1
$\begingroup$

I have two large data sets that I do several calculations on. In turn, I end up with a regression line for each data set. Now, I am not interested in comparing them to each other, but I am interested in comparing them to the identity line, since that is my "gold standard". If there are close to the identity line, my data fits perfectly. I am, however, not sure about which test I should use in order to analyse this, and see if one or the other is significantly different from the identity line.

$\endgroup$
3
$\begingroup$

A linear regression with one independent variable $x$ fits a line $y=\beta_1 x +\beta_0$ to the data $y$. The identity line between these two variables is given by $y=x$. Thus, to obtain exactly the identity line, the estimated regression coefficients must be $\hat{\beta_1}=1$ & $\hat{\beta_0}=0$, i.e. a slope of 1 and an intercept of 0. So to test whether your regression lines are significantly different from the identity line, I would use t-tests on your regression coefficients, to test whether they are significantly different from 1 (slope) and 0 (intercept).

$\endgroup$
0
$\begingroup$

If your original model is $y = \alpha_0 + \alpha_1 x$ then you could do the regression over $y' = y - x$ instead of $y$. In this way your model is $y' = \alpha_0 + (\alpha_1 - 1) x = \alpha_0 + \beta_1 x$, and you just have to check the significance test results for the fitted coefficient of $\beta_1$ to see if $\alpha_1$ is different from 1 (with some degree of confidence) or not.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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