I have a parenting questionnaire with three sub-scales. Within the sub-scales I have item level missing data. It is less than 5% of the data, but I would like to use multiple imputation to correct the missing data. I am using the MICE package in R, and the mice() function. I wanted to ask what method of imputation should I use? I know the package has built in imputation methods (e.g., pmm or norm), but I wasn't sure which imputation method was appropriate for item level missingness? Or how I should choose between the multiple options? I know that I do not what to use any of the two-level imputation methods, as they do not apply. But beyond that I am at a loss for how to choose between the multiple imputation methods that the package offers. For any of you who have experience with R or the MICE package, any and all feedback would be greatly appreciated! Thank you so much in advance!

  • $\begingroup$ What do you mean "item level missingness"? The variable is missing for everyone? $\endgroup$ – AdamO Dec 16 '16 at 22:41
  • $\begingroup$ Please put some links in to various terms, the terms you are using are so specific to R that it limits the potential readership. $\endgroup$ – Carl Dec 16 '16 at 23:39

The two methods you mention (predictive mean matching, and normal) are suitable for imputing continuous variables. Your reference to items in a sub-scale leads me to believe that these have only two or a small number of categorical values. In that case you would choose a different method logistic regression for binary and proportional odds for ordered categorical. My feeling is that with less than 5% missing the results are unlikely to be affected much but it is worth doing as a sensitivity analysis.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.