Assessment of the reliability of physical measures.

Assessment of the reliability of physical measures that yield continuous data (e.g., physiological measures such as blood pressure and anthropometric measures such as weight) should include the following statistics and procedures: mean, minimal and maximal differences; standard deviation of the net differences; technical error of measurement; and clinically meaningful indices of agreement. Displaying differences graphically and using Exploratory Data Analysis are excellent methods of summarizing differences. In those situations where any measurement error is important, regardless of its direction, the absolute value of the differences should be used to describe average differences. Correlation coefficients are not recommended because they are not adequate tests of reliability.