摘要: (c) Choy, Chan, and Yam:" Robust analysis of Salamander data, generalized linear model with random effects"(contributed). I discuss these papers now in reverse order. Choy et al. use a scale mixture of normals approach to allow for t-distributed random effects in a logistic regression example. They analyze a well-known data set (at least to Biometrics readers!), the Salamander mating data, analyzed in Karim and Zeger (1992). Computations are done with Win-BUGS. While the general approach taken is certainly interesting, a rather extreme prior is used for the degrees of freedom of the t-distribution, a gamma (0.001, 0.001) prior (incidentally, the WinBUGS default for precision parameters). This, however, corresponds to a prior probability of 0.9937 (!) that the degrees of freedom are smaller than unity, ie, that the t-distribution is more extreme than a Cauchy distribution! Unfortunately, the authors call this prior" non …