09-19-2014 03:46 PM
I have a dataset with missingness of two different types.
The data consists of Likert-type items, so an ordinal scale (excellent, good, satisfactory, marginal, fail), however there is also an option "not-applicable". Items coded as not-applicable are the first type of missingness.
There are also individual items that are just missing (blank) - so this is the second type of missingness.
I have identified these in the data set using the special missing values of .N and .M.
Each of these types of missingness has a different cause and I would like to be able to model them separately. I am particularly interested in conducting a sensitivity analysis on the blank items (.M) and was thinking of using PROC MI to do multiple imputation.
Does anyone know if PROC MI can distinguish the two sorts of missing values so that I can control how it handles each of them?