{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihutjkbyme3e6n6yszqhc2g4hkkchw4jq3j4dslsagi7wrnxizsge",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3mfword3oknd2"
  },
  "path": "/t/never-seen-this-sort-or-primary-endpoint-and-analysis-in-an-rct/28564#post_16",
  "publishedAt": "2026-02-27T09:51:44.000Z",
  "site": "https://discourse.datamethods.org",
  "textContent": "I listened to that “In the Interim” episode yesterday, so it’s fresh in my mind. I didn’t take such a negative message from it - I think the main issues with the proportional odds model were (a) it wasn’t a good model if you expected the intervention to increase the outcomes at both ends of the scale (e.g. full recovery and death); (b) they had had a specific criticism of the utility-weighted approach that weighting of the ordinall categories was not a good idea because everyone would differ in their utilities (and therefore a proportional odds model is preferable). Scott’s argument was that in fact the proportional odds model is also weighted in ways that you might not like, but it’s just hidden,\n\nScott did say (I think) that they did use proportional odds models in e.g. REMAP-CAP (presumably the issue of increasing both ends of the scale doesn’t apply??).",
  "title": "Never seen this sort or primary endpoint and analysis in an RCT"
}