As my understanding:
The imputation in MI procedure includes many methods. The bottom line is to impute missing through introducing randomness (with a set of plausible values that represent the uncertainty about the right value to impute). That is, instead of using one dataset, you create a number of datasets (nimpute=). NIMPUTE= default as 5. Of course, you need to adjust this randomness by using MIANALYZE procedure to summarize your analysis results.
In fact, if the results from the incomplete dataset (original) and the imputed dataset differ a lot, it is a big warning to review your imputation methods.
Jingju
|