[Rd] ks.test calculations incorrect (PR#7330)

[Peter Dalgaard]

tavery at mun.ca writes:
 
> Full_Name: t. avery
> Version: 2.0.0
> OS: windows xp / Linux
> Submission from: (NULL) (131.162.134.159)
> 
> 
> ks.test does not produce the correct output.
> If given the script:
> 
> d1 <- c(53.63984674,0.383141762,1.915708812,0.383141762,10.72796935,6.896551724,20.30651341,5.747126437,0)
> d1
> 
> d2 <- c(76.43312102,15.2866242,3.821656051,1.27388535,0,0.636942675,1.27388535,0.636942675,0.636942675)
> d2
> 
> ks.test(d1,d2,alternative="two.sided",exact)
> 
> The output generates:
>         Two-sample Kolmogorov-Smirnov test
> 
> data:  dd and ddd 
> D = 0.3333, p-value = 0.6994
> alternative hypothesis: two.sided 
> 
> Warning message: 
> cannot compute correct p-values with ties in: ks.test(dd, ddd, alternative =
> "two.sided", exact)
> 
> Which is far from the correct D value of:
> 
> D = 0.404934475 with a p-value < 0.01

According to what definition? Referring to what distribution?
Reference?

Considering that D is the distance between two CDFs, both of which
take values in (0:9)/9, I do find it a bit unconvincing that you claim
that D should be something that is not a whole number of ninths...
 
> I recommend that noone use this test. I have tested it with various datasets
> from publications and texts using by-hand calculations.

Well, I'm no fan of the KS test either, but you have to work harder to
convince us that there's a problem in R's implementation!

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907