mrrobust: Stata package for two-sample Mendelian randomization analyses

Latest updates

To obtain the latest update please see the instructions below.

  • September 2025:
    • Reran certification scripts on StataNow 19.5
  • May 2025:
    • Reran certification scripts (a.k.a. tests) on StataNow 18.5
  • July 2024:
    • Updated website to mention OpenGWAS instead of MR-Base
    • In Stata 18 on a graph twoway plot the default legend position appears to have changed from the 6 o’clock to the 3 o’clock position. Amendments have been made to mreggerplot, mrfunnel, and mrmodalplot to use the 6 o’clock position as the default again.
  • January 2024:
    • Reran certification scripts under Stata 18.0
  • August 2023:
    • Added a record of the Stata version in the certification scripts
  • April 2023:
    • Improved the alt text descriptions for the images in the README and package website, and also centred the images
    • Remade the mrrobust website using Quarto
    • Updated the Sanderson et al., bioRxiv, 2020, doi reference to its published version Sanderson et al., Statistics in Medicine, 2021, doi
    • Reran certification scripts under latest Stata 17.0
  • February 2023:
    • Updated R Markdown example to use the CRAN version of the Statamarkdown package
  • September 2022:
    • Updated manual installation instructions
  • February 2022:
    • Ran cscripts under Stata 17.0
    • Updated website examples to run under Stata 17.0
  • September 2021:
    • Changed relevant http: URLs to https:
    • Minor edits to the helpfiles
  • June 2021:
    • Published an interactive Code Ocean capsule demonstrating the use of the mrrobust package here
    • By default mrforest now specifies a fixed effect standard error for its IVW estimate
  • April 2021:
    • Added the I-squared statistic and its 95% CI to the mregger ..., heterogi output
  • February 2021:
    • Fixes to mrforest and mrleaveoneout related to the recent update to metan. mrforest and mrleaveoneout now use metan9 instead of metan because of the changes to metan syntax. No change was necessary in the dependency scripts because metan9 is also installed with ssc install metan
    • Checked cscripts pass
    • Checked examples on website run. And changed the 2 examples which use the TwoSampleMR R package to use the new ID code for the exposure data
    • Updated dependency.do to make it more robust to the more frequent updates to metan
    • Updated mrdeps to make it more robust to the more frequent updates to metan
  • October 2020:
    • dependency.do and mrdeps now install the updated version of the moremata package
    • Checked the cscripts run under Stata 16.1
    • Added description of Q-statistic as Cochran’s and Ruecker’s for the IVW and MR-Egger models respectively
    • In various helpfiles added clarification that genotype-disease stands for SNP-outcome (or indeed instrument-outcome) and that genotype-phenotype stands for SNP-exposure (or indeed instrument-exposure) respectively; i.e. the estimates required for the top and bottom of the IV Wald ratio estimate
  • August 2020:
    • Added html versions of the helpfiles to the website. These are available from the Helpfiles website menu bar item
    • Added extra decimal places examples to helpfiles of mrforest and mrleaveoneout
    • mrfunnel now includes a legend on its plot
  • July 2020:
    • Added gxse() option to mrmvivw to return instrument strength QA statistic for instrument validity in e(Qa) (Sanderson et al. 2019)
    • The gxse() option additionally returns the Qx and conditional F-statistics for each phenotype for instrument strength in e(Qx) and e(Fx) (Sanderson et al. 2020)
    • Added tdist option to mrmvivw and mrmvegger
    • mrmvivw and mrmvegger now ereturn the RMSE in e(phi)
    • mregger, ivw now displays the square root of the residual variance (residual standard error) and ereturns this is e(phi)
    • Checked that examples on website still run
    • Added mrleaveoneout command to perform leave one out analysis
  • June 2020:
    • Simplified the outcome variable name in mregger b and V e-returned matrices. Turn this off with new oldnames option
    • Added basic multivariable MR-Egger command, mrmvegger
    • Added basic multivariable IVW command, mrmvivw (currently command names mvmr and mvivw also work)
  • February 2020:
    • Updated contact details
    • Minor edits to helpfiles, to show examples setting seed() option where helpful
    • Fixed mregger bug where r(table) was not returned with the gxse or heterogi options. The output for these options now appears before the coefficient table.
    • Minor amendments to formatting of mregger gxse output
    • mregger now ereturns e(phi), the scale parameter, in some cases
  • January 2020:
    • mregger now additionally returns r(table)
    • Certification scripts: added master.do and renamed and edited a few scripts
    • Added mr command. Commands may now be run as either mr egger ... or as previously mregger ....
    • Best of IJE 2019!
    • mrmedian, mrmedianobs, mreggersimex, mrmodal, and mrratio now additionally return the r(table) matrix (the information from the coefficient table)
    • Added an example showing how you can save and export your estimates using r(table), see here
  • December 2019:
    • Added Q_GX to ereturn and display output when gxse() option specified to mregger
    • Changed Q_GX and I^2_GX output to use first order weights in mregger output. This matches the output from the mr_egger() function in the MendelianRandomization R package. Use the unwi2gx option to report the unweighted statistics.
  • July 2019:
    • Checked that examples on website still run
  • December 2018:
    • Improved compatibility with the github Stata package, i.e., mrrobust and its dependencies can be installed simply by issuing: gitget mrrobust, if you have the github package installed. See below for instructions.
    • mrdeps command added for conveniently installing dependencies
  • November 2018:
    • Example showing the use of the TwoSampleMR R package and mrrobust in the same R Markdown script (.Rmd file) is here
    • Example showing the use of the TwoSampleMR R package and mrrobust in the same Stata Markdown script (.stmd file) is here
  • September 2018:
    • IJE paper published online here
  • August 2018:
    • Click here for the example code and output from our IJE article
  • May 2018:
    • Click here for code and output from the examples in the helpfiles
    • This page is now rendered on GitHub Pages here
  • April 2018:
    • mregger now has option radial which implements the radial formulation of the MR-Egger model, and of the IVW model when used with option ivw

Short video introduction

Click here for a short video demonstrating the use of the package.

A screenshot of a video demonstrating the use of the mrrobust package.

Helpfile examples

Click here for some of the code and output from the examples in the helpfiles.

Once the package is installed, there is a summary helpfile which can be viewed in Stata with:

help mrrobust

This has links to the helpfile for each command, which has an example near the bottom. In these examples you can click on the code to run it.

Overview

The mrrobust package is a collection of commands for performing two-sample Mendelian randomization analyses using summary data of genotype-phenotype and genotype-outcome associations.

Such data can be obtained from repositories such as MR-Base https://www.mrbase.org (Hemani et al. 2016).

The package contains the following commands:

  • mrdeps installs dependencies for the package
  • mrratio implements the standard instrumental variable ratio (Wald) estimate with a choice of standard errors/confidence intervals
  • mrivests automates calling mrratio on all the selected genotypes in your dataset
  • mregger implements the IVW and MR-Egger regression approaches introduced in Bowden et al. (2015)
  • mreggersimex implements the simulation extrapolation algorithm for the MR-Egger model
  • mreggerplot implements a scatter plot with fitted line (either from IVW, MR-Egger, or weighted median estimators) and confidence interval
  • mrmedian and mrmedianobs implement the unweighted, weighted, and penalized weighted median IV estimators robust to 50% invalid instruments in Bowden et al. (2016)
  • mrmodal implements the zero modal estimator of Hartwig et al. (2017)
  • mrmodalplot plot of density used in modal estimator
  • mrforest implements a forest plot of genotype specific IV estimates and estimates from models (e.g. IVW and MR-Egger)
  • mrfunnel funnel plot of genotype specific IV estimates
  • mr acts as a primary command, e.g. so the other commands can be run as mr egger ... as well as mregger ...
  • mrmvivw (mvmr, mvivw) implements the multivariable IVW model
  • mrmvegger implements the multivariable MR-Egger model
  • mrleaveoneout implements leave one (genotype) out analysis

Installing and updating mrrobust

To install mrrobust in Stata versions 13 and later you have two choices.

1. Use net install

net install mrrobust, from("https://raw.github.com/remlapmot/mrrobust/master/") replace
mrdeps

In this code mrdeps installs the dependencies. These are addplot, kdens, and moremata packages (all by Ben Jann), the heterogi command (Orsini et al.), the metan command (Harris et al.), and the grc1leg command (Wiggins).

If you have previously installed the package and the net install command above fails with an error message that there are two copies of the package installed simply run adoupdate.

To check if there is an update available to any of your user-written Stata packages run adoupdate. To update mrrobust run:

adoupdate mrrobust, update

To uninstall mrrobust, issue in Stata:

ado uninstall mrrobust

If this fails with an error message mentioning that you have “multiple citations/instances of the package installed” simply issue adoupdate mrrobust. This should leave you with the most recent version of the package you previously installed. You can then run ado uninstall mrrobust.

2. Use the github package

net install github, from("https://haghish.github.io/github/")
gitget mrrobust

This automatically installs the dependencies.

To update the package issue:

github update mrrobust

To uninstall mrrobust issue:

github uninstall mrrobust

Installation instructions for Stata version 12 and earlier versions (and perhaps Stata version 13)

The net install syntax for installing mrrobust does not work under Stata version 12 and earlier because this webpage has an address starting with https rather than http. In such cases you need to do a manual installation.

To download and install mrrobust manually:

  • Click the green “Clone or download” button at the top of the GitHub repository here and download as a zip archive or click this direct link.

  • In your file explorer extract the zip archive and find its filepath, e.g. C:\Users\tom\Downloads\mrrobust-master\mrrobust-master

  • In Stata run

    net install mrrobust, from("C:\Users\tom\Downloads\mrrobust-master\mrrobust-master") replace

The installation commands for the other dependencies should work. However, if you need to install them manually their zip archives are available at the following links (extract the files from the downloaded zip archives and place them in your PERSONAL directory on your adopath):

  • the moremata package is available as a zip file here

  • the addplot command is available here

  • the heterogi command is available here

  • the kdens command is available here

  • the metan command is available here

  • the grc1leg command can be installed with

    net install grc1leg, from("https://www.stata.com/users/vwiggins")

Unit tests

As far as I know, and unlike R which has the testthat package (and other testing packages), there is no recognised standard for writing unit tests for Stata commands. StataCorp. refer to do-files with tests as cscripts (certification scripts). I publish my cscripts (and their log files of output) in the cscripts directory.

Authors

Tom Palmer , Wesley Spiller, Neil Davies

How to cite the mrrobust package

Spiller W, Davies NM, Palmer TM. Software Application Profile: mrrobust — A tool for performing two-sample summary Mendelian randomization analyses. International Journal of Epidemiology, 2019, 48, 3, 684–690. https://doi.org/10.1093/ije/dyy195.

Thank you to all our users who have cited mrrobust. We made The Best of IJE 2019!

Collaboration

If you would like to extend the code or add new commands I am open to receiving pull requests on GitHub or send me an email to .

References

Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International Journal of Epidemiology, 2015, 44, 2, 512–525. https://dx.doi.org/10.1093/ije/dyv080.

Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genetic Epidemiology, 2016, 40, 4, 304–314. https://dx.doi.org/10.1002/gepi.21965.

Hartwig FP, Davey Smith G, Bowden J. Robust inference in two-sample Mendelian randomisation via the zero modal pleiotropy assumption. International Journal of Epidemiology, 2017, 46, 6, 1985–1998. https://doi.org/10.1093/ije/dyx102.

Hemani G et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife, 2018, 7:e34408. https://doi.org/10.7554/eLife.34408.001.

Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. International Journal of Epidemiology, 2019, 48, 3, 713–727. https://doi.org/10.1093/ije/dyy262.

Sanderson E, Spiller W, Bowden J. Testing and correcting for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization. Statistics in Medicine, 2021, 40, 25, 5434–5452. https://doi.org/10.1002/sim.9133.

Acknowledgements

Thanks for helpful feedback and suggestions to (in no particular order): Jasmine Khouja, Michael Holmes, Caroline Dale, Amy Taylor, Rebecca Richmond, Judith Brand, Yanchun Bao, Kawthar Al-Dabhani, Michalis Katsoulis, Ghazaleh Fatemifar, Lai-Te Chen, Sean Harrison, Emma Anderson, Cassianne Robinson-Cohen, Alisa Kjaergaard, and Steve Burgess.