Tamás Rudas

Title: Effects and interactions (Wednesday 14:00-15:00)

Abstract: Not only are association and causation different, but as the talk
demonstrates, effects (causation) and interactions (association) need to
be measured differently. Therefore, the conventional wisdom of
interpreting association between two variables, in the case when one
variable cannot have an effect on the other one, as if it established the
existence of an effect of the second one on the first, is often wrong. The
relationship between ways of measuring effects and interactions depends on
whether the analysis is carried out in an experimental or in an
observational setting and the relevant questions and the implied measures
will be discussed. To explain effects and interactions among variables,
other variables are often taken into account. This is a useful practice,
but conditioning (selection) and marginalizing (mixing) may have
counter-intuitive consequences, as illustrated by the paradoxes of Berkson
and Simpson. It will be shown, that there is very limited possibility of
designing measures of effects or interactions which avoid these paradoxes. 

Tamás Rudas is Professor of Statistics and Head of the Department of Statistics at the RTbEötvös Loránd University, Budapest. He is also an Affiliate Professor in the Department of Statistics at the University of Washington. His main accomplishments include being an elected Fellow of the European Academy of Sociology and serving currently as the President of the European Association of Methodology. He has received degrees in mathematics and sociology, including a DSc from the Hungarian Academy of Sciences. His main research contributions are in the area of testing model fit for categorical data and in the theory of marginal modeling. In addition to these topics, he also works on methods of treatment selection.