Joseph Hilbe:
If you want to model the log fraction ln(y/ytotal) as a linear
function of X and y may be zero in your data, then a natural model is
ln(y)=ln(ytotal)+Xb+e
or
y=exp(ln(ytotal)+Xb+e)
or a Poisson regression of y on X with ytotal specified as exposure,
right? But ytotal equal to zero means you cannot glean any
information from that case, since y must also be zero and the fraction
y/ytotal is undefined.
On Fri, Mar 28, 2008 at 12:43 PM, <[email protected]> wrote:
> I've never seen such a data set. Strange. The exposure variable typically
> gives us the time frame or area/group in which counts occur; ie the model with
> exposure tells us, for each covariate pattern, how many counts are in a given
> range of time or area/group. If the time or area/group is 0 (exposure==0),
> then there can be no meaningful counts. I cannot see how observations
> associated with an exposure of 0 can be kept in the model.
>
> Try modeling a Poisson or negative binomial model with a zero exposure. I'll
> bet that the algorithm will not let you proceed. I suspect that the student
> does not know the meaning of exposure, and needs to recode the data so that
> it reflects a meaningful relationship. Maybe its too early in the morning, and
> I'm missing something. But I don't think so.
>
> Joseph Hilbe
>
> ====================
> A student comes in with a poisson model. The response variable is the
> number of seeds of a certain species. There is an exposure variable
> which is the total seeds of all species. The problem is that there
> are six exposure values of zero. There are three other predictor
> variables and 72 total observations. Is there any way of dealing with
> this problem other than dropping those six values? Any suggestions?
*
* 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/
