{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihxwm7faovzwe4vsmf5r4swhqxhw3wlzlab6wjuih267mcpv7u3aq",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3micdhlio3aq2"
  },
  "path": "/t/trying-to-understand-statistical-methods-through-the-lens-of-replication/28665#post_6",
  "publishedAt": "2026-03-30T16:54:11.000Z",
  "site": "https://discourse.datamethods.org",
  "textContent": "EvZ:\n\n> you define the z-statistic of the replication study as\n>\n> z_\\text{repl} = b_\\text{repl} / \\sqrt{v_1+v_2}\n\nI am not going to enter into a debate about formulae, but I would like to correct a misunderstanding in your comment about page 24 in Appendix 1 of my paper. The expression 𝑧_repl =𝑏_repl/√(𝑣1+𝑣2) was not described nor intended to be a classical z-statistic (I took on board your point about this potential source of confusion from our previous discussion). To be clear, 𝑧_repl =𝑏_repl/√(𝑣1+𝑣2) is a predictive standardisation reflecting uncertainty in both original and replicating studies conditional on the result of the original study. However, in Appendix 2 lower down on page 24 and onwards, a z-statistic and P value are used in their conventional sense to describe probabilities under a specified null hypothesis. This is distinct from the predictive framework used in the main text, where standardisation is based on the distribution of possible replication outcomes.",
  "title": "Trying to understand statistical methods through the lens of replication"
}