Comment: Bayesian multinomial probit models
with a normalization constraint
by Agostino Nobile
McCulloch, Polson and Rossi (2000) have proposed a prior for the
Bayesian analysis of the multinomial probit model which incorporates
the identification (or normalization) constraint $\sigma_{11}=1$.
Some empirical evidence on the performance of the prior and related
sampler is provided. Direct simulation
from Wishart and inverted Wishart distributions, conditional on
one of the elements on the diagonal, is then considered.
This suggests an alternative way of
imposing the normalization constraint
in a Bayesian multinomial probit model.
Keywords:
Identification, Inverted Wishart distribution, Wishart distribution.