Strange parametrization in polr
The problem is not in the book nor the code, but in Mr Carstensen not looking up the actual reference given in ?polr. There was a + in early printings of MASS3, and that difference is in the on-line Errata. But both the DESCRIPTION file and ?polr are explicitly to the fourth edition. As Thomas says, the minus seemed a more natural parametrization and so we changed to it.
On Thu, 8 Jan 2004, Thomas Lumley wrote:
On Thu, 8 Jan 2004, BXC (Bendix Carstensen) wrote:
In Venables \& Ripley 3rd edition (p. 231) the proportional odds model is described as: logit(p<=k) = zeta_k + eta but polr apparently thinks there is a minus in front of eta, as is apprent below. Is this a bug og a feature I have overlooked?
If there is really a bug I would guess that it was in the book rather than the code. This is not an unusual parametrisation for this model. It is the parametrisation that reduces to logistic regression for binary data, and makes the regression coefficients positive when the association is positive.
Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595