On the evaluation of structural equation models