An Improved Model and Computer Program for Bayesian M-Group Regression
Abstract
This paper presents a modified model and computer algorithm for the estimation of regressions in m-groups. The major features of the new approach are: (1) The use of parameters with constant values for slopes across groups whenever the prior or the data indicate that this is desirable, (2) the use of a residual variance constant across groups, (3) independent priors for regression parameters, (4) scales of the predictors at the grand mean rather than at the so-called 'Ideal scaling points', (5) transformation of all variables including the criterion to mean zero and variance one at the start of the calculations, (6) an improved search procedure for finding the mode of the posterior distribution. This new solution provides a greatly speeded-up algorithm which defines the posterior mode precisely.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1979
- Accession Number
- ADA075393
Entities
People
- Charles Lewis
- Ivo W. Molenaar
Organizations
- University of Iowa