目次

ベイズ信用区間

信頼区間と結局一緒

以下抜粋

Just like a ‘confidence’ interval, a Bayesian posterior-probability (‘credible’) interval can be treated as a compatibility interval, showing effect sizes most compatible with the data under the background model and prior distribution used to compute the interval (Greenland, 2019a). If the data, model, or prior do not inspire ‘confidence’ or ‘credibility’, the interval should not either (McElreath, 2020, Ch: 3). In contrast, ‘compatibility’ does not depend on how correct or incorrect the model assumptions are; it is just a mathematical statement about a relation between the data and the model, however questionable the data or model may be.
1)
【CC-BY】Journal of Information Technology Vol37, Issue 3, September 2022,316-320 DOI