I am running two-part models for semicontinuous outcomes with two classes (males and females). I have 5 time points. I understand that the variance of the slope growth factor (su@0) is fixed at zero for simplicity in the example in the user's guide. However, I would like the variance of the slope growth factor to be estimated. When I free it, I get the following error message: THE ESTIMATED COVARIANCE MATRIX IN CLASS 1 COULD NOT BE INVERTED. COMPUTATION COULD NOT BE COMPLETED IN ITERATION 122. CHANGE YOUR MODEL AND/OR STARTING VALUES.
How can avoid getting an error without fixing su@0?
Have you run the two parts separately to see if the problem occurs then. I would try that first. Other than that, you should send your input, data, output, and license number to email@example.com so we can see the full picture.
linda beck posted on Friday, June 13, 2008 - 6:01 am
Hello! I'm relatively new to growth modeling and also conducting a two part model. As it is often the case in such models, I have to fix both quadratic factor variances in both model parts to zero. So there is only the linear slope in both parts left (sig. variance) to predict change with covariates. But their means are not significant (means of quadratic factor are sig. in both parts). What does the (in-)significance of linear factor means telling us since quadratic change and linear change are confounded and both parts of the two part model are confounded too!? And can one say for expample, that being a boy leads to more change on the linear slope of both parts (gender significantly predicts both linear slopes)?
Your model has three growth factors. Each has a mean and variance. For all growth factors, the test for the means is if they are significantly different from zero. This is not an interesting test for the intercept growth factor. For the slope growth factor, if the mean is not significantly different from zero, this means there is no linear trend on average. For the quadratic, it is the quadratic component. It is not inconceivable that the linear trend not be signficant and the quadratic trend be signficant. Early development may be flat with an increasing or decreasing trend later in development. If the regression of the linear slope on gender is significant, this means that for the group scored 1 where the other group is scored 0 has a slope that is signficantly different from zero.
linda beck posted on Monday, June 16, 2008 - 1:28 am
ok, but the linear factor mean is positive, so it should be an early increasing trend. And albeit this trend is not significant on the average, it is different from zero for the group scored 1, am I right? Thank you!