I am just starting to learn Mplus. I am interested in using MPLUS to model twin data. Right now I am trying to fit a standard ACE model but including a covariate that moderates the path from the latent variables to the observed phenotypes (similar to the model presented in Purcell, 2002, Twin Resarch) Is XWITH appropriate for this?
bmuthen posted on Monday, August 08, 2005 - 1:16 pm
I am using simulated twin data to test a GxE interaction model. I'm using XWITH to specify an interaction between an observed moderator (m1 and m2) and latent genetic (A) and environmental (E) factors:
I have no problem running the model using multiple groups and the EM algorithm. However, eventually I want to include interactions between latent factors which requires using numerical integration (=montecarlo)
The problem is I get the error message shown below, regardless of what my simulation conditions are. If I take out the interactions the model runs fine, so the problem must have something to do with them. Do you have any tricks you could recommend?
Thank you, Carol
RANDOM STARTS RESULTS RANKED FROM THE BEST TO THE WORST LOGLIKELIHOOD VALUES
Unperturbed starting value run did not converge.
10 perturbed starting value run(s) did not converge.
Loglikelihood values at local maxima, seeds, and initial stage start numbers:
1 perturbed starting value run(s) did not converge.
THE ESTIMATED COVARIANCE MATRIX IN CLASS 2 COULD NOT BE INVERTED. COMPUTATION COULD NOT BE COMPLETED IN ITERATION 1. CHANGE YOUR MODEL AND/OR STARTING VALUES.
I would need more information to answer this. Please send the input that ran without the interaction, the input for the model that received the above message, the data, the two outputs, and your license number to firstname.lastname@example.org.
Hello, I would like to know if it is possible to test models that include 2 latent variable interactions in Mplus 5. For example, if I have a model of f1 --> f2 --> f3, can I include a moderator f4 between f1/f2 and a moderator f5 between f2/f3? Thanks for any assistance you can provide.
Can XWITH be used to determine a point of diminishing returns?
For example, if I think that Perceived Control (PC) has a point of diminishing returns in its positive effect on health, I can use XWITH to create a latent "PC Squared" term. If this squared term is significant, that means that there is a significant curvilinear component to the effect of PC on health. Can I determine the value of PC Squared at which the change in slope is zero? Does Mplus have functionality for this?
Thanks, Linda. Can I determine where the point of diminishing returns for this squared term would be? This would be something like looking for a marked decrease in the marginal effect. I also see this Mplus discussion page (http://www.statmodel.com/discussion/messages/23/828.html?1358295422) suggesting that you may have an Excel sheet that offers estimates of marginal effects?
Or, is there another way to graph the impact of this squared term created using XWITH on the y variable--to, again, determine the point of diminishing returns?
I am trying to use XWITH in a Montecarlo power analysis to get an interaction. However, I cannot get the model to run. Below is the model I tried. If you remove the XWITH the model runs just fine. Any ideas on what I could do?
model montecarlo: outcome BY d1@1 d2-d7*.7; d1-d7*.5; f1 by y1@1; y1@0; f1*1; f2 by y2@1; y2@0; f2*1; y1 WITH y2@0; f1 WITH f2@0; int | f1 XWITH f2; outcome ON f1*.3 f2*.3 int*.3; outcome*.73; model: outcome BY d1@1 d2-d7*.7; d1-d7*.5; f1 by y1@1; y1@0; f1*1; f2 by y2@1; y2@0; f2*1; y1 WITH y2@0; f1 WITH f2@0; int | f1 XWITH f2; outcome ON f1*.3 f2*.3 int*.3; outcome*.73; Analysis: type = random; ALGORITHM = integration; processors = 2;