Chunling Lu --
My example demonstrates that you can't do what you propose--assuming
the dummy for positive visits is generated by a visits variable
distributed Poisson can easily lead to a very poor estimate of mean
visits--and you have no way of assessing how badly you've done in your
data. If the only information you have is that 87% have zero visits,
how will you estimate the mean of the nonzero values? If you assume
Poisson, using only the fact that 87% have zero visits, your estimate
of lambda is 0.137, but the true mean is 1.42, so you'd be off by a
factor of ten using my data and your method. If you assume neg bin,
how will you model both the mean and dispersion?
Just find some better data on visits.
On 5/25/07, Chunling Lu <chunling_lu@harvard.edu> wrote:
Dear Austin,
Thanks very much for the feedbacks. I guess I didn't express my question
clearly. What I want to get is the "mean" of number of visits, not the
number of visit for each individual. So the lamda value in your data will be
the mean of visits if we assume the distribution is poisson. Of course,
that's a conveneient assumption. As David pointed out, the real distribution
will skew to the right and we may need to use negative binomial. Looking
forward to hearing your comments. Thanks again. Chunling
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/