PROC MIXED testing asymptotic vs inverted u-shaped

[emaneman]

Excellent!  I just tried it out. It works perfectly and as you say the results are very similar to that I get with the spline models. It is quite clear that at high levels of PROGRESS the two slopes differ, and now I can back up the claim with all of this. Thank you. 

[emaneman]

Dear Dave,

I played around with the PROC GLIMMIX / spline solution, on a different DV, and I am not sure whether I am specifying the estimate correctly. As you can see from the graphs of the effectplot below (I get that the GENMOD syntax that I include),  now the critical areas are the slopes for segment between -0.5 and +0.5 (negative for IV=2; quite flat for IV=1) and between 1 and 1.5 (positive for IV=2;quite flat for IV=1). Could you help get the correct estimates GLIMMIX for these two areas, separately for the two levels of IV?

In case you are curious, these data are from two types of novels,. IV=2 are popular novels - some would call them lowbrow fiction. The slope is exactly as it's been predicted, in terms of how positive emotions vary over the course of the novel (each novel is cut into 10 segments, of equal length, from the beginning to the end of the novel; variable PROGRESS). Lot's of good things, then a crisis, then things are good again! IV=2 are novels that are considered literary fiction - much less positive emotion overall, and not much of an emotional rollercoaster at all.

 

  proc genmod data=b ;
class id iv;
model emo_pos= iv|progress|progress|progress;
repeated subject=id;
effectplot slicefit(x=progress sliceby=iv);