Non-Positive Definite Between Level M...
Message/Author
 Anonymous posted on Saturday, April 23, 2005 - 2:53 pm
I am trying to develop multiple level path analysis. When I have a restricted number of variables (n=6), I am able to perform the analysis. However, I discovered theoretical reasons to disaggregate a few of the measures, but not enough for Multilevel SEM. When I do this, I end up with a non-positive definite matrix at the between level. Any suggestions as to why and how to fix this problem?

My level-2 n=29 and my level-1 n=447. There are two variables at level-2 and eight level-1 variables. I can't understand why the disaggregation would generate such problems. Any thoughts?
 Anonymous posted on Saturday, April 23, 2005 - 5:24 pm
I was able to figure out a proper start value that did the trick.
 Maren Schulze posted on Monday, May 27, 2019 - 6:06 am
I have a model with four factors with three indicators each (one with four), collected in 16 organisations.

I am using TYPE=COMPLEX to account for the nesting of the data.

I have received the following warning:
"THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS -0.142D-16. PROBLEM INVOLVING THE FOLLOWING PARAMETER:
Parameter 16, PSC BY PSC1_3

THIS IS MOST LIKELY DUE TO HAVING MORE PARAMETERS THAN THE NUMBER OF CLUSTERS MINUS THE NUMBER OF STRATA WITH MORE THAN ONE CLUSTER."

This happens both when I use all cases (i.e. including missing values, N = 1730) and complete cases (i.e. restricted to those with complete data, N = 1521).

What does this mean - what kind of "strata" does Mplus refer to?

Your help is very much appreciated.

Thanks a lot!
 Maren Schulze posted on Monday, June 24, 2019 - 2:40 am
Dear Linda,
dear Bengt,

referring to my post from May, 27th - do you have a suggestion of why this happens? To what kind of "strata" does Mplus refer to?

I have meanwhile run the model applying WLSMV because my indicators are 5-point-Likert type items. In this model, I do not get the above message. Can I consider this an indication that WLSMV is more appropriate for my data? Would this answer my above question?

Thanks & kind regards!
 Bengt O. Muthen posted on Monday, June 24, 2019 - 3:44 pm
Your run probably has more than 16 parameters which is the number of clusters - this gives the message. If you have less than 16 Between-level parameters you should be ok and can use MLR or WLSMV.