Bounding the effects of measurement error in regressions involving dichotomous variables