Print

Print


It's very fiddly.

If you are feeling up to it, you might consider doing it as a
multilevel model (SPSS calls them a mixed model).   This can be
equivalent to a repeated measures anova, but is much more flexible.

Converting the data into a randomized groups design is wrong, because
you violate the independence assumption.  But the idea of the mixed
model is basically that you do exactly that, except you account for
non-independence by telling SPSS that it is the same subject on
different lines.  In that way, session 1 can be a covariate for the
appropriate group.

You would have  5 variables:

id (which tells you the subject)
session1
condition
session (which would have 2 or 3 in it)
score

Your syntax would be something like:

MIXED
score with session1 by condition session
/fixed session1 condition session condition*session
/random = intercept | subject (id)
/print = solution.


You should also run:
MIXED
score with session1 by condition session
/fixed session1 condition session condition*session
/random = intercept session1 | subject (id) covtype(un)
/print = solution.

And see if the random effect of session1 is significant (if it is,
that means that the effect of the covariate varies across subjects,
and you should keep it in).

I wrote a blog entry which very, very briefly describes the syntax for
this sort of model (I wrote it assuming you knew how to do a
multilevel model in another program, but not SPSS).

http://www.jeremymiles.co.uk/regressionbook/2005/08/longitudinal-multilevel-models-in-spss.html

There's a book called something like "Multilevel models: it's just
regression" which covers mixed models in SPSS, and the 3rd edition of
Andy Field's book (Discovering statistics using SPSS) will also cover
multilevel models, but that's not out yet, so it's not much use.  I
don't know of any other books that cover multilevel models in SPSS,
although it's likely that newer editions of books like Tabachnik and
Fidell will (anyone else know?).

See how you get on, and let the list know how far you get.  (If you
let the list know, then everyone else can learn from it).



Jeremy




.

On 24/04/2008, Alexa Hoyland <[log in to unmask]> wrote:
> Hi All
>
>  I have a fully repeated measures design with two RM factors, condition (5
>  levels) and session (3 levels). Session 1 is a measure of baseline
>  performance taken on each of the test days (each condition). I would like
>  to use this as a varying covariate but am unsure how to do this in SPSS.
>
>  I am aware how to do it if my design only involved a baseline measure
>  followed by one measure of performance - this is possible using syntax in
>  SPSS. Alternatively, I could convert the data into a randomised-groups
>  design, where measurements for each case for each condition are on a
>  separate line. However, as I have two RM factors I can't work out how to
>  do it.
>  Any suggestions?
>
>  Thanks for your help
>
>
>  Alexa
>


-- 
Jeremy Miles
Learning statistics blog: www.jeremymiles.co.uk/learningstats
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com