Heteroscedastic variables with linear regression


I'm a stat beginner and learning about linear regression.

Regarding heteroscedastic variables (i.e. variables with different levels of residuals throughout the population).

I can identify them by looking at the scatter or plotting residuals.

A couple of questions I have are :

Is there a systematic way of detecting heteroscedastic variables?

Secondly, is there anything else we can do other than to discard the variables?

Thanks in advance.


No cake for spunky
http://en.wikipedia.org/wiki/Breusch–Pagan_test Breusch-Pagan is a test for heteroskedacity.

There are many options. In some case logging will address the issue. If you have a sense of what causes the problem and know weighted least squares you can use that (normally you won't know what causes it). You can use White's standard errors which correct for the problem.


TS Contributor
When doing regression it is not heteroskedasticity of the variables per se that is a problem, but heteroskedasticity of the errors.