I know of two methods but I've applied neither. One way to do it is
to include the aggregate measures of the surrounding census tracts as
dependent variables - I have seen examples of this method used by
economists. Another method is to use Geographically Weighted
Regression and it's actually developed by geographers and it's
available in STATA (findit geographical), but you need quite a bit of
details about how your census tracts are located away from one another
if you're going to do GWR. The GWR also has its own program and can
be easily found by a bit of Googling!
Hope this helps!
Ada
On 1/3/06, Dick Campbell <[email protected]> wrote:
> I am doing work in which I have a binary outcome (late stage diagnosis of
> breast cancer)
> and a number of variables at the level of the census tract. It is common
> practice in
> epidemiology to use aggregate measures of this kind, in addition to
> person-level data,
> to model binary outcomes. In many cases, including mine, the census tracts
> are from
> a relatively small area (e.g. Cook County Illinois) and, of course, some
> tracts are
> contiguous and others separated by some distance. This raises the
> possibility of some
> sort of spatial dependence in outcomes. I have not seen any attempt to
> correct for spatial
> auto-correlation of this kind in the epidemiology literature. For
> continuous outcomes,
> doing so has become a fairly straight forward task, but a search of both
> Stata space and
> more generally has not turned up any software for the logistic or probit
> case. I am new to this,
> and could easily have missed something, but could anyone direct me to
> either published work
> or software resources?
>
>
*
* 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/
