On 15 Nov 2003 at 17:53, Albert le Curieux wrote: This souns strange. The Shapiro-Wilk test observator is essentilly the Pearson correlation coefficient in the quantile- quantile normal plot, so measures linearity in this plot. You would reject normality on the base of a too high correlation coefficient only if you had some suspicion that the data was cooked up to look normal.
Kjetil Halvorsen > Hi, > I read in a book (Modelisation et estimation des erreurs de mesure, > publishef by M. Neuilly et CETAMA, France) that the Shapiro and Wilks > test of normality is bilateral. It looks strange for me as we know > that the W statistic is a ratio of two estimations of the variance > equal to 1 in case of normality. Kendal and Stuart precise that > critical values are small values. > Is it possible that, for this test, there exist (as tells the > book), > depending on alpha and n, two constants a and b, for which, when W > belongs to (a,b) we dont't refuse normality normality, and when W is > outside this interval, we refuse normality? ( 0<a< b<1 ) > Thank for your response > > > . > . > ================================================================= > Instructions for joining and leaving this list, remarks about the > problem of INAPPROPRIATE MESSAGES, and archives are available at: . > http://jse.stat.ncsu.edu/ . > ================================================================= . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
