proc calis

[Lindy]

Hi folks,

I have a SEM model with 7 variables and I used proc calis to test it. 

my code:

proc calis data=texas.offenderandfamily pshort plots=pathdiagram;
path
post_famisuppreality        post_financialstrain--->strain,
strain--->post2_famisuppreality,
strain--->post2_financialstrain,

post_famisuppreality <-> post2_famisuppreality,
post_financialstrain <-> post2_financialstrain,
post2_famisuppreality             u2gender->i3balc
;

run;

However, it will not give me a result and I got the warning in log

WARNING: The estimation problem is not identified: There are more parameters to estimate ( 18 )
than the total number of mean and covariance elements ( 15 ).

ERROR: LEVMAR Optimization cannot be completed.
NOTE: LEVMAR needs more than 50 iterations or 500 function calls.
NOTE: Due to optimization failure, statistics in the Fit Summary table might not be appropriate.
NOTE: Due to optimization failure, fit statistics in path diagrams might not be appropriate.
WARNING: Critical N is not computable for df= -3.
NOTE: Due to optimization failure, standard error estimates are not computed.
WARNING: Number of zero or negative variances encountered when computing standardized results:
1. Standardization with these values was not done.
NOTE: Because standard error estimates are not available, flagging for significant estimates is
not done in path diagrams.

Anyone know what is happening? I do not think I have over-specification. 

[PaigeMiller]

---

Anyone know what is happening? I do not think I have over-specification. 

---

Why do you think this? Explain. SAS thinks you have over-specified the model.

[Lindy]

But I have 7 variables. I calculated and I feel like I should not have a d.f. small than 0 as the SAS log showed. And I saw from text books SEM models with panel data of three waves, for example, x1, x2, x3, y1, y2, y3,  and the researchers can have paths between each x and y, and at the same time, there are paths between x1 and x2, x2 and x3, y1 and y2, y2 and y3. Even that, the model is still with d.f. bigger than 0...

Need someone to help me figure it out...Thank you!

[PhilWood]

At the beginning of your sas output sas will tell you which parameters it thinks are in the model. Looking at that can help you identify where the overspecified parameters are. Sometimes this results when more means are being estimated than there are means in the model, so it doesn't just apply to the variance/covariance matrix.