Five highly accurate algorithms that evaluate the normal distribution function are presented. The algorithms, which are based on either numerical integration or series expansions, are explicated, and applications of Simpson's rule are discussed. Computer programs that implement the algorithms were compared with respect to accuracy and speed. The programs attain from 11-to I5-decimal-place accuracy in finding one-tailed areas of the normal curve, and most execute very quickly. Recommendations are made for selection of algorithms and programs to approximate the normal distribution.uses an inequality that provides either the maximum possible error that can occur given a specified number of segments into which f(z) is partitioned, or, following algebraic rearrangement, the number of segments required to mathematically guarantee any specified level of accuracy. The first version of the inequality follows, cited from Anton (1981, p. 534).
M(b-a)5Ierror I~180N4 (I) In these inequalities, N is the number of segments into which an interval of f(z) is divided, and M is the maximum absolute value offiV(z) within that interval; b is. the upper bound, and a is the lower bound. Values ofrV(z) can be obtained from Pearson and Hartley (1972, Table 2) or by evaluating the derivative directly as shown below.The maximum value of /v(z) is 1.196827, occurring atStudy ofInequality 2 provides the keys to efficient numerical integration off(z) by Simpson's rule. Specifically, N is minimized (and a computer program therefore runs most quickly) when the range of numerical integration (i.e., b-a) is short and the fourth derivative of f(z) is small. N is especially reduced when (b -a) < 1, because fractions become smaller when raised to powers larger than 1, and when M < 1. Parenthetically, Simpson's rule requires that N be an even integer; therefore, the value of N is increased to the smallest even integer that is greater than or equal to N as calculated with Inequality 2.Highly accurate computation of the value of the normal integral can be achieved with numerical integration (Cyvin, 1964;Smits, 1981;Wood, 1985Wood, , 1987 and with polynomial and rational functions, continued fractions, and series expansions (Johnson & Kotz, 1970, chap. 13; Kennedy & Gentle, 1980, chap. 5). In this paper, we review five highly accurate normal curve algorithms, three of which are based on numerical integration, and two on series expansions. We also present, explicate, and compare programs that enact these algorithms, and we describe their performance characteristics. In addition, we explore the efficacy of Simpson's rule in finding normal curve areas with 10-and 15-decimal-place accuracy.The accuracy of numerical integration is a function of the number of points at which the integral is evaluated. In routines applying Simpson's rule to evaluate the normal integral in the interval (O,z), where z is a standard normal deviate, Smits (1981) used 1,000 segments to attain lO-decimal-place accuracy, and Wood (1985) used 52 segments to attain 6-decimal-pla...