mrrobust: Stata package for two-sample Mendelian randomization analyses

Latest updates
Short video introduction
Helpfile examples
Overview
Installing and updating mrrobust
Unit tests
Authors
How to cite the mrrobust package
References
Collaboration
Acknowledgements

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.

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 simulatio

Mrrobust

Install / Use

README

mrrobust: Stata package for two-sample Mendelian randomization analyses

Latest updates

Short video introduction

Helpfile examples

Overview