I am trying to perform a regression of GPR measurement response (X) vs concrete core density (Y). There is a known, constant error associated both these measurements. How would I incorporate this error into a simple linear regression?


1 Answer 1


If it is a known error in both $X$ and $Y$, you can subtract errors from both $X$ and $Y$ to get $X'$ and $Y'$, and then fit the regression on $X'$ and $Y'$.

So, lets say there is a known constant error of $c_x$ in your $X$ and a known constant error of $c_y$ in your $Y$. (These are signed errors). Then you define $X' = X - c_x$ and $Y' = Y - c_y$, and fit the regression on $X'$ and $Y'$. The model would then be $$Y' = a + bX' + \epsilon, $$

where $a$ is the intercept, $b$ is the slope and $\epsilon$ is the random error with zero mean. Just to see how this regression related back to the old one, $$Y - c_y = a + b(X - c_x) + \epsilon \Rightarrow Y = c_y + a - bc_x + bX + \epsilon. $$

Thus we get that if you fit the regression on the original $X$ and $Y$, then the slope is unaffected by the constant errors and the $y$-intercept changes according to the constants.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.