Print

Print
Hi Donald, I really appreciate your response. I've got a better picture of the math and stats due to this reply. However, I've got a few additional questions that might clarify some other things for me. Please check the inline response to your comments. 

On Mon, Mar 2, 2015 at 11:01 AM, MCLAREN, Donald <[log in to unmask]> wrote:
Please see inline responses below.

On Sun, Mar 1, 2015 at 12:56 PM, Lisa McDermott <[log in to unmask]> wrote:
Hello all, 

I feel I am close to understanding the definition of contrasts, but I am still confused about some issues. I was wondering if I could ask your guidance for defining contrasts in my fMRI experiment. Here are the onsets and durations of my experiment: 

Onset Duration
Fixation 0 8
Recrangle 8 20
Fearful_f 28 20
Circle 48 20
Neutral_f 68 20
Triangle 88 20
Fearful_m 108 20
Circle 128 20
Neutral_m 148 20
Rectangle 168 20
Neutral_f 188 20
Triangle 208 20
Fearful_m 228 20
Rectangle 248 20
Neutral_m 268 20
Triangle 288 20
Fearful_f 308 20
Circle 328 20
Fixation 348 2

I've color coded as well as I can. In the 1st level analysis, I modeled the following (in order): 
Shapes (black, not bolded): abbreviated S from now on
FearfulFaces (color coded red): abbreviated FF from now on
NeutralFaces (color coded purple): abbreviated NF from now on
And I am calling the set of all faces (both fearful and neutral) as F

I am not concerned about the differences between males and females right now. 

However, I am having a lot of confusion with the contrasts. I want to do the following contrasts: 
FF vs. S
FF vs. NF
NF vs. S
F vs. S

In the contrast manager, I defined the contrasts as follows: 
.tcon.name = 'FearfulFaces vs. Shapes';
.tcon.convec = [-1 1 0];

.tcon.name = 'FearfulFaces vs. NeutralFaces';
.tcon.convec = [0 1 -1];

.tcon.name = 'NeutralFaces vs. Shapes';
.tcon.convec = [-1 0 1];

.tcon.name = 'Faces vs. Shapes';
.tcon.convec = [-2 1 1];


>> These all look correct.
 
I am not sure as to the blood flow implications of what I am measuring when I am defining the above contrasts. I would appreciate it if someone could explain the blood flow implications for each contrast to me. 

>> They mean that one condition has a larger BOLD response than the other condition, or less of a deactivation. Blood flow is only one component of the BOLD response, so I would not intermix the two phrases.

Let's consider the 2nd contrast: FF vs. NF [0 1 -1] 
Let us consider that a single voxel shows activation during this comparison. 
Does this measure whether the BOLD signal for FF is greater than the BOLD signal for NF? If this is true, can we say that, this voxel is more activated during FF than during NF? 

And what about deactivations? What contrast should I use to see if a voxel deactivated? 

Basically, I am unsure of what you mean when you say "one condition has a larger BOLD response than the other condition, or less of a deactivation." Can you please elaborate on that? How can one contrast measure both? 


In particular, I am very confused about the 4th contrast: Faces vs. Shapes
I am told that the sum of the terms in the contrasts should be zero (I was previously using [-1 1 1] for the Faces vs. Shapes contrast before I was corrected by my colleague). Why is that? 

If one were to use [-1 1 1], this means that the sum of the BOLD response for FF and NF is greater than that for shapes. Hypothetically, if shapes and faces both had a BOLD response of 2, then you would incorrectly conclude that faces had a larger BOLD response than shapes because the contrast*b would be positive: -1*2+1*2+2*2=2. With the new contrast [-2 1 1], you would correctly get 0 as all three stimuli had the same response. Note, the correct choice really depends on the question; although in this case, I'm not sure what hypothesis would lead to using the initial contrast.

Is the above (marked in red) a typo? Did you mean to say -1*2+1*2+1*2=2? That would make a lot more sense to me. 
 

I would suggest that instead of defining the contrasts, you should first define the null hypothesis. In the above case:
Ho: shapes=faces

then make that equal to 0

Ho: shapes-faces=0 or faces-shapes=0

now since faces is actually 2 stimuli, you need to replace faces with what the actual value. Faces is the average of NF and FF.

Ho: (NF+FF)/2-shapes=0

Now, get the coefficients for each term in the model. Any term not in the null hypothesis has a weight of 0.

NF --> 1/2
FF --> 1/2
S --> -1

So the contrast would be [-1 1/2 1/2]. This will give yo the same t-statistic as [-2 1 1], but the contrast value will be half of [-2 1 1]. The contrast value will match the individual conditions better with [-1 1/2 1/2]. 
 
I do understand that [-2 1 1] and [-1 1/2 1/2] will have the same spmT images. However, what do you mean when you say that the "contrast value will match the individual conditions better" with the latter contrast? Should I change my contrast to the latter value? 


When there is a simple activation paradigm, say finger tapping vs. rest, I know the contrast is just [1, 0], which doesn't sum to zero. So why does the 4th contrast have to sum to zero? What is the criteria for the sum of the digits in the contrast vector summing to zero? 

It all depends on the null hypothesis, which should be set to 0 before making the contrast. Null hypothesis that are already equal at the beginning of the process to 0 won't sum to 0. All other should sum to 0.

Hope this helps.

I am also covarying for a variable. I used 1-sample t-test for my 2nd level and added this variable as a covariate. I took great care to ensure that the order of subjects matched the order in which I presented the covariates. I am trying to find which area shows negative correlation with that variable. In other words, when the score of the variable increases, which areas show a corresponding decrease in activation? The script ran successfully, but I am having trouble with the contrast I am to use in the 2nd level SPM.mat file. Is it [0 -1] (where 0 is the contrast vector, and -1 is the covariate vector)? Also what does the contrast of [0 1] mean for this analysis? 

Another thing is, I am having a lot of trouble finding the difference between a perceived decrease in activation vs. deactivation. What contrast should I use to distinguish the two? 


Thank you. 
- Lisa