Print

Print


Dear Ludo and Will,


In answer to your last question, yes, a 1 over the parametric regressor and a zero elsewhere will identify your parametric effect. SPM will show you voxels where the BOLD signal varies as a function of willingness to pay (WTP).


Your WTP values should be zero-mean'ed so that they don't explain variance that could be explained by the main effect (choice response). Otherwise your choice response activations will disappear.


You can then use masking eg. with other contrasts (such as those that identify the choice response regions). So then, you'll have voxels that are both (i) involved in the choice, (ii) change parametrically with WTP.


You can also e.g. plot your parametric effect. See e.g. "Modelling parametric responses" in Chapter 31.3 of the SPM manual (Face fMRI data).


All of these "1's somewhere, zero's everywhere else" contrasts show you increases in the BOLD signal over and above the average value at that voxel.


Best,


Will.


________________________________
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of Mitchener, Ludo <[log in to unmask]>
Sent: 31 January 2017 18:49
To: [log in to unmask]
Subject: [SPM] Contrast specification confusion


Dear SPM helpers,

We are having a bit of trouble with how to specify contrasts with a parametric modulator variable. If you would like to skip the lengthy description of the experiment and go straight to the key questions I have highlighted them in bold at the end. Thanks in advance for your help!

We ran an experiment investigating value-based decision-making in addicted and non-addicted cigarette smokers. Our aim was to locate a ‘value signal’ for cigarette rewards and voucher rewards. And to examine whether these were different between addicted and non-addicted cigarette smokers. We based our study design on this paper:
http://www.jneurosci.org/content/29/39/12315.short

Before the imaging stage of the experiment, participants rated various rewards in terms of their subjective value (or more specifically, how much they would be willing to pay for each reward). The reward was either: a number of cigarettes  or a number of vouchers.

In the scanner, participants were faced with a choice: a set amount of money and one of the reward options. The set amount of money stayed the same throughout the whole task in the scanner. The reward changed every time, e.g. 3 Marlboro Cigarettes, 7 Amazon Vouchers.

So the task design was very simple: a choice (that showed for 3s), followed by an inter trial interval (up to 9s). We also know what their subjective value (or Willingness To Pay - WTP) is for each reward that is an option in the choice (the other options being the constant amount of money).

Now, onto the bit where we are confused:

We specify the model so that it has the following terms: Cigarette-Choice, Cigarette-Choice parametrically modulated by WTP, Voucher-Choice, Voucher-Choice parametrically modulated by WTP, the 6 movement regressors.

We have not modelled the ITI, so that is considered the implicit rest.

When we form the contrasts, if we want to know what is going on when they are faced with a Cigarette-Choice we would specify [1]. That would give us the BOLD response for when they see the choice compared to implicit rest (the ITI). Therefore, it gives us lots of visual activation.

And we could do the same for the voucher choice – which should give a similar result.

Now, if we are interested in finding the region of the brain which tracks the value of the reward (cigarette or voucher) which is available, I would imagine we would use these contrasts: [0 1] for cigarette value signal and [0 0 0 1] for voucher value signal.

The first key question is: Is parametric modulation done by simply multiplying the presence of the event by the parametric modulator? i.e. If we had a list of choices that went: [cig, cig, vouch, cig, vouch] and they had WTPs of: [1, 2, 0.5, 0.25, 1.5] would SPM basically be fitting this model to the data (in order to isolate the cigarette value signal): [1, 2, 0, 0.25, 0]? Or, is it much more complicated than this and you can’t equate it to: the event * the parametric modulator?

The second key question is this: If you specify the contrast of [0 1], such that you are looking only at the cigarette-choice parametrically modulated by cigarette WTP, are you comparing regions that are activated proportionately to the cigarette WTP relative to rest. I was under the impression that any contrast you specify must be relative to something – and if you just have a 1 specified – then our comparison is the implicit rest. Or are you just looking at regions that have activation which is correlated with cigarette WTP – and you are not comparing it to rest. If it is the latter – then why is it not relative to rest – or relative to anything?

The third key question is related to the second: What contrast should I specify in order to get the cigarette value signal? The pure value signal – without any of the other aspects of the choice presentation in it. Would it be [0 1] or would it be [-1 1]? My reasoning for the latter would be I am minusing out the brain response to the presentation of the choice, and leaving myself with only the value signal. But, I have a feeling I have misunderstood something – and really, just specifying [0 -1] will give me the value signal I want.

Apologies for the lengthy description. I hope someone can help!

I suppose it all boils down to this:

When you’ve specified a parametric modulator in your model, do you just need to give it a [1] and everything else a [0] in the contrast manager in order to find the region of the brain which has activation correlated with the parametric modulator – or do you need to do something more complicated?


All the best,

Ludo & Will