Multilevel moderation PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Ewelina Lacka posted on Saturday, March 29, 2014 - 8:42 am
Is it possible to test the moderating effect of M (M is a moderators which is subject to influence of N) on the path from X (independent variable) to Y (dependent variable). I think this will be a multilevel moderation on path from X to Y. (X-->Y is moderated by N-->M). Can you please guide me through the process?
 Bengt O. Muthen posted on Sunday, March 30, 2014 - 4:45 pm
I don't understand your model. What role exactly does N play? What does "X-->Y is moderated by N-->M" mean? Why do you say it is multilevel?

Please write out the full model in some way.
 Ewelina Lacka posted on Monday, March 31, 2014 - 6:54 am
HI, I am sorry I did not explain it well. You can see the visual representation of my model following the link http://postimg.org/image/g0ex0qvtr/. The model has 3 IVs (lets say A, B and C) and 1 DV (Z). The paths between IVs and DV is moderated my M and P. Moderator M is further influenced by N (e.g. M can represent 'trust' and N is 'trust determinant').
 Bengt O. Muthen posted on Monday, March 31, 2014 - 6:30 pm
With a continuous moderator such as your M, you create interaction variables (products) in Define. So for instance

am = a*m;

Apart from that you simply say

Z on a-c am .... ;

M on n;
 Ewelina Lacka posted on Friday, April 04, 2014 - 8:42 am
I used the DEFINE command but I have an error message 'undefined variable used in transformation'
 Linda K. Muthen posted on Friday, April 04, 2014 - 8:48 am
Please send the output and your license number to support@statmodel.com.
 Dan R. posted on Thursday, May 28, 2015 - 12:01 pm
For the time being, I've dropped the 1-1-1 mediation component from my model to first look at the level 2 moderation effect. My question is whether it looks like I set up my level 2 moderation effects (Y ON XW & Y ON MZ) appropriately (i.e., if these regressions are significant, can I claim level two moderation is supported). Here is the model with simplified variable names:

Usevariables are
Y X M
W Z
XW MZ;

WITHIN = X M;
BETWEEN = W Z;
CLUSTER = Subject;

DEFINE:
XW=X*W;
MZ=M*Z;

Analysis:
TYPE = TWOLEVEL;
ESTIMATOR=MLR;

Model:
%WITHIN%
Y ON X M;

Y M;

%BETWEEN%
Y W Z XW MZ;

Y ON W;
Y ON XW;
Y ON Z;
Y ON MZ;

Output: sampstat stdyx Tech1 Tech8;
 Bengt O. Muthen posted on Thursday, May 28, 2015 - 2:02 pm
Looks ok so far.
 Dan R. posted on Thursday, May 28, 2015 - 8:39 pm
Thank you, Bengt. And for mediation tests at level 1 and level 2, have I set the following up correctly? Given the interaction effects, my understanding is that the overall chi-square and traditional fit statistics aren't available. So, should I report the BIC and SRMR for comparisons? I actually do get a Chi-square x(12)=6.213, but I'm not sure if this is for the overall multilevel model or just a portion of the model. Any insights you could provide would be greatly appreciated.

Usevariables are
Y X M
W Z
XW MZ;

WITHIN = X M;
BETWEEN = W Z;
CLUSTER = Subject;

DEFINE:
XW=X*W;
MZ=M*Z;

Analysis:
TYPE = TWOLEVEL;
ESTIMATOR=MLR;

Model:
%WITHIN%
Y ON X (cw);
Y ON M (bw);
M ON X (aw);

Y M;

%BETWEEN%
Y W Z XW MZ;

Y ON W;
Y ON XW (cb);
Y ON Z;
Y ON MZ (bb);
MZ ON XW (ab);

MODEL CONSTRAINT:
New(indirw directw totalw indirb directb totalb);

indirw = aw*bw;
directw = cw;
totalw = cw+aw*bw;

indirb = ab*bb;
directb = cb;
totalb = cb+ab*bb;

Output: sampstat stdyx Tech1 Tech8;
 Bengt O. Muthen posted on Friday, May 29, 2015 - 8:20 am
Your Between model looks strange. The "MZ" variable name sounds like an interaction term and it is not declared as Within or Between.
 Dan R. posted on Friday, May 29, 2015 - 12:21 pm
The MZ variable is an interaction - M is measured at level 1, and Z is measured at level 2. If I try to specify MZ on the between = line, I run into the following error:

*** ERROR
One or more between-level variables have variation within a cluster for one or more clusters. Check your data and format statement.

Between Cluster ID with variation in this variable
Variable (only one cluster ID will be listed)

AMG_NEGA 531
DLP_EF 531

Any thoughts?
 Bengt O. Muthen posted on Saturday, May 30, 2015 - 8:33 am
You typically don't have an interaction variable as a dependent variable as you do on Between saying MZ ON XW.

MZ is not a between-level variable given that M is measured on the within level.

I don't see which kind of interaction model you are after for Between, that is, it isn't clear if you want to moderate the x->m, x->y, or m->y paths. Putting aside that you are doing this on Between for 2-level model, maybe you want to study the Preacher et al (2007) MBR article on moderated mediation (single-level) to see how such interaction models are set up.
 dummyvariable123 posted on Tuesday, November 21, 2017 - 1:10 pm
Dear Dr. Muthen,
I want to test whether L2 continuous variable (Z) moderates (inter=Z*X) the effects of L2 continuous variable X on changes in L1 dependent variable Y.

1. Is my syntax correct? If s on inter is significant does this suggest moderation?

cluster = class id;
within = wave0 Qwave;
between = X Z inter;
TYPE = COMPLEX TWOLEVEL RANDOM;
MODEL:
%within%
s | Y ON wave0 Qwave;
%between%
Y;
s ON X Z inter;

2. What is the difference in conclusions if I omit "random" and inter is significant (Y is measured at 3 time points)?
TYPE = COMPLEX TWOLEVEL;
MODEL:
%within%
Y;
%between%
Y;
Y ON X Z inter;

Thank you for your help.
 Bengt O. Muthen posted on Tuesday, November 21, 2017 - 2:09 pm
Approach 1. and 2. should be the same if the residual variance of s is zero. The s residual gets multiplied by the within predictor of Y.

Note that your syntax for approach 1. has 2 predictors of Y on within - but you can only define a random slope for one at a time. Note also that if on Between you regress s on predictors, the same predictors typically influence Y as well.
 dummyvariable123 posted on Wednesday, November 22, 2017 - 1:36 am
Dear Dr. Muthen,
Thank you for your quick answer and guidance. Just to clarify, the "between" variances (L2) of both "S" and "Y" are significant for all 3 options of Random:

%within%
Y;
option 1) S | Y ON wave;
option 2) S | Y ON Qwave;
option 3) S | Y ON wave Qwave;

%between%
Y;
S;

Should I go with approach 1. from my previous post to test moderation by L2 variable "Z" (inter=Z*X) of the effects of L2 variable "X" on changes in L1 variable "Y"?
 Bengt O. Muthen posted on Wednesday, November 22, 2017 - 3:45 pm
Yes, approach 1 is good but your input isn't correct. First, as I mentioned you cannot say

s | Y ON wave0 Qwave;

because the random slope statementrefers to one predictor, not two as you have.

Second, on between you should let Y be regressed on X Z and inter.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: