On computer sampling from trivariate and multivariate discrete distributions
S. Loukas, C. D. Kemp · Journal of Statistical Computation and Simulation · 1983
Sampling methods for computer generation of i.i.d. bivariate discrete random variables given in Kemp and Loukas (1978a, 1981) are extended to the trivariate case, and some comparative timings and storage requirements are given. These indicate that although general methods which use large arrays of constants remain fast, they are usually impracticable because of their storage requirements. It is therefore necessary to use methods which are specific to the particular distribution being sampled and usually involve sampling from several univariate distributions and combining the sampled values in various ways. Storage and set—up problems appear to almost completely rule out the possible use of general methods when sampling from s—variate distributions (s > 3), but the specific methods generalize straightforwardly.