

COVARIANCE MATRIX NOT POSITIVE DEFINITE 

Message/Author 

Arne Floh posted on Thursday, August 30, 2007  11:20 am



Linda & Bengt, I wanna run a unconditional LGM with 7 data points. The outcome is a cumulative profit variable: y1= profit t1, y2= profit t1 + t2, y3 = profit t1  t3. Unfortunately, I got the following error message. WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR AN OBSERVED VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO OBSERVED VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO OBSERVED VARIABLES. CHECK THE RESULTS SECTION FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE P2001. Help appreciated. 


I would look to see if p2001 has a negative residual variance or one of the other issues outlined in the warning. If you can't figure this out, please send your input, data, output, and license number to support@statmodel.com. 


dear Linda, I´ve just seen this post and I have the same problem. The variable specified in the warning does have a negative residual variance!!! what I should do in this case, or better, is there anything i can do? thanks a lot in advance luca 


This means you need to change your model or if it is very small and not significant, you may want to fix it at zero. 

Lea Dipper posted on Wednesday, January 27, 2016  8:21 am



Dear Linda and Bengt, I'm running a CFA of items with four variables on one factor, three variables on a second and five on a third. I'm getting errors on the factor with three variables. Here is my error: WARNING: THE RESIDUAL COVARIANCE MATRIX (THETA) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR AN OBSERVED VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO OBSERVED VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO OBSERVED VARIABLES. CHECK THE RESULTS SECTION FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE FEL_POS. The variable specified in the warning has a negative residual variance (and keeps it if I reanalyze the data with the recoded problematic variable). You recommend a model change or, if the residual variance is small, I may fix the problematic variable to zero, is this correct? The residual variance in my MODEL RESULT section is 0.391 and it is not significant (p = 0.148). Is this a small residual variance? And if so, is it correct to set the problematic variable to zero by writing this in the model = ...section: fel_pos@0;? And my last question: What does this imply for my CFA? 


Please send the output and your license number to support@statmodel.com. 

Back to top 

