I found the solution.
coeftest does not work with na.exclude but with na.omit only, i.e. one
needs to omit missing values from the residual matrix.
Cheers!
On Thu, Aug 8, 2013 at 10:19 AM, ivan <i.petzev@gmail.com> wrote:
> Dear R Community,
>
> I am trying to build a very simple function which uses lm and coeftest to
> return a coefficient matrix with heteroskedasticity robust standard errors.
> The function is the following:
>
> reg=function(formula,data,na.action){
> res=lm(formula=formula,data=data,na.action=na.action)
> hc3=coeftest(res, vcov = vcovHC(res, type = "HC3"))
> residuals=resid(res)
>
return(list("coef.hc3"=hc3,"R2"=summary(res)$r.squared,
> "R2.adj"=summary(res)$adj.r.squared,
"residuals"=residuals))
> }
>
> The function works perfect as long as the data contains no missing values.
> I.e.
>
> test1=seq(1,30,1)
> test2=seq(1,30,1)
> testdata=data.frame(test1,test2)
> reg(formula=test1~test2, data=testdata, na.action=na.exclude)
>
> However, as soon as I have a missing value, it does not work any more (the
> error message is: Error in estfun(x)/X : non-conformable arrays):
>
> test1=seq(1,30,1)
> test2=seq(1,30,1)
> test2[5]=NA
> testdata=data.frame(test1,test2)
> reg(formula=test1~test2, data=testdata, na.action=na.exclude)
>
> My feeling is that it has something to do with na.exclude being a function
> itself and hence R having a problem with inputing a function into another
> function. Does anyone have an idea?
>
> Thanks a lot!!!
> Regards
>
>
>
[[alternative HTML version deleted]]