[R] mvtnorm and noncentrality parameters

Paul Sweeting mail at paulsweeting.co.uk
Mon Oct 11 10:18:25 CEST 2010


  Hi

I'm trying to calculate densities for the multivariate noncentral t 
distribution.  For the avoidance of doubt, and becuase there do seem to 
be at least two definitions for a noncentral t distribution, This is a 
noncentral t distribution of the sort described in McNeill Frey and 
Embrechts (2005) and elsewhere that can be constructed as a normal 
mean-variance mixture. It is not simply a shifted t distribution. The 
univariate noncentral t distribution density can be calculated using the 
function dt in the TDist package. The univariate version seems to be do 
what I expect, i.e. give a distribution with a different shape.

The obvious approach for calculating a multivarate result seems to be to 
use dmvt in the mvtnorm package.  However, the noncentrality parameters 
here appear actually to be means.  In other words, they seem to simply 
shift the distribution rather than calculating a noncentral mutivariate 
distribution as described by McNeill et al.  So, a few questions:

1. Is my understanding of what dmvt is doing correct?
2. Is this what it is supposed to be doing?
3. Is there any way of deriving the densities that I want in R (short of 
writing my own function...)?

Thanks

Paul



More information about the R-help mailing list