Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: ksmirnov

From   Nick Cox <>
Subject   Re: st: ksmirnov
Date   Tue, 21 Aug 2012 14:44:34 +0100

As usual, I agree with Maarten when he agrees with me, and versa vice.
The strategic issues in my view include attempting to summarize lack
of fit in one number and conflating an estimation problem with a
decision problem.

About a decade ago I persuaded StataCorp to include a quotation from
Rupert Miller on the limitations of Kolmogorov-Smirnov  within the
manual entry [R] diagplots. That quotation still seems cogent to me.

It's quite widely documented that K-S with parameters estimated from
the data is problematic. Houda started this thread apparently
_knowing_ that df = 8 but ended it wanting to know how to estimate it,
so the aims here are a moving target.

Incidentally, fashions change on transliteration, but as far as I know
Kolmogorov or possibly Kolmogoroff (resp. Smirnov and Smirnoff)  are
the only widely accepted renderings of those names in English. (I am
not ignoring other languages; I just don't know anything about their
rules here.) At least hereabouts there is a vodka brand named
Smirnoff. There are several Russian speakers on this list (e.g Stas
Kolenikov) who may be able to comment.


On Tue, Aug 21, 2012 at 1:55 PM, Maarten Buis <> wrote:
> On Tue, Aug 21, 2012 at 1:59 PM, Guessab Houda wrote:
>> Nick told me that it's not correct to use kormogolv smirnov to compare the shape of series with a student distribution, is there another numerical method (not graphical)to do that?
> Long standing members of this list will not be surprised that Nick and
> I agree on this point. To put it in another way, you seem to want to
> summarize difference between the shape of a variable and a theoretical
> curve in one number. By necessity this will throw away a huge amount
> of information. Even if you thought that that was acceptable, than
> you'd still have to consider very carefully what that number is
> measuring and what it is not measuring. In all likelihood it is not
> measuring what you think it measures.
> -- Maarten
> ---------------------------------
> Maarten L. Buis
> Reichpietschufer 50
> 10785 Berlin
> Germany
> ---------------------------------
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index