement error on exposures. arXiv:2203.08668. 2022

    work page arXiv 2022

  17. [17]

    Disentangling the effe cts of traits with shared clustered genetic predictors using multivariable Mendelian randomiza tion

    Batool, F, Patel, A, Gill, D, and Burgess, S. Disentangling the effe cts of traits with shared clustered genetic predictors using multivariable Mendelian randomiza tion. Genet Epidemiol. 2022;46:415–429

    2022

  18. [18]

    Using gene tic association data to guide drug discovery and development: Review of methods and applic ations

    Burgess, S, Mason, AM, Grant, AJ, Slob, EA W, et al. Using gene tic association data to guide drug discovery and development: Review of methods and applic ations. Am J Hum Genet. 2023;110(2):195–214

    2023

  19. [19]

    Use of allele scores as instrumen tal variables for Mendelian randomization

    Burgess, S and Thompson, SG. Use of allele scores as instrumen tal variables for Mendelian randomization. Int J Epidemiol . 2013;42(4):1134–1144

    2013

  20. [20]

    The many weak instruments problem and Mendelian randomization

    Davies, NM, von Hinke Kessler Scholder, S, Farbmacher, H, Bur gess, S, et al. The many weak instruments problem and Mendelian randomization. Stat Med. 2015;34(3):454–468

    2015

  21. [21]

    Causal inferenc e for heritable phenotypic risk factors using heterogeneous genetic instruments

    Wang, J, Zhao, Q, Bowden, J, Hemani, G, et al. Causal inferenc e for heritable phenotypic risk factors using heterogeneous genetic instruments. PLoS Genet . 2021;17(6):1–24

    2021

  22. [22]

    Weak instruments in multivar iable mendelian random- ization: methods and practice

    Patel, A, Lane, J, and Burgess, S. Weak instruments in multivar iable mendelian random- ization: methods and practice. arXiv:2408.09868v1. 2024. 16

    work page arXiv 2024

  23. [23]

    A Bayesian approach to Mendelian randomization with multiple pleiotropic variants

    Berzuini, C, Guo, H, Burgess, S, and Bernardinelli, L. A Bayesian approach to Mendelian randomization with multiple pleiotropic variants. Biostatistics. 2020;21(1):86–101

    2020

  24. [24]

    Inferring the directio n of a causal link and estimating its effect via a Bayesian Mendelian randomization approach

    Bucur, IG, Claassen, T, and Heskes, T. Inferring the directio n of a causal link and estimating its effect via a Bayesian Mendelian randomization approach. Stat Methods Med Res . 2020;29 (4):1081–1111

    2020

  25. [25]

    Mendelian randomization accounting for correlated and uncorrelated pleiotropic effects us ing genome-wide summary statistics

    Morrison, J, Knoblauch, N, Marcus, JH, Stephens, M, and He, X. Mendelian randomization accounting for correlated and uncorrelated pleiotropic effects us ing genome-wide summary statistics. Nat Genet . 2020;52(7):740–747

    2020

  26. [26]

    Mendelian randomizat ion accounting for complex correlated horizontal pleiotropy while elucidating shared genetic et iology

    Cheng, Q, Zhang, X, Chen, LS, and Liu, J. Mendelian randomizat ion accounting for complex correlated horizontal pleiotropy while elucidating shared genetic et iology. Nat Commun . 2022;13:6490

    2022

  27. [27]

    Improving bias and coverage in in strumental variable anal- ysis with weak instruments for continuous and binary outcomes

    Burgess, S and Thompson, SG. Improving bias and coverage in in strumental variable anal- ysis with weak instruments for continuous and binary outcomes. Stat Med . 2012;31(15): 1582–1600

    2012

  28. [28]

    A bayesian app roach to Mendelian ran- domisation with dependent instruments

    Shapland, CY, Thompson, JR, and Sheehan, NA. A bayesian app roach to Mendelian ran- domisation with dependent instruments. Stat Med. 2019;38(6):985–1001

    2019

  29. [29]

    Selecting likely c ausal risk factors from high-throughput experiments using multivariable Mendelian randomiz ation

    Zuber, V, Colijn, JM, Klaver, C, and Burgess, S. Selecting likely c ausal risk factors from high-throughput experiments using multivariable Mendelian randomiz ation. Nat Commun . 2020;11(1):29

    2020

  30. [30]

    GMM is inadmissible under weak ident ification

    Andrews, I and Mikusheva, A. GMM is inadmissible under weak ident ification. arXiv:2204.12462. 2023

    work page arXiv 2023

  31. [31]

    A Bayesian approach to Mendelian ra ndomization using summary statistics in the univariable and multivariable settings with correlated pleiotropy

    Grant, AJ and Burgess, S. A Bayesian approach to Mendelian ra ndomization using summary statistics in the univariable and multivariable settings with correlated pleiotropy. Am J Hum Genet. 2024;111(1):165–180

    2024

  32. [32]

    Multivariab le Mendelian randomiza- tion: the use of pleiotropic genetic variants to estimate causal effe cts

    Burgess, S, Dudbridge, F, and Thompson, SG. Re: “Multivariab le Mendelian randomiza- tion: the use of pleiotropic genetic variants to estimate causal effe cts”. Am J Epidemiol . 2015;181(4):290–291

    2015

  33. [33]

    A more robust approach to multivaria ble Mendelian random- ization

    Wu, Y, Kang, H, and Ye, T. A more robust approach to multivaria ble Mendelian random- ization. arXiv.2402.00307. 2025

    work page arXiv 2025

  34. [34]

    Noise-augmented directional clustering of genetic association data identifies distinct mechanisms underlying obesity

    Grant, AJ, Gill, D, Kirk, PDW, and Burgess, S. Noise-augmented directional clustering of genetic association data identifies distinct mechanisms underlying obesity. PLoS Genet . 2022;18:e10009975

    2022

  35. [35]

    An atla s of genetic correlations across human diseases and traits

    Bulik-Sullivan, B, Finucane, HK, Anttila, V, Gusev, A, et al. An atla s of genetic correlations across human diseases and traits. Nat Genet . 2015;47(11):1236–1241. 17

    2015

  36. [36]

    Methods for meta-analysis of multiple tr aits using GW AS summary statistics

    Ray, D and Boehnke, M. Methods for meta-analysis of multiple tr aits using GW AS summary statistics. Genet Epidemiol . 2018;42(2):134–145

    2018

  37. [37]

    Determ ining the relationship between blood pressure, kidney function, and chronic kidney disea se: Insights from genetic epidemiology

    Staplin, N, Herrington, WG, Murgia, F, Ibrahim, M, et al. Determ ining the relationship between blood pressure, kidney function, and chronic kidney disea se: Insights from genetic epidemiology. Hypertension. 2022;79(12):2671–2681

    2022

  38. [38]

    Mild-to-moderate kid ney dysfunction and cardiovascular disease: Observational and Mendelian randomizatio n analyses

    Gaziano, L, Sun, L, Arnold, M, Bell, S, et al. Mild-to-moderate kid ney dysfunction and cardiovascular disease: Observational and Mendelian randomizatio n analyses. Circulation. 2022;146(20):1507–1517

    2022

  39. [39]

    Disorders of lipid met abolism in chronic kidney disease

    Bulbul, M, Dagel, T, Afsar, B, Ulusu, N, et al. Disorders of lipid met abolism in chronic kidney disease. Blood Purif. 2018;46(2):144–152

    2018

  40. [40]

    Causa l relationship between kidney function and cancer risk: A Mendelian randomization study

    Dobrijevic, E, van Zwieten, A, Grant, AJ, Loy, CT, et al. Causa l relationship between kidney function and cancer risk: A Mendelian randomization study. Am J Kidney Dis . 2024;84(6):686–695.e1

    2024

  41. [41]

    A catalog of genetic loci associated with kidney function from analyses of a million individuals

    Wuttke, M, Li, Y, Li, M, Sieber, KB, et al. A catalog of genetic loci associated with kidney function from analyses of a million individuals. Nat Genet . 2019;51(6):957–972

    2019

  42. [42]

    Genome-wide asso ciation meta-analyses and fine-mapping elucidate pathways influencing albuminuria

    Teumer, A, Li, Y, Ghasemi, S, Prins, BP, et al. Genome-wide asso ciation meta-analyses and fine-mapping elucidate pathways influencing albuminuria. Nat Commun . 2019;10(1):4130

    2019

  43. [43]

    Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease

    Lambert, JC, Ibrahim-Verbaas, CA, Harold, D, Naj, AC, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat Genet . 2013;45 (12):1452–1458

    2013

  44. [44]

    Con sistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator

    Bowden, J, Davey Smith, G, Haycock, PC, and Burgess, S. Con sistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol . 2016;40(4):304–314

    2016

  45. [45]

    Pleiotropy robust methods for mult ivariable Mendelian random- ization

    Grant, AJ and Burgess, S. Pleiotropy robust methods for mult ivariable Mendelian random- ization. Stat Med. 2021;40(26):5813–5830

    2021

  46. [46]

    Robust multivariable Mendelian random ization based on constrained maximum likelihood

    Lin, Z, Xue, H, and Pan, W. Robust multivariable Mendelian random ization based on constrained maximum likelihood. Am J Hum Genet . 2023;110(4):592–605

    2023

  47. [47]

    MendelianRandomization: an R pa ckage for performing Mendelian randomization analyses using summarized data

    Yavorska, OO and Burgess, S. MendelianRandomization: an R pa ckage for performing Mendelian randomization analyses using summarized data. Int J Epidemiol . 2017;46(6): 1734–1739

    2017

  48. [48]

    Broadbent, JR, Foley, CN, Grant, AJ, Mason, AM, et al. Mende lianRandomization v0.5.0: updates to an R package for performing Mendelian randomization an alyses using summa- rized data [version 2; peer review: 1 approved, 2 approved with res ervations]. Wellcome Open Res. 2020;5(252). 18

    2020

  49. [49]

    GRAPPLE: R Package for MR Framework GRAPPLE

    Wang, J and Zhao, Q. GRAPPLE: R Package for MR Framework GRAPPLE . . 2024. R package version 0.2.2

    2024

  50. [50]

    rjags: Bayesian Graphical Models using MCMC

    Plummer, M. rjags: Bayesian Graphical Models using MCMC . . 2024. URL https://CRAN. R-project.org/package=rjags . R package version 4-16

    2024

  51. [51]

    R2jags: Using R to Run ’JAGS’

    Su, YS and Yajima, M. R2jags: Using R to Run ’JAGS’ . . 2024. URL https://CRAN. R-project.org/package=R2jags . R package version 0.8-9. 19 θ1 0.0 0.2 0.4 0.6 0.8 0.50 0.75 1.00 1.25 0.0 0.4 0.8 1.2 0.80 0.85 0.90 0.95 1.00 0.00 0.25 0.50 0.75 1.00 5 10 15 20 ρ θ2 Mean estimateMean biasCoverageRejection rateF statistic 0.0 0.2 0.4 0.6 0.8 −0.6 −0.4 −0.2 0....

    2024