Thanks to the four people who responded to my model diagnostics query.
Summarises follow:
Richard Stevens wrote:
"
I recommend "Regression modeling strategies" by Frank Harrell, published by
Springer, which gives excellent advice on model-building. It specialises in
linear, logistic, and Cox regression models.
"
Jim Hodges wrote:
"
As far as I know, there's no software out there allowing you to routinely
do these kinds of diagnostics.
My favorite paper on the subject is (surprise!) my own:
Hodges JS. ``Some algebra and geometry for hierarchical models, applied
to diagnostics (with discussion)", Journal of the Royal Statistical
Society, Series B, 60:497-536, 1998.
It gives a survey current as of 1998 on methods for outliers and other
diagnostics. The novelty in the paper is a unified system of diagnostics
analogous to the one for regular linear models. To my knowledge there
haven't been any other systems of diagnostics proposed since then, though
there have been a few more one-off methods. My web page
http://www.biostat.umn.edu/~hodges/
has a link to some S code that implements my diagnostics but frankly if
you have access to any kind of half-decent programming talent you'd be
safer to write your own.
Regarding outlier deletion, I don't believe there's a consensus about this
issue. In my own statistical practice, if there's an outlier that appears
to be correct and that has a noteworty influence on the results, I present
as the main analysis the analysis *including* the outlier, but also
explain its effect on the results. My reason is that absent some
compelling reason to delete an outlier, it represents bona-fide variation
and it would thus be misleading to delete it. However, I am not a
dogmatist about this and I'd have to see what you're doing to decide how I
wanted to handle them.
I would be interested to see what other responses you get.
"
Duncan Hedderley wrote:
"
I don't recall if SMART ( http://www.bioss.ac.uk/smart/ ) has a module
ondiagnostics,but I've found it useful as an online crib fro
beyond-the-basics techniques
"
Jenny Freeman suggested the following book looked promising, 'Regression
analysis of count data' by A. Colin Cameron and Pravin K. Trivedi,
published by Cambridge University
Press - has sections on negative binomial and zero inflated models
Once again thanks to these people for responding,
Steve
Stephen Kay
Adelphi Group Products
www.adelphigroup.com
DISCLAIMER: The information in this message is confidential and may be
legally privileged. It is intended solely for the addressee. Access to this
message by anyone else is unauthorised. If you are not the intended
recipient, any disclosure, copying, or distribution of the message, or any
action or omission taken by you in reliance on it, is prohibited and may be
unlawful. Please immediately contact the sender if you have received this
message in error. Thank you.
|