{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiestvz4bkmmlvljl5odjxvfovklc7ewtgmszrxo2t2vflnbtad37u",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3mj75spep7vt2"
  },
  "path": "/t/rms-titanic-binary-logistic-case-study/4801#post_13",
  "publishedAt": "2026-04-10T08:59:56.000Z",
  "site": "https://discourse.datamethods.org",
  "tags": [
    "Here"
  ],
  "textContent": "I have a basic question regarding clustered data:\n\nIf we fit a blrm() model on clustered data using `cluster()`. (e.g., repeated measurements _and_ we think they are exchangeable), and use the predict from from rms/rmsb, it sets the random intercept to 0.\nAm I right that these are then predictions for a **typical observation**?\nWhen the prediction model will be used for a new patient, I assume we should integrate out the random effect?\nIf we create calibration plots, should these then also use predictions where we integrated out the random effects?\n\nHere the authors just set the random effects to 0 because they were tiny, but if they are not, what would the recommended approach? MC integration would obviously work, but that would takes a long time…",
  "title": "RMS Titanic Binary Logistic Case Study"
}