A medication-adherence survey finds high adherence among respondents. Device logs show that respondents overstate use slightly, and invitees who never answer have much lower logged use than respondents. How do the two findings affect the survey conclusion?