We would like to test a moderated mediation model. We have a level 1 X variable, and level 1 mediator (M), and a level 1 Y variable. Our moderator is a level 2 variable. I am aware of the following method to test the 1-1-1 indirect effects with level 2 moderators:
Bauer, D. J., Preacher, K. J., & Gil, K. M. (2006). Conceptualizing and testing random indirect effects and moderated mediation in multilevel models: New procedures and recommendations. Psychological Methods, 11, 142-163.
However, we were wondering if it is a straightforward extension to also include a test of the 2-1-1 indirect effect? We want to now whether our level 2 variable also influences Y through M(under certain conditions of X).
Is it possible to test this model with Mplus, and how?
This is for a 2-1-1 model in which the 'b' path of the mediation model is random, and moderated by a Level-2 variable W. I tried it out on real data and it seems to work:
TITLE: 2-1-1 moderated mediation model DATA: FILE IS bpg_example_data_modmed.dat; VARIABLE: NAMES ARE id x m y xm mm ym w; USEVARIABLES ARE id m y xm w; CLUSTER IS id; BETWEEN ARE xm w; ANALYSIS: TYPE = TWOLEVEL RANDOM; MODEL:
!2-1-1 moderated mediation with random b slope !in which b is moderated by W. %WITHIN% m y; sb | y ON m; %BETWEEN% xm m y sb w; [sb](b1); sb WITH xm m y; w WITH xm m; m ON xm(a); y ON m(b1); y ON xm w; sb ON w(b3); MODEL CONSTRAINT: NEW(ind1 ind2 ind3); ind1=a*(b1+b3*.1); ! indirect effect conditional on W = .1 ind2=a*(b1+b3*.5); ! indirect effect conditional on W = .5 ind3=a*(b1+b3*.9); ! indirect effect conditional on W = .9 OUTPUT: CINTERVAL;
Additional Mplus code for similar models, including the 1-1-1 model from Bauer et al. (2006) can be found here: