PHYSTAT: making the most of statistical techniques

8 February 2006

Roger Barlow looks at the development of the PHYSTAT series of meetings, which bring statisticians together with astronomers, cosmologists and particle physicists.

Statistics has always been an essential tool in experimental particle physics, and today this is truer than ever. In the early days emulsions and bubble-chamber photographs were scanned slowly by hand; now modern electronic detectors perform equivalent processing quickly and automatically. However, physicists still classify and count their events and then, eagerly or reluctantly, turn to statistical methods to decide whether the numbers are significant and what results are valid.

As the subject has progressed, new themes have emerged. The high numbers of events obtained by CERN’s Large Electron-Positron collider (a Z-factory), the B-factories PEP-II and KEKB at SLAC and KEK respectively, and the experiments at DESY’s HERA collider, mean that statistical errors below 1% are now common. Many areas have become dominated by systematic effects, a relatively untrodden and much less well understood field.

On the theoretical side, the high precision of theories such as quantum electrodynamics and quantum chromodynamics means that the tiny uncertainties in their predictions have to be carefully studied and understood. Supersymmetry and other “new physics” models predict signals that depend on several parameters of the theory, and when an experiment fails to see such a signal the restrictions this places on possible values for these parameters has to be worked out. When different experiments probe the same basic theory, we need to evaluate the combined implication of their results.

The science of statistics is also developing fast. The availability of large amounts of processing power opens new possibilities for evaluating statistical models and their predictions. Bayesian statistics is a rapidly growing field in which a great deal of progress has been made. Machine learning techniques, such as artificial neural networks and decision trees, are flourishing, with further applications continually being found that open up new possibilities for exploiting data.
Astronomers and cosmologists are also developing the power and sophistication of their statistical techniques. Telescopes are becoming larger and more powerful, and the readout from their instruments with charge-coupled detectors produces a torrent of data. Observations at different wavelengths, from gamma rays to radio waves, from ground-based observatories and satellites are combined to yield clues about the nature of distant objects, the processes that power them and other features of the universe. Details of the distribution of the cosmic microwave background will, when properly interpreted, tell us what happened in the Big Bang at energies beyond the reach of man-made accelerators.

The PHYSTAT series of conferences and workshops provide a forum in which different communities can meet and exchange ideas. A first workshop of particle physicists at CERN in 2000 was followed by one at Fermilab in 2001, and then a full conference in Durham in 2002, which benefited from the presence of statisticians as well as physicists. At SLAC in 2003, astronomers and cosmologists were included (see CERN Courier March 2004 p22). This was so successful that it was repeated at the most recent conference, “Statistical Problems in Particle Physics, Astrophysics and Cosmology”, held in Oxford in September 2005 and organized by Louis Lyons.

PHYSTAT 2005 consisted of a wide-ranging programme of parallel and plenary talks. One of the most influential statistical thinkers of the 20th century, David Cox of Oxford University, gave the opening keynote speech, in which he provided an authoritative account of the Bayesian and frequentist approaches to inference. The official programme was supplemented by intense discussions in corridors, coffee lounges and local pubs, as the participants thrashed out ideas that ranged from the philosophical abstractions of the meaning of probability to the pragmatic and technical details of different computer systems.

These techniques are being fed back into the community through the activities of the participants, many of whom are active in analysis on various different experiments, through further meetings (a follow-up afternoon meeting in Manchester attracted 80 particle physicists from the UK), through the academic training programmes offered at CERN and other laboratories, and through graduate conferences and summer schools. There are developing plans for a repository of software that performs these increasingly sophisticated statistical tests. Further workshops are planned for 2006 and beyond.

Further reading

More information can be found at and at


Copyright © 2020 by CERN
bright-rec iop pub iop-science physcis connect