If you have a within subjects design then it's a little tricky.
Are your variables really Poisson? Are they counts?
I think, but I might be wrong, that you need to use something fairly
sophisticated to handle this - something like a multilevel model, a
complex sample, or a generalized estimating equation, and I don't
think you can do any of those in SPSS with a Poisson distribution.
How horribly non-normal are your data Could you get away with
repeated measures ANOVA? Don't forget it's not the variables that
need to be normally, but the differences / residuals.
Jeremy
2008/8/15 Sarah Ellis (Biological Sciences) <[log in to unmask]>:
> Dear All,
> I am hoping someone can help me. I have frequency data (DV) with an IV with
> 4 levels in a within subjects design. I have been advised to analyses this
> data using Generalized Linear Models with a Poisson Error Distribution due
> to the lack of normality with the frequency data. From reading up on
> Generalized Linear Models, I think Loglinear is the option to go for but I
> cannot figure out how to conduct this test on SPSS with a within subjects
> design and with only 1 IV. Can anyone help?
>
> Many thanks in advance
>
> Sarah
>
>
--
Jeremy Miles
Learning statistics blog: www.jeremymiles.co.uk/learningstats
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|