MI Impute and Tobit Analysis

[John Willett]

We need to use multiple imputation to resolve a missing-data problem in a covariate (X, say), whose values have a "tobit" distribution That is, about 75% of sampled cases have X-values that are positive (X>0) and follow a normal distribution. The remaining 25% of cases (where X<0) -- which would typically fill out the lower tail of the normal distribution -- are censored at X=0, causing the lower tail of the sample distribution of X to be truncated, with a big spike at X=0. To conduct MI in this sample, it seems that we need to use tobit regression in the imputation step. However, we can find no mi impute tobit command available in STATA. We seek general guidance and, in particular, wonder if anyone has already created a user-defined routine ("mi impute tobit") that performs a suitable imputation step?