Imputing sequence data: extensions to initial and terminal gaps, stata’s mi

In a previous paper (Halpin, 2012) (downloadable) I have described a strategy for multiple imputation for missing data in sequence data (that is, in categorical time series), where missingness tends to be consecutive and take the form of gaps. In this note I document two improvements made to the initial approach: the use of Stata’s built-in mi multiple-imputation framework for prediction, and the imputation of gaps at the beginning and end of the sequence (which was allowed for, but not implemented, in the original version). I also document the implementation of the algorithm in Stata, in more detail than previously (see section Code).