scientific article; zbMATH DE number 1251892

From MaRDI portal
Publication:4229308





From 1981 to 1985 a seminar part of the Ph.D. series Cognitive processes at the University René Descartes was devoted to statistical inference. The present book is the outgrowth of this seminar. It consists of three major parts. The first part (``Statistics for researchers and ``Statistical practice revisited by \textit{Henry Rouanet}) is devoted to a critical discussion of current statistical practice. The second part (``And... What about the researcher's point of view? by \textit{Marie-Paule Lecoutre}) presents experiments done among researchers. The third part (``Introduction to combinatorial inference by \textit{Henry Rouanet} and \textit{Marie-Claude Bert}; ``From significance tests to fiducial Bayesian inference by \textit{Bruno Lecoutre}; ``Bayesian inference for categorized data by \textit{Jean-Marc Bernard} and ``Geometric data: From Euclidean clouds to Bayesian MANOVA by \textit{Henry Rouanet}, \textit{Brigitte Le Roux}, \textit{Jean-Marc Bernard} and \textit{Bruno Lecoutre}) introduces set-theoretic inference, then Bayesian and fiducial-Bayesian inference. This book is directed to researchers in behavioural and social sciences (in the first place) who use statistical procedures. The book is interesting because of the junction of cognitive psychology and methodological statistics. It contains both experimental work done with researchers using statistical analysis and concrete proposals to renew statistical practice. The methods developed by the authors are fully implemented by software. The authors call for ``restructuring statistical practice. They write: ``By calling for a renewal of statistical practices, we do not invite researchers to renounce the techniques they are familiar with; what we rather suggest is to utilize them in a more pertinent way on the one hand, and to complement them by more efficient methods on the other hand. In brief, there are two sorts of situations where a renewal is desirable. Firstly, there are situations where the conventional probabilistic model - random sampling, normal distribution, etc. - either is hardly reasonable or rests upon unverifiable assumptions. In such situations, the set-theoretic framework, which is a straight extension of descriptive methods, will permit one to preserve the familiar algorithms, while evaluating the generalization potential of data in ~nonprobabilistic terms, such as the typicality of one data set with respect to a reference population, or the homogeneity of several groups of observations. Secondly, there are situations where the conventional model appears reasonable; then the usual procedures leading to conclusions in terms of significant vs nonsignificant effects can be retained and usefully complemented by Bayesian procedures, permitting further conclusions in terms of large vs small effects.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4229308)