HLM and different measures at different times

[sgcraig]

---

Members don't see this ad.

So I am doing my thesis on Conduct Disorder and Depression from early adolescence to early adulthood (3 time points) and have run into a conceptual to statisical issue I am hoping to get help on. As symptoms change for disorders over time, so do the measures. Is there a way to run a HLM (or MLM or LGC) on different measures of the same construct over time?

 

[Ollie123]

There are MANOVA-like adaptations of HLM, but if I'm understanding correctly that won't help...you have 3 different time points with 3 different "measures" of the constructs?

This will likely be a mess however it is done. There are likely ways to get at overall question you want to ask, but I don't think its not going to be anything remotely straightforward. You certain'y can't run a traditional repeated-measures-like model on variables operating on completely different metrics...you can't interpret a "change" over time as actual change given the different measures used. You could still justify co-varying out Time 1 and if you have groups you could then examine differences. I can't currently think of any way you could legitimately look at trajectories or "time" with that sort of design.

 

[JockNerd]

There's some intelligence testing lit on this using instruments that have been updated over time or that change when the participant reaches a certain age (e.g., WAIS, WIAT) that you might check as a guide.

You're going to be screwed by having only three time points though.

You might be able to swing something if your outcome was Diagnosed vs Not Diagnosed (assuming all measures have a cut point for diagnosis), because that reflects a slightly different construct than raw scores on the symptom inventory or whatever and potentially more comparable.

Wait, this is thesis data? Is this archived? You can't possibly be doing a longitudinal study for a thesis yourself....