1-tailed test and alpha

[Helga Walz]

Your assistance, please

 

SPSS provides probabilities for 2-tailed tests only.

With a one-tailed test with alpha 0.05, I understand that I need to
divide

the Sig. (2-tailed). Thus if SPSS lists the p-value  = 0.881, that means
for the

1-tailed test the p-value is 0.4405.  

Does the alpha level have to be changed in SPSS, from 0.05 to 0.10 for
the same

1-tailed test.

Also, is the 95% Confidence Interval affected at all whenever we conduct
a

1-tailed test i.e., the lower and upper values?.

 

Thank you for your assistance,

 

Helga Walz

 

Helga S. Walz, Ph.D.

Division of Applied Behavioral Sciences

University of Baltimore

1420 North Charles Street

Baltimore, Maryland  21201

 

Tel.: [phone]

Fax.:[phone]

 

[Richard Ristow]

At 09:50 AM 10/23/2006, Helga Walz wrote:

>SPSS provides probabilities for 2-tailed tests only.
>
>With a 1-tailed test and alpha 0.05, I understand that I need to
>divide the Sig. (2-tailed).
>
>Thus, if SPSS list the p-value as 0.881 [I assume you mean .0881],
>that means for the 1-tailed test the p-value is 0.04405.

To give a side piece of advice: consider VERY carefully before doing a
one-tailed test. 95% of the time, stay with the two-tailed test. The
one-tailed test has more statistical power, but very marginally so. And
the one-tailed test is valid only if a result in the unexpected
direction is truly of no interest - a rare case.

[Stephen Brand]

Helga,

If you can justify using a one-tailed test, then you can read the output
from a two-tailed test as follows:  A correlation that is significant at
alpha < .10 in the predicted direction is significant at alpha < .05 in
the one-tailed sense.  I stress that the correlation must be in the
predicted direction.  Any correlation in the opposite direction, no
matter how huge, counts against your alternative hypothesis.

As Richard pointed out, you must be very careful about choosing to use a
one-tailed test.  Can you really justify why the results would only go
in one direction?  Journal reviewers are very reluctant to accept a
one-tailed test, even when there is a great deal of evidence to justify
why an effect should go one way instead of another.

HTH,

Steve

Stephen Brand, Ph.D.
Associate Professor (Research)
NCPE-SP, University of Rhode Island
Kingston, Rhode Island

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Helga Walz
Sent: Monday, October 23, 2006 9:51 AM
To: [hidden email]
Subject: 1-tailed test and alpha

Your assistance, please



SPSS provides probabilities for 2-tailed tests only.

With a one-tailed test with alpha 0.05, I understand that I need to
divide

the Sig. (2-tailed). Thus if SPSS lists the p-value  = 0.881, that means
for the

1-tailed test the p-value is 0.4405.

Does the alpha level have to be changed in SPSS, from 0.05 to 0.10 for
the same

1-tailed test.

Also, is the 95% Confidence Interval affected at all whenever we conduct
a

1-tailed test i.e., the lower and upper values?.



Thank you for your assistance,



Helga Walz



Helga S. Walz, Ph.D.

Division of Applied Behavioral Sciences

University of Baltimore

1420 North Charles Street

Baltimore, Maryland  21201



Tel.: [phone]

Fax.:[phone]