R help - Oct 2007 - R segmented package

Most of the data sets I'm dealing with exhibit a time trend.
We would like to get rid of the time trend.
The plot shows in some cases a monotonic increase of the dependent variable
with time. This is the easiest case.
In some other cases the plot shows a time trend where the dependent variable
changes slope 4-5 times along the observations measurement period.

I've attempted a segmented regression and estimated the break-point
empirically. This is a tedious error-prone process.
I just found out that R includes a package, called "segmented",  which
can
estimate the break-point discriminating between two regression models.
I have browsed through the on-line documentation. It is stated multiple
break-points are supported by "segmented".
However I cannot understand if a break-point per independent variable
(regressor) is handled when there are many independent variables in the
model, or if "segmented" can handle many breakpoints for the same
independent variable (time in my case) ... ?

I would greatly appreciate some explanations and suggestions on the use of
"segmented".

Thank you very much in advance.

Regards,

-- 
Maura E.M

	[[alternative HTML version deleted]]

Dear Maura,
For package-specific questions, it is probably advisable to contact 
privately the author..

BTW:
 > However I cannot understand if a break-point per independent variable
 > (regressor) is handled when there are many independent variables in the
 > model, or if "segmented" can handle many breakpoints for the
same
 > independent variable (time in my case) ... ?

Yes, both the statements are true. Virtually segmented is able to handle 
GLMs with any number of (linear) exaplanatory variables + any number of 
`segmented' variables, and multiple breakpoints are allowed for *each* 
`segmented' variable. See the relevant help files.

All the best,

vito


PS: I am sending you some further documentation via off-list

best,
vito


Maura E Monville ha scritto:
> Most of the data sets I'm dealing with exhibit a time trend.
> We would like to get rid of the time trend.
> The plot shows in some cases a monotonic increase of the dependent variable
> with time. This is the easiest case.
> In some other cases the plot shows a time trend where the dependent
variable
> changes slope 4-5 times along the observations measurement period.
> 
> I've attempted a segmented regression and estimated the break-point
> empirically. This is a tedious error-prone process.
> I just found out that R includes a package, called "segmented", 
which can
> estimate the break-point discriminating between two regression models.
> I have browsed through the on-line documentation. It is stated multiple
> break-points are supported by "segmented".
> However I cannot understand if a break-point per independent variable
> (regressor) is handled when there are many independent variables in the
> model, or if "segmented" can handle many breakpoints for the same
> independent variable (time in my case) ... ?
> 
> I would greatly appreciate some explanations and suggestions on the use of
> "segmented".
> 
> Thank you very much in advance.
> 
> Regards,
> 
-- 
===================================Vito M.R. Muggeo
Dip.to Sc Statist e Matem `Vianelli'
Universit? di Palermo
viale delle Scienze, edificio 13
90128 Palermo - ITALY
tel: 091 6626240
fax: 091 485726/485612