IBM SPSS Modeler Cookbook

Using random imputation to match a variable's distribution

This recipe imputes missing values with actual values (selected at random) from the variable with missing values needing to be imputed. It is valuable when one does not want to impute with a constant but the variable has a distribution that isn't replicated well by a normal or uniform random imputation method.

In this recipe we will impute values for a missing or blank variable with a random value from the variable's own known values. This random imputation will therefore match the actual distribution of the variable itself.

Getting ready

This recipe uses the following files:

• Datafile: cup98lrn_variable cleaning random impute recipe.sav
• Stream file: Recipe – impute missing with actual values.str ...