Andie R posted on Friday, November 30, 2007 - 6:00 pm
I am trying to run a simple 1-factor model using item-level data with 150 categorical variables, sample size is appr. 35,000. However, when I ran the 1F-model with the WLSMV estimator the program aborts after about 2 hours with an error message saying that not enough memory space is available (the same problem occurs when I tried to run the model in EQS). I've tried various computers, all 32-bit machines with up to 3GB, and I also tried the faster ULS estimator but the same problem occurs each time.
Do you have any suggestions or explanations for these memory limitations and do you think, this problem will persist in the new version 5 of Mplus?
Is there any way of finding out how much memory one needs for analyses this large?
I also tried to work around this problem by using a tetrachoric correlation matrix as input, but that won't allow me to get st.errors or model chi-square with the ULS estimator, which seems to be the only estimator available in this context. (I need model chi-squares for model comparisons.) Is there an alternative specification that allows me to use summary data with categorical variables?
I don't know what version of Mplus you are using. The weight matrix needed for model estimation becomes very large as the number of variables increases particularly for ordered polytomous variables. Because you have only one factor, you may find maximum likelihood estimation faster.