Skip to content

restricted bootstrap

5 messages · Grant Gillis, Brian Ripley, ONKELINX, Thierry

#
I see nothing here to do with the 'bootstrap', which is sampling with 
replacement.

Do you know what you mean exactly by 'randomly sample'?  In general the 
way to so this is to sample randomly (uniformly, whatever) and reject 
samples that do not meet your restriction.   For some restrictions there 
are more efficient algorithms, but I don't understand yours.  (What are 
the 'rows'?  Do you want to sample rows in space or xy locations?  How 
come 'dist' is not symmetric?)  For some restrictions, an MCMC sampling 
scheme is needed, the hard-core spatial point process being a related 
example.
On Wed, 3 Sep 2008, Grant Gillis wrote:

            

  
    
#
Grant,

Have you considered a gls model instead of a lm model? In a gls model
one can model the correlation between the measures. So you won't need to
select a subset of your data. You can kind gls in the nlme package.

HTH,

Thierry


------------------------------------------------------------------------
----
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature
and Forest
Cel biometrie, methodologie en kwaliteitszorg / Section biometrics,
methodology and quality assurance
Gaverstraat 4
9500 Geraardsbergen
Belgium 
tel. + 32 54/436 185
Thierry.Onkelinx op inbo.be 
www.inbo.be 

To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to
say what the experiment died of.
~ Sir Ronald Aylmer Fisher

The plural of anecdote is not data.
~ Roger Brinner

The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of
data.
~ John Tukey

-----Oorspronkelijk bericht-----
Van: r-help-bounces op r-project.org [mailto:r-help-bounces op r-project.org]
Namens Grant Gillis
Verzonden: donderdag 4 september 2008 14:57
Aan: r-help op r-project.org
Onderwerp: Re: [R] restricted bootstrap

Hello Professor Ripely,

Sorry for not being clear.  I posted after a long day of struggling.
Also
my toy distance matrix should have been symmetrical.

Simply put I have spatially autocorrelated data collected from many
points.
I would like to do a linear regression on these data.  To deal with the
autocrrelation I want to resample a subset of my data with replacement
but I
need to restrict subsets such that no two locations where data was
collected
are closer than Xm apart (further apart than the autocrrelation in the
data).

Thanks for having a look at this for me.  I will look up the hard-core
spatial point process.

Grant

2008/9/4 Prof Brian Ripley <ripley op stats.ox.ac.uk>
the way
samples
more
'rows'?
is
needed,
a
would
that
autocorrelation.  I
times
stuck on
restriction
http://www.stats.ox.ac.uk/~ripley/<http://www.stats.ox.ac.uk/%7Eripley/>
______________________________________________
R-help op 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.

Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document.%CRLF%The views expressed in  this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document%CRLF%
1 day later
#
On Thu, 4 Sep 2008, Grant Gillis wrote:

            
That is impossible.  Resampling with replacement will give duplicated
locations (with a very high probability) and those have distance zero.

If you want a subsample (necessarily without replacement) you have a 
hard-core point process on a discrete set.  It's possible that the MCMC 
methods we used for Strauss processes can be made to work in that case, 
but it is also possible that the state space is reducible and so more 
elaborate algorithms are needed.

I do think it would be much easier to take autocorrelation into account in 
your linear model fit.  There are many ways to do that, e.g. MASS::lm.gls, 
and in fact uless the correlations are very high OLS is likely to be quite 
efficient (but you need to use e.g. a sandwich estimator to get reliable 
standard errors).