As long as you’re using a multilevel model I don’t see why having uneven observations would be a problem? I’m assuming plenty of participants have <4 per day, no? Unless you unlocked the secretly to perfect compliance! :)
Comments
Log in with your Bluesky account to leave a comment
As Whitney says, the different numbers of obs per participant won't be an issue. I guess I'm wondering 1) how could participants complete more than 4 questionnaires (unless including event-contingent?) and 2) whether there is something different about participants who completed >4?
Ah, that’s very helpful context. Yeah, agree with Olivia. I’d just add though that I’d err on the side of keeping responses if there is no good way to evaluate validity of responses. There should be really strong justification to throw out data. Why assume those at the end are less valid?
And I’d say whatever validity filter you decide to apply, if you do, should extend to all observations, not just the “extra” ones. If you have timing of responses recorded that could be a more proximal and relevant way to eval validity…
I’ll also add that one part of the survey is about social interactions so extra records aren’t a problem. But I want a clean dataset for all future analyses that may use the 4x/day design to do some within day slopes.
And to cut records clearly just shoved in at the end to get study credit.
Sounds like maybe keeping responses closest to prompt could work & also thinking about some careless responding analyses. Probably worth considering some sensitivity analyses too, but it sounds like careless responding may be more of an issue, e.g., cramming in responses to hit credit threshold.
Comments
It was 4 emails sent at 11am,1pm,3pm,5pm.
My mistake was a reusable link that they can use as much as they chose. So for some it was clear they just wanted to get to 28 to be done.
And to cut records clearly just shoved in at the end to get study credit.