Spatial Autogressive Models for Binary Outcomes?

[Brenda Philips]

Hi Folks,

I was reading through about Stata's spatial autogressive models (https://www.stata.com/features/overv...essive-models/) especially the capability to create a W matrix that is not necessarily based on geographical location. From the manual "Spatial analysis is about accounting for spillover effects. Consider an analysis of test scores of students. There may be spillover effects among friends for no other reason than friends share similar but relevant unmeasured characteristics. Or you might hypothesize more direct effects. Such data are known as social network data." I am interested in modelling how drivers and influenced by other nearby drivers, and this seems to offer me a great way to account for autocorrelation. The problem however, it that these models seem to only apply to linear regression not logistic regression.

My question is: Are there any stata packages that address spatial autogressive model with binary outcomes? I don't think I can use XTGEE because my clusters and are not independent, or I don't have enough clusters, or they are not balanced. Thanks in advance.

[Justin Niakamal]

Are there any stata packages that address spatial autogressive model with binary outcomes?

I don't believe there are any explicit packages for what you're looking for but in the [SP] manual there's an example of how you can use spgenerate to create a spatial lag of a covariate and then use that in a logistic regression. I think this is as close as you'll get. See the "Use with other datasets" section of the manual. Hope this helps.

[Brenda Philips]

Thanks...

[Mike Lacy]

There might be some other possibilities described in the work of some criminologists. Some of them work with city blocks as a unit of analysis, with variables like "Crime on this block y/n?," and I recall that even about 10 years ago, when fancier software was less available, they were working on solutions to spatial autocorrelation in such contexts.

[Brenda Philips]

Thanks, I'll do a search on that topic and see what I can find. I was also looking at social network analysis in Stata, but the nwcomands don't seem to address autocorrelation.