Shapiro-Wilk Normality test results are slightly different between the UNIVARIATE procedure and the MODEL procedure


The Shapiro-Wilk test statistic and associated p-value produced by the NORMAL option on the FIT statement in PROC MODEL might be slightly different than the Shapiro-Wilk test statistic and p-value produced by the NORMAL option on the PROC UNIVARIATE statement in releases earlier than SAS® 9.4M0 (TS1M0). They are slightly different because PROC UNIVARIATE uses an updated algorithm that better approximates the coefficients required to compute the W statistic. This new algorithm has been incorporated into PROC MODEL starting from the SAS 9.4M0 release.