In method-comparison studies, regression analysis is often used to estimate the systematic difference
between measurements from two different methods1
. Ordinary linear regression should only be used if
the measurements from one of the methods are without random error, which rarely occurs. The Deming
method of regression analysis, which accounts for measurement error in both methods, is often more
appropriate, and is requested by the U.S. Food and Drug Administration (FDA) in medical device
submissions. Though PROC NLP and CALIS in SAS® can accommodate Deming regression models,
these procedures are not widely used, and may not be easily understood by many applied statisticians.
This paper will present a macro that uses multiple DATA steps and PROC MEANS statements to
calculate the slope and intercept of the Deming regression line. Since it is difficult to derive a formula for
the standard deviations of these estimates, the non-parametric jackknife method is employed to construct
confidence intervals.