Standardized technical procedures as well as statistical methods can be used to calculate the magnitude of measurement errors in studies related to anthropometric measures. OBJECTIVE: To compare two statistical approaches to evaluate interviewers´ calibration during the training phase before data collection in an epidemiological study. METHODS: Thirteen interviewers, divided into sub-groups were trained to take two measures of weight and two of height from 10 volunteers in each training section. Training was completed in six sections for weight and eleven for height measurements. Digital scales were used to measure weight and wall stadiometers to measure height. In order to evaluate interviewers´ calibration, two statistical methods were compared: a) the Intraclass Correlation Coefficient (ICC); and b) the precision and accuracy as proposed by Habicht. RESULTS: On average, interviewers were submitted to 2 sections for weight measure calibration, and to 3 sections for height measure calibration, and precision was reached before accuracy. ICC values showed that measures were strongly correlated since the first section. CONCLUSION: The Habicht method seems to be better than ICC, given it allowed for corrections not only of interviewer discrepancies related to precision but it also calculated the magnitude of deviations between measurements of interviewers and supervisor (accuracy).
Reliability; Intraclass Correlation Coefficient; Calibration
