scientific article; zbMATH DE number 4007437

From MaRDI portal
Publication:3757136





[For the entire collection see Zbl 0608.00012.] After having asserted that Bayesian statistics is growing rapidly and vigorously, the author maintains that it remains imperative for Bayesians to continue to propagandize their viewpoint effectively and to attempt to improve their arguments. In this perspective, the paper is adressed not to the problem of convincing the newcomers in statistics but instead of that to changing the views of established non-Bayesians, a problem that the author rightly sees to be more difficult than the first one. To this end the author argues that this can be best accomplished by: i. arguing for the need to think conditionally on the data and following the likelihood principle; ii. maintaining that there is no need to presume that prior distributions can be specified completely and accurately; iii. admitting that non-Bayesian techniques are of interest to Bayesians. The author's preference goes to the robust Bayesian approach, that is, he argues that no single prior should be chosen but that one learns by passing various plausible priors over the likelihood function. While not disputing the value of other Bayesian approaches, the author maintains that the robust Bayesian one is more suitable on pragmatic and scientific grounds. The paper is a very good, non dogmatic and convincing work in favour of the Bayesian viewpoint. In the reviewer's opinion, its sole weakness is a tiresome use of terms like sell and sale techniques in treating a theoretical problem.











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 Q3757136)