Year: 1993
Type: preprint
Abstract: This paper shows new ways to discriminate between regression models.The most appealing aspect is that the criteria come from all possible prediction exercises one can imagine, but may be computed directly from the residual sums of squares.The computational clue is the Binet-Cauchy theorem from linear algebra, whlch points the way to new possibilities, encompassing several recent proposals.The context is Bayesian with noninformative priors, but the basic results are also of interest from a Classical viewpoint.The use of predictions allows to derive Bayes Factors circumventing the problem that noninformative priors lead to probability statements of incomparable dimensions.These Bayes Factors perform well in simulation studies, but a number of questions remains.
Author: A.F. Vos
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1544156720