can you conclude causation from an observational study