I'm dealing with a mediation model and am using the PROCESS module in SPSS. Due to SPSS and PROCESS being limited in the imputation methods - being unable to handle multiple imputation - the other obvious choice is using EM.
What I noticed however is that for participants that didn't respond at all to the questionnaires, EM estimated 16 participants as having exactly the same scores on each item when trying to impute.
So that leads me to the question of when you can use EM to replace missing data, and when there is too much missingness in a particular case that produces the estimates to all be insufficiently variable (or in my case, exactly the same for 16 cases). The literature seems to mainly talk about overall missingness on a variable, such as 5% missingness on a particular variable, but I can't seem to find things that deal with it on a participant by participant basis. I.e. if a participant completed 5/100 questions, can you impute their missing responses?
Really struggling to find a reference to actually support any interpretation on how to handle this.
Hi,
You can refer to this particular book which deals with missing values in detail especially SPSS and SAS.
You can try it on researchgate search.
MISSING VALUES ANALYSIS & DATA IMPUTATION
Web: www.statisticalassociates.com
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS