Hi All,

I'm using regression analysis (multilevel linear and logistic) on a measure of children's health behaviours over 5 time-points of data. I didn't collect the data, so I'm having to contend with rather odd decisions made by the people who did collected the data.

The main problem is that the DV was measured from a parent at waves 1-3, but then collected directly from the child at waves 4-5. Additionally, the actual tool used changed when the changed reporters, and while the tools are quite similar, there is no getting around the fact that they are different.

Initially, I just wanted to add a dummy variable for the measurement type, but since ALL of the data at waves 4-5 is different, it seems like this would be inappropriate.

I've tried running it as two models (one for each measurement type), but since this limits it to 2 waves at the later stage (which is the time period I'd expect to see the biggest change) it almost seems like a waste of longitudinal data.

Does anyone have any experience trying to harmonise different measurement strategies across longitudinal data.

Many thanks in advance.