[R] Heteroscedasticity and mgcv.

[Simon Wood]

> (1) Am I correct in understanding that Heteroscedasticity is a problem for
> Generalized Additive Models as it is for standard linear models?  I am
> asking particularly about the GAMs as implemented in the mgcv package.
> Based upon my online search it seems that some forms of penalized splines
> can address heteroscedasticity while others cannot and I'm not sure what
> is true of the methods used in mgcv.
- Yes, the mgcv implementation estimates the models via penalized 
likelihood maximisation, and will be as sensitive to violation of the 
assumed mean variance relationship as any GLM fitted by MLE.

>
> (2) Assuming that heteroscedasticity is a problem for the mgcv GAMs, can
> anyone recommend a good test implementation?  I am familiar with the
> ncvTest method implemented in the car package but that applies only to
> lms.
- I tend to check for heteroscedasticity graphically using the usual 
plots of residuals vs fitted values, predictors (and possibly 
combinations of predictors). I like the way that plots often point 
towards a solution to any problem they show.

best,
Simon




>
> 	Thank you,
> 	Collin Lynch.
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>


-- 
Simon Wood, Mathematical Science, University of Bath BA2 7AY UK
+44 (0)1225 386603               http://people.bath.ac.uk/sw283