Just because correlation doesn't equal correlation doesn't mean that correlation can't provide evidence that a model of behavior is correct.

When you can't measure something directly you make predictions of what it causes and then measure them. If that occurs it provides evidence that model is correct.