regression Q

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

regression Q

abdalla alsmadi
Hi evrybody
When comparing test scores, a regression proceadure is preffered for determining how large a difference score must be to obtain significance, correlations between the two scorse must be known & reliability for both scores are available .... would you please tell me what is that formula, how can we compute that???
thanks a lot
 
Prof. Abdalla Alsmadi
college of graduate studies
Arabian Gulf University - Bahrain
Tel: 0097317239999  ext. 676
Reply | Threaded
Open this post in threaded view
|

Re: regression Q

Rich Ulrich
This is a very confused request. 

The Question seems to suggest that you want some
version of a Power Analysis -- "... how large a difference
score must be to obtain significance."

Now, a basic power analysis has 5 specifications:
  a test (like, "two-tailed t-test"),
  a test size (say, "5%"),
  N,
  effect size ("difference"), and
  power (probability of achieving the test size).


For a power analysis --
Given any four of these, you can find the other one.
Given three of them, you can look at a curve or set
of results that represents the other two.  This is properly
done before starting an experiment, to confirm that
the design has a chance of finding something useful.

The formulas for power analyses look at cutoff
points or intersections of non-central distributions.
Mostly, people use a textbook like the one by J. Cohen,
or try to use a computer program, instead of dealing
with the equations themselves.

The other pieces of information you mention, the
reliabilities and correlations, can be important for
figuring out something like the "effect size" for a
particular complicated analysis; or correlations
might be an object of a test. 

I can't guess at all why you mention regression analysis,
unless you are trying to say that you want a power
analysis for a regression.  You *can*  use a regression
to test for a two-group difference -- However, a difference
would usually be assessed by ANOVA or ANCOVA. 
The difference can be translated to R^2  if the test is
going to be performed by multiple regression.

I hope this tells you what you need, or helps you to
frame your question better.

--
Rich Ulrich


Date: Wed, 9 May 2012 01:47:40 -0700
From: [hidden email]
Subject: regression Q
To: [hidden email]

Hi evrybody
When comparing test scores, a regression proceadure is preffered for determining how large a difference score must be to obtain significance, correlations between the two scorse must be known & reliability for both scores are available .... would you please tell me what is that formula, how can we compute that???
thanks a lot
 
Prof. Abdalla Alsmadi
college of graduate studies
Arabian Gulf University - Bahrain
Tel: 0097317239999  ext. 676
Reply | Threaded
Open this post in threaded view
|

Automatic reply: regression Q

Kreischer,Resha M
I will be out of the office until May 15th. I will respond to your email upon my return.

Sincerely,

Resha

=====================
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