Question

when it comes to testing for heteroscedasticity. what is meant by the following statement "divide the sample observations based on the values of predicted y?

Answer 1

This means if you have in total 40 values of Y variable and corresponding set of X variable.

Set the data in ascending order of Y and say you need to divide the data in two sets, then pick first 20 observations of Y along with the respective set of X with that, and the remaining 20 is the another set.

Basically, divide the sample observations based on the values of predicted y is using only the set of Y observation to split the data keeping the respective values of X same for each Y as earlier.