How meaningful and sensitive are selection models and pattern-mixtures models?