Using self-generated identification codes to match questionnaires in panel studies of adolescent substance use.

The usefulness of self-generated codes for anonymously linking data in panel studies of adolescent substance use was investigated in a study of Irish post-primary students and sample bias resulting from this procedure considered. A seven element code exactly matched 71% of questionnaires over one month when school absences were taken into account. Allowing codes to differ on one element to compensate for respondent errors increased matching success to 88% without resulting in appreciable mismatching. Unmatched compared with exactly matched respondents tended to be male, lower SES, have more spending money, and were less closely bonded to school and religion. They also were more involved in smoking, drinking, and drug use and had more favorable beliefs toward these behaviors. Off-one respondents generally were intermediate on these measures. However, the differences were small and the characteristics of the combined matched respondents closely resembled those of the total sample. When predicting substance use, the regression coefficients were quite similar for the matched and unmatched groups and the total sample. The data thus provide evidence for the usefulness of self-generated codes in panel and longitudinal studies of adolescents when anonymity and confidentiality are of concern.