“which may assume bizarre shapes, as happens when a variety of different experiments search for neutrino oscillations”. Credit: CERN Courier May 2000 p17
On 16 January, physicists and statisticians met in the CERN Council Chamber to celebrate 25 years of the PhyStat series of conferences, workshops and seminars, which bring together physicists, statisticians and scientists from related fields to discuss, develop and disseminate methods for statistical data analysis and machine learning.
The special symposium heard from the founder and primary organiser of the PhyStat series Louis Lyons (Imperial College London and University of Oxford), who together with Fred James and Yves Perrin initiated the movement with the “Workshop on Confidence Limits” in January 2000. According to Lyons, the series was to bring together physicists and statisticians, a philosophy that has been followed and extended throughout the 22 PhyStat workshops and conferences, as well as numerous seminars and “informal reviews”. Speakers called attention to recognition from the Royal Statistical Society’s pictorial timeline of statistics, starting with the use of averages by Hippias of Elis in 450 BC and culminating with the 2012 discovery of the Higgs boson with 5σ significance.
Lyons and Bob Cousins (UCLA) offered their views on the evolution of statistical practice in high-energy physics, starting in the 1960s bubble-chamber era, strongly influenced by the 1971 book Statistical Methods in Experimental Physics by W T Eadie et al., its 2006 second edition by symposium participant Fred James (CERN), as well as Statistics for Nuclear and Particle Physics (1985) by Louis Lyons – reportedly the most stolen book from the CERN library. Both Lyons and Cousins noted the interest of the PhyStat community not only in practical solutions to concrete problems but also in foundational questions in statistics, with the focus on frequentist methods setting high-energy physics somewhat apart from the Bayesian approach more widely used in astrophysics.
Giving his view of the PhyStat era, ATLAS physicist and director of the University of Wisconsin Data Science Institute Kyle Cranmer emphasised the enormous impact that PhyStat has had on the field, noting important milestones such as the ability to publish full likelihood models through the statistical package RooStats, the treatment of systematic uncertainties with profile-likelihood ratio analyses, methods for combining analyses, and the reuse of published analyses to place constraints on new physics models. In regards to the next 25 years, Cranmer predicted the increasing use of methods that have emerged from PhyStat, such as simulation-based inference, and pointed out that artificial intelligence (the elephant in the room) could drastically alter how we use statistics.
Statistician Mikael Kuusela (CMU) noted that Phystat workshops have provided important two-way communication between the physics and statistics communities, citing simulation-based inference as an example where many key ideas were first developed in physics and later adopted by statisticians. In his view, the use of statistics in particle physics has emerged as “phystatistics”, a proper subfield with distinct problems and methods.
Another important feature of the PhyStat movement has been to encourage active participation and leadership by younger members of the community. With its 25th anniversary, the torch is now passed from Louis Lyons to Olaf Behnke (DESY), Lydia Brenner (NIKHEF) and a younger team, who will guide Phystat into the next 25 years and beyond.
