[R] linear against nonlinear alternatives - quantile regression

[Roger Koenker]

Roger Koenker
rkoenker at illinois.edu




On Nov 5, 2011, at 1:02 PM, Julia Lira wrote:

>
> Dear David,
>
> Indeed rq() accepts a vector fo tau. I used the example given by  
> Frank to run
>
> fitspl4 <- summary(rq(b1 ~ rcs(x,4), tau=c(a1,a2,a3,a4)))
>
> and it works.
>
> I even can use anova() to test equality of slopes jointly across  
> quantiles. however, it would be interesting to test among different  
> specifications, e.g. rcs(x,4) against rcs(x,3). but it does not work.

Probably because the models aren't nested...
>
> Thanks for all suggestions!
>
> Julia
>
>> From: dwinsemius at comcast.net
>> Date: Sat, 5 Nov 2011 13:42:34 -0400
>> To: f.harrell at vanderbilt.edu
>> CC: r-help at r-project.org
>> Subject: Re: [R] linear against nonlinear alternatives - quantile  
>> regression
>>
>> I suppose this constitutes thread drift, but your simple example,  
>> Frank, made wonder if Rq() accepts a vector argument for tau. I  
>> seem to remember that Koencker's rq() does.. Normally I would  
>> consult the help page, but the power is still out here in Central  
>> Connecticut and I am corresponding with a less capable device. I am  
>> guessing that if Rq() does accept such a vector that the form of  
>> the nonlinearity would be imposed at all levels of tau.
>>
>> -- 
>> David
>>
>> On Nov 5, 2011, at 10:43 AM, Frank Harrell  
>> <f.harrell at vanderbilt.edu> wrote:
>>
>>> Just to address a piece of this - in the case in which you are  
>>> currently
>>> focusing on only one quantile, the rms package can help by fitting
>>> restricted cubic splines for covariate effects, and then run anova  
>>> to test
>>> for nonlinearity (sometimes a dubious practice because if you then  
>>> remove
>>> nonlinear terms you are mildly cheating).
>>>
>>> require(rms)
>>> f <- Rq(y ~ x1 + rcs(x2,4), tau=.25)
>>> anova(f)  # tests associations and nonlinearity of x2
>>>
>>> Frank
>>>
>>> Julia Lira wrote:
>>>>
>>>> Dear all,
>>>>
>>>> I would like to know whether any specification test for linear  
>>>> against
>>>> nonlinear model hypothesis has been implemented in R using the  
>>>> quantreg
>>>> package.
>>>>
>>>> I could read papers concerning this issue, but they haven't been
>>>> implemented at R. As far as I know, we only have two  
>>>> specification tests
>>>> in this line: anova.rq and Khmaladze.test. The first one test  
>>>> equality and
>>>> significance of the slopes across quantiles and the latter one  
>>>> test if the
>>>> linear specification is model of location or location and scale  
>>>> shift.
>>>>
>>>> Do you have any suggestion?
>>>>
>>>> Thanks a lot!
>>>>
>>>> Best regards,
>>>>
>>>> Julia
>>>>
>>>>   [[alternative HTML version deleted]]
>>>>
>>>> ______________________________________________
>>>> R-help@ 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.
>>>>
>>>
>>>
>>> -----
>>> Frank Harrell
>>> Department of Biostatistics, Vanderbilt University
>>> --
>>> View this message in context: http://r.789695.n4.nabble.com/linear-against-nonlinear-alternatives-quantile-regression-tp3993327p3993416.html
>>> Sent from the R help mailing list archive at Nabble.com.
>>>
>>> ______________________________________________
>>> 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.
>>
>> ______________________________________________
>> 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.
> 		 	   		
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> 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.