[R] Problems with ks.test

[Patrick Burns]

The distribution of p-values should be uniform under
the null hypothesis.  When I do:

 > jj <- numeric(10000)
 > for(i in 1:10000) jj[i] <- ks.test(rexp(2500, .4), 'pexp', .4)$p.value
Warning messages:
1: cannot compute correct p-values with ties in: ks.test(rexp(2500, 
0.4), "pexp", 0.4)
2: cannot compute correct p-values with ties in: ks.test(rexp(2500, 
0.4), "pexp", 0.4)
3: cannot compute correct p-values with ties in: ks.test(rexp(2500, 
0.4), "pexp", 0.4)
4: cannot compute correct p-values with ties in: ks.test(rexp(2500, 
0.4), "pexp", 0.4)
 > hist(jj, 50, col='yellow'); abline(h=200, col='green')

I get a histogram that looks reasonably flat to me.

Patrick Burns
patrick at burns-stat.com
+44 (0)20 8525 0696
http://www.burns-stat.com
(home of S Poetry and "A Guide for the Unwilling S User")


Emanuele Mazzola wrote:

>Hi everybody,
>
>while performing ks.test for a standard exponential distribution on samples 
>of dimension 2500, generated everytime as new, i had this strange behaviour:
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0147, p-value = 0.6549
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.019, p-value = 0.3305
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0171, p-value = 0.4580
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0143, p-value = 0.6841
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0145, p-value = 0.6684
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0123, p-value = 0.8435
>alternative hypothesis: two.sided
>
>  
>
>>data<-rexp(2500,0.4)
>>ks.test(data,"pexp",0.4)
>>    
>>
>
>	One-sample Kolmogorov-Smirnov test
>
>data:  data
>D = 0.0186, p-value = 0.3532
>alternative hypothesis: two.sided
>
>
>It seems kind of strange to me that max p-value obtained is 0.8435 and all 
>the best i can have from the rest is a 0.66-0.68.
>I'm probably not so expert in running this kind of test, but am I doing 
>something wrong?
>I would expect p values ranging from 0.75 (to be kind) to 0.9, 0.95. How is 
>this possible?
>
>Thank you in advance for your answers.
>See you soon
>EM
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://stat.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>
>
>
>  
>