Spectroscopic confirmation of galaxies at z ~ 7 and above has been extremely difficult, owing to a drop in intensity of Ly alpha emission in comparison with samples at z ~ 6. This crucial finding could potentially signal the ending of cosmic reionization. However, it is based on small data sets, often incomplete and heterogeneous in nature. We introduce a flexible Bayesian framework, useful to interpret such evidence. Within this framework, we implement two simple phenomenological models: a smooth one where the distribution of Ly alpha is attenuated by a factor epsilon sub(s) with respect to z ~ 6 and a patchy one where a fraction epsilon sub(p) is absorbed/non-emitted while the rest is unabsorbed. From a compilation of 39 observed z ~ 7 galaxies, we find epsilon sub(s) = 0.69 + or - 0.12 and epsilon sub(p) = 0.66 + or - 0.16. The models can be used to compute fractions of emitters above any equivalent width W. For W > 25 Angstrom, we find X super(25) sub(z = 7) = 0.37 + or - 0.11 (0.14 + or - 0.06) for galaxies fainter (brighter) than M sub(UV) = -20.25 for the patchy model, consistent with previous work, but with smaller uncertainties by virtue of our full use of the data. At z ~ 8 we combine new deep (5sigma flux limit 10 super(-17) erg s super(-1) cm super(-2)) Keck/NIRSPEC observations of a bright Y-dropout identified by our Brightest of Reionization Galaxies Survey, with those of three objects from the literature and find that the inference is inconclusive. We compute predictions for future near-infrared spectroscopic surveys and show that it is challenging but feasible to constrain the distribution of Ly alpha emitters at z ~ 8 and distinguish between models.