This paper is from Session 4C: Rank-based inference, association measures and nonparametric statistics

which comes under Topic 4: Statistics education at the post-secondary level

Is the real world normal?

Presenter

• Catherine Dehon (Free University of Brussels, Belgium)

Abstract

In recent times, we have become increasingly confronted with high dimensional data sets. Statistical methods have had to adapt themselves to more complex questions from many different scientific disciplines, notably in the social sciences. But in spite of this evolution, statistics courses still rely too often on artificial examples that contribute to the myth that the real world is quite simple. Classical statistics based on parametric models also feature in undergraduate and graduate curricula. Nevertheless apparent deviations of the model cannot always be ignored. For example, we often expect large datasets to contain a small number of unusual observations, which renders classical procedures unreliable. The theory of robust statistics deals with small deviations from the model, and can be viewed as a compromise between parametric and nonparametric analysis. Should we consider introducing these modern concepts into undergraduate statistics courses?