Effects of interview mode on bias in survey measurements of drug use: do respondent characteristics make a difference?