s-news
[Top] [All Lists]

[S] Bootstrap Question

To: s-news@wubios.wustl.edu
Subject: [S] Bootstrap Question
From: Cox.Bill@epamail.epa.gov
Date: Thu, 11 May 2000 16:37:55 -0400
Sender: owner-s-news@wubios.wustl.edu


       I am fitting a generalized model to daily data that extends over a number
of years.  I want to be able to bootstrap the regression coefficients (and
perhaps other stats) within groups (years) but also by blocks of say consecutive
three day periods to preserve serial correlation known to exit.   The data looks
something like this:

   Date       Yr        Ozone         Temp           Wspd      Other variables
   June 1     94           100             72                3.4
........
   June 2     94             85             88                4.3
........
   June 3     94             .                   .                    .
.....
     ........
   Oct 30     99            120            76                4.5
.........

 One form of the model being evaluated is:

  Ozfit <- glm(Ozone ~ ns(temp,5) + ns(wspd,5 ) + .... + ns(yr,5),
family=quasi(link=log),  data= ..  )
I can bootstrap the reg coefficients nicely with

  Ozfit.boot <-     bootstrap(data = ..., statistic = coef(eval(Ozfit$call)), B
= 200, group = yr)

Is there some way I can use the bootstrap function (or the boot function from
statlib), to handle the serial correlation, e.g. bootsrap blocks of days so that
each block of days would be sampled rather than each day?    Any other
suggestions for handling this issue are welcome.  I am using Splus Pro 2000
release 2.  Thanks.


-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message:  unsubscribe s-news

<Prev in Thread] Current Thread [Next in Thread>
  • [S] Bootstrap Question, Cox . Bill <=