# MR Assumption 3 (Exclusion Restriction)

**URL:** <https://isgw-forum.colorado.edu/t/mr-assumption-3-exclusion-restriction/896>\
**Category:** Day 7\
**Created:** [June 11, 2026, 3:47pm UTC](https://isgw-forum.colorado.edu/t/mr-assumption-3-exclusion-restriction/896 "2026-06-11T15:47:22Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![anonymous](https://isgw-forum.colorado.edu/letter_avatar_proxy/v4/letter/a/8491ac/32.png) [@anonymous](https://isgw-forum.colorado.edu/u/anonymous)\
**Post date:** [June 11, 2026, 3:47pm UTC](https://isgw-forum.colorado.edu/t/mr-assumption-3-exclusion-restriction/896/1 "2026-06-11T15:47:22Z")

</div>

By third assumption, you mean SNP marginally not associated with outcome?

> Replying to “By third assumption, you mean SNP marginally not …”:

Marginally, it is expected to be associated (via the exposure variable). But the SNP is not associated with the outcome conditional on the exposure (i.e., over and above the SNP-outcome association mediated through the exposure)

> Replying to “By third assumption, you mean SNP marginally not …”:

That’s correct- one subtlety is that the SNP must not be associated with the outcome conditional on BOTH the exposure AND confounders. Conditioning just on the exposure will actually open a path from SNP to confounders to outcome since exposure is a collider (i.e. caused by the SNP and confounders)- therefore important to condition on both exposure and confounders. See here for more info: [Collider scope: when selection bias can substantially influence observed associations - PubMed](https://pubmed.ncbi.nlm.nih.gov/29040562/)