Hi all, I would like to implement Conley standard errors to allow for spatial correlation, but I'm not sure which code is most appropriate.
My data is repreated cross-sections, and I'm implementing a Difference-in-Difference design, by birth-year cohort and a continuous measure of worm infection:
Y = Infection + Years of treatment + Infection*Years of treatment + Birth-year fixed effects + State fixed effects + Survey round fixed effects
Each cross-section has outcome data recorded at the individual level, and data at the survey cluster level (i.e. about 30 households). These clusters change for each survey round, so I'm not sure if I can use standard errors clustered at the survey cluster level. However, I have each survey cluster's latitude/longitude, so I was hoping to utilise this to use Conley standard errors.
I tried ols_cluster_hac but I don't think it is appropriate. My data is not really a panel (Each individual only appears once. The survey clusters don't appear in every survey round, although they would appear in every birth-year). Also, it does not let me include factor variables for the fixed effects.
Does anyone know of a good way to deal with this type of data? One option might be to cluster at the survey cluster level. However, is this a problem if the survey clusters are not repeated in each survey round? (In contrast, each state appears in each survey round.)
Thank you kindly for your advice,
Lucas
0 Response to Spatial Correlation in Standard Errors
Post a Comment