Hi Sara,

PE can be used as parametric regressor, but I would input the other as standard variable, thus partialling out there effects on the variables of interest.

Cyril 

Sent from my HTC

----- Reply message -----
From: "Sara Garofalo" <[log in to unmask]>
To: <[log in to unmask]>
Subject: [SPM] negative parametric modulators and orthogonalization
Date: Fri, Jun 26, 2015 09:47

Thinking about the mean-centered values, I have another doubt about the use of dummy variables as PMs.

In my case, for one EV I am using three PMs: one is the prediction error (PE) (calculated with a computational model), and two more are dummy variables used to define three trial types. So, for example I have:

PE     reward    punishment
.3           1                0
.6           1                0
.2           1                0
.2           1                0
.4           0                1
.6           0                1
.6           0                0
.8           0                0
.2           0                0

(the third trial type of the last three trials is automatically defined, given that there are no ones in the first two dummy variables, am I right?)

after being mean-centered they are transformed in

           PE      reward       punishment
 -0.13333333   0.5555556  -0.2222222
  0.16666667   0.5555556  -0.2222222
 -0.23333333   0.5555556  -0.2222222
 -0.23333333   0.5555556  -0.2222222
 -0.03333333  -0.4444444   0.7777778
  0.16666667  -0.4444444   0.7777778
  0.16666667  -0.4444444  -0.2222222
  0.36666667  -0.4444444  -0.2222222
 -0.23333333  -0.4444444  -0.2222222

Of course, in this case I am interested in looking at the PE and taking out the variability explained by the trial type, but does a dummy variable still make sense after being transformed in this way?

Many many thanks,

Sara