3 Comments
User's avatar
Arqiduka's avatar

Would it be enough to run the regressions for just the P25 and P75 to visually test? If there's any systematically changing variance, these (or even one of them) should be enough to highlight? And if it shows up at, say, the P90 but not the P75, is it really an issue?

Expand full comment
Emil O. W. Kirkegaard's avatar

Sure, any centile comparison will work in theory. You could fit the models for P25 and P75 and check they have a ~constant gap size across values of the predictors.

Expand full comment
Bazza's avatar

Data analytical magik.

Expand full comment