road12fg@NETSCAPE.COM wrote back:
>On Wed, 6 Sep 2006 11:19:45 -0700, davidlcassell@MSN.COM (David L
> >road12fg@NETSCAPE.COM wrote:
> >> >I'm just learning to use the SAS software and am confused about one of
> >> >the readouts. After putting in a set of univariate data and running an
> >> >analysis on it, the goodness of fit for a normal distribution (under
> >> >the K-S method) reads out a P value that states Psub r > D >0.150.
> >> >How do I interpret this? The null should be that the data can be
> >> >modeled by a normal distribution and should be rejected if the Test
> >> >statistic ( D ) ( 0.088 ) exceeds the critical value . Is the P value
> >> >the same thing as the critical value? So is the read out " >0.150 "
> >> >mean the critical value is 0.150 or that the statistic is less that
> >> >the critical value by 0.150 or what? thanks for any help here. john
> >>Also what is the presumed alpha? thanks jk
> >First off, I do not recommend the K-S test in this situation. Use the
> >Shapiro-Wilk test instead. It is more 'reliable'. None of these tests
> >is perfect, and different sorts of deviaitons from the null-hypothesis
> >distribution can lead to different performance issues with different
> >In fact, I wrote a bunch about this in SAS-L just a couple months ago.
> >Obviously, you didn't see that.
> >That said, D is (probably) your test statistic. I say probably, because
> >you may (or may not) actually want D+ or D-, depending on your
> >chosen alternative hypothesis. This makes a difference in the p-value
> >So you know that your p-value is greater than .15 . That's all you get
> >here. (Well, I told you to use a different test, didn't I?) Your alpha
> >is whatever YOU chose it to be. If you chose alpha to be .01 or .05
> >or .10, then clearly you didn't reject the null hypothesis. (That does
> >not mean your distribution is normal, it may just mean that the test
> >didn't find the problems effectively.)
> >And finally, why are you testing that your data are normal? I see people
> >doing this all the time, and in most of these cases, they are doing the
> >wrong thing.
>Thanks, this is pretty what I needed. But one question... Does the
>Shapiro -Wilk give a specific P value rather than the "greater than "
>that the K-S test? thanks jk
K-S only gives that "P > ..." bit under certian circumstances. Don't
count on that.
Similarly, I canot guarantee that the Shapiro-Wilk statistic will *never*
give you such an output. It is designed to handle both large and
small n, and it can handle a wide range of p-values. But there may be
circumstances where someday, somewhere, it will give you such a
message. Expect that they will be the extreme cases, such as really
minute or really enormous p-values which are so far away from your
cutoff values that you don't need to know an exact number.
Which is basically what happened with your K-S statistic with your data.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Call friends with PC-to-PC calling -- FREE