Retrospective information from questionnaires. II. Intrarater reliability and comparison of questionnaire types

Medical and dietary consumption histories obtained at two times from cases or controls and their next-of-kin as part of a study of colon cancer in the five Pennsylvania counties of the Philadelphia metropolitan area were compared. The case population was confined to whites aged 45-69 years who had resided in the region for at least two years prior to diagnosis and were diagnosed with colon cancer after July 1, 1976. Controls were selected using an area probability sampling scheme and were frequency-matched to the case group. Questionnaires for randomly selected subsamples of cases and controls were administered by interviewers; questionnaires for next-of-kin were randomly allocated to be self- or interviewer-administered. Agreement when respondents received the interviewer-administered questionnaire at both interviews was greater than when the self-administered questionnaire was used. Medical variables exhibited high agreement, the percentage agreement exceeded 80 for over 80% of the comparisons, the kappa statistic exceeded 0.6 for half of the comparisons. Diet histories were more variable (average agreement ranging from 56% to 67% and average kappa values from 0.16 to 0.40 for the different comparisons).