Alternative for mi impute intreg Is there a free alternative available for the Stata procedure mi impute intreg (Impute using interval regression)? For example as an R package. I have not found any yet.
 A: If you want to impute the missing data from an interval censored regression model (either Cox-PH or proportional odds, no aft at this time), you can use the R package icenReg (admission of bias: this is the author). This allows imputation for both fully-parametric and semi-parametric models. 
In contrast to my standard bias, I suggest using the fully-parametric models for imputation (after model checking). There a two reasons for this: one is that the semi-parametric model, like the non-parametric model (think KM curves if you're not familiar), does not assign positive density to all points on $[0, \infty)$, and thus you may want to impute into an interval that has no density. icenReg tries to do the best it can there, imputing at the edge of the given interval. It's worth noting that this will not happen if you're imputing on the data for which the model was fit, but could happen if you're imputing for new data. 
Secondly, while fully parametric models have your standard asymptotic normality for the baseline parameters, this is not the case for semi-parametric model. In fact, there's currently no characterized limiting distribution for the baseline. Because of this, resampling of the baseline parameters for the non-parametric model is ignored and only the regression parameters are resampled during imputation. For the fully parametric model, both the baseline and regression parameters are resampled. 
See icenReg::imputeCens for more details.
A: You might be able to roll your own with the R package mice (using mice.impute.passive, etc) and the package intReg. Stata's got some fairly unique features, though.
A: I am using gretl for interval regression now. The gretl can be run from the command line, so it is easy to integrate it in R processing by writing data out, generating the gretl script by cat(), running gretlcli by shell() and reading the gretl result in by read.table().
It is still a long way to go till the implementation of multiple imputation though.
Here is an example of gretl integration in R:
# write out for gretl ####
setwd(dir.data)
write.csv(dat.mod, file = "dat.mod.csv", row.names = F)

# Script
# vrs2 is the list of variable names to be used in a model
cat('# EHSIS 2012\n',
    '# Martins Liberts\n',
    '# Generated file from R\n',
    '\n',
    'open dat.mod.csv\n',
    '\n',
    'intreg lb ub const ', paste(vrs2, collapse = " "), '\n',
    '\n',
    'yhat1 = $yhat\n',
    'yres1 = $uhat\n',
    '\n',
    'matrix b = $coeff\n',
    'err = mwrite(b, "coeff.txt")\n',
    '\n',
    'store "ymod.txt" yhat1 yres1\n',
    file = "rungretl.inp", sep = "")

# run gretl ####
shell("gretlcli -b rungretl.inp > outgretl.txt",
      shell = "powershell", flag = NULL)

# Read gretl results ####
setwd(dir.data)
# Regression coefficients
coeff <- read.table("coeff.txt", skip = 1)[[1]]
# yhat and uhat
ymod <- data.table(read.table("ymod.txt", header = T))

A: mi is the official command in Stata but there is also the user written command mim in Stata. For the difference between these two, please see here
A: Check out the Amelia package for R.
Also has a standalone version available for those who want to do MI but can't/won't do R.
