One crazy thing about period fitting (which we are implementing because we are searching for exoplanets in the Kepler data) is that for every period you discover, there is a chance you are only discovering it because it is twice, or three times, or five times the fundamental period. Today, Foreman-Mackey and I worked on ruling out integer fractions of the best-fit period. We came up with some nice heuristics; there is a chance we can find new exoplanets in Kepler without too many false positives. More testing is needed! On a walk in the park with Marshall (visiting this week) and Fadely, we realized that what we are doing isn't even frequentist. Following along discussions we started on twitter, we should entitle our paper on the subject "Incorrect" or "Approximately Frequentist" exoplanet discovery!
Any heuristic can be made frequentist by doing simulations and evaluating the frequency properties of whatever you're doing, as a function of the true period.
ReplyDeleteYou can even make it Bayesian by coming up with a heuristic posterior-probability recipe and saying that it's just the conditional part of a Bayesian model.
Any ad-hoc procedure can be interpreted in these ways! BTW say hi to Marshall for me, and remind him of the recent cricket results.