Many thanks for your help Donald!

I have some follow-up questions below:

Both PMs and regressors take away variance from the signal. If your regressor was identical to the PM term, then it wouldn't matter if you called it a PM or a regressor.

I'm surprised that it wouldn't matter to my result whether it's a PM or a regressor, since the regressor isn't convolved with the HRF, and the PM is convolved with the HRF..... could you please explain a bit more?

I have a single behavioural value per trial (not per timepoint).

So if I put in a PM with 10 values (1 per trial), it would be the same as putting in a regressor with each of these 10 values repeated many times (ie from trial1 start timepoint to trial 1 end timepoint copy paste the trial 1 behavioural value) ? I wouldn't think so since the regressor will not be convolved with HRF, and the PM will be convolved with the HRF.

If you had a behavioral regressor with 1 value per timepoint, you could put it in as a regressor and get a contrast value for it. However, I'm not sure how you'd interpret a behavioral covariate at each timepoint that isn't convolved with the HRF as you need something that would align to the BOLD signal (BOLD signal is the neural activity filtered with a hemodynamic response function (not necessarily the canonical HRF).

Ok, so I guess the regressor maybe only makes sense when I really want to take something away (ie motion parameters), and doesn't make sense to specifically focus on a regressor with a contrast?

I guess I'm still struggling a bit with how to include behaviour in my model. My hypothesis is that behaviour is relevant to activation (the HRF) (so I don't want to regress it out). So the best option seems to be to include it as a PM. Another thing I was thinking about is to do a PPI. But from my understanding that looks at connectivity and behaviour, rather than activation and behaviour.

What do you think?

Thanks,

Joelle

On Thu, Aug 20, 2015 at 12:08 PM, Joelle Zimmermann <[log in to unmask]> wrote:
Hi Donald - thanks very much!

So by adding behavioural as a regressor, I'm taking any effects of behaviour (on the signal) away? I'm more interested in how behaviour across the trials actually modulates activation, which is what is done with a parametric modulator, right?

I wonder what it means if I do use behaviour as a regressor, and then create a contrast highlighting this regressor, and take that to the group level, where I do a one sample t-test with these .con images... would you be able to give me your thought about that?

On Thu, Aug 20, 2015 at 11:52 AM, MCLAREN, Donald <[log in to unmask]> wrote:
See below.

Best Regards,
Donald McLaren, PhD

On Thu, Aug 20, 2015 at 9:59 AM, Joelle Zimmermann <[log in to unmask]> wrote:
Hi SPMers,

I have a couple of questions about parametric modulators. I'd appreciate thoughts about any/all of them :-)

1) When entering a parametric modulator at the first-level, where I have one condition with 10 onsets, am I expected to have 10 parametric modulator values? (ie one per onset)? The PM values I want to enter are behavioural performance values, and indeed I have 1 performance value per onset, so logically this would make sense, I am just not sure what the program expects.

PMs are meant to modulate each trial, thus you need to have 1 value per trial.

2) Can I use raw behavioural performance scores as covariates? These performance scores are all numbers between 0 and 1. Or is some adjustment to the values advised?

As long as you think the BOLD response changes linear over the range of performance, no adjustment is necessary. The program should remove the mean from the values.

3) MOST IMPORTANTLY: What's the difference between adding behavioural performance as a parametric modulator, and adding it as a regressor?

Regressors aren't convolved with the HRF.

Thanks in advance,
Joelle