Model assessment
Murray A. Aitkin · 2022
We have so far assumed that the probability models specified in the analyses are appropriate. We now examine ways in which this assumption can be investigated. The first approach, for continuous distributions, is through the agreement between the empirical cdf (ecdf) and the model cdf, described several times previously. For this we construct a credible region for the true model cdf from the set of credible intervals from the ecdf at each distinct data point, based on the binomial model and the Beta posterior distribution. We give the formal construction.