Date:         Mon, 26 Feb 1996 14:14:55 GMT
Reply-To:     m.voeten@ped.kun.nl
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         Rinus Voeten <m.voeten@PED.KUN.NL>
Organization: K.U. Nijmegen
Subject:      Re: Regression problem

Helen Mok <hhymok@psy.cuhk.hk> wrote:
 
>Could someone kindly advise me as to some appropriate procedure regarding
>the following problem:
 
>I've a data set collected from 30 different schools with different levels of
classes.  I want to
>use regression to predict the students' average score(Y) using a psychological
variable(X).
>However, different schools and classes may affect their scores.  So, is it
reasonable to use the
>standardized values of X and Y to do the regression?  If so, how can I produce
these standardized
>values and run the regression in SPSS?
 
 
Probably the best thing to do would be to use software for multilevel
analysis, like VARCL, HLM or MLn. That would allow you to properly
account for the two levels in your data (pupil and class). Multilevel
regression allows for the possibility of class differences between
regression lines and it will give you better estimates of standard
errors.
 
Having to deal with different class levels, you should indeed use some
form of standardization. Four possibilities come to my mind:
(1) if you have norms (or simply means and standard deviations from
some other representative data set) available for the measures of your
variables you could use these to compute standard scores.
(2) if you have data on age of subjects you could compute
age-equivalent scores.
(3) you can compute standard scores separately on your data for the
various class levels, setting the mean for each class level to 0 and
the standard deviation to 1.
(3) you can compute standard scores for your whole sample, setting the
sample mean to zero and the sample standard deviation to 1 (or other
convenient numbers).
 
I don't know what would work best for your problem.