Correlation in crossover studies
Chris
I have some data from a simple 2x2 crossover study with 20 subjects.
I have been asked to look for correlations between two variables. But
I am unsure how best to go about this. I'm pretty sure I cannot just
use all the data and do a simple Spearman's correlation (i.e. with 40
observations). So should I split the data so I do a correlation on
each treatment group or use multiple regression? What would be the
arguments for and against these two methods?
If I use multiple regression I would have var1 as the dependent
variable and var2 as an independent variable presumably I just add
subject (not treatment?) as a second independent variable?
Many thanks for your help.
Regards
Chris
Steve Simon, P.Mean Consulting
I didn't see any answers to your question. If you are just satisfying a
request to get someone off your back, then two separate correlations,
one for treatment A and one for treatment B is the simplest and most
palatable option. As a quick check make sure that the correlation is
about the same for those who get A first versus those who get B first.
If you want to explore this, and think it is a suggestion worth a bit
more of your time and energy, then go for it! I think your thoughts
about using a multiple regression model would be great. Look for effects
where treatment and treatment order influences the slope and intercept
of your regression model. You have to account for correlated errors
using a mixed linear model, of course. It's also a bit tricky sometimes
to decide which of the two measures is the dependent variable. But a
regression model allows far greater flexibility in modeling than a
correlational model.
I hope this helps.
--
Steve Simon, Standard Disclaimer
Second free statistics webinar, Wed, Nov 4, 11am-noon CST.
"The first three steps in data entry, with examples in PASW/SPSS"
Details at www.pmean.com/webinars