*Geek Box: Measurement Errors
Measurement error is a reality of any epidemiological study. In nutritional epidemiology, the goal is to capture dietary intake as accurately as possible; the extent to which an instrument achieves this is known as ‘validity’. To ascertain the validity of a dietary assessment method, validation studies are conducted, by taken a subsample of the cohort and conducting a 7-day measured food record, which is then used to compare the accuracy of the FFQ against.
There are two main types of error: random and systematic. Random error results from difficult or inconsistencies in taking a measurement [for example, different styles of questionnaire, or human memory]. Systematic errors result from predictable inaccuracies in the measurement instrument used, that are consistent in the direction of the error. For example, in nutritional epidemiology, the systematic error from FFQ is to underestimate true dietary intake.
The random error potential in nutritional epidemiology is generally human memory. The overall effect, in a prospective cohort study, any measurement errors tend to bias the associations between diet and a given outcome towards the null (i.e., ‘no association).
One way to overcome the potential influence of measurement error is with very large sample sizes. Therefore, for prospective cohort studies in nutrition science, ‘bigger is better’. In any cohort study, however, the most important means of dealing with potential error is validation and calibration studies: this allows for systematic error to be addressed, by understanding the size of the error, and being able to correct for this measurement error in statistical analysis.