# fizinfo AT lists.kfki.hu

**Subject:** ELFT HÍRADÓ

## List archive

**From**: "Laszlo E. Szabo" <leszabo AT hps.elte.hu>**To**: mafla <mafla AT hps.elte.hu>, Multiple recipients of list <koglist AT cogpsyphy.hu>, fizinfo <fizinfo AT sunserv.kfki.hu>**Subject**: [Fizinfo] Philosophy of Science Colloquium, Tamas Rudas**Date**: Tue Nov 12 01:06:00 2002**List-archive**: <http://sunserv.kfki.hu/pipermail/fizinfo/>**List-id**: ELFT HRAD <fizinfo.lists.kfki.hu>

Department of HISTORY AND PHILOSOPHY OF SCIENCE

Eotvos University, Budapest

Pazmany P. setany 1/A Budapest

Phone/Fax: (36-1) 372 2924

Department's Home Page:http://hps.elte.hu

Philosophy of Science Colloquium

Room 6.54 (6th floor) Monday 4:00 PM

____________________________________

18 November 4:00 PM 6th floor 6.54

Instead of the canceled lecture of 21 October!

Tamas Rudas

Department of Statistics, Institute of Sociology, Eötvös University, Budapest

Measurement and modelling of association in contingency tables

Association between two variables is defined in the talk as the information in their joint distribution not present in the univariate distributions. Therefore, a measure of association, together with the marginal distributions, has to parameterize the joint distribution and has to be variationally independent from the marginals. These requirements point to the odds ratio as the only appropriate measure of association.

For higher dimensional contingency tables, a possible generalization is the system of conditional odds ratios. The conditional odds ratios, on an ascending class of subsets, are variationally independent from the marginal distributions on the complement descending class and together parameterize the joint distribution. Depending on the class of subsets used, one obtains a flexible class of parametereizations that can be used to model the conditional association structure. The models obtained by assuming lack of conditional association on an ascending class of subsets are of the log-linear type.

The concepts discussed in the talk and the analyses based on these concepts suggest that association has a hierarchical structure. The assumption of multivariate normality, routinely applied in our thinking about multivariate data structures, is equivalent to assuming that only

first order interactions exist is therefore, is an oversimplification of reality.

------------------------------------------------------------------------

The 60-minute lecture is followed by a 10-minute break. Then we held a

30-60-minute discussion.

------------------------------------------------------------------------

The organizer of the seminar: László E. Szabó

<http://hps.elte.hu/~leszabo> (email:

leszabo AT hps.elte.hu)

--

L a s z l o E. S z a b o

Theoretical Physics Research Group of the Hungarian Academy of Sciences

Department of History and Philosophy of Science

Eotvos University, Budapest

H-1518 Budapest, Pf. 32, Hungary

Phone/Fax: (36-1)372-2924

Mobil/SMS: (36) 20-366-1172

http://hps.elte.hu/~leszabo

**[Fizinfo] Philosophy of Science Colloquium, Tamas Rudas**,*Laszlo E. Szabo, 11/12/2002*

Archive powered by MHonArc 2.6.19+.