Dear Rob,
Rob Knell ha scritto:> Hello everyone
> Not strictly an R question but close... hopefully someone will be able
> to help. I wish to compare the switchpoints in two switchpoint
> regressions. The switchpoints were estimated using the segmented
> library
It's a package.
running in R, and I have standard errors for the estimates.
I> initially thought I could just bootstrap confidence intervals for the
> difference between the switchpoints, but I have been having trouble
> with getting this to work because for about 25% of the bootstrap
> samples the algorithm in segmented fails to converge. So I had another
> think, and I thought that maybe I could just do a t-test: knowing the
> estimated switchpoints and their
> standard errors I can easily calculate the SE of the difference, so I
> can calculate a t-value using that. My question is whether there is
> anything wrong with doing it this way, and if not, how many degrees of
> freedom should I use? I would guess at df=n1-5+n2-5 5 df lost for each
> sample because two slopes, two intercepts and one switchpoint have
> been estimated, but I'm not sure: I'm but a humble biologist and
not
> very good at this sort of thing.
The SE() of the breakpoints are reliable only for large samples and/and
with clear-cut relationships. Having said that, I think that you can
compare them by using the approximate t-like studentized statistic, but
results have to taken with care.. The relevant sampling distribution is
unknown (it depends on the location of the breakpoint, (standardized)
difference-in-slope, sample size and, to some extend, distribution of
the `segmented variable') . Therefore quantiles of a t distributions are
not appropriate in theory, but in practice someone uses them..BTW the
model parameters are 4 (intercept+two slopes+ breakpoint, the regression
lines are joined at the breakpoint).
Hope this helps you,
best,
vito
>
> Any help gratefully received
>
> Thanks
>
> Rob Knell
>
>
>
>
> School of Biological and Chemical Sciences
> Queen Mary, University of London
>
> 'Phone +44 (0)20 7882 7720
> Skype Rob Knell
>
> Research: http://webspace.qmul.ac.uk/rknell/
>
> "The truth is that they have no clue why the beetles had horns,
it's
> the researchers who have sex on the brain and everything has to have a
> sexual explanation. And this is reasearch?!" Correspondent known as
> FairOpinion on Neo-Con American website discussing my research.
>
>
>
>
> [[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.
>
>
--
===================================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