If you write ADMISSIBILITY OFF before End of problem this should turn the default admissibility check off and allow you to examine the solution/parameter estimates. Perhaps this model is misspecified and results in an improper solution (Heywood case). You should be able to see this in the output. For example, you may get a negative error variance estimate for one or more of the observed variables (indicators).

Dear Mr. Geiser,

I have corrected my path as following:

Model 10

Raw Data from file 'C:\Users\svato\Desktop\LISREL 4\Zkouška 5.psf'

Sample Size = 424

Latent Variables vero maro zuzo

Relationships

V1 = vero

V2 = vero

V3 = vero

V3_A = vero

V5 = vero

M1 = maro

M2 = maro

M3 = maro

M4 = maro

M5 = maro

Z1 = zuzo

Z2 = zuzo

Z3 = zuzo

Z4 = zuzo

Z5 = zuzo

Path Diagram

ADMISSIBILITY OFF

End of Problem

Unfortunatelly, the same fault is ocurred:

W_A_R_N_I_N_G: PHI is not positive definite

W_A_R_N_I_N_G: THETA-DELTA is not positive definite

W_A_R_N_I_N_G: The solution was found non-admissible after 255 iterations.

The following solution is preliminary and is provided only

for the purpose of tracing the source of the problem.

Setting AD>255 or AD=OFF may solve the problem

The observations V1 - V5, M1 - M5, Z1 - Z5 are ordinal data from 1 to 5.

I would like to kindly ask you, what I can do next. I am a beginner in Lisrel, so I am very grateful for any help.

Thank you very much.

You should take a look at the "preliminary solution" that LISREL provides, that is, the parameter estimates in the output. It is likely that there is at least one inadmissible (improper) residual variance estimate (i.e., a variance that is estimated to be negative). Then, you should think about why this may have happened. Improper solutions (Heywood cases) are usually due to one of the following problems:

- model misspecification (e.g., too many or too few factors, omitted parameters, overfitting, underfitting, etc.)

- too small a sample size

- random sampling error (bad luck).

For more explanations and guidance, check out the paper below:

Article Improper Solutions in Structural Equation Models

Dear Mr. Geiser,

thank you very much for your valuable help. I have expanded number of observations and number of latent variables (it is just only testing file). The model has been set up, nevertheless, with no values, as you can see attached.

Why this happened?

Thank you very much for your help. You are saving me.

Is there not some sort of numerical output (text file with parameter estimates) that gives you the "preliminary solution" that is mentioned in the warning message? If not, I would try to use a different software that would provide you with more information about the specifics of the problem.