Hello all ---
I am using Intercooled, v. 8.2.
I hope someone can shed some light on a problem concerning technique
(or the appropriateness thereof). I have a dataset wherein I want to
quantify the effect that a few covariates have on a presumably rare
event, a surgical burn. Each subject reported the number of surgeries
he/she did in the previous five years and the corresponding number of
surgical burns. The number of surgical burns ranges from 0 to 15 and
the number of surgeries ranges from 1 to 11,000. My initial attempt at
modeling this involved expanding the data to -long- then running a
logistic model on a dichotomous outcome that denotes whether the surgery
resulted in a wound burn, clustered by surgeon. Parenthetically, the
total number of surgeries exceeds 75,000, the total number of burns is
75, there are 80+ unique surgeons, of which only 20 reported any burns.
My second attempt at modeling this involved the use of a Poisson model
wherein I emulated the example provided on page 207 of the Stata N-R
reference manual. In this example (as in my attempt), I collapsed the
dataset to the 20 surgeons that reported wound burn and optioned the
-poisson- command with -exposure(surgeries)-. Both approaches seem
reasonable, albeit I obtain odds ratios and incidence rate ratios that
are, for a few covariates, contradictory.
Does anyone have any advice or insight into the appropriateness (or
violent inappropriateness) of my approaches??
Many Thanks,
Clint
*
* 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/
