That's a complex model. A key issue is the moderation of the b path which needs special care already in the single-level case. See the Preacher, Rucker, Hayes (1997) MBR article, Figure 2, Model 3, where you add WITH statement between M and W, MW to avoid poor fit. So that needs to be added to your input on Within and something similar to the Between level as well. I asked Kris Preacher if he had written specifically about the multilevel version of this, and he said no, but offered this modified input (TYPE = TWOLEVEL suffices):
MODEL: %WITHIN% m z y mz; y ON m(b1) mz(b2) z; m WITH z mz; z WITH mz;
%BETWEEN% x w m z y xw mz; m ON x(a1) xw(a2) w; y ON m mz z x xw w; x WITH w xw z mz; w WITH xw z mz; xw WITH z mz; z WITH mz; m WITH z mz;
Sven Lohrey posted on Wednesday, December 10, 2014 - 2:37 am
Dear Dr. Muthen, thank you (and Dr. Preacher) very much! This is very helpful!
Sven Lohrey posted on Monday, February 16, 2015 - 7:50 am
Dear Dr Muthen,
I have some question regarding the WITH statements in the BETWEEN part of the above syntax:
1) Why is "m" only included in the last line with z and mz, but not in any of the previous WITH statements?
2) If I were to expand the model to test effects on an additional group-level outcome (i.e. 2-2-2 in addition to the 2-1-1 model discussed above) and want to add a group-level moderator "v" to the path from group-level mediator "m2" to group-level outcome, how would I have to change the WITH statement? My thoughts are:
x WITH w xw z mz m2 v m2v; w WITH xw z mz m2 v m2v; xw WITH z mz m2 v m2v; z WITH mz m2 v m2v; mz WITH m2 v m2v; m2 WITH v m2v; v WITH m2v; m WITH z mz;
I am trying to run a 2-1-1 (x,m,y) moderated mediation model. This model includes a cross-level interaction (m and w) on path b. However, I am having difficulty declaring the whole model since I have to declare the m variable as a WITHIN variable for the cross-level interaction and group centering, which will then not allow me to use it for the BETWEEN part of the mediation analysis. Is a 2-1-1 moderated-mediation model including a cross-level interaction possible in MPLUS? Or would you suggest a 2-step process for this? I wanted to ask you this first initial question before posting my code. I appreciate your time.
Thank you for your response! I have written the model and it runs, but I wanted to see if I was not too far off. It is a 2-1-1 model with a a level 2 variable w moderating both paths a and b (b as the cross-level int). I created mm as a between-level variable with the cluster mean of m as suggested. Any help would be appreciated.
USEVARIABLES ARE x m w y mm xw; CLUSTER = SupID; WITHIN = m; BETWEEN = x w mm xw; MISSING ARE .;
DEFINE: CENTER x (GROUPMEAN); xw = x*w; ANALYSIS: TYPE = TWOLEVEL RANDOM; MODEL: %WITHIN% bij|y ON x; ! [T1LMX@0]; I was going to add this since m was group-mean centered and its mean should be fix to zero to avoid any possible unconflated results but it but gave me error. Not sure if this is even an issue
%BETWEEN% mm ON x(a1) w xw(a3) ; y ON mm(b1) w x ; bij ON w(b3); y WITH bij;
Thank you again Dr. Muthen. Because of the length and size, I cannot seem to paste a whole run in this forum. If there is another way for me to send it let me know. If not, I understand and still thank you for your time.
Regarding multilevel modeling in Mplus, I am trying to understand why we are finding a discrepancy in the coefficients when introducing variables at the between level. We have two variables one at the between level (level 2) and one at the within level (level 1), and when we run a simple regression with the two variables, whether in SPSS or Mplus, we get a significant coefficient of around .33. However, when we label each variable as a between variable, and then label the regression as a between regression, the coefficient drops to about .17. I understand the power level changes at the second level due to the smaller N (level 2 is supervisors, level 1 is employees). I don't understand why the coefficient would decrease by 50%. We ran the regression by introducing a variable that is the group means of the level 1 variable, and ran everything at level 2 in Mplus. To my understanding, this is the recommended process when dealing with a 2-1-1 model, which is what we have. At this point, we are only looking at the first half of the model, say between X and M. Any thoughts on this?
Sorry about the misunderstanding...it is my understanding that introducing a between level (level 2) variable that contains only group means is the appropriate method when dealing with a 2-1-1 mediation model and testing indirect effects (Zhang, Zyphur, & Preacher, 2009). My understanding was that introducing the group mean variable for the level 1 variable creates a between level variable since the only variation is at the between level. Then run the between level regression and labeling the between level X variable and the between level group mean variable for M and have only a between level regression to represent the relationship between X and M even though M is actually a lower level variable.
I want to run a 2-1-1 with a level 2 moderator of the a path. I am receiving an error “TYPE=TWOLEVEL requires specification for the CLUSTER option”. I am assuming there is a small error I am not seeing, but I haven’t been able to figure it out. Thank you as always for the help! Robert USEVARIABLES ARE ID X Y M XW W; CLUSTER = ID; WITHIN=Y M; BETWEEN = X W XW; ANALYSIS: TYPE = TWOLEVEL RANDOM; MODEL: %WITHIN% Y M; Y ON M(b1);
%between% X W XW M Y; M ON X(a1) XW(a2) W; Y ON M X XW W;
MODEL CONSTRAINT: NEW (IND1 WMODVAL); WMODVAL=3.5; !+1SD IND1=(a1+a2*WMODVAL)*b1;
We need to see the full output to say - send to Support along with your license number.
Mercy Oyet posted on Sunday, November 12, 2017 - 2:15 am
Hello Dr. Muthen,
I want to test a 1-1-1 model with a level-2 variable w moderating path a (this is a repeated measures analyses). I have three level 2 co-variates (l, m, n) and one level 1 co-variate (p). Could you have a look at my syntax and let me know if this looks okay? Thank you!
I have not used Mplus before but recently purchased the license. I would really appreciate your advise on this.
For a MLM 2-1-1 mediation model, is it possible to include a level-1 demographic moderator like a race dummy variable on the 2-1 (a path) portion of the model?
I see that there is code by Preacher et al for 2-1 moderation, but I'm having a hard time applying the logic to the above example since I can't see how there would be a between part to individual-level (level-1) race. Or am I missing something?