R help - Dec 2003 - bootstrap pValue in DClusters

Hello R-List

I use DClusters package (I work in a cancer regestry). I have 2 questions
about it:

1-how is it possible to get back the bootstrap pValue? I mean the pValue of
the calculated statistic with respect of the distribution of this statistic
under the null hypothesis.

2-how is it possible to test an overdispersion in the poisson model? for
choosing a best model I need this mesure of dispersion. Should I build a
glm(cases~expected,family=quasipoisson)$sig2 or is it possible directly in
DClusters?

Thank you, for advance for answers. Best regards, Erik A.


===========================================Erik-André SAULEAU

SEAIM
Hôpital du Hasenrain
BP 1070
87, Avenue de Altkirch
68051 MULHOUSE Cédex

Tel: 03-89-64-79-95
Fax: 03-89-64-79-71
Mél: sauleauea@ch-mulhouse.fr
Web: http://www.ch-mulhouse.fr
===========================================



************************************************************************************
Afin d'éviter toute propagation de virus informatique, et en complément des
dispositifs en place, ce message (et ses pièces jointes s'il y en a) a été
automatiquement analysé par un antivirus de messagerie.
***********************************************************************************


	[[alternative HTML version deleted]]

Salut Erik,

Apologyses for not writing to you before.


> 1-how is it possible to get back the bootstrap pValue? I mean the
> pValue of the calculated statistic with respect of the distribution of
> this statistic under the null hypothesis.
Let's suppose that you perform 100 bootstrap replicates. Then you should
sort this 100 values and see how many of them are higher than your
observed value. If we call N to this quantity, then the p-value is
N/100.
> 2-how is it possible to test an overdispersion in the poisson model?
> for choosing a best model I need this mesure of dispersion. Should I
> build a glm(cases~expected,family=quasipoisson)$sig2 or is it possible
> directly in DClusters?
We plan to incorporate Dean's test in the near future:

@article{Dean:1992,
   author    = {C. B. Dean},
   title     = {Testing for Overdispersion in Poisson and Binomial
Regression Models},
   journal   = {Journal of the American Statistical Association},
   pages     = {451-457},
   year      = {1992},
   volume    = {87},
   number    = {418}
}

In fact, we already have the code but not the manual page.

Pearson's chi-square and Potthoff-Wittinghill can be used to test data
homogeneity, to see any departure from the Poisson distribution (which
MAY be due to overdispersion).

For those interested in package DCluster,  I hope to resume its
development soon. Probably next week.

Best regards,

-- 
             Virgilio G?mez Rubio

Grup d'Estad?stica espacial i temporal 
en Epidemiologia i medi ambient 

Dpto. Estad?stica e I. O. - Facultat de Matem?tiques
Avda. Vicent A. Estell?s, 1 - 46100 Burjassot
Valencia - SPAIN

http://matheron.uv.es/~virgil

TLF: 00 34 96 354 43 62 - FAX: 00 34 96 354 47 35