Dear Mplus team, I would like to run a 1-factor model (CFA) with my data (n=177) for the only reason to reject that model - but the model doesn't run. I get the following error: NO CONVERGENCE. NUMBER OF ITERATIONS EXCEEDED
I have collected data on two different time points. Unfortunately, my N is quite low (n=95 per time point). During a SEM course the teacher suggested that I should do multiple group analysis so I can take the two time points in my model and see if my respondents differ in their answers over time. I tried running my CFA model with my data (n=190) using the grouping variable. However, I get the following error: 'No convergence. Number of iterations exceeded'. I have increased the number of iterations to 50,000 and it still won't converge.
I assume that you have different subjects at the two time points so that multiple-group analysis is relevant. As a first step to solve your non-convergence problem you should analyze each group separately. And if that is problematic, switch to EFA.
If this doesn't help, you need to send your output, input, data, and license number to email@example.com.
Eiko Fried posted on Monday, June 15, 2015 - 4:50 am
We allow for correlated item residuals across time (x1_t1-x28_t1 PWITH x1_t2-x28_t2), seeing that otherwise modindices show massive misfit, and that it seems to be recommend in the literature.
M1 baseline, M3 strong (thresholds), and M4 strict (residual variances) models converge fine without problems (and fit a lot better than not allowing item residuals to be correlated). Only M3 strong invariance model (loadings) does not converge:
NO CONVERGENCE. SERIOUS PROBLEMS IN ITERATIONS. CHECK YOUR DATA, STARTING VALUES AND MODEL.
Why could that be? (cannot really post the output seeing that is it way too long)
If you have not already done so, you should fit the model at each time point as a first step. The same model should fit well at each time point before you test for measurement invariance.
Eiko Fried posted on Monday, June 15, 2015 - 6:31 am
Thanks Linda. When we originally tried the CFA framework, the CFA model we derive from the baseline measurement point does not fit well to the follow up time point.
This is why we decided to use the EFA/ESEM framework instead, to which my question refers.
Therefore I'm not sure what you mean——since we are using an ESEM approach, the EFA model fits equally well (because it's completely free) to both measurement points (although the parallel analysis suggests 4 factors at baseline and only 3 at follow up)
I am trying to run a chi-square difference test between a five-factor model CFA and a Harmon single-factor model. The five-factor CFA looks great, but the single-factor has trouble converging. I have tried multiple things from the user guide to help with convergence, but the convergence or the results are still an issue (fixing the factor variance to 1, freeing the different items, etc.).
Is there something else that is recommended or is it sufficient to say that since their is no convergence, there is evidence that a single-factor doesn't fit and shouldn't be used?