Response to bootstrapping question, from our stat. consultants
Beatrice L. Rogers
BRogers at infonet.tufts.edu
Thu Jun 25 08:54:04 PDT 1998
On Fri, 12 Jun 1998, Robert HouserJr. wrote:
> We recently received a question from a student regarding bootstrapping.
> Her question is as follows ...
> "I have read a study where a procedure called "bootstrapping" to estimate
> a 99% confidence interval is mentioned. Could anybody explain for me
> what "bootstrapping" means?"
> In this context would it be fair to say that bootstrapping involves
> generating a large number of subsamples of the data set so that the true
> 99% confidence interval can be more accurately estimated?
> What can you add/change in my description? Thanks. - Bob
Here is an extended description:
Bootstrap methods are computer-intensive methods of statistical analysis
that use simulation(resampling methods) to calculate standard errors,
confidence intervals and significance tests. The methods apply for any
level of modelling, and can be used for fully parametric, semiparametric
and completely nonparametric analysis. Careful application of the
methodogy is essential. Bootstrap methods do not "fix" problems or
may work poorly in settings associated with incomplete data, dependent
data, dirty data and poorly balanced designs.
> P.S. I've heard that there is a Sage publication on bootstrapping. Have
> you seen it?
No. But is likely a go introduction.
>Is it any good? Also, I heard about a book by Erron &
> Tibshirani (1993) An introduction to the bootsrap. Have you heard of this
> one? Any good? Any recommendations?
Yes. These guys are well known in this area! I have read much of their
research and they have a good handle on the issues. Also of interest is:
Bootstrap Methods and their Application, A.C. Davison and D.V.Hinkley,
Cambridge Series in Statistical and Probabilistic Mathematics,
This is a fine work. Another choice nonmathematical *intro-article* for
Statistical Data Analysis in the Computer Age, B. Efron and R. Tibshirani,
Science, 1991, Vol. 253, pg 390.
Software is another story. The best collection of routines are free but
you have to use them with S-plus. Also, StatXact also does a form of
bootstrapping(sort of) called exact inference. SAS will bootstrap some
things, but the code interface is not easy. Amos will bootstrap Liserl
models. Spss ver. 8 has some resampling as well.
Durwood Marshall-Tufts University
Statistical and Research Computing Consulting
Tufts Computer and Communications Services
617.628.5000 x2180 : Fax 617.627.3667 : E-Mail: dmarshal at tufts.edu
More information about the PHNUTR-L