The paper examines seven methods of numerical integration, including
both special purpose algorithms designed for the multivariate normal density
and general algorithms such as Gauss—Legendre and Newton—Cotes methods.
With the aid of some five functions, the accuracy of these methods and their
computational cost are compared in matched experiments on an IBM 370/3081
Model K and a 2-pipe CYBER 205. The effect of vectorised computation is
also examined.