Bayesian analysis of Bell inequalities

Statistical tests are needed to determine experimentally whether a hypothetical theory based on local realism can be an acceptable alternative to quantum mechanics. It is impossible to rule out local realism by a single test, as often claimed erroneously. The ``strength'' of a particular Bell inequality is measured by the number of trials that are needed to invalidate local realism at a given confidence level. Various versions of Bell's inequality are compared from this point of view. It is shown that Mermin's inequality for Greenberger-Horne-Zeilinger states requires fewer tests than the Clauser-Horne-Shimony-Holt inequality or than its chained variants applied to a singlet state, and also than Hardy's proof of nonlocality.