Propensity score matching with zip code/geographical distance I have a test group and a control group with a bunch of covariates; one of them is ZIP code.
Is there a methodology that I can use to perform a propensity score matching based on ZIP code or other geographical information such as county, state, or latitude-longitude?
 A: Ridgeway and MacDonald (2009) suggest to use non-linear basis functions (e.g. splines) to estimate non-linear effects in the logistic regression equation in estimating the propensity score. This is amenable to including geographic terms in the equation (e.g. projected latitude and longitude coordinates).
You can also include other terms in the equation that characterize the extent of the spatial matching you want to accomplish, such as neighborhoods of matching (e.g. Ridgeway and MacDonald (2009) match stops within the same police precinct using dummy variables - this would work the same for county) or distance from a particular point if that distance is predictive of selection into treatment (Chagas et al. 2011).

Citations

*

*Chagas, André, Rudinei Toneto & Carlos Azzoni. 2012. A spatial propensity score matching evaluation of the social impacts of sugarcane growing on municipalities in Brazil. International Regional Science Review 35(1):48-69.


*Ridgeway, Greg & John MacDonald. 2009. Doubly robust internal benchmarking and false discovery rates for detecting racial bias in police stops. Journal of the American Statistical Association 104(486):661-668. Online PDF Here.
A: Thoemmes has written a useful SPSS user interface that uses R protocols to perform propensity score matching (http://arxiv.org/pdf/1201.6385.pdf). You will need to install R and a few add-ons, but it works quite well. You will need to be familiar with propensity score matching though, if you are not already. If you provide more details regarding your design, perhaps we can help more. 
If you are familiar with R, then you can see the R packages that Thoemmes cites and skip using SPSS altogether. 
