[R] Anderson-Darling GoF (re-sent)

Shiazy shiazy at gmail.com
Fri May 18 09:53:55 CEST 2007

I'm not a statistician so sorry for possible trivial questions ...

I want to perform a GoF test on sample data against several distribution 
(like Extreme Value, Phase Type, Pareto, ...).

Since I suspect a long-tailed behaviour on data I want to use 
Anderson-Darling (AD) GoF test because it's well known it's more 
sensible to tail data.

Looking at R packages the only AD test is the AD normality test 
("ad.test") in the "nortest" package. So I think this function is not 
for me since long-tailed samples aren't normally distribuited (right?!)

I've found the Marsaglia article ("Evaluating the Anderson Darling 
distribution") where it seems I can consider the ECDF (empirical CDF) 
and the theoretical as a uniformly [0,1] distributed data and then 
perform the test like I had to compare two uniform distribution. The 
problem is the theoretical CDF ( i.e. the parameters of theoretical 
distribution) has been estimated from the data against which I want to 
make the test. I've read somewhere it's not a good technique to compare 
the distribution with the above way because the resulting AD test might 
be biased.

So, finally, I don't know how to proceed ...

Can anyone give me a help or any reference (please remember I'm not a 
statistician so do not write too technically)??

Thanks a lot to everyone!!

-- Marco

More information about the R-help mailing list