I am using lagsarlm in spdep package in r to estimate a spatial Durbin (mixed) model by
m1 <- lagsarlm(f, data = d, wlist, type = "mixed")
where f is my designed model of dependent variable and independent variables, d is a set of data points loaded from csv file, and wlist is the spatial weight list. The regression of SDM works fine for me. And I used predict function:
pred <- predict(m1,newdata = d, listw = wlist)
with the original data and spatial weight list to estimate the dependent variable. The estimation is different from the fitted.value in model object m1.
From my understanding, the fitted.value in an Sarlm object is estimated with assumption that dependent variables are known. I wonder how does the predict function estimate a spatial mixed model with lag on both dependent and independent variables.
I checked the documents of sarlm and it provides the information of predicting with spatial lag model. I could not find a prediction formula for the spatial mixed model.
