Even though multiverse theories usually involve unobservable elements, they may give testable predictions for observable elements if they include a well defined measure for observations. One can then analyze them by Bayesian means, using the theory-dependent typicality of the result of observations as a likelihood for the theory, though there is still an inherent ambiguity in assigning prior probabilities to the theories.
One can try to avoid specifying the equations or other properties of an individual universe by assuming that there is an ensemble of different universes, but this replaces the question of the equations with the question of the measure for the different universes in the ensemble. There is no apparent way to avoid having non-trivial content in a testable theory fully describing all of reality.
Was this article helpful?