Dear Jarrod,
I have a data set where residual have a heavy-tailed distribution with
some extreme residual values and consequently the distribution deviates
from the Gaussian one.
Is it possible to include an skewed-normal density for the residual in
MCMCglmm package?
I have done the analysis of this data with both ASReml & MCMCglmm. The
results are similar and outcome from MCMCglmm analysis doesn't change
with change in prior.
While MCMCglmm with large numbers of iterations (400000) provides an
appropriate posterior distribution for the variance components which
include additive animal additive effect and common full-sib effects, the
chain doesn't mix well as it is judged by the autocorrelation.
The Box-Cox transformation of data suggests a square root
transformation. When I analysed the square root transformed data the
chain mix well, distributions are OK, but results change when different
prior.
Alex
Dr Alex Safari
Aquaculture and Genetic Improvement
The WorldFish Center, Penang, MALAYSIA
Telephone: 60 4 620 2175
Fax: 60 4 626 5530 (GMT +8)
e-mail: a.safari@cgiar.org
http://www.worldfishcenter.org <http://www.worldfishcenter.org/>
Reducing poverty and hunger by improving fisheries and aquaculture
[[alternative HTML version deleted]]
