{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiealpdw7lqrle5k5iy6shu4sysau3rhkrg7wuvrftnd2jxfsg2cxq",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3mfucszeh62o2"
  },
  "path": "/t/never-seen-this-sort-or-primary-endpoint-and-analysis-in-an-rct/28564#post_15",
  "publishedAt": "2026-02-26T20:56:27.000Z",
  "site": "https://discourse.datamethods.org",
  "tags": [
    "post about his Belz Lecture",
    "@f2harrell",
    "berryconsultants.com",
    "50: The Fallacy of Ordinal Endpoints | Podcast Episode"
  ],
  "textContent": "A couple of recent things that I’ve come across on this topic.\n\nI was re-reading Prof Lumley’s post about his Belz Lecture and there was a section that I either missed the first time around or was added later. One of the points very briefly mentions @f2harrell (**bold** emphasis mine)\n\n> Issues that came up in questions or afterwards\n>\n>   * I should have made it clearer that it may not be the _statistician_ who evaluates the tradeoff. For example, in medical treatment it would ideally be the individual patient. In government it may be the Minister rather than the official stats agency. The statistician’s job is to make sure that people know evaluating the tradeoff is needed.\n>\n>   * This _is_ a lot like other utility tradeoffs: eg equity vs number of people helped in public health. Someone needs to decide; the problem can’t just delegated to maths.\n>\n>   * **Someone (?Cameron Patrick) asked about how this differs from Frank Harrell’s views and why. I think (having given a version of this talk at his department and talked to him about this) that Frank regards failures of ordering as less important and less plausible than I do. He has also been more interested in ordinal scales as a way of pushing back against dichotomisation in medical statistics.**\n>\n>\n\n\nAlso, Berry Consultant’s latest “In the Interim” podcast episode is quite negative about proportional odds models for ordinal outcomes. Their main contention seems to be that (as opposed to explicitly assigning values to the ordinal levels and analysing it as an interval measure), the PO model implicitly assigns these values as a function of the observed frequency of the outcomes. The article presenting this is still in development, but I think the argument is based on comparing the PO model’s score statistic to that from a t-test of the weights. My initial reaction is that this argument isn’t overly compelling, but I need to think about it a bit more…\n\nberryconsultants.com\n\n### 50: The Fallacy of Ordinal Endpoints | Podcast Episode\n\nIn this episode of \"In the Interim…\", Dr. Scott Berry and Dr. Lindsay Berry investigate the statistical foundations and clinical implications of analyzing ordinal endpoints, drawing on experience from major stroke and COVID-19 trials.",
  "title": "Never seen this sort or primary endpoint and analysis in an RCT"
}