Multilevel data and mediation/path model
Message/Author
 Karin Truijen posted on Tuesday, March 30, 2010 - 2:44 am
I have multilevel data (two nested levels: individual and team) and an upper level and lower level mediation. I will hypothesize that an individual-level predictor (X) affects a group level mediator (M1), and affects another individual level mediator (M2), which, in turn affects a group level outcome (Y). Is there a way to test this model?

X (individual) - M1 (group level) - M2 (individual) - Y (group level)
 Linda K. Muthen posted on Tuesday, March 30, 2010 - 8:31 am
You would specify the model you show as follows where x and m2 are not put on the WITHIN list and m1 and y are put on the BETWEEN list:

%WITHIN%
%BETWEEN%
m1 ON x;
m2 ON m1;
y ON m2;
 Franziska Reinhardt posted on Wednesday, January 15, 2020 - 12:04 pm
Hello,
I find this forum extremely helpful and have already learned a lot. But now I have a problem where I can't find a solution:
I am having a multilevel 2-1-1 mediation with 2 predictors (x1, x2) and a dichotomous mediator (m) and a number of control variables (cv).
The model works very well, but if I would specify m as Categorical, the model does not work. What can I do?

Within= cv1 cv2 cv3 cv4;
between= X1Index AYoS X2index income;
CLUSTER=origin;
Define: Center cv1 X1Index X2index AYoS (Grandmean);
Analysis: TYPE= TWOLEVEL;
Estimator= Bayes;
fbiteration=10000;
Processors=2;
MODEL:
%Within%
y ON cv1 cv2 cv3 cv4;
y ON M; !(bw)
y M;
%Between%
M y X2 X1;
X1 BY X1Index AYoS;
X2 BY income X2index;
M ON X1 (a1);
M ON X2 (a2);
y ON M(b);
y ON X1 (c1);
y ON IN (c2);
X1 with X2;
MODEL CONSTRAINT:
New(indX1 indX2);
indX1=a1*b;
indX2=a2*b;
 Bengt O. Muthen posted on Wednesday, January 15, 2020 - 1:04 pm
We need to see your full output and, preferably, the data as well - send to Support along with your license number.
 Franziska Reinhardt posted on Saturday, January 18, 2020 - 3:02 am
Dear Mr Muthen, thank you for your answer, I have contacted my university because I use a general university access and they do not have a license number that I can use as an individual. Therefore I can not use the support. May I therefore ask a short question of understanding.
Do I have to specify a Binary Mediator for a multilevel analysis? Which estimator do I use for this?
 Bengt O. Muthen posted on Monday, January 20, 2020 - 8:07 am
Q1: Yes.

Q2: You can use ML (or MLR) or Bayes or WLSMV.
 Franziska Reinhardt posted on Tuesday, January 21, 2020 - 12:12 am
Thank you very much, I appreciate your response.
 Mplusquestions posted on Wednesday, February 05, 2020 - 10:56 am
Hi Mplus Team,

I have a three-level data-set (individual, team and organization) and I was hoping to test a moderated mediation in which a team level variable (coh) mediates a three-way interaction between two team (Tid and Tdiv) and one organization-level variables (cOid) on an individual outcome (OCB). X Y Z are control variables on the individual level.

1. Is this possible with MPLUS; Type = Threelevel random? Estimator = MLR.

2. Disregarding the model constraints would the below code fit with the description above?

MODEL:
%WITHIN%
OCB ON X Y Z

%BETWEEN Tcode%
OCB ON coh (b1);
OCB ON Tid (cdash);
a | coh ON Tid (a1);
b | coh ON Tdiv (c1);
c | coh ON TdivxTid (c2);

%BETWEEN Ocode%
coh ON cOid (c6);
a ON cOid (c3);
b ON cOid (c4);
c ON cOid (c5);

 Bengt O. Muthen posted on Thursday, February 06, 2020 - 12:35 pm
Yes on both.
 Mplusquestions posted on Thursday, February 06, 2020 - 1:26 pm
Great, thank you for the reply!

Does this seem accurate in terms of constraints to calculate the indirect effect?

MODEL CONSTRAINT:
NEW (Low_cOid HI_cOid Low_tdiv HI_tdiv
IND_LOWcOidLOWdiv IND_LOWcoidHIdiv IND_HIcOidLOWdiv IND_HIcOidHIdiv);

Low_cOid=-1
HI_cOid=1;
Low_tdiv=-1;
HI_tdiv=1;

IND_LOWcOidLOWdiv = a1*b1 + b1*c1*c2*c3*c4*c5*LOW_cOid*LOW_tdiv;
IND_LOWcOidHIdiv = a1*b1 + b1*c1*c2*c3*c4*c5*LOW_cOid*HI_tdiv;
IND_HIcOidLOWdiv = a1*b1 + b1*c1*c2*c3*c4*c5*HI_cOid*LOW_tdiv;
IND_HIcOidHIdiv = a1*b1 + b1*c1*c2*c3*c4*c5*HI_cOid*HI_tdiv;
 Bengt O. Muthen posted on Thursday, February 06, 2020 - 3:43 pm
The top 2 levels are similar to twolevel mediation with random slopes so you can check how that is done in our Short Course Topic 7, slides 81 and on.
Post:
This is a private posting area. Only registered users and moderators may post messages here.