The effects of message framing on behavioral prevalence assumptions