Please delete e7 and e8 from the model. Then, fix the variances of the kalite and yapi to the 1.

Try to run your model after making corrections that I suggested.

I can't tell anything about the variables from your path diagram but won't Volkan's suggestion change your hypotheses?

AMOS requires a lot statistical assumptions associated with normality, homoscedasticity, independence, among others.

Software based on Partial Least Squares performs better when those assumptions are not met or have small samples.

I use WarpPLS and it has given me good results.

Dear Volkan,how can I delete e7 and e8.They are error terms for endogenous variables. Don't we have to add error terms? It is CFA ,it is hybrid structural model

Your model shows that you didn't used double headed arrow which used to draw the covariance between Independent variables (SMEANS1,2,3). secondly I wonder why you used mean substitution while you could run CFA and then u could applied Data imputation.

there are no errors for your exogenous variables. and in line with Jorge Luis I recommend you to use PLS instead of Amos. for so many reasons.

there are some assumptions that you may check before using Amos. also PLS is friendly use.

I agree with the suggestion of Dr Volkan.

By removing these two error terms, you will have an assumption that the latent constructs are measured free-laden. Although it is not suggested, you have to follow this advice to test your model. Furthermore, people should avoid using PLS-SEM, which is variance-based SEM. I have two reasons on this issue. First, latent constructs exists regardless of their items. Thus, latent construct should be antecedent of the item, which is incongruent with the PLS-SEM. Second, it is not possible test whether your measurement model adequately fits data in PLS-SEM. This variance-based SEM is not disparate from the OLS-Regression.

Thanks all.I have to use SEM because my thesis subject is SEM,cannot use PLS.

Dear Alireza what do you mean? You mean I should have used data imputation method other than mean substitution? Or do you say I must run dfa model with incomplete data and then run structural model with subtituted data?

I am trying to use two step approach:first I tested cfa model and got good indices. Then in second stage now I am trying to run stuctural model which I attached.What did I wrong?

Thanks for your interest.That's very nice of you all.

It indicates that the data distribution and settings simply do not make sense. Need to check properly the input. Most of the times, this problem occur due to "copy and paste" data.