Message/Author 


Hi! I need some additional help following up on a previous post of mine. I basically have 2 treatments. treatment a and treatment b. I am exploring moderators of treatment outcome. I previously used POST TX SCORE ON TYPE OF TREATMENT POST TX SCORE ON PRE TX SCORE and the MODERATOR (e.g. SEX as the GROUPING VARIABLE). I THEN TRIED PRODUCT TERMS BC I WAS TOLD THAT MULTIPLE GROUP REQUIRES LARGE SAMPLES. Then my professor advised another student to create a difference score = prepost and regress type of treatment into it if she wanted to get the interaction with treatment. I am back at square 1 thinking that it is this difference score that I need to regress onto the product term between type of tx and the moderator. But I am not sure which way is the correct way. I am esentially looking at a 3 way interaction. I even created files and looked at the means separately for each groups and the means I get by doing this in each group are consistent with the difference score approach and the findings make theoretical sense. But when I use post on type of treatment controlling for pre and using multiple group it is not the same. Any help as to which is the correct way will be appreciate it. Thank you!! 


I would not use a difference score. I would use the following model: post ON tx pre inter; where the pretest is centered and inter is created in DEFINE as follows: DEFINE: pre = pre  a; where a is the mean of pre inter = pre*tx; 

JHS posted on Tuesday, March 01, 2016  9:29 am



Hello, I have a related question regarding moderation. I have 4 continuous variables: a personality variable (X), baseline sad affect before a lab task (M1), sad affect after the lab task (M2), and a symptom variable (Y). My hypothesis is that magnitude of change in affect will moderate the relationship between the personality variable and the symptom variable (i.e. X will predict Y in those participants who showed a large increase in sad affect after the task and X will not predict Y in those participants who showed no change in affect between M1 and M2.) I have read about the problems with using difference scores and would like to avoid using those. Would this be the appropriate code to test my hypothesis? MODEL: Y BY y1* y2 y3; X BY x1* x2 x3; M1 BY m11* m12; M2 BY m21* m22; Y@1 X@1 M1@1 M2@1; XM2 X XWITH M2; M2 ON M1; Y ON X M2 XM2; 


If you are not an advanced Mplus user I would take the difference score route. Otherwise try to create a new latent variable (f) that is the M2M1 difference: f on m2@1 m1@1; and is uncorrelated with every other factor. Then create the interaction int  x XWITH f; But I haven't tried it so you are on your own... 

Back to top 