- Should all the event onsets (e.g., onset for stimulus arrays, cue, delay, probe) be included as regressors? As this is an event-related design, for other analyses, the natural answer would be yes. However, for DCM do the events matter or only the conditions matter?

 

The design considerations for DCM are identical to those for the standard SPM analysis. If you model each of those parts of the trial, you may find there is collinearity between the parameters (I’m assuming the onset, cue, delay and probe are all presented on every trial with similar timing). You could verify this by clicking Review in SPM, opening a single subject’s SPM, click Design in the grey window, and click Design Orthogonality. With highly collinear parameters (dark squares in that GUI, with values greater than around 0.7), it will be hard to get significant results.

 

That said, you said you’ve already found results that you’re happy with, so it may be not be causing a big problem for you. One further caveat. The basic deterministic DCM assumes all effects are driven by the stimuli – there are no endogenous brain dynamics modelled. If you have a long delay period, that could cause problems for you.

 

Here’s one way to model this design, which is probably what I would try first. Have 4 regressors in the GLM: Trials (blocks covering the whole trial period), Color Delay (blocks or events for just the Color delay periods), Shape Delay (blocks or events for just the Shape delay periods), and Color+Shape Delay. Use this design to extract ROIs based on the Trials regressor for the driving input region, and use contrasts comparing color and shape delay periods to select the other ROIs. Then for the DCM, have Trials as the driving input and the other regressors as modulatory inputs.

 

(Another option would be to form a Color-Shape regressor that represents the difference between color and shape, and use that as a modulatory input, alongside Color+Shape.)

 

I hope that helps!

 

Best

Peter