The authors investigated the effect of collecting food frequency intake data using questionnaires that record response intervals rather than exact frequencies of consumption. Measures of energy and 24 nutrients were calculated using both types of frequency data for subjects' mean intakes, rank classifications and group mean values. Frequency data obtained between 1987 and 1989 using the open-ended Health Habits and History Questionnaire (HHHQ) developed by Block and associates at the National Cancer Institute were recoded into the interval response formats used by the computer-scannable version of the HHHQ and into the format used in the food frequency questionnaire developed by Willett for the Nurses' Health Study and other studies, Compared with the open-ended HHHQ, for otherwise identical data sets, the closed-ended HHHQ and Willett response categories produced significantly different (p < 0.05) measures of intake on the individual level for 18 (72%) (HHHQ) and 16 (64%) (Willett) of the 25 nutrient and energy measures studied, and they ranked 13-53% (HHHQ) and 16-52% (Willett) of subjects in different quintiles for the various measures. Use of food frequency questionnaires with closed-ended response categories causes nondifferential misclassification that could bias study results. To reduce such misclassification in epidemiologic studies, the authors recommend that food frequency questionnaires obtain exact frequencies of intake for measurement of diet exposure, and they describe an open-ended questionnaire layout which does so and also permits computer scanning of data.
