Message/Author |
|
Isaac posted on Wednesday, April 29, 2009 - 8:17 pm
|
|
|
I'm running 4 class splitplot with 2 knots. I am trying to follow Flora's recent article in SEM. I've run the same model with two different data sets and each time I get error of a no-positive cov matrix. Also, one class drops to 0% with and I and S at 0. I'm posting my syntax. Any thoughts? MODEL: %OVERALL% I BY j1-j8@1; s1 BY j1@-2 j2@-1 j3@0 j4@0 j5@0 j6@0 j7@0 j8@0; s2 BY j1@0 j2@0 j3@0 j4@1 j5@2 j6@2 j7@2 j8@2; s2 by j1@0 j2@0 j3@0 j4@0 j5@0 j6@1 j7@2 j8@3; [j1-j8@0]; s1@0; s2@0; s2@0; |
|
Isaac posted on Thursday, April 30, 2009 - 8:18 am
|
|
|
Sorry, for some reason I wrote splitplot and I meant piecwise. I wrote this kind of late at night. Anyways, exchange the word piecewise for split plot and my question is the same. |
|
|
I am unclear about why you have two s2 BY statements. |
|
Isaac posted on Thursday, April 30, 2009 - 9:07 am
|
|
|
I'm sorry, that's an error in how I posted it. I have a s1 s2 and s3. |
|
|
I would need to see the full output and your license number at support@statmodel.com. This is usually due to a mismatch between the model and data. |
|
Isaac posted on Monday, May 11, 2009 - 7:30 pm
|
|
|
Hi linda, I've been able to work out the previous problem however a new problem has emerged and I wanted to get your advice. When running a piecewise model, the final class consistantly displays a Mean for the slopes as: S1 0.000 0.000 999.000 999.000 S2 0.000 0.000 999.000 999.000 This occurs regardless of which class is the final class and does not change if I use starting values. Any thoughts? Also, I have a large class(both in percentage and number of subjects) with a significant slope that is relatively flat. There's a decrease in estimated means of roughly .4 on a 9 point scale over 4 time points. Could the significance of the slope be effected by the size of the class? If so, is there a way to adjust for size or ascertain an effect size for the slope? |
|
|
Please send your full output and license number to support@statmodel.com. |
|
Back to top |