Multiple imputation
Creates several completed data sets by fully conditional specification and pools a t test or regression across them with Rubin's rules.
Menu: Analyze → Multiple Imputation → Impute Missing Data Values…
When should I use it?
- More than a few percent of cases incomplete and data plausibly missing at random.
What are the assumptions?
- Missing at random given the variables in the imputation model.
- Include the outcome and predictors of the analysis in the model.
How do I run it in StatLab?
- Choose the variables, the number of imputations (5–20; more when much is missing) and a seed.
- Choose the pooled analysis and its variables.
- Put categorical variables (e.g. gender, education) in 'Categorical variables'; they are imputed with logistic or multinomial models.
- Tick 'Create a stacked data set' to get all imputations in one file with an Imputation_ variable (0 = original).
How do I read the output?
| Output | Meaning |
|---|---|
| Imputed Variables: Means | Original vs each imputation. |
| Pooled row | Estimate, SE and df by Rubin's rules. |
| Fraction Missing Info. | Share of the variance due to missing data. |
| Categorical Variables: Percentages | Share of each category in the observed data and in each imputation. |
Worked example
On the practice survey dataset (File → Sample dataset), this command — which the dialog's Paste button also produces — runs the procedure:
PROC miImpute /vars=(score age) /m=5 /iter=10 /seed=20261029 /method=regression /analysis=none /dep='' /group='' /preds=(pre post) /br=NO /cats=(group) /stack=NO.
Output tables: Imputation Specifications · Imputation Results · Imputed Variables: Means.
Computed by StatLab's validated procedure code when this page was built. The example only illustrates the steps; choose variables that fit your own research question.
What are common mistakes?
- Analysing only one imputed data set.
- Leaving the outcome out of the imputation model.