Great, just sent it. Also, I have MPlus 6 installed on my Dell laptop. When I save output or input files, I'm not able to open them. I get an eror message that says the file "is not a valid Win32 application." Any suggestions?
Try freeing the first factor loading of each factor and fixing the factor variances to one. It may be that the first factor loading which is fixed to one as the default would not be estimated close to one or is negative.
Thank you so much! The suggestion worked when I let all the second factor loadings free (not the first). But the fit is not good... I suppose this has to do with the model and is not changeable? However, maybe the fit will be better when the following problem is solved:
"WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE.THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR AN OBSERVED VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO OBSERVED VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO OBSERVED VARIABLES. CHECK THE RESULTS SECTION FOR MORE INFORMATION.PROBLEM INVOLVING VARIABLE ES."
John Perry posted on Wednesday, September 19, 2012 - 9:57 am
in terms of non-convergence, would I be right in assuming that this is probably because of bad starting values?
If so, is it wise to see what starting values are and then perhaps identify very high ones and constrain these to one in the model? I've got a couple of models to converge by doing this but would like your advice.
I would not think that starting values is a big problem. We have good default starting values. Often it can be related to large variances so we recommend keeping variances of continuous variables between one and ten. It can also be caused by variances or residual variances approaching zero. If you have a problem, send it and your license number to firstname.lastname@example.org. I do not think constraining values even if it works is a good idea.
thank you for the advice. I ran each group separately, and had no convergence problems there. Some models are better than others (these are different data across different years), but all are good or exceptional fits.
And I replicated the analysis using the data of two waves of my survey. The model is identified in one case but it does not converge when I replicate the analysis for the second wave. In particular I get this message:
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 135, D BY A
THE CONDITION NUMBER IS 0.151D-10.
FACTOR SCORES WILL NOT BE COMPUTED DUE TO NONCONVERGENCE OR NONIDENTIFIED MODEL.
Do you have an idea why this is happening? Thank you very much in advance
Marianne SB posted on Saturday, August 15, 2015 - 3:01 am
Hi Linda, I am trying to create a developmental measurement model with some shifting indicators of the same latent variable over time. Configural and metric invariance models run OK, but the scalar invariance model lead to convergence problems: NO CONVERGENCE. NUMBER OF ITERATIONS EXCEEDED.
I have rescaled the variables so the residual variances are between 0 and 10, but this did not resolve the problem. Neither did increasing the number of iterations.
Do you have any suggestions on how this might be resolved? Can it be a problem that the response categories for the items are on a categorical scale (aka: 0 = no, 3 = sometimes, and 6 = yes). I have used the MLR estimator, but the same happens when using the WLSMV estimator.