How should data be treated in multilevel modeling when they are not available because they are contingent upon how a previous question is answered?
For example, if the participant answers "no" on the first item during a given measurement occasion, they will not be asked to answer questions 2-5. In the output for this measurement occasion, items 2-5 will appear to be missing (because the participant was not prompted to answer these items), but they're not actually "missing" in the way we think of missing data.
Thanks!
More detail about my study:
Participants will be randomized to one of three treatment conditions and will be asked to respond to 4 (signal contingent) EMA surveys daily for one week (28 surveys total) following the intervention. Each of these surveys consist of 5 items. Responses to these items will serve as dependent variables in my study.
The first item will ask whether or not the participant has encountered a tempting food item in the past 30 minutes. If participants answer "yes" to this item, they will be asked (2) how strong the temptation is (1-7 scale), (3) how strongly eating the food would conflict with their long-term goals (1-4 scale), (4) if they attempted to resist eating the food (Y/N), and (5) whether or not they actually ate the food (Y/N).
However, if they answer "no" to the 1st item, they will not be asked questions 2-5 during that particular measurement occasion (since the relevant event--encountering a tempting food item--did not occur in the time-frame of that occasion). This is where the issue of missing data (due to intentionally skipped items) comes in. I'm not sure how to deal with these, and there may be quite a bit to deal with if my participants aren't confronted with a ton of tempting foods throughout the day.
In my multilevel models, measurement occasions (level 1) are nested within participants (level 2). I'm also including treatment condition (condition 1, 2, or 3) as a level 2 predictor (in case this is relevant).
Also, I'm planning to use SPSS to conduct my analyses. I'm using multilevel linear modeling for #2 & 3 and multilevel logistic regression for #4 & 5.
Thanks again!