Reply With Quote
I think it depends on what is causing heteroskedacity. Transformation or weighted least squares is often recommended.
Treating Heteroskedasticity in Logistic regression
I want to fit a probit model:
Phi^-1(p)=aln(x)+b
The Pearson residuals appear to be heteroskedastic as confirmed in the attached figures. The heteroskedasticity in this case is not due to influential points . MY question is how can I treat this?DO i need a different link function, do I need to add parameters (if so how) or do I need to add more predictor variables? Any ideas?
I think it depends on what is causing heteroskedacity. Transformation or weighted least squares is often recommended.
"Facts are stubborn things, but statistics are more pliable." Mark Twain
Posting Permissions
|
|