Key points are not available for this paper at this time.
Shapiro and Wilk's (1965) Wstatistic has been shown to provide a superior omnibus test of normality (Pearson et al., 1977). It has recently been extended to cope with samples of size up to 2000 (Royston, 1982a). The purpose of the present algorithm is to enable the calculation of Wand its significance level for any sample size between 3 and 2000. The full description of the theory behind this algorithm is given by Royston (1982a). Using Monte Carlo simulation, Royston (1982a) showed that the transformation y = (1-W)2 (1)
J. P. Royston (Fri,) studied this question.