*Geek Box: Correlation Coefficients
The correlation coefficient, also known as the ‘coefficient of variation’ [CV] is the result of a statistical analysis of the results of the food-frequency questionnaire used in a study compared to the results from the validation study for that FFQ. The most robust validation occurs when measured food records are kept by a random sample of participants in the cohort; this provides a more objective assessment of diet to compare the accuracy of the answers from participants in the FFQ. The result of these correlation analysis is a score from -1.o to 1.0: -1 is a perfect negative correlation, 0 would be no correlation at all, and 1 would be a perfect positive correlation.
For example, if the CV for fibre was 0.58, that would indicate that a strong positive correlation between the FFQ correlated and the measured food records. A good paper in nutritional epidemiology will state what the CV for the foods or nutrients of interest was, giving an idea of how the findings may relate to a more ’true’ level of intake of the exposure of interest.
The CV varies across dietary constituents. The CV for fat and carbohydrate in most major cohort studies is >0.60, while the CV for micronutrients, for example zinc, is often <0.50. The CV is something to think about in assessing prospective cohort studies, because if the exposure of interest has a low CV [for example, <0.50], then it may indicate that the particular nutrient is prone to measurement error in that cohort. It would yield a degree of caution in interpreting the findings. Conversely, a CV of >0.60 would indicate a good level of correlation, that has sufficient granularity to capture a relationship between the exposure and the outcome. In the NIH-AARP, the CV for carbohydrates, fibre, fat, and saturated fat, were all >0.70.