Number of dimensions/integration points
Message/Author
 Patchara Popaitoon posted on Thursday, September 15, 2011 - 3:38 am
Hello,

I followed example 9.8 to specify the model and got this error message and I am not sure how to solve this.

THERE IS NOT ENOUGH MEMORY SPACE TO RUN THE PROGRAM ON THE CURRENT
INPUT FILE. THE ANALYSIS REQUIRES 7 DIMENSIONS OF INTEGRATION RESULTING
IN A TOTAL OF 0.17086E+09 INTEGRATION POINTS. THIS MAY BE THE CAUSE
OF THE MEMORY SHORTAGE. YOU CAN TRY TO REDUCE THE NUMBER OF DIMENSIONS
OF INTEGRATION OR THE NUMBER OF INTEGRATION POINTS.

Could you please explain: what are the number of dimensions and the nubmer of integration points?; and how could I locate them in the model?. Thanks.

The analysis and model command are as follows:

Analysis: TYPE = Twolevel random;
ALGORITHM=INTEGRATION;

Model:
%within%
A by A1-A7;
B by B1-B7;
C by C1-C4;
XW by X1-X6;

AX | XW on A;
BX | XW on B;
CX | XW on C;

%Between%
XB by X1-X6;
Y by Y1-Y6;
Z by Z1-Z5;

Y on XB AX BX CX;
Z on Y;

Please note that there is no command re integration (ALGORITHM=INTEGRATION;) in example 9.8 but the system suggested adding this in the analysis command.
 Linda K. Muthen posted on Thursday, September 15, 2011 - 5:40 am
The number of dimensions of integration depends on the model. I assume some of your factors have factor indicators that are not continuous. This would result in each factor requiring one dimension of integration. Each latent variable interaction adds one dimension of integration. The number of integration points depends on the INTEGRATION option. See the user's guide under numerical integration for further information.

I would suggest running the model with only one latent variable interaction at a time to see if they are in fact needed. We recommend no more than four dimensions of integration.
 Patchara Popaitoon posted on Friday, September 16, 2011 - 4:34 am
Thanks Linda. I have successfully tested this by applying one latent variable interaction at a time. Indeed, I don't need them in the between level analysis as none of them shows significant impact on organizational level outcomes.

In terms of interpretation, can I say that A B and C influenced Y through X and that the perceived levels of A B and C are invariant across clusters ? A, B and C are employee perceptions at the individual level. Thanks.
 Bengt O. Muthen posted on Saturday, September 17, 2011 - 10:47 am
I don't see that interpretation being valid here. Do you really need the XWITH statements, i.e. the random slopes? If you are interested in how A-B-C influence Y, why don't you instead simply work with random intercepts for A, B, C, and X, that is, define these 4 factors on between as well and let their between parts influence Y.
 Patchara Popaitoon posted on Saturday, September 17, 2011 - 11:59 am
Thanks. This is much more straightforward.
 EFried posted on Monday, February 13, 2012 - 3:25 am
I'm also trying to reduce number of integration points in a mixture model in the exploratory stages.

Even with integration=standard(5) the models run for 2 hours+ with only 3 latent factors. Is there a way to reduce time further, e.g. by limiting the number of drawn bootstraps down from 100 to 30?

The "bootstrap = 30" command is not available for mixture models as far as I understand.

Thank you
 Linda K. Muthen posted on Monday, February 13, 2012 - 9:52 am
Please send your output and license number to support@statmodel.com so we can see why this is taking so long.