Hi Alberto,
> reaction times differences among conditions are, in average, lower than 100ms
> (mean response time vary from 744ms to 840ms with a standard deviation
> lower than 200ms) should I consider them NOT strong differences?
The issue is not whether these differences are "strong" or "weak", but rather whether they are detectable. The answer is usually yes (e.g. see Menon et al, 1998 for BOLD differences due to 50 ms onset asynchonies). Moreover, if you assume that you are dealing with a linear time invariant system then it behooves you to model this variance explicitly.
> what should I fill in as EV duration "without modelling the RT" a RT mean value?
> a fixed interval (cue duration+blank duration+approximate target stimulus processing maybe 300ms)?
You should enter actual RT values, not RT mean values. If you use RT means, then you are inflating your residual variance. Also, you cannot predict how such mismodeling will affect the variance attributed to other correlated regressors.
> in target2 there aren't any effect, it's just a control stimulus to assure that subjects correctly remember cue stimulus characteristics.
Your regressors are correlated. If you don't explicitly model all of the neural events, then you may mistakenly attribute target2 activity to target1.
cheers,
jack
|