How do I identify in my output the parameter that has a problem? i.e. THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING PARAMETER 310. Is there a sequence, so that I should go through the output, starting with loadings first, then structural parameters, correlated errors, variances, etc until I reach this number? Thanks
I have an indicator in my measurement model that is behaving very strangely. There is an outlandishly high covariance with the other indicators and TECH 1 indicates a very high start value for that parameter. I have examined the variable and cannot find any coding issues or univariate distribution issues with it. I even cheated and tried to use it as the fixed parameter but that didn't work either. Any suggestions as to another set of diagnostics for it?
I am trying to use a bi-factor model in an SEM framework. I keep receiving the following error:
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 -0.444D-17. PROBLEM INVOLVING PARAMETER 48.
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 X2.
MODIFICATION INDICES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED.
The bi-factor model does not have this error itself; the error only occurs when extending to SEM. Importantly, the model estimation terminated normally and model fit statistics were produced. Should I be concerned about this error? Do you have any recommendations to fix the error if it is of concern?
The following message was sent to you this morning:
The error message is caused because you have two negative residual variances. The model should be changed.
The license you give is registered to Shaine Blanco. Support is available to one registered user per license.
Paula Vagos posted on Thursday, December 15, 2016 - 9:33 am
Dear Doctors Muthen, I am testing for the measurment invariance of a bifactorial model, with one general measure and four group factors. I am still only testing the baseline model and got the following error:
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 394, Group FEMALE: LIM
This parameter refer to the psi matrix associating one of the four group factors with it self in one of the groups... I don't know how to make sense of this or how to fix this problem.
Please advice. Thank you in advance. Paula
Paula Vagos posted on Thursday, December 15, 2016 - 9:40 am
Just in case, my syntax is as follows:
VARIABLE: NAMES ARE group U1-U50; GROUPING IS group (1 = male, 2 = female);
analysis: estimator is ML; Model=nocovariances
MODEL: A by u2 u7 u9 u10 u15 u22 u25 u28 u31 u33 u34 u39 u41 u48 u50; B by u4 u11 u12 u13 u17 u18 u23 u30 u36 u40 u42 u46 u47 u49; C by u1 u3 u5 u26 u27 u32 u38 u44 u45; D by u6 u8 u16 u19 u21 u24 u37 u43; All by U1-U50; [A@0B@0C@0D@0All@0]; MODEL female: A by u2 u7 u9 u10 u15 u22 u25 u28 u31 u33 u34 u39 u41 u48 u50; B by u4 u11 u12 u13 u17 u18 u23 u30 u36 u40 u42 u46 u47 u49; C by u1 u3 u5 u26 u27 u32 u38 u44 u45; D by u6 u8 u16 u19 u21 u24 u37 u43; All by U1-U50; [u1-u50];!allow intercepts to differ
OUTPUT: STANDARDIZED MODINDICES.
I also have a sample size of over 2000 participants in each group.
One problem I see is that you mention the first factor indicator in the group-specific MODEL command which frees it from its default of being fixed at one to set the metric of the factor.
Also, in a model with a general and specific factors, specific factor should be uncorrelated with each other and the general factor.
Paula Vagos posted on Friday, December 16, 2016 - 5:18 am
Dear Doctor Muthen,
Thank you for your quick and helpful response. Either by fixing the 1st indicator of each factor to 1 or removing it from the syntax solved the problem. Might I just ask if this is something that must be done always when testing for measurement invariance or is it a specific case of bifactorial models?
As for the uncorrelation between factors, I had thought that the command Model=nocovariances determined that...?