# Multiple Imputation in SPSS for RCT

I have conducted a randomised controlled trial design (2 groups - experimental and control) with data collection at two time points (T1 and T2). I want to use the Multiple Imputation Method in SPSS to address item non-response (i.e. item-level missing data) at T1 and T2. It makes sense to me that MI will insert data points at T1 and T2 for questions that a participant has missed out. But it does not make sense to me that MI will insert data points at T2 for respondents who didn't even open up the second survey - they are systematic missing data at T2. If I allowed MI to do that then it would 'fill in' all the responses for the entire survey for some participants at T2. My datafile is set up in long format with each row as a participant and the T1 variables in columns and then the T2 variables in columns.

My questions: 1. Am I correct to think that MI should not insert data into T2 cells for participants who didn't do any of the questionnaire (i.e. systematic missing)? 2. If yes, how can I stop SPSS addressing those particular respondents? I can see how I can target different variables for MI but not different participants.

My question isn't really about the appropriateness of MI or my missing data. I'm ok up to the point of actually running the MI itself.

I'm using SPSS v224.

Antonia

1). You are correct.

2). Before you run MULTIPLE IMPUTATION, create an identifier variable coded 1 for "keep" and 0 to "drop", and then select out the "drops." I assume that people who did not respond to the survey at T2 are system-missing for all the questions asked at that time.

Here is an example. Ignore the warnings about missing data when you run it.

data list free /v1t1 v2t1 v3t1 v1t2 v2t2 v2t3.
begin data.
9 8 7 1 2 3
6 5 4 . . .
3 2 1 4 5 6
end data.

count t2=v1t2 v2t2 v2t3 (sysmis).
list.
compute keep=1.
if t2=3 keep=0.
select if keep.
list.


If you didn't leave the vars sysmis but used a missing value, change SYSMIS to MISSING in the COUNT command.

You could use TEMPORARY to select them out for the next procedure only:

temporary.
select if keep.
multiple imputation /etc.