Bayesian versus frequentist upper limits
Abstract
While gravitational waves have not yet been measured directly, data analysis from detection experiments commonly includes an upper limit statement. Such upper limits may be derived via a frequentist or Bayesian approach; the theoretical implications are very different, and on the technical side, one notable difference is that one case requires maximization of the likelihood function over parameter space, while the other requires integration. Using a simple example (detection of a sinusoidal signal in white Gaussian noise), we investigate the differences in performance and interpretation, and the effect of the "trials factor", or "look-elsewhere effect".
Cite
@article{arxiv.1103.2987,
title = {Bayesian versus frequentist upper limits},
author = {Christian Röver and Chris Messenger and Reinhard Prix},
journal= {arXiv preprint arXiv:1103.2987},
year = {2015}
}
Comments
http://cdsweb.cern.ch/record/1306523/files/CERN-2011-006.pdf, http://indico.cern.ch/materialDisplay.py?contribId=52&materialId=paper&confId=107747