Limitations of the Correlation Coefficient in the Validation of Diet Assessment Methods

Dietary assessment methods usually are validated one against the other, using the correlation coefficient between the methods to assess agreement. The correlation coefficient, however, is only an indicator of relative agreement and depends on the range of intakes. As this dependence can lead to misinterpretation, we propose that the standard deviation of the difference between methods should be used as an indicator of absolute agreement and the standard deviation of their average as an indicator of the range of intakes. We illustrate this method in a validation study of a diet history against a 3-day diet record in 58 non-insulin-dependent diabetes mellitus patients. A re-analysis of studies from the literature validating food frequency questionnaires against diet records showed that the absolute agreement between methods is very consistent and that the differences in the correlation coefficients are mostly due to the selection of subjects.