Skip to content
Prev 5990 / 20628 Next

Inverse Wishart: rIW, riwish, and Sorensen-Gianola

On Tue, May 3, 2011 at 2:36 AM, Jarrod Hadfield <j.hadfield at ed.ac.uk> wrote:
Thanks Jarrod,

It appears that there is some inconsistency in the notation used by
some authors.
For example, Sorensen-Gianola write T ~ IW_p(Sigma, n) and say
E[T|Sigma,n] = Sigma^-1 / (n - p - 1).

Other authors (B. Walsh, e.g.) write T ~ IW_n(Sigma^-1), where
T = 1/W, and W ~ W_n(Sigma). So the confusion is over what is the
parameter of the Inverse Wishart: Sigma or Sigma^-1?

If the "Sigma" in the Sorensen-Gianola definition is replaced by Sigma^-1,
then you get E[T|Sigma,n] = Sigma/(n - p - 1), the formula that appears
in the MCMCpack man page for riwish.

Dominick