I have conducted an online survey with 70 questions from 8 different scales; I have around 5% missing data. It looks like some of my data drops off for the last block of questions as the missing data increases towards the end of my survey, so the last 3 scales have the majority of missing values.
I am looking to do a t-test and linear regression analysis, and I am wondering the best approach to take for the missing data. I have read that MI can be used, but I am wondering if this is acceptable when I have entire scales that have 0% data completed and whether a pairwise would be better? Any recommendations on the best approach?
The right thing to do would be to simply remove the responses with the missing values. Every other effort is just an educated guess that could be wrong.
Hi,
I don't know the capabilities of SPSS when applying multiple imputation in giving you diagostics whether the imputation process makes sense, but I guess this will not be much of a problem. The main issue is whether there is enough information in the overall dataset that helps to impute the missing values.
Another option would be to switch to R (which you should do anyway) and use the mice package by Stef van Buuren. Here's his online book: https://stefvanbuuren.name/fimd/
And finally, you could do the regression in R's lavaan/SEM package which has the full information maximum likelihood estimator.
Throwing away cases with missings is the least optimal choice. And you should learn from this. If people skip that later part of the questionnaire, than this problematic and you could have known this after some form of pretesting. I would recommend checking the "social validity" of a questionnaire prior to a study.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers