How measures of perception from survey data lead to inconsistent regression results: evidence from adolescent and peer substance use