Dear Allstat members
I have a query regarding some medical data and would value your
opinions.
The absence or presence of three radiographic features
(osteophites, sclerosis and disc narrowing) has been recorded for a
group of individuals at four levels in the lumbar spine. So for each
individual there are 12 observations, ie the presence / absence of
three features, each measured at four sites. We are interested in
the relationships between the features, say, at a particular level,
but also the patterns of the presence/absence of a feature down
the spine.
If we had a single level of the lumbar spine then I could use log-
linear modeling to investigate the patterns in the presence/absence
of the features, and the parameters could then be converted to
odds ratios for ease of interpretation. However, with the 3 features
measured at multiple sites within a spine, which can obviously not
be taken as independent measures as there will be correlation in
the features within a person (or spine), an extra level of complexity
is added.
On a similar note, is it possible to fit a log-linear model in which all
two-way interactions are assumed to be equal?
I would be very grateful for ideas of how to analyse the data and
which statistical packages would be the most appropraite to use.
Best wishes
Elaine Thomas
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|