ABSTRACT
Figure 5.1 Functions for rejection sampling. Thin line: envelope G(Y); heavy line: density g(Y); broken lines: squeezing functions a(Y) and b(Y).
Zeger and Karim (1991) and Carlin and Gelfand (1991) propose rejection sampling for multivariate full conditional distributions, using multivariate normal and multivariate split-t distributions as envelopes. A difficulty with these methods is in establishing that the proposed envelopes are true envelopes. Bennett et al. (1995: this volume) use rejection sampling for multivariate full conditional distributions in nonlinear models. For the envelope function G, they use the prior distribution multiplied the likelihood at the maximum likelihood estimate.
