s-news
[Top] [All Lists]

[S] Homogeneity test

To: S-news@wubios.wustl.edu
Subject: [S] Homogeneity test
From: Sixten.Borg@astrazeneca.com
Date: Fri, 15 Sep 2000 13:32:28 +0200
Sender: owner-s-news@wubios.wustl.edu

        Dear all,

This is probably a basic statistics question, I would none the less like to
ask for any hints as to choice of method / test.

I want to analyze a study of events taking place in patients subjected to
either of two interventions, with patients enrolled in any of five
countries. I regard my data as a table of one row for each country, and one
column for each intervention. In each table cell, a number of patients
reside which either have experienced {zero events} or {one or more events}.

In each cell, the data x_{country,intervention} is considered to be
Binomial(n = n_{country,intervention}, p = p_{country,intervention}).

I would like to perform a homogeneity test, to test the hypothesis of p
being independent of country.

There is no obvious reason to believe in a difference between countries, but
there may be evidence of a difference between interventions.

A chi-square test sounds like the obvious solution, but I have not managed
find any formulation of the chi-square test that suits my data.

May plan is to perform some kind of permutation test, with a test variable Q
= Sum ( (observed - expected)^2/expected), where 
observed = x_{country,intervention},
expected = n_{country,intervention} * p_{intervention},
p_{intervention} = estimate of the probability of an event under the given
intervention, using all countries' data.

In the permutation test, the estimate of p_{intervention} is held fixed at
its observed value, and the country labels are permuted, giving a
distribution Q*. Comparing the observed Q with the tails of Q*,  a p value
is estimated.

I feel however that I am probably overlooking an obvious, and much simpler
solution to this.

I am grateful to any hints or advice,

Kind regards,
Sixten Borg.
============================================
Sixten Borg
Statistician / Modeller
Health Economics & Outcomes Research
Clinical Science
AstraZeneca R&D Lund, S-221 87  Lund, Sweden
Phone +46 46 33 77 13.  Fax +46 46 33 75 53.
Email: Sixten.Borg@astrazeneca.com

-----------------------------------------------------------------------
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] Homogeneity test, Sixten . Borg <=