Power Estimation for Intervention Study PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Rich Jones posted on Tuesday, September 26, 2000 - 1:10 pm
In regard the computation of the noncentrality parameter (N-1*FCN) when using the Satorra-Sarris method to power estimation for intervention study using a Muthen and Curran (1997) and technical appendix latent growth model approach:

(1) Is the correct function (FCN) to use in computing the noncentrality parameter the "quasi-Newton" function?

(2) Is the 'N' the number of subjects in both the intervention and control groups?

My second question is motivated by noticing that the minimum of the fitting function varies with NOBSERVATIONS given the same covariance matrix, but yet the SAS programs distributed with the Technical Appendix consider power of as a function of N given a FCN.

Thanks,

Rich
jones@mail.hrca.harvard.edu
 bmuthen posted on Wednesday, September 27, 2000 - 11:34 am
Yes, on both questions.
 Anonymous posted on Wednesday, January 26, 2005 - 8:49 am
I tried to copy Mplus code from Muthen & Muthen (2002) "How to use a monte carlo study to decide on sample size and determine power" into Mplus 3, but got some error messages. Is the syntax chaged a lot? One of the error I got is "*** ERROR in Montecarlo command
Unknown option:
NCLASSES"

Thanks!
 Linda K. Muthen posted on Wednesday, January 26, 2005 - 1:36 pm
The language has changed slightly since that paper was written. The option NCLASSES is now CLASSES. See the Mplus Version 3 User's Guide for current language.
 Marc Glassman posted on Thursday, March 15, 2007 - 3:09 pm
In "How to Use a Monte Carlo Study to Decide on Sample Size and Determine Power", (SEM v9, n4, 2002) a monte carlo analyis for a growth model with a covariate is discussed (p. 604). The mean and variance of the covariate, x, are 0.5 and .25, respectively (p. 604). However, when setting up the monte carlo analysis in MPLUS (p. 614) the mean and variance for the covariate, x, are fixed to 0 and 1 (p.614).

My question is simply why 0 and 1 instead of the values given on p. 604 of the article, i.e., .5 and .25? Does it possibly have to do with centering the covariate so that the intercept growth factor will estimated at the average value of the covariate ?

Any help/clarification would be much appreciated.
 Bengt O. Muthen posted on Thursday, March 15, 2007 - 5:19 pm
This was answered under your Monte Carlo post.
 Emily Blood posted on Friday, November 09, 2007 - 10:04 am
In your article "How to Use a Monte Carlo Study to Decide on Sample Size and Determine Power" you define the effect size of the treatment variable on the latent intercept as the value of the coefficient of the treatment variable divided by the standard deviation of the latent intercept. I am trying to describe the power of a study where there is a mediating variable between the treatment indicator and the latent intercept, so there is a direct and indirect effect of the treatment on the latent intercept. In this setting, is it appropriate to define a treatment effect as the value of the total effect (a+b*c) divided by the standard deviation of the latent intercept? Is this still true if there are additional covariate predictors in the model? Thanks.
 Bengt O. Muthen posted on Saturday, November 10, 2007 - 7:47 am
Yes, I think so. Yes.
 Dustin Pardini posted on Sunday, October 19, 2008 - 9:04 am
I am trying to calculate an effect size estimate for a single group treatment study using an unconditional growth curve model. I am assuming that I can simple calculate the estimated mean change from pre to post using the growth curve model parameters, but am unsure what SD I should use (estimated pre SD?) or how to actually calculate this using the output. It seems that calculating effect sizes using the observed means may be beneficial given that most studies use this methodology and therefore would make it easier to compare to previous studies. Any suggestions?
 Linda K. Muthen posted on Monday, October 20, 2008 - 8:38 am
You can create the mean difference using MODEL CONSTRAINT and thereby obtain the correct standard error.
 Dustin Pardini posted on Tuesday, October 21, 2008 - 9:10 am
Linda,

I am not exactly sure how to use the command in this manner. Could you provide an example of how this could be used with a growth model to get the mean difference between time 1 (pre) and 2 (post), with the standard error for the means at time 1 and 2? If this is too time consuming for you I understand.
 Bengt O. Muthen posted on Tuesday, October 21, 2008 - 5:28 pm
With only 2 time points you would have to have a very simple growth model like

Model:
i s |y1@0 y2@1;
s@0;

where the y2-y1 mean difference is simply the mean of s, [s].

It would seem reasonable to simply divide [s] by the sample SD for the time 1 outcome to get the effect size.
 ywang posted on Friday, May 14, 2010 - 1:49 pm
Dear Drs. Muthen:
I used Monte Carlo for power analyses using the same setting as the example on Mplus website (http://www.statmodel.com/power.shtml). However, % Sig coeff for mean s is 0 (see input below), but it should be 0.862. Anything wrong? Thanks!
MONTECARLO: NAMES ARE y1-y4;
Nobservations=50;
NREPS=1000;
Model population: i BY y1-y4@1;
s BY y2@1 y3@2 y4@3;
[y1-y4@0];
[i@0 s@.2];
i@.5;
s@.1;
i WITH s@0;
y1-y4@.5;
Model:
i BY y1-y4@1;
s BY y2@1 y3@2 y4@3;
[y1-y4@0];
[i@0 s@.2];
i@.5;
s@.1;
i WITH s@0;
y1-y4@.5;
 Linda K. Muthen posted on Saturday, May 15, 2010 - 8:07 am
I think the problem is in the MODEL command where you fix all of the values using the @ symbol. You should use the asterisk (*) instead for the free parameters.
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: