FITTING STRAIGHT LINES TO SCATTER GRAPHS WHEN BOTH VARIABLES ARE SUBJECT TO ERROR
Eddie Channon, Chirostat Statistical Consulting
PSI/EFSPI conference May 2008
Linear regression with errors in both variables may be appropriate. There is a lot of literature in this area but none of the methods is easily available in SAS(R). A pragmatic approach is to fit regression lines for y on x and x on y and then to choose the bisector of the lines.
When correlation between the variables is high, there is no practical advantage of allowing for any error in x. In other cases, the most appropriate method depends on the interpretation to be made from the fitted line.