04-20-2015 12:10 AM
I originally asked this question on Stack Overflow/Exchange and was directed here. I don't remember if I posted this already (since the other post is likely in moderation if I did). If this is a double-post I apologize.
I'm trying to model the number of parks in a neighborhood as a function of education, land area (both continuous variables), and poverty perScentage (categorical). There may be other covariates included but for the sake of the question I will exclude them for now. I chose a Poisson model because to my understanding this is the model that is ideal for count data and my data has a lot of sampling zeroes.
My unit of analysis for the neighborhood is the census tract. Using ArcGIS, I joined geocoded point information about parks to a census tract shape file, exported this information into SAS, and generated counts. I then ran the following Poisson model.
proc genmod data = parks;
class povper / descending;
model cnt = povper edu area / link=log dist=poi;
My issue with this model is that it doesn't take into account the number of parks in neighboring census tracts. I've been reading about possibly creating a spatially lagged dependent variable, but according to this answer, spatial lag and poisson don't necessarily mix. All of the census tracts in my table have x/y coordinates. Does anyone have advice on how I could incorporate the park counts from neighboring census tracts into my model using SAS?
04-21-2015 08:41 AM
A possibility here would be PROC GLIMMIX, which allows for spatial covariance structures. Check out the thread at
where this is discussed, looking at counts.