The Human Side

I've talked a lot about particle physics and astrophysics, but one thing I haven't discussed much is the human side of an experimental hard science. In medical research, you've heard of the double blind study, in which neither the patients nor the doctors involved know who is in the testing or control groups, to eliminate or minimize the role of the placebo effect in the study. A less-known fact is that other experiments, even those with no human subjects, still have the possibility of human bias. It's called experimenter's bias, and it usually appears in the form of an analyst who unconsciously chooses various parameters to bring the final result closer to some expectation, whether that is in agreement with or contrary to theory. For instance, in particle physics, an analysis typically entails choosing a threshold and the values of several other parameters, which can affect the final result in subtle ways. If it is clear to the experimenter what the effect of each parameter is, then the analysis can be tuned to obtain certain results.

Experimenter's bias can be counteracted by a variety of blinding techniques, very much dependent on the experiment. Some experiments, like both CMS and ATLAS's searches for the Higgs, block out the region of interest and allow the analysis to develop based on the remaining data, un-blinding only when the analysis methods are finalized. Others multiply certain results by either 1 or -1 (chosen randomly experiment-wide), so that analysts fiddling with parameters can't tell whether those parameters increase or decrease the final result.

Muon g-2 (it had to come back to this experiment eventually, right?) is aiming to measure the anomalous magnetic moment of the muon, which can be found as the ratio of two values, a spin precession frequency called $\omega_a$ and the value of the magnetic field throughout the storage ring. As a result, the experiment's collaboration (that is, all the people working on it) is divided into three groups: the accelerator group, which deals with providing the muon beam to the storage ring, the field team, which tackles the magnetic field measurement, and the detector team, which deals with the detectors necessary to measure the precession frequency and monitor the beam. The final analysis will involve the measurement of $\omega_a$ and the magnetic field, and in order to ensure no bias in the analysis, these are blinded measurements. In this case, the way that the analysis is blinded is by a slight offset in the clocks provided to the two components of the experiment. In the naive case, or the final analysis (after blinding is removed), it is crucial that all of the systems are precisely synchronized. But in the preliminary analyses, when all the parameters are still being calibrated and the analysis finalized, the clock sent to the detector team has a frequency of, for instance 45.000xxx (where those x's are hidden values), and the clock sent to the field team has a frequency of perhaps 52.670yyy (again, the y's are hidden values) so the actual numbers obtained have no bearing on the final result. The hidden values, then, are only revealed when the analysis is finalized shortly before the result is released. It's fascinating how such a technically complex experiment still has to account for human bias. That's science.