Finding multiple breakpoints - 'segmented' ?

[Peter Hoitinga]

Hello,

I'm attempting to find multiple breakpoints in an association of my
response variable (R.AUC) with two explanatory variables ('s.size' and
'bedekking'). The association between 's.size' and 'R.AUC' shows a
plateau, but the value when this plateau is reached is differs for
different values of 'bedekking'.

Initially I thought these different values could be assessed by the
'segmented' function in the package of the same name, but this only
seems to give one value of a breakpoint for each explanatory variable.
A similar issue was treated in the following message from 2006:
https://stat.ethz.ch/pipermail/r-help/2006-January/086446.html

Does anybody know whether there are any packages that do have a
function for this? Or might it be a possibility to split up the data
per value of 'bedekking' and then do the segmented function?

______________________________________________
[hidden email] 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.

[vito.muggeo]

dear Peter,
Currently segmented handles multiple breakpoints for several variables
(the limit discussed in the msg 2006 has been fixed..).

However you are looking for a somewhat complicated model where the
breakpoint of the relationship 's.size' and 'R.AUC' depends on another
covariate 'bedekking'. Namely it is an interaction in the breakpoint
parameter (rather then in a linear parameter).

Currently segmented is not able to deal with these interactions, and
splitting up the data into two or three groups according the values of
bedekking could be a feasible alternative.

Hope this helps you,

vito




Il 01/06/2012 11.35, Peter Hoitinga ha scritto:

> Hello,
>
> I'm attempting to find multiple breakpoints in an association of my
> response variable (R.AUC) with two explanatory variables ('s.size' and
> 'bedekking'). The association between 's.size' and 'R.AUC' shows a
> plateau, but the value when this plateau is reached is differs for
> different values of 'bedekking'.
>
> Initially I thought these different values could be assessed by the
> 'segmented' function in the package of the same name, but this only
> seems to give one value of a breakpoint for each explanatory variable.
> A similar issue was treated in the following message from 2006:
> https://stat.ethz.ch/pipermail/r-help/2006-January/086446.html
>
> Does anybody know whether there are any packages that do have a
> function for this? Or might it be a possibility to split up the data
> per value of 'bedekking' and then do the segmented function?
>
> ______________________________________________
> [hidden email] 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.
>

--
====================================
Vito M.R. Muggeo
Dip.to Sc Statist e Matem `Vianelli'
Università di Palermo
viale delle Scienze, edificio 13
90128 Palermo - ITALY
tel: [phone]
fax: [phone]
http://dssm.unipa.it/vmuggeo

______________________________________________
[hidden email] 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.

[Peter Hoitinga]

Dear Vito,

Thanks for the quick reply.

In the case of splitting up the data, would it be advisable to
independently fit a GLM on the subsets of the data, or use one single
GLM for comparison in the 'segmented' function?

Kind regards,

Peter

On Fri, Jun 1, 2012 at 2:21 PM, Vito Muggeo (UniPa)
<[hidden email]> wrote:

> dear Peter,
> Currently segmented handles multiple breakpoints for several variables (the
> limit discussed in the msg 2006 has been fixed..).
>
> However you are looking for a somewhat complicated model where the
> breakpoint of the relationship 's.size' and 'R.AUC' depends on another
> covariate 'bedekking'. Namely it is an interaction in the breakpoint
> parameter (rather then in a linear parameter).
>
> Currently segmented is not able to deal with these interactions, and
> splitting up the data into two or three groups according the values of
> bedekking could be a feasible alternative.
>
> Hope this helps you,
>
> vito
>
>
>
>
> Il 01/06/2012 11.35, Peter Hoitinga ha scritto:
>>
>> Hello,
>>
>> I'm attempting to find multiple breakpoints in an association of my
>> response variable (R.AUC) with two explanatory variables ('s.size' and
>> 'bedekking'). The association between 's.size' and 'R.AUC' shows a
>> plateau, but the value when this plateau is reached is differs for
>> different values of 'bedekking'.
>>
>> Initially I thought these different values could be assessed by the
>> 'segmented' function in the package of the same name, but this only
>> seems to give one value of a breakpoint for each explanatory variable.
>> A similar issue was treated in the following message from 2006:
>> https://stat.ethz.ch/pipermail/r-help/2006-January/086446.html
>>
>> Does anybody know whether there are any packages that do have a
>> function for this? Or might it be a possibility to split up the data
>> per value of 'bedekking' and then do the segmented function?
>>
>> ______________________________________________
>> [hidden email] 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.
>>
>
> --
> ====================================
> Vito M.R. Muggeo
> Dip.to Sc Statist e Matem `Vianelli'
> Università di Palermo
> viale delle Scienze, edificio 13
> 90128 Palermo - ITALY
> tel: [phone]
> fax: [phone]
> http://dssm.unipa.it/vmuggeo
> ====================================

______________________________________________
[hidden email] 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.