Dear Peter,

Thank you for your quick and helpful response. I've checked my design orthogonality. Except for a very small numbers of squares with the cos values (I'm assuming these are the ones you referred to) exceeding 0.7, the majority are < .2.

Regarding your recommendation on creating the 4 regressors in GLM, just so I know I understand you correctly, for the Trials (blocks covering the while trial period), the vector would have all 1's, each 1 is for an event. If I have the following 12 events: color stim_array, color cue, color delay, color probe, shape stim_array, shape cue, shape delay, shape probe, color+shape stim_array, color+shape cue, color+shape delay, color+shape probe, then the vector would look like this:

[1 1 1 1 1 1 1 1 1 1 1 1]. Would this be correct? Somehow I have a suspicion that it isn't.

I guess the Color Delay is more straightforward, I would just have '1' for the color delay event so it would be something like this [0 0 0 0 0 0 1 0 0 0 0 0]

Also, for the effects of interest, each event would also get the value of 1 so the identity matrix would be:
[1 1 1 1 0 0 0 0 0 0 0 0]
[0 0 0 0 1 1 1 1 0 0 0 0]
[0 0 0 0 0 0 0 0 1 1 1 1]

Could you please confirm? Again, thank you for your kind advice and support!

Best,

Miranda E.

On Wed, Oct 5, 2016 at 9:07 AM, Zeidman, Peter <[log in to unmask]> wrote:

Dear Miranda

 

- 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