On Fri, 06 Dec 2002 16:04:50 +0100, "Koen Vermeer" <[EMAIL PROTECTED]> wrote:
> On Fri, 06 Dec 2002 09:24:16 +0000, Herman Rubin wrote: > > >>I am not very familiar with the Anderson-Darling test. I only know it is > >>an alternative to KS and it is not distribution free. The last thing was > >>the mean reason I wasn't considering it for the moment. Maybe I'll have a > >>look at it in th near future. > > The Anderson-Darling test is as distribution free as the > > Kolmogorov-Smirnov. > > Hmm, I thought I read somewhere that the KS test statistic is independent of the > distribution that is tested against. Maybe I'm mistaken... The URL below has a description that confirms the thought. Their modification of the K-S makes use of particular distributions to make a more powerful test. http://www.itl.nist.gov/div898/handbook/eda/section3/eda35e.htm Below is a comprehensive posting to sci.stat.math by "Ken K" on April 25, 2001. It cites the address above, and it quotes D'Agostino & Stephens. This is the main text of his post, which I found by googling in the sci.stat.* groups for Anderson-Darling. ============= start of post Try the NIST/Sematech On-line Engineering Statistics Handbook, and specifically the EDA section. Here is a link to the main handbook, which is pretty nice: http://www.itl.nist.gov/div898/handbook/index.htm The EDA section is listed under the Explore link. If you go to http://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm you'll find near the bottom of the page links to explanations covering the Anderson-Darling, Chi-Square, and Kolmogorov-Smirnov tests. And a word of warning from D'Agostino & Stephens' "Goodness-of-Fit Techniques" book (published by Marcel-Dekker, Inc.), page 406: "6. For testing for normality, the Kolmogorov-Smirnov test is only a historical curiosity. It should never be used. It has poor power in comparison to the above procedures**. 7. For testing normality, when a complete sample is available the chi-square test should not be used. It does not have good power when compared to the above tests**." **Shapiro-Wilks W test, D'Agostino-Pearson K^2 test, Bowman-Shenton version K(2,S), and Anderson-Darling edf test A^2 - also the R test and D'Agostino's D test. =========== end of Ken's comments. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
