Hi All,

I'm performing a data analysis on a dataset I didn't do the collection for, and I am contending with some odd decisions made by the people who collected the data.

I'm using multilevel regression to test associations with children's health behaviours over 5 data collection periods. For the first 3 waves, the dependent variable was answered by the parent, and for the last 2 waves it was answered by the child. Additionally, the tool used to measure the DV changed as well. The two methods are very similar, but there is no getting around the fact that they are different.

My initial thoughts were that I could add a dummy variable to represent the measure type, and include it in the regression. However, since "controlling" relies on there being some data to work with, and ALL of the later to waves are measured differently, this seems like it would be wrong, correct?

So, is there any method which allows me to harmonise this measurement difference? At the moment, my strategy is to model each measurement type separately (waves 1-3, waves 4-5), but these seems to almost defeat the purpose of having longitudinal data.

Many thanks for any help.