Mediation PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 jonas helao posted on Tuesday, November 19, 2019 - 8:55 am
1) I have a binary X, a continuous M (standardized with its sample mean and standard deviation) and a continuous Y (which is also standardized with its sample mean and standard deviation). No missing values.

Is the following a correct interpretation of the indirect effect?

“there is a 'z' standard deviation decrease of Y (in terms of a sample standard deviation of Y) when X goes from 0 to 1”

2) a continuous X (standardized with its sample mean and standard deviation), a binary M and a continuous Y (standardized with its sample mean and standard deviation). I give two values of the continuous X (0 and +1 SD).

Is the following a correct interpretation of the total natural indirect effect?

“there is a “z” sample standard deviation decrease of Y (in terms of Y) when X goes from 0 to 1 sample standard deviation of X”
 Bengt O. Muthen posted on Wednesday, November 20, 2019 - 4:06 pm
Looks ok. Note that for the two continuous X values, they should be in the order +1SD, 0.
 Stefan Poier posted on Wednesday, June 17, 2020 - 4:26 am
Dear Dres M&M,

I would like to conduct a mediation analysis on the impact of the Big Five on a dichotomous outcome with three mediators.
Each Big Five factor is determined by three observed variables. All Big Five should be independent latent variables simultaneously, because I want to control the effects of the remaining as covariates. Now I get the warning message that too many Integration Points are needed. Monte Carlo takes hours and hours. What else can I do?
1) calculate the average of the observed variables for each of the 5 factors, thus virtually eliminating the latent factors
2) Perform 5 individual analyses for the Big Five, each with one factor (although this would no longer be controlled for the other four factors)

Many thanks and best regards
Stefan
 Bengt O. Muthen posted on Thursday, June 18, 2020 - 4:00 pm
Your screen printing says how many dimensions of integration you have. Perhaps it is 5 which is a lot. This means that you don't want to use the default of 15 points per dimension - instead try

integration = montecarlo(5000);

With integration, the computation speed is also affected by the number of observations.
 Stefan Poier posted on Saturday, June 20, 2020 - 2:21 am
Thanks for the hint, Bengt,

the calculation time is now about 2 hours (which is great).
But what I don't understand (and what is not described in the UG or in the Mediation and Regression Book):
a) what exactly do "integration points" mean?
b) what is the meaning of "X points per dimension"?
c) does the estimate become less accurate if Monte Carlo is chosen?
d) can I also get Bootstrap CI with "Integration: Monte Carlo"?

Many thanks and best regards
Stefan
 Bengt O. Muthen posted on Sunday, June 21, 2020 - 4:01 pm
a-b) See e.g. the 1981 Psychometrika article by Bock & Aitkin.

c) Typically not.

d) Yes, but it will be slow. You can instead use Estimator = Bayes for which you get non-symmetric CIs which I assume is the reason you want bootstrapping.
 Stefan Poier posted on Sunday, June 21, 2020 - 11:53 pm
Thank you very much.
 Bengt O. Muthen posted on Monday, June 22, 2020 - 5:49 am
Numerical integration is also described in our User's Guide on pages 526 and on. We also discuss it in one of our Short Course Topic videos.
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: