Multiple Mediations. Troubles with fi...
Message/Author
 Nils Henker posted on Monday, September 26, 2011 - 8:34 am
Hallo,

i conducted a path analysis with Mplus.

I used this syntax:

model:

B on A;
C on AB;
E on ABC;

model indirect:

A ind B A;
E ind C B A;

This model had a perfect fit.

Then I added a second mediator for the relation between B and E. I used this syntax:

model:

B on A;
C D on AB;
E on ABCD;

model indirect:

C ind B A;
D ind B A;
E ind C B A;
E ind D B A;

Now the model fit indices are really bad.
Does anyone know why the model fit droped from perfect to very bad just because I added a second mediator?

Best regards,

Nils
 Linda K. Muthen posted on Monday, September 26, 2011 - 11:48 am
The first model is just-identified so model fit is not relevant. The second model has one degree of freedom. Leaving out that path causes the model not to fit the data.
 Nils Henker posted on Wednesday, September 28, 2011 - 2:40 am
Thank you very much.
 Ann Chen posted on Wednesday, September 28, 2011 - 9:33 am
Dear Dr. Muthén,

I have a multiple mediations model with zero CFI and negative TLI. I'm not sure what does this mean? Poor model fit? But I got aceeptable RMSEA and SRMR. The following is the tests of model fit:

TESTS OF MODEL FIT

Chi-Square Test of Model Fit
Value 12.217
DF 11
P-value 0.3476

Chi-Square Test of Model Fit for the
Baseline Model
Value 20.001
DF 20
P-value 0.4579

CFI/TLI
CFI 0.000
TLI -2009.133

RMSEA
Estimate 0.029

SRMR
Value=0.051

I appreciate any comments and suggestions.

Thank you.

Best,
Ann
 Linda K. Muthen posted on Wednesday, September 28, 2011 - 10:40 am
 Jordan Gross posted on Wednesday, November 25, 2015 - 1:22 pm
Hi Dr. Muthen,

I have a model with the following specification:

VARIABLE:
Names are V1-V24;
MISSING ARE ALL (-99);
CATEGORICAL ARE V5 V6;

ANALYSIS:
ESTIMATOR= ML;
ITERATIONS = 3000;
CONVERGENCE = 0.00001;

MODEL:
precoll BY V1-V4;
V2 WITH V1;
instemp BY V7-V10;
V9 WITH V7;
peers BY V11-V13;
fit BY V14-V17;
V17 WITH V15;
attract BY V18-V21;
V19 WITH V18;
intent BY V22-V24;
V5 V6 ON precoll;
fit ON V5 V6 instemp peers;
attract ON fit;
intent ON precoll attract;

1. What is the best method to estimate my model given that I have missing data and both categorical and continuous variables?

2. When I ran the model, it did not produce model fit statistics and it said that it couldn't produce MODINDICIES because of the binary variable. Is there a way to gather this information?

3. I would like to estimate indirect, direct, and total effects between all of the variables. What is the best way to do that? What would I have to include beside IND or VIA statements?

Any help would be greatly appreciated!
 Bengt O. Muthen posted on Wednesday, November 25, 2015 - 4:08 pm

Estimator choices with categorical outcomes

2. Fit and Modind are only obtained using WLSMV. Fit is obtained also with Bayes.

3. Model Indirect with IND and VIA should do it.
 Jordan Gross posted on Monday, November 30, 2015 - 2:19 pm
Hi Dr. Muthen,

Thank you for your helpful response. I tried using both ML and Bayes estimation methods and received the following error messages:

*** ERROR
MODEL INDIRECT is not available for analysis with ALGORITHM=INTEGRATION.

*** ERROR
MODEL INDIRECT is not available for analysis with ESTIMATOR=BAYES.

How can I avoid getting this message? Thank you!
 Jordan Gross posted on Monday, November 30, 2015 - 2:49 pm
Also, when I ran the model using ML without the model indirect commands, all of the Betas were 0.000. Is there a way to get path coefficients when there are continuous and categorical variables in the model?
 Bengt O. Muthen posted on Monday, November 30, 2015 - 5:34 pm
Model Indirect with Bayes is available in Version 7.4.

For your zero beta run, send to support along with license number.
 Jordan Gross posted on Tuesday, December 01, 2015 - 7:47 am
Thank you for all of your help.

Is there a way to get Model Indirect with ML?
 Linda K. Muthen posted on Tuesday, December 01, 2015 - 9:13 am
When MODEL INDIRECT is not available, you can use MODEL CONSTRAINT.
 Kathleen Kenny posted on Wednesday, December 13, 2017 - 9:35 am
Hello!

I am conducting a moderation-mediation path analysis and building a model with a binary exposure, a binary outcome, 2 binary mediators, and 1 binary moderator. I am also adjusting for 4 confounders. I am using longitudinal data and my sample has 466 individual clusters and a total of ~ 2000 observations over time.

One of the confounders I am adjusting for is also hypothesized to be on causal path however, from a theoretical perspective it is not a mediator of interest to my research question since it is at the individual-level and my question is focused on social-structural mediators. When I remove this confounder/mediator from my model and do not adjust for it, the hypothesized indirect mediated pathways are strongly significant. When I adjust for it, pathways are only marginally significant at best.

Fit indices for model where I do not adjust the coufounder are below. RMSEA looks ok but TLI is negative (see more below).

Chi-Square Test of Model Fit
Value 43.533*
Degrees of Freedom 5
P-Value 0.0000

RMSEA

Estimate 0.064
90 Percent C.I. 0.048 0.082
Probability RMSEA <= .05 0.079

CFI/TLI
CFI 0.715
TLI -0.196

Do you have any suggestions?
Thank you!
 Bengt O. Muthen posted on Wednesday, December 13, 2017 - 2:08 pm
We need to see the relevant outputs - send to Support along with your license number.

Note also that mediation with binary DVs call for special considerations as explained on our Mediation web page under counterfactual effects.