RANT: NPTESTS vs NPAR TESTS

classic Classic list List threaded Threaded
11 messages Options
Reply | Threaded
Open this post in threaded view
|

RANT: NPTESTS vs NPAR TESTS

Bruce Weaver
Administrator
[ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]

Hello folks.  Our university just started using v18.  A colleague who was attempting to perform  Mann-Whitney U tests on a couple of variables asked me why SPSS was advising her to "retain the null hypothesis" when the p-value was well below .05.  I said I didn't know, and asked to see the data.  

It turns out that there are now so-called "legacy dialogs" under Analyze - Nonparametric.  If you fail to notice that, and you use the new-fangled dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The output (the default output, at least) for NPTESTS is VERY sparse -- it does not give any descriptive statistics (including the sample sizes), nor does it report the value of the test statistic.  It only gives a statement of the null hypothesis, says which test was used, gives a p-value, and a decision (i.e., reject or retain H0).  

IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics really want to market itself by saying things like, "You don't even have to understand statistics to use it!"?  The term GIGO comes to mind.

And there's another problem:  For one of my colleague's two variables, the "decision" was to retain H0, despite  the fact that the p-value was .013.  See below.  (And yes, I do know that there are far too many ties here, which is very problematic for rank-based tests.  But will the targeted user who "doesn't even have to understand statistics?" know that?  I doubt it.)

data list list / Group Y1 Y2 (3f5.0).
begin data
0  0 34
0  0 16
0  0 33
0  0  0
0  0 22
0  0  0
0 46 22
0  0  0
0  0  0
0  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
end data.

variable level Y1 Y2 (scale).

* The default GUI options for Analyze - Nonparametric
* now generate NPTESTS syntax.

*Nonparametric Tests: Independent Samples.
NPTESTS
  /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
  /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
  /CRITERIA ALPHA=0.05  CILEVEL=95.

* Notice that the output is VERY sparse.
* There is no reporting of descriptive stats -- not even
* the sample sizes; nor is the test statistic reported.
* A "decision" is reported (reject or retain H0); but
* for variable Y2, the "decision" does not agree with
* the reported p-value.

* To obtain the old-fashioned NPAR TESTS syntax, one
* must now use the "legacy dialogs" under Analyze - Nonparametric.

NPAR TESTS M-W =  Y1 Y2 BY group(0,1).

--
Bruce Weaver
bweaver@lakeheadu.ca
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.
Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

Marta Garcia-Granero
Hi Bruce:

Bruce Weaver wrote:
> [ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]
>
> Hello folks.  Our university just started using v18.  A colleague who was
> attempting to perform  Mann-Whitney U tests on a couple of variables asked
> me why SPSS was advising her to "retain the null hypothesis" when the
> p-value was well below .05.  I said I didn't know, and asked to see the
> data.
>

Given the small sample sizes, you must pay attention to the exact
p-value (although it will be somewhat biased towards non significance
due to the high number of ties), and it is p=0.063. The asymptotic
approximation is not reliable when both sample sizes are below 20, like
in this case.

Anyway, Mann-Whitney's U test needs that both sample distributions are
similar in shape and spread. This is clearly not the case with your
data. I would recommend using median test instead:

NPAR TESTS
  /MEDIAN=Y1 Y2   BY Group(0 1)
  /MISSING ANALYSIS.


The result is significant (using and exact p-value, BTW).

I DON'T like NPTESTS either. I use a collection of macros that are more
complete (IMHO)

HTH,
Marta GG

> It turns out that there are now so-called "legacy dialogs" under Analyze -
> Nonparametric.  If you fail to notice that, and you use the new-fangled
> dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
> output (the default output, at least) for NPTESTS is VERY sparse -- it does
> not give any descriptive statistics (including the sample sizes), nor does
> it report the value of the test statistic.  It only gives a statement of the
> null hypothesis, says which test was used, gives a p-value, and a decision
> (i.e., reject or retain H0).
>
> IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
> really want to market itself by saying things like, "You don't even have to
> understand statistics to use it!"?  The term GIGO comes to mind.
>
> And there's another problem:  For one of my colleague's two variables, the
> "decision" was to retain H0, despite  the fact that the p-value was .013.
> See below.  (And yes, I do know that there are far too many ties here, which
> is very problematic for rank-based tests.  But will the targeted user who
> "doesn't even have to understand statistics?" know that?  I doubt it.)
>
> data list list / Group Y1 Y2 (3f5.0).
> begin data
> 0  0 34
> 0  0 16
> 0  0 33
> 0  0  0
> 0  0 22
> 0  0  0
> 0 46 22
> 0  0  0
> 0  0  0
> 0  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> end data.
>
> variable level Y1 Y2 (scale).
>
> * The default GUI options for Analyze - Nonparametric
> * now generate NPTESTS syntax.
>
> *Nonparametric Tests: Independent Samples.
> NPTESTS
>   /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
>   /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
>   /CRITERIA ALPHA=0.05  CILEVEL=95.
>
> * Notice that the output is VERY sparse.
> * There is no reporting of descriptive stats -- not even
> * the sample sizes; nor is the test statistic reported.
> * A "decision" is reported (reject or retain H0); but
> * for variable Y2, the "decision" does not agree with
> * the reported p-value.
>
> * To obtain the old-fashioned NPAR TESTS syntax, one
> * must now use the "legacy dialogs" under Analyze - Nonparametric.
>
> NPAR TESTS M-W =  Y1 Y2 BY group(0,1).
>
>
>
> -----
> --
> 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/RANT-NPTESTS-vs-NPAR-TESTS-tp2807847p2807847.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
>
>


--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

Kornbrot, Diana
Re: RANT:  NPTESTS vs NPAR TESTS I recommend the legacy dialogues for n on-parameteric to all my friends ans students
The new part is complicated to use for INPUT and seriously lacking for OUTPUT
Yet another example of PASW letting their designers drive the interface
Typically they have no idea about statistics [as eg shows] and no idea about hci as evidenced by lack of user testing
Best
Diana


On 08/09/2010 14:44, "Marta García-Granero" <mgarciagranero@...> wrote:

Hi Bruce:

Bruce Weaver wrote:
> [ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]
>
> Hello folks.  Our university just started using v18.  A colleague who was
> attempting to perform  Mann-Whitney U tests on a couple of variables asked
> me why SPSS was advising her to "retain the null hypothesis" when the
> p-value was well below .05.  I said I didn't know, and asked to see the
> data.
>

Given the small sample sizes, you must pay attention to the exact
p-value (although it will be somewhat biased towards non significance
due to the high number of ties), and it is p=0.063. The asymptotic
approximation is not reliable when both sample sizes are below 20, like
in this case.

Anyway, Mann-Whitney's U test needs that both sample distributions are
similar in shape and spread. This is clearly not the case with your
data. I would recommend using median test instead:

NPAR TESTS
  /MEDIAN=Y1 Y2   BY Group(0 1)
  /MISSING ANALYSIS.


The result is significant (using and exact p-value, BTW).

I DON'T like NPTESTS either. I use a collection of macros that are more
complete (IMHO)

HTH,
Marta GG
> It turns out that there are now so-called "legacy dialogs" under Analyze -
> Nonparametric.  If you fail to notice that, and you use the new-fangled
> dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
> output (the default output, at least) for NPTESTS is VERY sparse -- it does
> not give any descriptive statistics (including the sample sizes), nor does
> it report the value of the test statistic.  It only gives a statement of the
> null hypothesis, says which test was used, gives a p-value, and a decision
> (i.e., reject or retain H0).
>
> IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
> really want to market itself by saying things like, "You don't even have to
> understand statistics to use it!"?  The term GIGO comes to mind.
>
> And there's another problem:  For one of my colleague's two variables, the
> "decision" was to retain H0, despite  the fact that the p-value was .013.
> See below.  (And yes, I do know that there are far too many ties here, which
> is very problematic for rank-based tests.  But will the targeted user who
> "doesn't even have to understand statistics?" know that?  I doubt it.)
>
> data list list / Group Y1 Y2 (3f5.0).
> begin data
> 0  0 34
> 0  0 16
> 0  0 33
> 0  0  0
> 0  0 22
> 0  0  0
> 0 46 22
> 0  0  0
> 0  0  0
> 0  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> end data.
>
> variable level Y1 Y2 (scale).
>
> * The default GUI options for Analyze - Nonparametric
> * now generate NPTESTS syntax.
>
> *Nonparametric Tests: Independent Samples.
> NPTESTS
>   /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
>   /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
>   /CRITERIA ALPHA=0.05  CILEVEL=95.
>
> * Notice that the output is VERY sparse.
> * There is no reporting of descriptive stats -- not even
> * the sample sizes; nor is the test statistic reported.
> * A "decision" is reported (reject or retain H0); but
> * for variable Y2, the "decision" does not agree with
> * the reported p-value.
>
> * To obtain the old-fashioned NPAR TESTS syntax, one
> * must now use the "legacy dialogs" under Analyze - Nonparametric.
>
> NPAR TESTS M-W =  Y1 Y2 BY group(0,1).
>
>
>
> -----
> --
> Bruce Weaver
> bweaver@...
> 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/RANT-NPTESTS-vs-NPAR-TESTS-tp2807847p2807847.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> LISTSERV@... (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
>
>


--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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




Professor Diana Kornbrot
email: 
d.e.kornbrot@...    
web:    http://web.me.com/kornbrot/KornbrotHome.html
Work
Centre for Lifespan & Chronic Illness Research, CLiCIR
School of Psychology
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
voice:  +44 (0) 1707 28 46 26
Home
19 Elmhurst Avenue
London N2 0LT, UK
 landline:  +44 (0) 2084 44  36 57
 mobile:   +44 (0)
7403 18 16 12
fax:        +44 (0) 8707 06 49 97






Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

SPSS Support-2
In reply to this post by Bruce Weaver
Hi Bruce,

With regard to the disagreement between the significance displayed in the initial output table and the decision recommended, there's a bug, as described in the resolution included below.

David Nichols
Statistical Support
SPSS, an IBM Company

Resolution #90896

Problem Summary:

NPTESTS Mann-Whitney test output significance and decision don't agree

Problem Description:

I'm running a Mann-Whitney test in the NPTESTS procedure in IBM SPSS Statistics. In the resulting output, I sometimes see a p value below my specified critical level in the Sig. column, along with Retain the null hypothesis in the Decision column, or vice versa. Why is this happening?

Resolution Summary:

This problem has been reported to SPSS Development - it occurs because significance is based on asymptotic results, decision on exact results

Resolution Description:

If the sample size is sufficiently small, you get both asymptotic and exact results for the Mann-Whitney test. The results shown in the main initial output in the Viewer should be consistent in using the exact significance levels, but are currently using significance levels from the asymptotic test and basing the decision on the exact test. The problem has been reported to SPSS Development and will be fixed in a future release. In the meantime, you can double-click on the output to activate the Viewer and look at the results in the right-hand screen for the variable(s) of interest and see both asymptotic and exact significance levels. We apologize for any resulting inconvenience.


From: Bruce Weaver <[hidden email]>
To: [hidden email]
Date: 09/08/2010 08:12 AM
Subject: RANT:  NPTESTS vs NPAR TESTS
Sent by: "SPSSX(r) Discussion" <[hidden email]>





[ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]

Hello folks.  Our university just started using v18.  A colleague who was
attempting to perform  Mann-Whitney U tests on a couple of variables asked
me why SPSS was advising her to "retain the null hypothesis" when the
p-value was well below .05.  I said I didn't know, and asked to see the
data.

It turns out that there are now so-called "legacy dialogs" under Analyze -
Nonparametric.  If you fail to notice that, and you use the new-fangled
dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
output (the default output, at least) for NPTESTS is VERY sparse -- it does
not give any descriptive statistics (including the sample sizes), nor does
it report the value of the test statistic.  It only gives a statement of the
null hypothesis, says which test was used, gives a p-value, and a decision
(i.e., reject or retain H0).

IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
really want to market itself by saying things like, "You don't even have to
understand statistics to use it!"?  The term GIGO comes to mind.

And there's another problem:  For one of my colleague's two variables, the
"decision" was to retain H0, despite  the fact that the p-value was .013.
See below.  (And yes, I do know that there are far too many ties here, which
is very problematic for rank-based tests.  But will the targeted user who
"doesn't even have to understand statistics?" know that?  I doubt it.)

data list list / Group Y1 Y2 (3f5.0).
begin data
0  0 34
0  0 16
0  0 33
0  0  0
0  0 22
0  0  0
0 46 22
0  0  0
0  0  0
0  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
end data.

variable level Y1 Y2 (scale).

* The default GUI options for Analyze - Nonparametric
* now generate NPTESTS syntax.

*Nonparametric Tests: Independent Samples.
NPTESTS
 /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
 /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
 /CRITERIA ALPHA=0.05  CILEVEL=95.

* Notice that the output is VERY sparse.
* There is no reporting of descriptive stats -- not even
* the sample sizes; nor is the test statistic reported.
* A "decision" is reported (reject or retain H0); but
* for variable Y2, the "decision" does not agree with
* the reported p-value.

* To obtain the old-fashioned NPAR TESTS syntax, one
* must now use the "legacy dialogs" under Analyze - Nonparametric.

NPAR TESTS M-W =  Y1 Y2 BY group(0,1).



-----
--
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/RANT-NPTESTS-vs-NPAR-TESTS-tp2807847p2807847.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

Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

Bruce Weaver
Administrator
Thanks David.  I didn't know about the "model viewer" that pops up when you double-click on the output.  The output shown there is actually quite nice -- I like the plot showing the distributions for the two groups, for example.  If this had appeared as part of the standard output, I would have avoided a rant.  ;-)

Cheers,
Bruce


SPSS Support-2 wrote
Hi Bruce,

With regard to the disagreement between the significance displayed in the
initial output table and the decision recommended, there's a bug, as
described in the resolution included below.

David Nichols
Statistical Support
SPSS, an IBM Company

Resolution #90896

Problem Summary:
NPTESTS Mann-Whitney test output significance and decision don't agree

Problem Description:
I'm running a Mann-Whitney test in the NPTESTS procedure in IBM SPSS
Statistics. In the resulting output, I sometimes see a p value below my
specified critical level in the Sig. column, along with Retain the null
hypothesis in the Decision column, or vice versa. Why is this happening?

Resolution Summary:
This problem has been reported to SPSS Development - it occurs because
significance is based on asymptotic results, decision on exact results

Resolution Description:
If the sample size is sufficiently small, you get both asymptotic and
exact results for the Mann-Whitney test. The results shown in the main
initial output in the Viewer should be consistent in using the exact
significance levels, but are currently using significance levels from the
asymptotic test and basing the decision on the exact test. The problem has
been reported to SPSS Development and will be fixed in a future release.
In the meantime, you can double-click on the output to activate the Viewer
and look at the results in the right-hand screen for the variable(s) of
interest and see both asymptotic and exact significance levels. We
apologize for any resulting inconvenience.


From:
Bruce Weaver <bruce.weaver@hotmail.com>
To:
SPSSX-L@LISTSERV.UGA.EDU
Date:
09/08/2010 08:12 AM
Subject:
RANT:  NPTESTS vs NPAR TESTS
Sent by:
"SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>



[ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]

Hello folks.  Our university just started using v18.  A colleague who was
attempting to perform  Mann-Whitney U tests on a couple of variables asked
me why SPSS was advising her to "retain the null hypothesis" when the
p-value was well below .05.  I said I didn't know, and asked to see the
data.

It turns out that there are now so-called "legacy dialogs" under Analyze -
Nonparametric.  If you fail to notice that, and you use the new-fangled
dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
output (the default output, at least) for NPTESTS is VERY sparse -- it
does
not give any descriptive statistics (including the sample sizes), nor does
it report the value of the test statistic.  It only gives a statement of
the
null hypothesis, says which test was used, gives a p-value, and a decision
(i.e., reject or retain H0).

IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
really want to market itself by saying things like, "You don't even have
to
understand statistics to use it!"?  The term GIGO comes to mind.

And there's another problem:  For one of my colleague's two variables, the
"decision" was to retain H0, despite  the fact that the p-value was .013.
See below.  (And yes, I do know that there are far too many ties here,
which
is very problematic for rank-based tests.  But will the targeted user who
"doesn't even have to understand statistics?" know that?  I doubt it.)

data list list / Group Y1 Y2 (3f5.0).
begin data
0  0 34
0  0 16
0  0 33
0  0  0
0  0 22
0  0  0
0 46 22
0  0  0
0  0  0
0  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
1  0  0
end data.

variable level Y1 Y2 (scale).

* The default GUI options for Analyze - Nonparametric
* now generate NPTESTS syntax.

*Nonparametric Tests: Independent Samples.
NPTESTS
  /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
  /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
  /CRITERIA ALPHA=0.05  CILEVEL=95.

* Notice that the output is VERY sparse.
* There is no reporting of descriptive stats -- not even
* the sample sizes; nor is the test statistic reported.
* A "decision" is reported (reject or retain H0); but
* for variable Y2, the "decision" does not agree with
* the reported p-value.

* To obtain the old-fashioned NPAR TESTS syntax, one
* must now use the "legacy dialogs" under Analyze - Nonparametric.

NPAR TESTS M-W =  Y1 Y2 BY group(0,1).



-----
--
Bruce Weaver
bweaver@lakeheadu.ca
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/RANT-NPTESTS-vs-NPAR-TESTS-tp2807847p2807847.html

Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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
--
Bruce Weaver
bweaver@lakeheadu.ca
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.
Reply | Threaded
Open this post in threaded view
|

Please remove me from SPSSX-L

necia
In reply to this post by SPSS Support-2
Thank you!  I've retired!

Necia A. Black, Ph.D.           e-mail:  [hidden email]
246 Computing Center            Web URL:http://www.acsu.buffalo.edu/~black
SUNY-Buffalo
Buffalo, New York 14260

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

Bruce Weaver
Administrator
In reply to this post by Marta Garcia-Granero
Hi Marta.  As it happens, I ended up at the same place you did.  I suggested converting things to 2x2 tables, with Group in the rows, and None vs Some in the columns.  Given that the grand median = 0 in both cases, my 2x2 tables are exactly the same as those used for the median test.  One difference is that I recommended using the N-1 chi-square in the case where all expected counts were large enough.  

   http://www.iancampbell.co.uk/twobytwo/calculator.htm

Cheers,
Bruce


Marta García-Granero-2 wrote
Hi Bruce:

Bruce Weaver wrote:
> [ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]
>
> Hello folks.  Our university just started using v18.  A colleague who was
> attempting to perform  Mann-Whitney U tests on a couple of variables asked
> me why SPSS was advising her to "retain the null hypothesis" when the
> p-value was well below .05.  I said I didn't know, and asked to see the
> data.
>

Given the small sample sizes, you must pay attention to the exact
p-value (although it will be somewhat biased towards non significance
due to the high number of ties), and it is p=0.063. The asymptotic
approximation is not reliable when both sample sizes are below 20, like
in this case.

Anyway, Mann-Whitney's U test needs that both sample distributions are
similar in shape and spread. This is clearly not the case with your
data. I would recommend using median test instead:

NPAR TESTS
  /MEDIAN=Y1 Y2   BY Group(0 1)
  /MISSING ANALYSIS.


The result is significant (using and exact p-value, BTW).

I DON'T like NPTESTS either. I use a collection of macros that are more
complete (IMHO)

HTH,
Marta GG
> It turns out that there are now so-called "legacy dialogs" under Analyze -
> Nonparametric.  If you fail to notice that, and you use the new-fangled
> dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
> output (the default output, at least) for NPTESTS is VERY sparse -- it does
> not give any descriptive statistics (including the sample sizes), nor does
> it report the value of the test statistic.  It only gives a statement of the
> null hypothesis, says which test was used, gives a p-value, and a decision
> (i.e., reject or retain H0).
>
> IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
> really want to market itself by saying things like, "You don't even have to
> understand statistics to use it!"?  The term GIGO comes to mind.
>
> And there's another problem:  For one of my colleague's two variables, the
> "decision" was to retain H0, despite  the fact that the p-value was .013.
> See below.  (And yes, I do know that there are far too many ties here, which
> is very problematic for rank-based tests.  But will the targeted user who
> "doesn't even have to understand statistics?" know that?  I doubt it.)
>
> data list list / Group Y1 Y2 (3f5.0).
> begin data
> 0  0 34
> 0  0 16
> 0  0 33
> 0  0  0
> 0  0 22
> 0  0  0
> 0 46 22
> 0  0  0
> 0  0  0
> 0  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> 1  0  0
> end data.
>
> variable level Y1 Y2 (scale).
>
> * The default GUI options for Analyze - Nonparametric
> * now generate NPTESTS syntax.
>
> *Nonparametric Tests: Independent Samples.
> NPTESTS
>   /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
>   /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
>   /CRITERIA ALPHA=0.05  CILEVEL=95.
>
> * Notice that the output is VERY sparse.
> * There is no reporting of descriptive stats -- not even
> * the sample sizes; nor is the test statistic reported.
> * A "decision" is reported (reject or retain H0); but
> * for variable Y2, the "decision" does not agree with
> * the reported p-value.
>
> * To obtain the old-fashioned NPAR TESTS syntax, one
> * must now use the "legacy dialogs" under Analyze - Nonparametric.
>
> NPAR TESTS M-W =  Y1 Y2 BY group(0,1).
>
>
>
> -----
> --
> Bruce Weaver
> bweaver@lakeheadu.ca
> 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/RANT-NPTESTS-vs-NPAR-TESTS-tp2807847p2807847.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> LISTSERV@LISTSERV.UGA.EDU (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
>
>


--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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
--
Bruce Weaver
bweaver@lakeheadu.ca
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.
Reply | Threaded
Open this post in threaded view
|

Re: RANT: NPTESTS vs NPAR TESTS

Maguin, Eugene
There's another minor angle on that new command which is that all the model
graphics, which look very nice, do not directly print. They print but only
after you have double clicked on result box and then select print on the
right hand pane. I'm not interested in ranting on this (and I won't say
anything more on this) but I vote against hidden output. I prefer everything
to be visable in the output file, whether viewed on a screen or on paper.

Gene Maguin



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Bruce Weaver
Sent: Wednesday, September 08, 2010 3:50 PM
To: [hidden email]
Subject: Re: RANT: NPTESTS vs NPAR TESTS

Hi Marta.  As it happens, I ended up at the same place you did.  I suggested
converting things to 2x2 tables, with Group in the rows, and None vs Some in
the columns.  Given that the grand median = 0 in both cases, my 2x2 tables
are exactly the same as those used for the median test.  One difference is
that I recommended using the N-1 chi-square in the case where all expected
counts were large enough.

   http://www.iancampbell.co.uk/twobytwo/calculator.htm

Cheers,
Bruce



Marta García-Granero-2 wrote:

>
> Hi Bruce:
>
> Bruce Weaver wrote:
>> [ posted to the SPSSX-L mailing list, and to comp.soft-sys.stat.spss ]
>>
>> Hello folks.  Our university just started using v18.  A colleague who was
>> attempting to perform  Mann-Whitney U tests on a couple of variables
>> asked
>> me why SPSS was advising her to "retain the null hypothesis" when the
>> p-value was well below .05.  I said I didn't know, and asked to see the
>> data.
>>
>
> Given the small sample sizes, you must pay attention to the exact
> p-value (although it will be somewhat biased towards non significance
> due to the high number of ties), and it is p=0.063. The asymptotic
> approximation is not reliable when both sample sizes are below 20, like
> in this case.
>
> Anyway, Mann-Whitney's U test needs that both sample distributions are
> similar in shape and spread. This is clearly not the case with your
> data. I would recommend using median test instead:
>
> NPAR TESTS
>   /MEDIAN=Y1 Y2   BY Group(0 1)
>   /MISSING ANALYSIS.
>
>
> The result is significant (using and exact p-value, BTW).
>
> I DON'T like NPTESTS either. I use a collection of macros that are more
> complete (IMHO)
>
> HTH,
> Marta GG
>> It turns out that there are now so-called "legacy dialogs" under Analyze
>> -
>> Nonparametric.  If you fail to notice that, and you use the new-fangled
>> dialogs, your analysis is done via NPTESTS rather than NPAR TESTS.  The
>> output (the default output, at least) for NPTESTS is VERY sparse -- it
>> does
>> not give any descriptive statistics (including the sample sizes), nor
>> does
>> it report the value of the test statistic.  It only gives a statement of
>> the
>> null hypothesis, says which test was used, gives a p-value, and a
>> decision
>> (i.e., reject or retain H0).
>>
>> IMO, this is a big step in the wrong direction.  Does IBM SPSS Statistics
>> really want to market itself by saying things like, "You don't even have
>> to
>> understand statistics to use it!"?  The term GIGO comes to mind.
>>
>> And there's another problem:  For one of my colleague's two variables,
>> the
>> "decision" was to retain H0, despite  the fact that the p-value was .013.
>> See below.  (And yes, I do know that there are far too many ties here,
>> which
>> is very problematic for rank-based tests.  But will the targeted user who
>> "doesn't even have to understand statistics?" know that?  I doubt it.)
>>
>> data list list / Group Y1 Y2 (3f5.0).
>> begin data
>> 0  0 34
>> 0  0 16
>> 0  0 33
>> 0  0  0
>> 0  0 22
>> 0  0  0
>> 0 46 22
>> 0  0  0
>> 0  0  0
>> 0  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> 1  0  0
>> end data.
>>
>> variable level Y1 Y2 (scale).
>>
>> * The default GUI options for Analyze - Nonparametric
>> * now generate NPTESTS syntax.
>>
>> *Nonparametric Tests: Independent Samples.
>> NPTESTS
>>   /INDEPENDENT TEST (Y1 Y2) GROUP (Group)
>>   /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE
>>   /CRITERIA ALPHA=0.05  CILEVEL=95.
>>
>> * Notice that the output is VERY sparse.
>> * There is no reporting of descriptive stats -- not even
>> * the sample sizes; nor is the test statistic reported.
>> * A "decision" is reported (reject or retain H0); but
>> * for variable Y2, the "decision" does not agree with
>> * the reported p-value.
>>
>> * To obtain the old-fashioned NPAR TESTS syntax, one
>> * must now use the "legacy dialogs" under Analyze - Nonparametric.
>>
>> NPAR TESTS M-W =  Y1 Y2 BY group(0,1).
>>
>>
>>
>> -----
>> --
>> 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/RANT-NPTESTS-vs-NPAR-TESTS-tp2
807847p2807847.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
>>
>>
>
>
> --
> For miscellaneous SPSS related statistical stuff, visit:
> http://gjyp.nl/marta/
>
> =====================
> 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
>
>


-----
--
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/RANT-NPTESTS-vs-NPAR-TESTS-tp2
807847p2810189.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
Reply | Threaded
Open this post in threaded view
|

Re: Please remove me from SPSSX-L

David Marso
Administrator
In reply to this post by necia
On Wed, 8 Sep 2010 15:35:26 -0400, necia <[hidden email]> wrote:

>Thank you!  I've retired!
>
>Necia A. Black, Ph.D.           e-mail:  [hidden email]
>246 Computing Center            Web URL:http://www.acsu.buffalo.edu/~black
>SUNY-Buffalo
>Buffalo, New York 14260
>
>=====================
>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

Congratulations Necia!!
I don't know if you would remember me, but we used to chat on the phone
on occasion something like 20 years ago when I did TekSport for SPSS!!!
You always had challenging questions!  I am sure your students will miss
your support!
Best regards, David

=====================
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
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"
Reply | Threaded
Open this post in threaded view
|

Re: Please remove me from SPSSX-L

Tesiny, Ed
 To those wishing to unsubscribe, this is the only way to do it:

Send a message to: [hidden email]

No Subject or Signature

In the body of the message put: SIGNOFF SPSSX-L


Ed Tesiny
[hidden email]

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
David Marso
Sent: Friday, September 10, 2010 7:18 AM
To: [hidden email]
Subject: Re: Please remove me from SPSSX-L

On Wed, 8 Sep 2010 15:35:26 -0400, necia <[hidden email]> wrote:

>Thank you!  I've retired!
>
>Necia A. Black, Ph.D.           e-mail:  [hidden email]
>246 Computing Center            Web
URL:http://www.acsu.buffalo.edu/~black
>SUNY-Buffalo
>Buffalo, New York 14260
>
>=====================
>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

Congratulations Necia!!
I don't know if you would remember me, but we used to chat on the phone
on occasion something like 20 years ago when I did TekSport for SPSS!!!
You always had challenging questions!  I am sure your students will miss
your support!
Best regards, David

=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: Please remove me from SPSSX-L

Bruce Weaver
Administrator
In reply to this post by David Marso
David Marso wrote
Congratulations Necia!!
I don't know if you would remember me...
.oO(Let me think about that...is David a forgettable type?  Hmmm.  Somehow, I doubt it.)
--
Bruce Weaver
bweaver@lakeheadu.ca
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.