I have a longitudinal data at three time points: baseline, y1, y2. Baseline responses were collected right before treatment/control but after recruitment. Subjects were included if the baseline observation was above certain value. y1 were collected right after treatment. y2 were collected one year after treatment completed.
I ran two models. One using baseline observations as response (M1). One using baseline observations as predictor (M2). The treatment effects in two models are similar, I think:
at y1: -1.09 vs -1.00
at y2: -0.90 vs -0.81
But the magnitude of p-values are 10x times differed.
at y1: 0.0004 vs 0.006
at y2: 0.004 vs 0.027
I read this post. It seems like either model is fine but is there a way to choose? Baseline adjustment in mixed models
One thing they mentioned is the normality assumption between baseline and t1 observations. Responses at both time points are mildly skewed to the right. Spearman correlation is 0.8.
Below are the plots of some predictions. I guess the difference between them is that I cannot set the baseline response in M1. Does that matter?
