A review of the effects of random measurement error on relative risk estimates in epidemiological studies.

Many articles in the recent epidemiological literature have discussed the effects of random error and misclassification on effect estimation, but many of these have been unclear and hard to follow. This paper reviews and interprets many of these and summarizes the use of the correlation coefficient in assessing the likely effect of measurement error on relative risk estimates for variables that are either continuous or ordered. A table of expected values of relative risks (RRs) calculated in different ways for different levels of random error is presented and the typically large expected attenuation in RR values is shown. The recommendation of taking repeated or multiple measurements whenever possible is endorsed.