Re: Using Poisson data

From: Bruce Southey <southey_at_UIUC.EDU>
Date: Wed, 27 Apr 2005 20:43:35 -0500

There is probably a good chance that you can assume normality. You probably
should at least do so. You may need to first use a square root or natural log
transformation to stabilize the variance.

This should provide some basis for using the variances and model checking. Your
results suggest the the model (including distribution) may not be correct.

Can you at least provide the relevant part of the .asr file?


---- Original message ----
>Date: Thu, 28 Apr 2005 10:41:32 +1000
>From: Rob Brooks <rob.brooks_at_UNSW.EDU.AU>
>Subject: Using Poisson data
>Dear ASREMLers
>I have recently used !POIS to estimate variances of poisson distributed
>data from a half-sib design. The dire and dam within sire components are
>significantly greater than zero but appear impossibly small relative to the
>within groups variance. I gather this is an issue to do with the scaling of
>poisson variances, and it seems from searching back through the ASREML
>discussions that there is no appropriate way to use these variance
>estimates to calculate heritabilities.
>Likewise, I gather that there is no way to estimate covariances where one
>or more traits follows a poisson distribution.
>Am I right about these points or am I missing something?
>Further, would it be appropriate to use the sire breeding values for each
>trait given in the .sln file to estimate the correlation among breeding
>values as a rough estimate of the genetic correlation?
>Your thoughts on these issues would be much appreciated.
>School of Biological, Earth and Environmental Sciences
>The University of New South Wales
>Kensington, Sydney 2052
>NSW, Australia
Received on Tue Apr 27 2005 - 20:43:35 EST

This webpage is part of the ASReml-l discussion list archives 2004-2010. More information on ASReml can be found at the VSN website. This discussion list is now deprecated - please use the VSN forum for discussion on ASReml. (These online archives were generated using the hypermail package.)