I have a dataset where I am trying to use multiple imputation with the packages mice, miceadds and micemd for a categorical/factor variable in a multilevel setting. I am able to use the method 2l.2stage.pois for a continuous variable, which works quite well. The imputation for the categorical variable also works with polyreg, but this does not make use of the multilevel data.
Here is replicable code that illustrates my problem.
dat <- data.frame(x=rep(c("A", "B"), 50),
y=rep(c(10:14), 20),
r=rep(c(1:50), 2),
z=rep(c(1:4), 25))
dat[40:44, c("x")] <- NA
dat[60:64, c("y")] <- NA
dat[80:84, c("r")] <- NA
predm <- mice::make.predictorMatrix(data=dat)
predm[, c("z")] <- -2
predm[, c("r")] <- 3
predm[row(predm) == col(predm)] <- 0
meth <- c("x"="polyreg", "y"="2l.2stage.pois", "r"="", "z"="")
imputed = mice(dat, method=meth, predictorMatrix=predm, m=1, maxit=1)
imp <- complete(imputed)
table(imp$x, dat$x, useNA = "always")
table(imp$y, dat$y, useNA = "always")
In essence, I am trying to substitute polyreg with something that should probably start with 2l.. I have tried 2l.2stage.bin, but that is for binary variables. Thanks!
