I have two conditions fast ER experiment, where for each trial (and TR) I have
a value, which varies. The vector of these values I would like to convolve with
basic functions and to estimate the model in order to get the activations. In
other words, the value of my resgressor will be not "1", but my variable, which
varies in each TR.
I though, parametric modulation (pmod) is exactly what I need. Once I put the
vector of my values in pmod, I get additional regressor for each of my
condition/sessions. Now, making t-contrast on only modulated regressors
results in total miss of activations. To ensure that these are not my values,
which are responsible I used in pmod some very close to "1" values (1.00001
etc). Although one would expect to get almost identical to non-modulated t-
contrast results, the result which I get is total mess. After debugging the code
(spm_get_ons.m) I understood that modulation regressor values are
orthogonolized with non-modulation regressor value ("1"); the first regressor
always remains as before, while the second is changing dramatically. So, it
explains why putting 1.00001 values resulted in nothing.
My next step was just to hook and to put my values instead of "1" regressor
(variable u). So, I have no modulations, but only two regressors, while they are
not "1", but my own values. When I doing the things this way, my estimation
activations results perfectly make sense.
1. How parametric modulation supposed to work?
2. Is it OK to substitute the native regressor values "1", with my variable
Thanks for help,