Say you have a longitudinal data set with 5 measurement points, which begins at 200-300 participants and declines to about 50. There is an intervention between time-point 2 and 3. There's 3 continuous IVs and 1 continuous DV measured at all timepoints. One group. The key interest is the predictive value of the IVs as far into the future as possible, not the effect of the intervention. What alternative types of modeling are applicable? Do you do a bunch of regressions of the IVs on the DV between all interesting 2 timepoint-spans? Or is there some more effective type of modeling which extracts the same information in a more compact manner using less analyses? Is there any idea to do anything but a linear model given the low number of time-points?
I am thankful for any answers.