R help - Feb 2012 - non linear quantile regression - Median not plotting where it should

Hi,


I'm attempting to calculate the 0.25 and 0.97 quantiles for tree height
(0-50 meters) against tree age (0-300 years) and I am running into some
difficulty with the plotted grafic. I've run the examples in the quantreg
help and can get those to work properly and by plugging in my data I can also
get the lines plotted on my dataset. Unfortunately I'm running into a
problem with the median and other regression lines with tree age younger than 50
years, basically the median is in this range overestimated and even comes out of
the rage of oberservations. here is the code I'm using.


# then fit the median using nlrq

spruce.nlrq <- nlrq(height ~ SSlogis(age, Asym, mid, scal),data=spruce,
tau=0.5, trace=TRUE)
lines(1:200, predict(spruce.nlrq, newdata=list(age=1:200)), col="2")


I believe  this has something to do with the SSlogis, as this gives the
parameters for an S curve.  My data set does not have the typical S curve shape,
instead you could image it as starting at the inflection point of an S curve.
This is what I expect the .025 quantile to be similar to:

x <- seq(1,100,1)
plot(log(x))


the 0,975 should also have a logarithmic shape but no such a steep incline.  Ive
tried using different self starting models as found under:

apropos("^ss")

      
however I have not gotten them to work, or at least to fix the problem. Curves
similar to mine look like these site index curves:
http://www.extension.umn.edu/distribution/naturalresources/components/3473-13.html

In the example from the nlrq help the lines of the median and the various
quantiles all start from the same location, tha is basically (x=0, y=0) in the
coodinate plane.  With my problem, the lines start to be drawn from various
different positions ( the lines always start at age(x)=0, but the height(y) can
range between 0 and 15).
Additionally, the data set is quite large. with about 50,000 oberservations on
age and height.
Does anyone have a suggestion on how to fix this problem?
Thanks in advance
Dan

	[[alternative HTML version deleted]]

Dan,

It is hard to say without being able to reproduce your example.  If you send me
the data
I could try to advise something.

Roger

url:    www.econ.uiuc.edu/~roger            Roger Koenker
email    rkoenker at uiuc.edu            Department of Economics
vox:     217-333-4558                University of Illinois
fax:       217-244-6678                Urbana, IL 61801

On Feb 13, 2012, at 7:50 AM, Dan Morovitz wrote:
> Hi,
> 
> 
> I'm attempting to calculate the 0.25 and 0.97 quantiles for tree height
(0-50 meters) against tree age (0-300 years) and I am running into some
difficulty with the plotted grafic. I've run the examples in the quantreg
help and can get those to work properly and by plugging in my data I can also
get the lines plotted on my dataset. Unfortunately I'm running into a
problem with the median and other regression lines with tree age younger than 50
years, basically the median is in this range overestimated and even comes out of
the rage of oberservations. here is the code I'm using.
> 
> 
> # then fit the median using nlrq
> 
> spruce.nlrq <- nlrq(height ~ SSlogis(age, Asym, mid, scal),data=spruce,
tau=0.5, trace=TRUE)
> lines(1:200, predict(spruce.nlrq, newdata=list(age=1:200)),
col="2")
> 
> 
> I believe  this has something to do with the SSlogis, as this gives the
parameters for an S curve.  My data set does not have the typical S curve shape,
instead you could image it as starting at the inflection point of an S curve.
This is what I expect the .025 quantile to be similar to:
> 
> x <- seq(1,100,1)
> plot(log(x))
> 
> 
> the 0,975 should also have a logarithmic shape but no such a steep incline.
Ive tried using different self starting models as found under:
> 
> apropos("^ss")
> 
>       
> however I have not gotten them to work, or at least to fix the problem.
Curves similar to mine look like these site index curves:
>
http://www.extension.umn.edu/distribution/naturalresources/components/3473-13.html
> 
> In the example from the nlrq help the lines of the median and the various
quantiles all start from the same location, tha is basically (x=0, y=0) in the
coodinate plane.  With my problem, the lines start to be drawn from various
different positions ( the lines always start at age(x)=0, but the height(y) can
range between 0 and 15).
> Additionally, the data set is quite large. with about 50,000 oberservations
on age and height.
> Does anyone have a suggestion on how to fix this problem?
> Thanks in advance
> Dan
> 
> 	[[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.