|
|
ESTIMATION REACHED A SADDLE POINT |
|
Message/Author |
|
M.O. posted on Tuesday, March 24, 2015 - 9:26 pm
|
|
|
Dear Dr Muthen, I am conducting mixture factor analysis on categorical data. My input statement is: ----- TITLE: LCA DATA: FILE = '35itemN888.dat'; VARIABLE: NAMES = u1-u35; USEVARIABLES = u18-u19 u33; CATEGORICAL = u18-u19 u33; CLASSES = C(2); ANALYSIS: TYPE IS MIXTURE; STARTS = 10000 1000; STITERATIONS = 1000; ALGORITHM=INTEGRATION MODEL:%OVERALL% f1 by u18-u19 u33; ---------- I receive warning message as follows. I suspect I should try with different starting value, but I am not sure how to select a right value. Would you kindly give suggestions? ----- WARNING: THE MODEL ESTIMATION HAS REACHED A SADDLE POINT OR A POINT WHERE THE OBSERVED AND THE EXPECTED INFORMATION MATRICES DO NOT MATCH. AN ADJUSTMENT TO THE ESTIMATION OF THE INFORMATION MATRIX HAS BEEN MADE. THE CONDITION NUMBER IS -0.185D-04. ----- Thank you for your help, |
|
|
This may be ignorable, but try a smaller mconvergence value than the one reported in the Summary. Also, delete your STITERATIONS line - the default is probably better. |
|
M.O. posted on Friday, June 05, 2015 - 11:38 pm
|
|
|
Dear Dr. Muthen, Thank you very much for your advice. If estimated parameters (factor loadings etc) differs with different MCONVERGENCE values, which is recommended? Result with smaller mconvergence value with warning message (THE MODEL ESTIMATION HAS REACHED A SADDLE POINT c), or one obtained by larger (more lenient) mconvergence value without warning message? In my case, the results were similar, with both solutions seem reasonable, but they were not exactly the same. I read your article (Tihomir and Bengt on the issuehttp://www.statmodel.com/download/SaddlePoints2.pdf) and thought about it, but was not clear. Sorry for stupid question. |
|
|
I would trust the solution based on a smaller mconvergence value. |
|
Back to top |
|
|