|
Message/Author |
|
Jamin Day posted on Saturday, March 12, 2016 - 5:30 pm
|
|
|
1. I'm doing mediation in Mplus with continuous mediator and outcome measured at T2 (post-intervention), and want to control for baseline differences (T1) in the model. Does the following code achieve this?: Analysis: Bootstrap = 10000; Estimator = ML; Model: y2 on y1; ! control for baseline differences in outcome m2 on m1; ! and mediator y2 on treat1 (c1) ! dummy-coded categorical treat2 (c2); m2 on treat1 (a1); treat2 (a2); y2 on m2 (b1); ! direct effect Model Indirect: y2 IND treat1; y2 IND treat2; Model Constraint: New (ind1 ind2) ind1 = a1*b1; ind2 = a2*b1; 2. I'm having difficulty testing the equality of the indirect effects shown in Model Constraint above. I tried: Model Test: 0 = ind1 - ind2; But the Wald test won't work with bootstrapping. I also tried Difftest to compare two models, one with the above Model constraint, and one with 0 = ind1-ind2 added, but this fails also (due to the ML estimator) and says I need to use WLSMV instead. However I'm using continuous outcome variables so ML seems the best approach. What would be the recommended way to test for equality of the indirect effects? |
|
|
1. Yes, you are using y1 and m1 as control variables correctly. 2. add to Model Constraint: New(diff); diff = ind1 - ind2; |
|
Jamin Day posted on Wednesday, March 16, 2016 - 7:14 pm
|
|
|
Thank you for the helpful advice. Can I also check whether requesting cint(bcbootstrap) in the output and using model constraints as above will work with multiply imputed data? |
|
|
We don’t allow CINT(boot) or CINT(BCBOOT) with type=imputation or data imputation. It's not clear how that would be done. |
|
Back to top |
|
|