[R] Implementing Jeffreys Invariant Prior to GLM

[Prof Brian Ripley]

On Mon, 30 Aug 2004, Martin Plöderl wrote:

> Hello!
> 
> In logistic regression I have a rare event as the dependent variable,
> i.e. the "successes" appear with a frequency of ca. 3%.
> In a Paper by King & Zeng "Logistic Regression in Rare Events Data" 
> (http://gking.harvard.edu/stats.shtml )
> a prior-correction method and/or a weighting procedure is suggested to 
> reduce the bias resulting from such analysis.

Bias of what?  
This is hardly news and is in all good books on classification, for example.

> The authors state that using Jeffreys invariant prior also works.
> Does anybody know who to implement this in R?

Take a look at package brlr (`who' being David Firth) for one approach.

-- 
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