SPSSX Discussion

Testing the null Vs directional hypotheis

Classic

List

Threaded

11 messages Options

Anthony James

Testing the null Vs directional hypotheis

Dear all,

When we state a null hypothesis (in case of a t-test or ANOVA), we can reject it if the obsrved t or F are greater than the critical values in the table. But if we state a directinal hypothesis and the t or F are greater than their critical values in the table we accpet the directional hypothesis. Am I right?

We don't want to reject the directional hypothesis.

Thanks

Anthony

ANDRES ALBERTO BURGA LEON

Repeating graph

Hello to everybody:

I need to do this graphic for other 11 variables other than Rasch_M_N (all remains the same)

I now I can't include this inside a macro, but is there another easy way to do this instead to copy / paste and edit this 12 times, changin the MEAN(Rasch_M_N) each time?

GGRAPH
/GRAPHDATASET NAME="data" VARIABLES=tiempo MEAN(Rasch_M_N)[name="MEAN_Rasch"] grupo
MISSING=LISTWISE REPORTMISSING=NO
/GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
SOURCE: s=userSource(id("data"))
DATA: tiempo=col(source(s), name("tiempo"), unit.category())
DATA: MEAN_Rasch=col(source(s), name("MEAN_Rasch"))
DATA: grupo=col(source(s), name("grupo"), unit.category())
GUIDE: axis(dim(1), label("Momento"))
GUIDE: axis(dim(2), label("Medida promedio"))
GUIDE: legend(aesthetic(aesthetic.color.exterior), label("Grupo"))
SCALE: cat(dim(1), include("0", "1"))
SCALE: linear(dim(2), include(0))
SCALE: cat(aesthetic(aesthetic.color.exterior), include("0", "1"))
ELEMENT: point(position(tiempo*MEAN_Rasch), color.exterior(grupo))
END GPL.

Andres

PS: I don't know anything about Phyton

Mg. Andrés Burga León
Coordinador de Análisis e Informática
Unidad de Medición de la Calidad Educativa (UMC)
Ministerio de Educación del Perú
Av.de la Arqueología s/n (cuadra 2)
Lima 41
Perú
Teléfono 615-5840 - 6155800 anexo 1212
http://www2.minedu.gob.pe/umc/

ViAnn Beadle

Re: Repeating graph

You can include GPL in a macro. The general problem with a macro and GPL is that a macro changes quotation marks to apostrophes and GPL requires quotation marks around strings. Thus you must use the CONCAT macro function to put the quotation marks back in. So you’ll need a !LET statement for each quoted string such as “data”, “grupo”, “Momento”, ….

A statement looks something like this:

!LET !qvarname=!CONCAT('"',!varname,'"')

!LET !LET !gdataset=!CONCAT('"', data,'"')

where varname is a macro variable inserted into the GPL for one of your substituted variables.

So you need a !LET for string “data”, the three DATA statements, the three GUIDE statements, and the two SCALE statements for the two included values.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of ANDRES ALBERTO BURGA LEON
Sent: Tuesday, January 17, 2012 10:04 AM
To: [hidden email]
Subject: Repeating graph

Bruce Weaver

Re: Testing the null Vs directional hypotheis

Administrator

In reply to this post by Anthony James

This question doesn't have anything to do with SPSS per se, which may be why answers are slow in coming. It would be more appropriate for a general stats newsgroup or list. Having said that...

First of all, one does not usually talk about a directional alternative hypothesis (H1) for an F-test, because the rejection region for the F-test is always in the upper tail. I.e., without resorting to some trickery, the F-test implies a non-directional alternative hypothesis. If there are only two groups (or paired scores), you can take the square root of F to get the corresponding t-value, and perform either a two-tailed or one-tailed test.

For a z- or t-test with a directional H1, you can only reject H0 if:

1. The difference is in the direction specified by H1, and
2. The computed value of z or t is further out in the tail than the critical value.

Whether "further out in the tail" means greater than or less than the critical value depends on which tail it is, obviously.

Finally, one should only be using a directional H1 (aka a one-tailed test) if a difference in the "wrong" direction, no matter how large, can be treated exactly the same as a difference of zero. I suspect that quite often, people use one-tailed tests when this is not really the case.

HTH.

Anthony James wrote

Dear all,
When we state a null hypothesis (in case of a t-test or ANOVA), we can reject it if the obsrved t or F are greater than the critical values in the table. But if we state a directinal hypothesis and the t or F are greater than their critical values in the table we accpet the directional hypothesis. Am I right?
We don't want to reject the directional hypothesis.
Thanks

Anthony

--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

PLEASE NOTE THE FOLLOWING:
1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above.
2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/).

Jarrod Teo-2

Re: Testing the null Vs directional hypotheis

Hi Anthony and Bruce,

First of all, I always enjoy reading Bruce's posting because I feel that he is a great Statistician.

I believe this discussion was held in the listing before because a researchers was trying to perform a 1 tailed 2 independent samples T-test as well in SPSS. In which I think this link below was provided and I am not sure if this information helps. Do note that the link is an old one and new updates has not been done to the page so please use this information cautiously.

http://www.ats.ucla.edu/stat/spss/faq/pvalue.html

One note from this webpage is something which I do find interesting:

"Many researchers argue that it is very rarely appropriate to do a one-tailed test. However, if it is logically impossible for the result to go in one direction (for example, for the mean height of 5-year-olds to be smaller than the mean height of 15-year-olds) or if such a result is of no practical importance (for example, the experimental medicine is less effective than currently used medicine), then a one-tailed test is appropriate."

Bruce, I had heard at least twice with regards to managing the results of 1-tailed 2 independent samples T-test in SPSS. Is it really that as according to the article,

Ho: mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Means that we have to take p/2 since the direction is logical

Ho: mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old < mean height of 5 years old

Means that we have to take 1-(p/2) since the direction is not logical

Warmest Regards
Dorraj Oet

__________________________________________________________________________________________________________________________________________________________________________

> Date: Tue, 17 Jan 2012 14:58:42 -0800

> From: [hidden email]
> Subject: Re: Testing the null Vs directional hypotheis
> To: [hidden email]
>
> This question doesn't have anything to do with SPSS per se, which may be why
> answers are slow in coming. It would be more appropriate for a general
> stats newsgroup or list. Having said that...
>
> First of all, one does not usually talk about a directional alternative
> hypothesis (H1) for an F-test, because the rejection region for the F-test
> is always in the upper tail. I.e., without resorting to some trickery, the
> F-test implies a non-directional alternative hypothesis. If there are only
> two groups (or paired scores), you can take the square root of F to get the
> corresponding t-value, and perform either a two-tailed or one-tailed test.
>
> For a z- or t-test with a directional H1, you can only reject H0 if:
>
> 1. The difference is in the direction specified by H1, and
> 2. The computed value of z or t is further out in the tail than the
> critical value.
>
> Whether "further out in the tail" means greater than or less than the
> critical value depends on which tail it is, obviously.
>
> Finally, one should only be using a directional H1 (aka a one-tailed test)
> if a difference in the "wrong" direction, no matter how large, can be
> treated exactly the same as a difference of zero. I suspect that quite
> often, people use one-tailed tests when this is not really the case.
>
> HTH.
>
>
> Anthony James wrote
> >
> > Dear all,
> > When we state a null hypothesis (in case of a t-test or ANOVA), we can
> > reject it if the obsrved t or F are greater than the critical values in
> > the table. But if we state a directinal hypothesis and the t or F are
> > greater than their critical values in the table we accpet the directional
> > hypothesis. Am I right?
> > We don't want to reject the directional hypothesis.
> > Thanks
> > �
> > Anthony
> >
>
>
> -----
> --
> 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/Testing-the-null-Vs-directional-hypotheis-tp5150899p5153231.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

Art Kendall

Re: Testing the null Vs directional hypotheis

Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Ho:� mean height of 15 years old is not statistically distinguishable from the mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Art Kendall
Social Research Consultants

On 1/17/2012 8:55 PM, DorraJ Oet wrote:

Hi Anthony and Bruce,
�
First of all, I always enjoy reading Bruce's posting because I feel that he is a great Statistician.
�
I believe this discussion was held in the listing before because� a researchers was trying to perform a 1 tailed� 2 independent samples T-test as well� in SPSS. In which I think this link below was provided and� I am not sure if this information� helps. Do note that the link is an old one and new updates has not been done to the page so� please use this information cautiously.
�
http://www.ats.ucla.edu/stat/spss/faq/pvalue.html
�
One note from this webpage is something which I do find interesting:
�
"Many researchers argue that it is very rarely appropriate to do a one-tailed test.� However, if it is logically impossible for the result to go in one direction (for example, for the mean height of 5-year-olds to be smaller than the mean height of 15-year-olds) or if such a result is of no practical importance (for example, the experimental medicine is less effective than currently used medicine), then a one-tailed test is appropriate."�
�
Bruce, I had heard at least� twice� with regards to managing the results of 1-tailed 2 independent samples T-test in SPSS. Is it really that as according to the article,
�
Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old
�
Means that we have to take p/2 since the direction is logical
�
Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old < mean height of 5 years old
�
Means that we have to take 1-(p/2) since the direction is not logical
�
Warmest Regards
Dorraj Oet�
�
__________________________________________________________________________________________________________________________________________________________________________
�

> Date: Tue, 17 Jan 2012 14:58:42 -0800
> From: [hidden email]
> Subject: Re: Testing the null Vs directional hypotheis
> To: [hidden email]
>
> This question doesn't have anything to do with SPSS per se, which may be why
> answers are slow in coming. It would be more appropriate for a general
> stats newsgroup or list. Having said that...
>
> First of all, one does not usually talk about a directional alternative
> hypothesis (H1) for an F-test, because the rejection region for the F-test
> is always in the upper tail. I.e., without resorting to some trickery, the
> F-test implies a non-directional alternative hypothesis. If there are only
> two groups (or paired scores), you can take the square root of F to get the
> corresponding t-value, and perform either a two-tailed or one-tailed test.
>
> For a z- or t-test with a directional H1, you can only reject H0 if:
>
> 1. The difference is in the direction specified by H1, and
> 2. The computed value of z or t is further out in the tail than the
> critical value.
>
> Whether "further out in the tail" means greater than or less than the
> critical value depends on which tail it is, obviously.
>
> Finally, one should only be using a directional H1 (aka a one-tailed test)
> if a difference in the "wrong" direction, no matter how large, can be
> treated exactly the same as a difference of zero. I suspect that quite
> often, people use one-tailed tests when this is not really the case.
>
> HTH.
>
>
> Anthony James wrote
> >
> > Dear all,
> > When we state a null hypothesis (in case of a t-test or ANOVA), we can
> > reject it if the obsrved t or F are greater than the critical values in
> > the table. But if we state a directinal hypothesis and the t or F are
> > greater than their critical values in the table we accpet the directional
> > hypothesis. Am I right?
> > We don't want to reject the directional hypothesis.
> > Thanks
> > �
> > Anthony
> >
>
>
> -----
> --
> 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/Testing-the-null-Vs-directional-hypotheis-tp5150899p5153231.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

Art Kendall
Social Research Consultants

Bruce Weaver

Re: Testing the null Vs directional hypotheis

Administrator

In reply to this post by Jarrod Teo-2

Thanks for those kind words, DorraJ. But now I feel I must offer my standard disclaimer: I'm not really a statistician--I just play one in a medical school. ;-)

I don't remember who it was just now, but someone with a background similar to mine (experimental psych) posted in one of the sci.stat.* usenet groups that he used the term "quantitative methodologist" rather than "statistician".

Cheers,
Bruce

DorraJ Oet wrote

Hi Anthony and Bruce,

First of all, I always enjoy reading Bruce's posting because I feel that he is a great Statistician.

--- snip ---

Ryan

Re: Testing the null Vs directional hypotheis

In reply to this post by Bruce Weaver

I would like to point out that there are situations in which a directional test via an F test is warranted. Suppose we have a 2X2 design:

Factor B

1 2

------------

1| a1b1 |a1b2 |

Factor A |------------|

2| a2b1 |a2b2 |

------------

Assuming one were to run an ANOVA on the design above, including both main effects and the interaction effect, one could employ a "directional test" on the interaction term. Note that the numerator df for the interaction test would be 1.

Ryan

On Tue, Jan 17, 2012 at 5:58 PM, Bruce Weaver <[hidden email]> wrote:

This question doesn't have anything to do with SPSS per se, which may be why
answers are slow in coming. It would be more appropriate for a general
stats newsgroup or list. Having said that...

First of all, one does not usually talk about a directional alternative
hypothesis (H1) for an F-test, because the rejection region for the F-test
is always in the upper tail. I.e., without resorting to some trickery, the
F-test implies a non-directional alternative hypothesis. If there are only
two groups (or paired scores), you can take the square root of F to get the
corresponding t-value, and perform either a two-tailed or one-tailed test.

For a z- or t-test with a directional H1, you can only reject H0 if:

1. The difference is in the direction specified by H1, and
2. The computed value of z or t is further out in the tail than the
critical value.

Whether "further out in the tail" means greater than or less than the
critical value depends on which tail it is, obviously.

Finally, one should only be using a directional H1 (aka a one-tailed test)
if a difference in the "wrong" direction, no matter how large, can be
treated exactly the same as a difference of zero. I suspect that quite
often, people use one-tailed tests when this is not really the case.

HTH.

Anthony James wrote

>
> Dear all,
> When we state a null hypothesis (in case of a t-test or ANOVA), we can
> reject it if the obsrved t or F are greater than the critical values in
> the table. But if we state a directinal hypothesis and the t or F are
> greater than their critical values in the table we accpet the directional
> hypothesis. Am I right?
> We don't want to reject the directional hypothesis.
> Thanks

> Â
> Anthony
>

-----
--
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/Testing-the-null-Vs-directional-hypotheis-tp5150899p5153231.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

Jason Schoeneberger

Re: Testing the null Vs directional hypotheis

Hi All,

I believe in Ryan’s case below, if the interaction term attains the direction expected, you also need to divide your obtained p-value by 2. Essentially, this is equivalent to a linear contrast set up to test the direction desired, yielding a t-value that, when squared, should equal the F value from the interaction term in ryan’s example.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of R B
Sent: Wednesday, January 18, 2012 12:58 PM
To: [hidden email]
Subject: Re: Testing the null Vs directional hypotheis

I would like to point out that there are situations in which a directional test via an F test is warranted. Suppose we have a 2X2 design:

Factor B

1 2

------------

1| a1b1 |a1b2 |

Factor A |------------|

2| a2b1 |a2b2 |

------------

Ryan

On Tue, Jan 17, 2012 at 5:58 PM, Bruce Weaver <[hidden email]> wrote:

This question doesn't have anything to do with SPSS per se, which may be why
answers are slow in coming. It would be more appropriate for a general
stats newsgroup or list. Having said that...

First of all, one does not usually talk about a directional alternative
hypothesis (H1) for an F-test, because the rejection region for the F-test
is always in the upper tail. I.e., without resorting to some trickery, the
F-test implies a non-directional alternative hypothesis. If there are only
two groups (or paired scores), you can take the square root of F to get the
corresponding t-value, and perform either a two-tailed or one-tailed test.

For a z- or t-test with a directional H1, you can only reject H0 if:

1. The difference is in the direction specified by H1, and
2. The computed value of z or t is further out in the tail than the
critical value.

Whether "further out in the tail" means greater than or less than the
critical value depends on which tail it is, obviously.

Finally, one should only be using a directional H1 (aka a one-tailed test)
if a difference in the "wrong" direction, no matter how large, can be
treated exactly the same as a difference of zero. I suspect that quite
often, people use one-tailed tests when this is not really the case.

HTH.

Anthony James wrote

> Dear all,
> When we state a null hypothesis (in case of a t-test or ANOVA), we can
> reject it if the obsrved t or F are greater than the critical values in
> the table. But if we state a directinal hypothesis and the t or F are
> greater than their critical values in the table we accpet the directional
> hypothesis. Am I right?
> We don't want to reject the directional hypothesis.
> Thanks

> Â
> Anthony
>

-----
--
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/Testing-the-null-Vs-directional-hypotheis-tp5150899p5153231.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

Jarrod Teo-2

Re: Testing the null Vs directional hypotheis

In reply to this post by Art Kendall

Hi Art,

Thanks for the correction in the wording of the hypothesis. :)

In any case, may I know then if the information provided in the link is beneficial to someone who would like to perform a directional 2 independent samples T-test then? Why is it that we could take p/2 on the following hypothesis in theory?

Ho: mean height of 15 years old is not statistically distinguishable from the mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Warmest regards
Dorraj Oet

Date: Wed, 18 Jan 2012 07:57:13 -0500
From: [hidden email]
Subject: Re: Testing the null Vs directional hypotheis
To: [hidden email]

Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Ho:� mean height of 15 years old is not statistically distinguishable from the mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old

Art Kendall
Social Research Consultants

On 1/17/2012 8:55 PM, DorraJ Oet wrote:

Hi Anthony and Bruce,
�
First of all, I always enjoy reading Bruce's posting because I feel that he is a great Statistician.
�
I believe this discussion was held in the listing before because� a researchers was trying to perform a 1 tailed� 2 independent samples T-test as well� in SPSS. In which I think this link below was provided and� I am not sure if this information� helps. Do note that the link is an old one and new updates has not been done to the page so� please use this information cautiously.
�
http://www.ats.ucla.edu/stat/spss/faq/pvalue.html
�
One note from this webpage is something which I do find interesting:
�
"Many researchers argue that it is very rarely appropriate to do a one-tailed test.� However, if it is logically impossible for the result to go in one direction (for example, for the mean height of 5-year-olds to be smaller than the mean height of 15-year-olds) or if such a result is of no practical importance (for example, the experimental medicine is less effective than currently used medicine), then a one-tailed test is appropriate."�
�
Bruce, I had heard at least� twice� with regards to managing the results of 1-tailed 2 independent samples T-test in SPSS. Is it really that as according to the article,
�
Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old > mean height of 5 years old
�
Means that we have to take p/2 since the direction is logical
�
Ho:� mean height of 15 years old=mean height of 5 years old
H1: mean height of 15 years old < mean height of 5 years old
�
Means that we have to take 1-(p/2) since the direction is not logical
�
Warmest Regards
Dorraj Oet�
�
__________________________________________________________________________________________________________________________________________________________________________
�

> Date: Tue, 17 Jan 2012 14:58:42 -0800

> From: [hidden email]
> Subject: Re: Testing the null Vs directional hypotheis
> To: [hidden email]
>
> This question doesn't have anything to do with SPSS per se, which may be why
> answers are slow in coming. It would be more appropriate for a general
> stats newsgroup or list. Having said that...
>
> First of all, one does not usually talk about a directional alternative
> hypothesis (H1) for an F-test, because the rejection region for the F-test
> is always in the upper tail. I.e., without resorting to some trickery, the
> F-test implies a non-directional alternative hypothesis. If there are only
> two groups (or paired scores), you can take the square root of F to get the
> corresponding t-value, and perform either a two-tailed or one-tailed test.
>
> For a z- or t-test with a directional H1, you can only reject H0 if:
>
> 1. The difference is in the direction specified by H1, and
> 2. The computed value of z or t is further out in the tail than the
> critical value.
>
> Whether "further out in the tail" means greater than or less than the
> critical value depends on which tail it is, obviously.
>
> Finally, one should only be using a directional H1 (aka a one-tailed test)
> if a difference in the "wrong" direction, no matter how large, can be
> treated exactly the same as a difference of zero. I suspect that quite
> often, people use one-tailed tests when this is not really the case.
>
> HTH.
>
>
> Anthony James wrote
> >
> > Dear all,
> > When we state a null hypothesis (in case of a t-test or ANOVA), we can
> > reject it if the obsrved t or F are greater than the critical values in
> > the table. But if we state a directinal hypothesis and the t or F are
> > greater than their critical values in the table we accpet the directional
> > hypothesis. Am I right?
> > We don't want to reject the directional hypothesis.
> > Thanks
> > �
> > Anthony
> >
>
>
> -----
> --
> 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/Testing-the-null-Vs-directional-hypotheis-tp5150899p5153231.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

Rich Ulrich

Re: Testing the null Vs directional hypotheis

Why take p/2?

- Because the 2-tailed t-test that you are looking at is
artificially constructed from two symmetrical 1-tailed
tests. It is a convention that says, "Make a 5% test by
using 2 1/2% from each end." So if a program is giving
you a 2-tailed result and you want 1-tailed, you divide
the stated p by 2.

Statistical theory starts with 1-tailed tests. A 1-tailed
test *can* be the "uniformly most powerful test" for
a given hypothesis and set of assumptions. A 2-tailed
test is *never* UMP -- You can always beat it, for some
outcomes, by (say) arbitrarily splitting the 5% into 1%
at one end and 4% at the other end. Or, you may find
a test with different mathematics that ends up with that
same sort of result.

If you ask your stat-pack for 1-tailed tests, you have to
be careful when reading the results.
- Is the program assuming that only *positive* differences,
Mean1-Mean2, will be significant for a t? (or the reverse).
- Is the program assuming for correlations that the
direction that you *see* is the direction that you "expected",
so that all extreme r's will be significant, or will it only flag
the positive ones as "significant"? (SPSS corr does provide
and option that will give 1-tailed results, IIRC, but I don't
recall what rule it follows.)

These days - in addition to the risk of applying bad inference
by claiming "one tail" when you should not, there is the
risk of failing to understand what your stat-pack is telling you.

--
Rich Ulrich

Date: Thu, 19 Jan 2012 01:08:52 +0000
From: [hidden email]
Subject: Re: Testing the null Vs directional hypotheis
To: [hidden email]