Page 1 of 1

MCMC residuals

Posted: Fri Feb 04, 2011 1:05 pm
by ewancarr
Hi all,

I'm estimating a two-level logistic regression model, with MCMC estimation. Previously, I've used IGLS estimation, and have used "Model >> Residuals" to plot the ranked residuals for each level-2 unit (the typical 'caterpillar plot').

I notice in the MCMC manual (Browne, 2009) that there are instructions to calculate residuals specifically for MCMC estimation (p. 51). My question is then:
  • Is the 'old' method of calculating residuals (using "Models >> Residuals") therefore invalid for MCMC estimation?
The above reference seems to imply that the new method is more accurate. An alternative question, therefore:
  • Are MCMC residuals calculated using "Model >> Residuals" sufficiently accurate in order to produce these caterpillar plots?
Any help at all would be much appreciated!

Ewan
--
CCSR, University of Manchester