The survey dataset I'm currently working on has "branching", where for example respondents that answer that they did not receive subsidies for item A, will not answer questions about item A, and those who answer that they did not receive subsidies for item B, will not answer for item B. This leaves intentionally systematic missing data.
As with all surveys, there are also other missing data, most likely missing at random. The plan is to conduct multiple imputations on the "real" missing data, but I can not get the analysis to "ignore" the intentionally missing systematic data (Working in SPSS) in a logical way.
As I understand to conduct the imputations on all missing data and then delete the intentionally missing data is not an option. I can conduct several multiple imputations and later combine the datasets, but this too has limitations, as I will have to group the imputations by respondents (ie. only people that received subsidies for only A, then do the same analysis for only B, then finally A and B), significantly lowering N. Or I could group them by variable, (impute only data in section A, then section B) but this ignores all connections between A and B, which likely exists, resulting in worse predictions.
Has anyone encountered similar issues, and if so, what was your solution?
Thank you for your answer Nicholas J. Parr
. I plan to use the data for two purposes, where the first is displaying descriptives, as we are comparing them to another study, and later do logistic regression.I considered doing FIML, but to the best of my understanding it is accepted that MI generally is a better alternative for missing data, as later analysis takes into account the uncertainty of the imputed datasets.
I also found the answer to my own question, and the answer is "use Stata", as you can there code soft and hard missing data, where the MI will ignore the hard missing data.
https://onlinelibrary.wiley.com/doi/book/10.1002/9780470316696
https://methods.sagepub.com/book/missing-data
https://www.guilford.com/books/Applied-Missing-Data-Analysis/Craig-Enders/9781606236390
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology