same data, same test with different procedures =different results USING 19.0

[msherman]

Dear List;  I just tested a sample distribution using the One Sample Kolmogorow-Smirnov Test under Non-Parametric Tests—One sample K-S test and get a K-S Z of .933 with a p value of .349.  Indicating that my sample is consistent with a population that is normally distributed.  I then went and used the Explore procedure and looked at the results of the K-S test and get some very different results.  The statistic reported is .070 and the p value is .034.  I am not sure what is happening here.  Does anyone have any ideas. Thanks,  martin

 

 

                                                                                                                                                                             
Martin F. Sherman, Ph.D.

Professor of Psychology

Director of Masters Education: Thesis Track
Loyola College of Arts and Sciences

 

[Marta Garcia-Granero]

El 22/10/2010 2:11, Martin Sherman escribió:

Dear List;  I just tested a sample distribution using the One Sample Kolmogorow-Smirnov Test under Non-Parametric Tests—One sample K-S test and get a K-S Z of .933 with a p value of .349.  Indicating that my sample is consistent with a population that is normally distributed.  I then went and used the Explore procedure and looked at the results of the K-S test and get some very different results.  The statistic reported is .070 and the p value is .034.  I am not sure what is happening here.  Does anyone have any ideas. Thanks,  martin

 

 

                                                                                                                                                                              
Martin F. Sherman, Ph.D.

Professor of Psychology

Director of Masters Education: Thesis Track
Loyola College of Arts and Sciences

 

Hi Martin:

This difference can be found since SPSS 4.0 (PC+ I think its name was). It is not specific of SPSS 19.

The difference is due to the fact that One sample KS (under Non parametric tests) computes Kolmogorov-Smirnv test WITHOUT the Lilliefors correction, and EXPLORE computes the test WITH Lilliefors correction (to adjust the test to the fact that the sample mean&sd are used, instead of the corresponding parameters mu and sigma). Don't use the first one, since it is over conservative. As a matter of fact, DON'T use the second (KS with Lilliefors correction) either, since Shapiro-Wilk test has been proven to be better (the best, I might add).

Best regards,
Marta GG

[Bruce Weaver]

msherman wrote

Dear List;  I just tested a sample distribution using the One Sample Kolmogorow-Smirnov Test under Non-Parametric Tests-One sample K-S test and get a K-S Z of .933 with a p value of .349.  Indicating that my sample is consistent with a population that is normally distributed.  I then went and used the Explore procedure and looked at the results of the K-S test and get some very different results.  The statistic reported is .070 and the p value is .034.  I am not sure what is happening here.  Does anyone have any ideas. Thanks,  martin

Hi Martin.  Marta has given a very thorough answer (as usual) to your question.  But I'm still wondering why you want to test for normality.  

<soapbox>
If you are doing it as a precursor to a t-test (or some other parametric procedure), I would advise you to not bother.  Sampling from (approximately) normal populations is most important when sample sizes are small.  It becomes less and less important as the sample size increases, because the sampling distribution converges on the normal.  Now consider the test of normality.  When sample sizes are small (and normality is most important for the t-test), the test of normality has very low power, and will fail to detect important departures from normality.  But when sample sizes are large (and normality is not so important for the t-test), the test of normality has too much power--i.e., it will throw up the red flag of non-normality for unimportant departures from normality.  So testing for normality as a precursor to a t-test or ANOVA is just about the most pointless statistical exercise one can engage in.  IMO, at least.  ;-)
</soapbox>

[msherman]

Agreed.  When I teach my students about the assumptions I show them a number of things that you can look at (e.g., Skewness, SE Skew, Kurtosis, SE Kurtosis, visual shape, K-S test, and most importantly the sample size (which ties into the CLT).  I show students how the various ways of going about looking at normality and then when all is said I talk about the Sampling distribution of same means which is the pivotal point of the entire discussion. Students recall from their undergraduate courses that the "sample" needs to be normal, when it fact it is not the sample but the population and more to the point the shape of the sampling distribution. It is an exercise of simply gathering data about an assumption and then how to go about making a decision. So when I use a test like the K-S it is for teaching purposes (Large N's allow for significance for slight variations from normality whereas small N's lack power).  All is used to get my students to think in terms of making 'good!
 ' decisions about their data.  But the issue at hand is why am I getting two different results with supposedly the same test but under different SPSS procedures.   mfs


Martin F. Sherman, Ph.D.
Professor of Psychology
Director of Masters Education: Thesis Track
Loyola College of Arts and Sciences

Loyola University Maryland
4501 North Charles Street
222 B Beatty Hall
Baltimore, MD 21210-2601

410-617-2417 office
410-617-5341 fax

[hidden email]

www.loyola.edu

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bruce Weaver
Sent: Friday, October 22, 2010 8:33 AM
To: [hidden email]
Subject: Re: same data, same test with different procedures =different results USING 19.0

msherman wrote:

>
> Dear List;  I just tested a sample distribution using the One Sample
> Kolmogorow-Smirnov Test under Non-Parametric Tests-One sample K-S test and
> get a K-S Z of .933 with a p value of .349.  Indicating that my sample is
> consistent with a population that is normally distributed.  I then went
> and used the Explore procedure and looked at the results of the K-S test
> and get some very different results.  The statistic reported is .070 and
> the p value is .034.  I am not sure what is happening here.  Does anyone
> have any ideas. Thanks,  martin
>

Hi Martin.  Marta has given a very thorough answer (as usual) to your
question.  But I'm still wondering why you want to test for normality.

<soapbox>
If you are doing it as a precursor to a t-test (or some other parametric
procedure), I would advise you to not bother.  Sampling from (approximately)
normal populations is most important when sample sizes are small.  It
becomes less and less important as the sample size increases, because the
sampling distribution converges on the normal.  Now consider the test of
normality.  When sample sizes are small (and normality is most important for
the t-test), the test of normality has very low power, and will fail to
detect important departures from normality.  But when sample sizes are large
(and normality is not so important for the t-test), the test of normality
has too much power--i.e., it will throw up the red flag of non-normality for
unimportant departures from normality.  So testing for normality as a
precursor to a t-test or ANOVA is just about the most pointless statistical
exercise one can engage in.  IMO, at least.  ;-)
</soapbox>


-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/same-data-same-test-with-different-procedures-different-results-USING-19-0-tp3231572p3232172.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD