Hi Fiona
Multiple slightly connected thoughts.
I don't think there exists, or could exist a simple guide to repeated
measures ancova. It's a horrible procedure. If statistics were being
reinvented today it wouldn't exist.
So what are you supposed to do? it's far more straightforward, and
for more interpretable, and better in many ways to do this as a
multilevel model. R (which is the newest stats package around)
doesn't (I don't think) have a procedure for repeated measures anova.
If you try asking on the R list how to do it, you sometimes get a
reply along the lines of "If you want to do repeated measures analysis
of variance, you obviously don't know enough about data analysis to be
analysing data".
I'm not sure what you mean by "takes out an interaction term for every
covariate", but if you click on model you should be able to specify
the model. (Except the thing I hate about SPSS repeated measures is
that some features of the model seem to be implicit).
Missing data are a PITA for repeated measures ANOVA, because if a
person misses one variable, they are thrown out. That can lose a lot
of your data, but a multilevel model (or a structrual equation model,
for that matter) doesn't care about missing data.
If your data are non-normal, or binary or ordinal or Poisson or
anything else, you are completely stuck if you want to do repeated
measures anova, for a multilevel model (or a structural equation
model) it's a pretty minor tweak.
As an example, for the past couple of days I've been looking at people
smoking at different ages. People were asked if they had smoked in
the past 30 days once a year for 5 years. First, I can't do repeated
measures anova because it's binary. So I'm stuck. Second, people
were aged from 15 to 25, and I'd like to know about smoking across
that range of ages - but that's 10 years, and the best I have for
anyone is 5 measures, so most of the data are missing. So even if my
data were normal and continuous, I couldn't do it because I've got too
much missing data - everyone is missing at least 5 data points, and
some people are missing as many as 8.
But as a multilevel model, it's surprisingly easy - I just tell my
stats package it's binary, and I'm away.
The continuous case is covered in SPSS in Andy Field's book
Discovering Statistics.
Jeremy
On 18 February 2010 07:47, Fiona Essig <[log in to unmask]> wrote:
> Hi,
>
> Can anyone point me towards a very simple (some hope!) guide to repeated
> measures ANCOVA? The GLM option in SPSS takes out an interaction term for
> every covariate, which I know can't be great! Also is it sufficient to
> report partial eta squared for effect size with this?
>
> Many thanks
>
> Fiona
>
> --
> Tel: 01707 284 761
> E-mail: [log in to unmask]
>
> Fiona Essig
> Research Student
> Room E384 Research Huts
> School of Psychology
> University of Hertfordshire
> College Lane
> Hatfield
> Herts AL10 9AB
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|