{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreifczqhrllznmvhlkx3er6glwiiapcbvnohpyyhwmu24keyi3lvyhq",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3miicszckq3x2"
  },
  "path": "/t/collider-in-rct-subgroup-analysis/28689#post_1",
  "publishedAt": "2026-04-02T00:22:51.000Z",
  "site": "https://discourse.datamethods.org",
  "tags": [
    "A proposal for capturing interaction and effect modification using DAGs"
  ],
  "textContent": "I came across this interesting article: A proposal for capturing interaction and effect modification using DAGs\n\n**Interaction** : they propose that this term should be used specifically when both treatment and “subgroup” have a direct causal effect on the outcome\n\n**Effect modifier** : they suggest reserving this strictly for situations where the second factor (the modifier) _does not_ have a direct causal effect on the outcome itself.\n\nI want to extend this framework to subgroup analysis in randomized trials, where one would focus on the parameter related to T \\times S to answer the question:\n\n> Is the treatment effect different between subgroups?\n\nI noticed that stratifying by subgroup and including the T \\times S parameter could open a backdoor path in both interaction or effect modifier frameworks if an unobserved variable influences both T \\times S and the outcome:\n\nie **T \\times S would act as a collider**.\n\nHas this phenomenon in subgroup analyses been described before in RCTs? How should one interpret such analyses if there is a risk of opening backdoor paths through unobserved variables?",
  "title": "Collider in RCT Subgroup Analysis"
}