Modelling data at two separate locations

Hello,

I am looking for advice on what kind of models I can use for my data.

The dependent (y) variable are count data (poisson) and are located at an irregular pattern. The
independent data (x1, x2, x3, ...) are based on regions (closed polygons) (but could be represented at the
centroids of these regions if this helped, but these centroid will not correspond to the location of the
y's). My problem is that I do not have an exact (other than geographical) data match between the y's and x's.
In particular I have about 8,000 y observation points and 20,000 x regions.

What I would like to examine is how the proximity of various regions (via their x1, x2, x3, ... values)
influenced the value of my y variable. I have looked at some common spatial models (SEM, SAR and GWR) but
they all require the y and x data to be observed at the same locations. What I have done is to aggregate my y
counts to the region geography, so that I have for each region a y aggregated count and the corresponding
x1, x2, x3, ... , but I have lots of missing values where a region has no y count, and my concern is that these
missing values will mean I lose the information about these regions in forming my relationship.

Thanks.

--
Stephen Clark,
First year PhD,
School of Geography
0113 343 6707

	[[alternative HTML version deleted]]