Acceptance/Rejection Methods for Beta Variate Generation
Bruce W. Schmeiser, Mohamed Ahmed Shalaby · Journal of the American Statistical Association · 1980
The computer generation of pseudorandom variates from the beta distribution is discussed. Three exact methods applicable for parameter values p > 1 and q > 1 are presented. All three are acceptance/rejection methods, based on functions of the location of the mode and points of inflection. The methods differ in the amount of the set-up required, with higher set-up times generally resulting in lower marginal generation times. Generation times are compared to four methods in current use: ratio of gamma variates, and the methods of Jöhnk, Ahrens and Dieter, and Cheng. Over a wide range of parameter values marginal generation times are reduced.