I'm trying to estimate various spatial models such as spatial autoregressive regression (SAR), Spatial Durbin Model (SDM), and Spatial Error Model (SEM) but have missing data throughout my variables. The mice package in R gives a straightforward solution when none of the variables with a spatial lag have missing observations (such as the SAR with missing observations confined to the independent variables).

However, I can't figure out how to use mice to perform multiple imputation if the missing observations are in the variables with a spatial lag (such as missing observations in the dependent variable of the SAR). I could simply use a regression equation that ignores the spatial aspect of the variable in my imputations, but I would like to use the spatial aspect in my imputation since I will have the spatial lags in my final regression analysis. I've tried using the passive imputation method with something like this for the SAR model where Xm has 3 columns and Ym has one column and W is the weight matrix

DATA = cbind(Ym, W %*% Ym, as.matrix(Xm))
mice(DATA, m=10, 
     method=c("norm", ~W %*% Y, "norm", "norm", "norm"),
     maxit=15, print=FALSE)

but without success.

Any thoughts on how to do this?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.