K-sample Anderson-Darling test

[Eric Chassande-Mottin]

hi,

I have coded the K-sample Anderson-Darling test [*].

It tests if the populations of several (more than two) random samples
are drawned from the same distribution. This test is an extension of
the Kolmogorov-Smirnov test and it is known to be more sensitive
to discrepancies in the tails of the distribution.

The implementation follows exactly [*] and it only provides the "continuous"
distribution case. Note that the compensation given in [*] for the discrete case
(with ties), is not provided here.

adt.m execute three basic checks of ad.m

Ref: [*] K-sample Anderson-Darling tests, Schotz and Stephens
          Journal of American Statistical Association, p 918, 1987.

cheers,
eric.



_______________________________________________
Octave-sources mailing list
Octave-sources@...
https://www.cae.wisc.edu/mailman/listinfo/octave-sources

[dbateman3]

Eric Chassande-Mottin wrote:

> I have sent this message to the octave-sources mailing list twice yesterday
> but somehow the two copies haven't gone through. I send it one more time
> with a copy to help-octave.
>
> ----------------------------------------------------------------------------------------------------------------
>
> hi,
>
> I have coded the K-sample Anderson-Darling test [*].
>
> It tests if the populations of several (more than two) random samples
> are drawned from the same distribution. This test is an extension of
> the Kolmogorov-Smirnov test and it is known to be more sensitive
> to discrepancies in the tails of the distribution.
>
> The implementation follows exactly [*] and it only provides the "continuous"
> distribution case. Note that the compensation given in [*] for the discrete case
> (with ties), is not provided here.
>
> adt.m execute three basic checks of ad.m
>
> Ref: [*] K-sample Anderson-Darling tests, Schotz and Stephens
>           Journal of American Statistical Association, p 918, 1987.
>
> cheers,
> eric.
>

Maybe these should be added to Octave-Forge in the statistics toolbox.
How do they compare to the similar functions on mathworks file exchange?

D.

_______________________________________________
Octave-sources mailing list
Octave-sources@...
https://www.cae.wisc.edu/mailman/listinfo/octave-sources

[Eric Chassande-Mottin]

> Maybe these should be added to Octave-Forge in the statistics toolbox.
> How do they compare to the similar functions on mathworks file exchange?

the code might have to be adapted to the standards of the statistic
toolbox (the present code doesn't produce a p-value). I just had a
look to the function in the mathworks file exchange. my implementation
is different (I used the method of kolmogorov_smirnov_test.m to get
the distribution function of the pooled set whereas they use hist.m)
however both functions should give the same result.
their input data structure is kind of complicated. cell array of
vectors is the natural choice. I see that they implemented the
discrete case (with ties).
i hope I answer your question.

eric.
_______________________________________________
Octave-sources mailing list
Octave-sources@...
https://www.cae.wisc.edu/mailman/listinfo/octave-sources