If the numbers of trials for all the conditions were the same then
you would get the right weighting but this is not guaranteed in
general. To adjust the weights you could just do:
S.c = [1 1 1 1]/4;
etc.
Best,
Vladimir
On Thu, Dec 13, 2012 at 7:39 AM, Deborah Talmi
<[log in to unmask]> wrote:
> Hi Vladimir,
>
> Thank you for your quick response!
>
> I *think* this is different to fmri? If I'm not mistaken, there you get
> identical results for
>
> 1 -1 1 -1 1 -1 1 -1
>
> 2 -2 2 -2 2 -2 2 -2
>
> And anything that sums to zero.
>
> I don't know how I would adjust the weights myself - what do you suggest?
>
> I tried weightave=1 and it seems to have done what I wanted (max value now
> 2, rather than 8).
>
> Please could you clarify?
>
> Thanks, d
>
> Sent from my Android phone using TouchDown (www.nitrodesk.com)
>
>
> -----Original Message-----
> From: Vladimir Litvak [[log in to unmask]]
> Received: Thursday, 13 Dec 2012, 6:12
> To: Deborah Talmi [[log in to unmask]]; [log in to unmask]
> [[log in to unmask]]
> Subject: Re: contrasting
>
> Hi Debbie,
>
> WeightAve refers to the weighting by the number of trials used to
> compute the original average waveforms that you are contrasting. So in
> any case it would not weight by the number of conditions. This option
> is useful when you use contrast to combine several conditions with
> different original trial numbers into one. If you want to compute the
> mean rather than the sum of some waveforms you need to adjust your
> contrast coefficients yourself as I think is also the case in fMRI
> contrasts.
>
> Best,
>
> Vladimir
>
> On Wed, Dec 12, 2012 at 7:44 PM, Deborah Talmi
> <[log in to unmask]> wrote:
>> Hi Vladimir,
>>
>>
>>
>> This is a cheeky question with regards to
>>
>> spm_eeg_weight_epochs
>>
>>
>>
>> I think I have misunderstood the WeightAve flag.
>>
>>
>>
>>
>>
>> It says “flag whether average should be weighted by number of replications
>> (yes (1), no (0))
>>
>>
>>
>> Which I have taken to mean that an average will be calculated, namely,
>> you’d
>> get a normal average with a value of 0. But that’s not the case, right?
>> Instead a sum is calculated, and an average only calculated with a flat=1?
>>
>>
>>
>> That’s the gist of my question but I’ll write down the whole story below
>> if
>> that’s of any help
>>
>>
>>
>> Thanks ,d
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> I used a contrast [1 -1 1 -1 1 -1 1 -1]
>>
>>
>>
>> It seem to have done the sum
>>
>>
>>
>> (Cond 1 + cond 3 + cond 5 + cond 7 ) - (Cond 2 + cond 4 + cond 6 + cond 8
>> )
>>
>>
>>
>> Instead of either of those two below, which is what I expected:
>>
>>
>>
>> ( (Cond 1 – cond 2) +(Cond 3 – cond4) +(Cond5 – cond6) +(Cond7 –
>> cond8))/4
>>
>>
>>
>> Or (obviously equal)
>>
>>
>>
>> Average ( (Cond 1 – cond 2) ,(Cond 3 – cond4),(Cond5 – cond6) ,(Cond7 –
>> cond8))
>>
>>
>>
>>
>>
>> I’m saying this because both my topography and the field intensity give me
>> a
>> value of 8, but my script, which does the average, gives me a value of
>> about
>> 2 .
>>
>> It would be helpful, if it is at all possible, to label the units of the y
>> axis in both of these plots.
>>
>>
>>
>>
>>
>>
>>
>> condition_labels=[2251 2250 2751 2750 251 250 751
>> 750];
>>
>> prefix='bfmaMMMcaedffM';
>>
>> file=[prefix 'spm8_2.mat']
>>
>> S.D =[file];
>>
>>
>>
>> S.c = [
>>
>> 1 -1 -1 1 1 -1 -1 1 %expectancy; low=1, high=-1
>>
>> 1 -1 1 -1 1 -1 1 -1 %outcome (omitted - received)
>>
>> 1 1 -1 -1 1 1 -1 -1 %expectancy by outcome
>>
>> ];
>>
>> S.c=[eye(8); S.c];
>>
>> S.label = {
>>
>> cellstr(num2str(condition_labels(1)))
>>
>> cellstr(num2str(condition_labels(2)))
>>
>> cellstr(num2str(condition_labels(3)))
>>
>> cellstr(num2str(condition_labels(4)))
>>
>> cellstr(num2str(condition_labels(5)))
>>
>> cellstr(num2str(condition_labels(6)))
>>
>> cellstr(num2str(condition_labels(7)))
>>
>> cellstr(num2str(condition_labels(8)))
>>
>> 'exp'
>>
>> 'out'
>>
>> 'exp_out'
>>
>> }';
>>
>> S.WeightAve = 0;
>>
>> D = spm_eeg_weight_epochs(S);
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> From: Deborah Talmi
>> Sent: 12 December 2012 18:27
>> To: 'Vladimir Litvak'
>> Subject: RE: clarification...
>>
>>
>>
>> Hi,
>>
>> Thanks for reminding me about the F test! I did know this at one point.
>> Glad
>> it will be resolved in spm12. Yes, the maxima appears more noisy.
>>
>>
>>
>> Cheers, d
>>
>>
>>
>>
>>
>> From: Vladimir Litvak [mailto:[log in to unmask]]
>> Sent: 11 December 2012 21:19
>> To: Deborah Talmi
>> Subject: Re: clarification...
>>
>>
>>
>> Hi Debbie,
>>
>>
>>
>> I think the reason you don't have cluster-level significance is that you
>> are
>> doing an F-test. There is no cluster-level correction for F-tests in SPM8
>> for any modality, but there will be in SPM12. Regarding the other issue
>> perhaps you should use CheckReg to look at the F/T, con and ResMS images
>> together to see what's going on. There might be more noise around the peak
>> difference which reduces the statistic value there. If you see something
>> unexpected, let me know.
>>
>>
>>
>> Best,
>>
>>
>>
>> Vladimir
>>
>>
>>
>> On 11 Dec 2012, at 20:10, Deborah Talmi wrote:
>>
>>
>>
>> Hi Vladimir,
>>
>> Just to clarify my previous query on cluster stats. The maxima of one of
>> the
>> clusters I’m reporting for the comparison A>B is, say, x ms after an
>> event.
>> Like I said, I only get the maxima without info on whether the entire
>> cluster is significant.
>>
>>
>>
>> When I extract the time course from that peak for the difference wave A-B
>> I
>> see a the peak somewhere else, say x+50 ms after the event. Firstly I’m
>> surprised that the maxima is not at the same place as the waveform peak,
>> but
>> I am sure this can happen; it makes less sense to me that the waveform
>> peak
>> wouldn’t even be significant.
>>
>> I hope this makes some sense!
>>
>> Hanuka sameach
>>
>> d
>>
>>
>>
>>
>>
>> -------------------------
>>
>> Deborah Talmi, School of Psychological Sciences, University of Manchester
>>
>> http://www.psych-sci.manchester.ac.uk/staff/talmi
>>
>>
>>
>>
|
