Hello Allstaters
A student approached me with the following problem, and much to my own
dismay I was unable to offer a solution. Perhaps someone else has an idea?
'I'm modelling longitudinal faecal egg count data in patients who are
infected by a gut parasite. Samples are collected at 12, 18, 22 and 72
months after patients have been treatment. Re-infection occurs at some rate
that varies considerably amongst individuals, depending on their level of
exposure to the parasite. Egg counts are characteristically very skewed and
essentially follow a negative binomial distribution at each time point. My
intention was to use a negative binomial random effects model, with repeated
measures clustered within individuals. However, the random intercepts are
not normally distributed (but very much skewed).'
Any suggestions of alternative approaches that allow for negative-binomial
random effects would be gratefully received. We've tried transforming the
egg-count data to something more symmetrical but to no avail, as there is
some considerable '0' inflation.
Cheers
Mark
------
Mark Booth PhD
Senior Research Associate
University of Cambridge
Dept of Pathology
Tennis Court Rd
Cambridge CB2 1QP
Tel: 01223 333338
Fax: 01223 333471
Web: www.path.cam.ac.uk/~schisto
www.matangini.org.uk
|