CURRENT RESEARCH TOPICS

‘Buehlerised’ conditional upper limits for the odds-ratio. Conditional upper limits for the log-odds ratio in a 2x2 table can be based on the tilted hypergeometric distribution of the number of successes in group 1 conditional on the last marginal total. These upper limits have optimality properties conditional on this last marginal total. However, evaluated unconditionally they are typically highly conservative. By conservative we mean that there coverage is much higher than nominal and (consequently) they tend to be larger than competing unconditional methods. Buehler (1957) showed how to construct upper limits that are unconditionally valid and also optimal (i.e. minimal) subject to a given ordering. The aim of this project is to use Buehler’s method to adjust the conditional limits so that they are unconditionally minimal. We will compare the results with Buehler’s limits based on the ordering induced by the signed root LR upper limit. (This is joint work with my Doctoral Student, Max Moldovan and a draft version is HERE.) A zipped set of Matlab functions is available HERE. The paper has been accepted by Statistics in Medicine and will appear in 2008.