Choosing between spatial fixed-effects and time fixed-effects in Dynamic Spatial Regression using Stata

[jayce le]

I am doing Dynamic Spatial Regression and encountered problems in choosing a spatial fixed-effects wtype(ind), or time fixed-effects wtype(time), or both wtype(both). What are the differences when we use these 3 types, which statements in Stata can test the above effect?

[George Ford]

Haven't done much of it, but here's what I think is happening.

(wtype(ind)) when you suspect heterogeneity across spatial units (cross sections)

(wtype(time)) to control for temporal shocks or trends (time only)

(wtype(both)) when heterogeneity exists in both dimensions or when not sure about it (like TWFE) Maybe this would help.

 xsmle y x1 x2, wmat(W) model(sdm) fe wtype(ind)
estimates store ind
xsmle y x1 x2, wmat(W) model(sdm) fe wtype(time)
estimates store time
xsmle y x1 x2, wmat(W) model(sdm) fe wtype(both)
estimates store both

lrtest ind time
lrtest ind both
lrtest time both

I suppose you could also compare the AIC/BIC across models.

estat ic

if available.