C. Lechner posted on Monday, February 24, 2014 - 10:11 am
Dear Drs. Muthén, I use piecewise models to see how X affects Y before and after an event, where s1 represents changes in Y before and s2 after the event: i_y BY y0-y6@1; s1_y BY y0@0 y1@1 y2@2 y3@3 y4@3 y5@3 y6@3; s2_y BY y0@0 y1@0 y2@0 y3@0 y4@1 y5@2 y6@3; [y0-y6@0 i s1 s2]; For X, I estimate some timescores: i_x BX x0-x6@1; s1_x BX x0@0 x1* x2* x3@1 x4@1 x5@1 x6@1; s2_x BX x0@0 x1@0 x2@0 x3@0 x4* x5* x6@1; [x0-x6@0 i s1 s2]; Two questions: 1) Would this model warrant valid inferences concerning how X affects changes in Y before/after the event? 2) Is it legit to model a regression path only from i_x on the i_y, s1_y and s2_y and and let the slopes correlate - or would I have to let them all correlate? Many thanks in advance!
1) Yes, if you regress the Y growth factors on the X growth factors.
2) You can do it either way.
C. Lechner posted on Tuesday, February 25, 2014 - 10:43 am
Please allow me two follow-up questions: 1) So would you say regressing slope factors on each other is ok, too? I know there are several posts on this question in this forum. At any rate, my impression is that this is rarely encountered in the literature - slopes in parallel process growth models are often regressed on intercepts but not on other slopes.
2) If I want to re-center the X growth model so that the intercept represents the level of X at the time of the event (i.e., at the fourth time-point), would the following syntax be correct?
1) You don't have temporal separation between the two slopes, but if your theory says that one predicts the other, it is alright to regress one slope on the other. The reason you see slopes regressed on intercepts is that you have temporal separation - the intercept refers to an earlier time than the slope.