Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: use of offset with xtlogit


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: use of offset with xtlogit
Date   Fri, 24 Apr 2009 16:04:00 -0400

Sorry, the prior graph is not illustrative of the right problem--if
the mean of y is exp(x) where x is a linear combination of some
observed RHS variables Xb plus distance, then the probability that y
is at least 1 looks like the solid line in

tw function y=gammap(1,exp(x)),ra(-5 5)||function y=invlogit(x),ra(-5 5)

where x varies with distance, and the dashed line is your logit model.

But I guess the real problem is that you don't think the Poisson
distribution is right for your model, and you don't know what family
of distributions the DGP uses, which suggests semiparametric
estimation...

On Fri, Apr 24, 2009 at 3:14 PM, Austin Nichols <austinnichols@gmail.com> wrote:
> <K.Boughey@uea.ac.uk>:
> In short, I think not.  If y is distributed Poisson with mean x, where
> length of transect enters x linearly (with coef 1), then the
> probability of seeing y>0 is gammap(1,x) and the logit link does not
> make a lot of sense...  look at:
>
> tw function y=gammap(1,x),ra(-5 5)||function y=invlogit(x),ra(-5 5)
>
> See also:
>   findit gammafit
> which models y as a function of x implicitly defined by the density
>   gammaden(ax,bx,0,y)
> integrating to one, I believe--I haven't used it (perhaps Nick or
> Stephen will comment).
>
> On Fri, Apr 24, 2009 at 9:20 AM,  <K.Boughey@uea.ac.uk> wrote:
>> Dear Statalisters,
>>
>> I'm investigating the relationship between the number of bats recorded
>> along a series of transects and the landscape characteristics of the
>> transects. I intend to model the count data using xtpoisson, then convert
>> the dependent variable to presence/absence and model using xtlogit.
>>
>> However, my transects vary in length, which will affect both the number of
>> bats encountered and to a lesser extent bat presence/absence. I intend to
>> include transect length as an offset when using xtpoisson, but should I
>> also include it as an offset when using xtlogit? Does an offset work in
>> comparable way with binary data?
>>
>> Many thanks,
>>
>> Katherine Boughey
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index