Harold Jeffreys was a brilliant statistician, but a terrible lecturer and a terror on a bicycle. Prof Cox's first encounter with Richard Doll was a telling off for having his shoelaces untied, and Bradford Hill* was a kind soul who took the trouble to issue a simple "Well done!" to a nervous Cox presenting his first paper, which Hill "could not have had any interest in!".
With a mix of anecdotes and history, Prof David Cox held the relaunched Oxford local group of the Royal Statistical Society enthralled on 14 October 2015. Over 100 academics and members of the public with an interest in statistics gathered to hear about the history, present, and future of statistics.
Harold Jeffreys, the eminent statistician, was still cycling around Cambridge at 95 - and getting into numerous accidents. He only stopped when his wife Bertha took the tyres off his bike. The couple wrote the book Methods of Mathematical Physics, which is idiosyncratic and ideal... - David Cox
Prof Cox cautioned against the "dangerous illusion" of a theory of statistics built on a neat set of axioms. Statistical problems come with too many issues and details for one simple set of axioms to apply to every problems. He suggested that statisticians concentrate on getting the initial concepts in a problem correct, rather than worrying about what mathematics will be needed next.
Prof Cox then used a controversial randomised controlled trial on bovine tuberculosis interventions on badgers to illustrate the ideal way to approach a problem with statistics: Statistics ideas were used to carefully plan the study and hold randomisation above reproach. Clear initial hypotheses were tested. And when the results disagreed with the hypotheses, a secondary study was conceived to determine why.
In his discussion of the future of statistics, Prof Cox touched on 'big data', which he assured us has been in play since the beginning of his long research career! Still, he acknowledged the need for ways to deal with estimating precision in big data.
Prof Cox's final exhortation to the research students in the audience was well-received: Ask questions your supervisor cannot answer. When your supervisor suggests one analysis method, use a different method, a better method. And by better we mean simpler and more intuitive.
Join us next time
*An earlier version of this article exchanged Richard Doll and Bradford Hill