Dear Mark ,
Thank you for your previous message. I have one further question about this issue.
When we include the behavioural data in the analysis these data are actually a composite score
based on how much time a person spent looking at a portion of the screen. We have 3 areas that we monitor,
let's call them A, B & C. We're interested in accounting for differences in the amount of time people spend looking
at the 3 areas of screen, so we can say something sensible about the actual stimuli presented in these locations.
I.e. if we want to look at effects of a fluffy kitten presented in area A, but subjects spent all their time looking at the
mouldy potato in area B, we would like to account for this. We're not interested in time spent looking at C, which is normally
a very small proportion of the trial time.
So my idea was to create 2 regressors A / (A+B+C) and B/(A+B+C) and demean them each individually, and include these
as nuisance regressors at the first level. Given that C is close to zero, then B can be predicted by A, does this cause
us any problems with modeling? Should we only include one regressor i.e. A *or* B?
Should we also include C = C/(A+B+C)?
Any help would be much appreciated,
Olivia
|