{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreievog36q6l7spyriuoija36owcxtxpzizaomwisuqd5qnlju3dzky",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3milokjdrcdq2"
  },
  "path": "/t/a-b-testing-for-information-gain/28690#post_2",
  "publishedAt": "2026-04-02T20:55:30.000Z",
  "site": "https://discourse.datamethods.org",
  "tags": [
    "http://dx.doi.org/10.1198/tast.2009.08196"
  ],
  "textContent": "If the response variable is semi-continuous and has somewhat of a normal distribution you could fit a unstructured covariance matrix and try to fit several correlation structures to it, then use the fitted structure to estimate efficiencies of certain contrasts. This paper may help: http://dx.doi.org/10.1198/tast.2009.08196",
  "title": "A/B Testing for Information Gain"
}