jonas.bilenas@CHASE.COM replied:
>
>On Fri, 11 May 2007 22:35:51 -0700, David L Cassell <davidlcassell@MSN.COM>
>wrote:
>
> >mytkolli@GMAIL.COM wrote:
> >>
> >>Hi,
> >>
> >>Is there anybody to tell me how to calculate the standard deviation of
> >>K-S statistics?
> >>
> >>Thanks,
> >>
> >>Hansi
> >
> >Why on earth would you want to do that kind of computation?
> >SAS already does the work to get the K-S statistic and the p-value
> >you need.  Plus, it's a royal pain to compute when things like ties
> >occur...
> >
> >And why are you using a K-S statistic anyway?  It is not generally
> >the best choice for many comparisons, and it usually has a lot less
> >power than other alternatives.
> >
> >So it might be useful for you to write back to SAS-L and explain
> >why you need this anyway.
> >
> >HTH,
> >David
> >--
> >David L. Cassell
> >mathematical statistician
> >Design Pathways

>
>David,
>
>What alternatives would you suggest for binary models?  KS stats are very
>popular in Credit Models (risk, response).  People go through enormous
>efforts to calculate KS statistics and lift plots in the credit industry
>which always confuses me since PROC NPAR1WAY will generate 2 sample KS and
>there are a number of macros available for ROC curves.

Well Jonas, that's why I asked what he wanted to use the K-S statistic
*for*.
As you know, it's popular in the credit industry and related fields where
people end up with lift data.  (It does bother me when people take their
data, turn them into deciles, and *then* perform the K-S work on the now-
incredibly-tied data.)

But you also know that people regularly write to SAS-L about the K-S
statistic
when they should *not* be using it.  For example, checking normality of
residuals after regression.  I've written a lot about alternatives and power
of alternatives under such circumstances.

So that's why I was concerned about the use.  The K-S statistic might be a
reasonable choice, or it may not.  But I doubt that the Original Poster
really
wants to compute the standard error when SAS will get the p-value for
you.

HTCT,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
More photos, more messages, more storage�get 2GB with Windows Live Hotmail.
http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migration_HM_mini_2G_0507

On May 15, 1:20 am, davidlcass...@MSN.COM (David L Cassell) wrote:
> jonas.bile...@CHASE.COM replied:
>
>
>
>
>
>
>
> >On Fri, 11 May 2007 22:35:51 -0700, David L Cassell <davidlcass...@MSN.COM>
> >wrote:
>
> > >mytko...@GMAIL.COM wrote:
>
> > >>Hi,
>
> > >>Is there anybody to tell me how to calculate the standard deviation of
> > >>K-Sstatistics?
>
> > >>Thanks,
>
> > >>Hansi
>
> > >Why on earth would you want to do that kind of computation?
> > >SAS already does the work to get theK-Sstatistic and the p-value
> > >you need.  Plus, it's a royal pain to compute when things like ties
> > >occur...
>
> > >And why are you using aK-Sstatistic anyway?  It is not generally
> > >the best choice for many comparisons, and it usually has a lot less
> > >power than other alternatives.
>
> > >So it might be useful for you to write back to SAS-L and explain
> > >why you need this anyway.
>
> > >HTH,
> > >David
> > >--
> > >David L. Cassell
> > >mathematical statistician
> > >Design Pathways
>
> >David,
>
> >What alternatives would you suggest for binary models?  KSstats are very
> >popular in Credit Models (risk, response).  People go through enormous
> >efforts to calculateKSstatistics and lift plots in the credit industry
> >which always confuses me since PROC NPAR1WAY will generate 2 sampleKSand
> >there are a number of macros available for ROC curves.
>
> Well Jonas, that's why I asked what he wanted to use theK-Sstatistic
> *for*.
> As you know, it's popular in the credit industry and related fields where
> people end up with lift data.  (It does bother me when people take their
> data, turn them into deciles, and *then* perform theK-Swork on the now-
> incredibly-tied data.)
>
> But you also know that people regularly write to SAS-L about theK-S
> statistic
> when they should *not* be using it.  For example, checking normality of
> residuals after regression.  I've written a lot about alternatives and power
> of alternatives under such circumstances.
>
> So that's why I was concerned about the use.  TheK-Sstatistic might be a
> reasonable choice, or it may not.  But I doubt that the Original Poster
> really
> wants to compute the standard error when SAS will get the p-value for
> you.
>
> HTCT,
> David
> --
> David L. Cassell
> mathematical statistician
> Design Pathways
> 3115 NW Norwood Pl.
> Corvallis OR 97330
>
> _________________________________________________________________
> More photos, more messages, more storage-get 2GB with Windows Live Hotmail.http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_m...- Hide quoted text -
>
> - Show quoted text -

Thanks all.

Actually I do modeling for the credit industry. I do know that SAS
produces KS and its p. Sometimes KS come too small and one might be
interested still on the model, and of course you can choose p to make
the decision, but sometimes, especially when the audience (as it is
mostly the case with the financial apps) is far from undersanding the
significance of p, a standard deviation could be more acceptable for
them.

Thanks anyway for all your thoughts.

Hansi