Hi Thomas. Thanks for your reply.
The difference between the Conjunction null and Global null hypothesis
surely makes sense from a statistical standpoint. However, and this may be
because of my own naivety, I'm wondering how the Conjunction null hypothesis
fits from a neuroanatomical standpoint. I'm assuming that the conjunctional
null allows you to infer that a specific voxel is activated in every single
subject's brain tested. However, since everyone's brain anatomy/topography
is different (no matter how much normalization/pre-processing we do in an
attempt to account for this) is the conjunction null the best way to
describe a brain's activation (i.e. if 9 out of 10 subjects have that same
voxel activated, however 1 out of the 10 does not, then the conjuction null
hypothesis holds true and no brain activation will be reported). Say for
example though that the 1 subject has a voxel activated 2-3 voxels away from
the other 9 of 10 (i.e. different voxel activated but still the same brain
area activated), the conjunction null will again report no brain activation,
eventhough all 10 subjects have activation in the same area. I guess my big
beef, and I may have the reasoning all wrong, is that the conjunction null
looks at a voxel to voxel comparison when I'm more interested in the
activity in a certain area. So would the Global null, since its more
sensitive, be more representative of area activation? Maybe people on the
list can correct my assumptions if need be....any thoughts would be good.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Nicholas Masse, M.Sc.
Research Technician
Neurovascular Research Lab
School of Kinesiology
University of Western Ontario
Rm. 3110 Thames Hall
519.661.2111 x88526
[log in to unmask]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
----- Original Message -----
From: "Thomas Ethofer" <[log in to unmask]>
To: "Nicholas Masse" <[log in to unmask]>
Sent: Monday, February 28, 2005 2:09 PM
Subject: Re: [SPM] Negative T Values and Conjunction Analysis
> Hi Nicholas!
>
> Before you use a conjunction analysis, it's crucial to think about what
> you want to test. Significant results obtained by using a Conjunctions
> based on the composite null hypothesis indicate that activation was found
> in two conditions or subjects or group of subjects.
> If you use a conjunction based on a global null hypothesis you can only
> say that you found activation in at least one condition or subject or
> group of subjects (whatever you use in your conjunction).
> A global null hypothesis offers the advantage that you can lower your T
> threshold for obtaining a significant effect.
> There is no lower limit for T values (they can even be negative!) if you
> calculate a conjunction over many subjects or conditions using a global
> null hypothesis. However, you pay for this increased sensitivity with loss
> in inference (you can only say that you found activivation in at least one
> of the contrasts).
> Tom Nichols and Matthew Brett mainly published their paper about
> conjunctions because many people used the increased sensitivity of a
> global null, but the inference of a composite null (namely that they found
> activation in each and every contrast).
> To demonstrate the absurdity of this approach, they showed that your
> threshold can be even negative if you use a lot of contrasts for your
> conjunction.
>
>
> Greetings
>
> Tom Ethofer
>
>
>
>
> Nicholas Masse wrote:
>> Yes I have, but what I gathered from the reports was that it basically
>> just
>> said negative t values can occur when the hypothesis is based on the
>> minimum
>> t statistic...it didn't really explain the validity or what the negative
>> t
>> values really meant. Although maybe I didn't get much out of the articles
>> since statistics is not my forte....I'm still learning!
>>
>> Any further explanation would be much appreciated, especially on
>> displaying
>> negative t values!
>>
>> Thanks
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> Nicholas Masse, M.Sc.
>> Research Technician
>> Neurovascular Research Lab
>> School of Kinesiology
>> University of Western Ontario
>> Rm. 3110 Thames Hall
>> 519.661.2111 x88526
>> [log in to unmask]
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
>> ----- Original Message -----
>> From: "Matthew Brett" <[log in to unmask]>
>> To: <[log in to unmask]>
>> Sent: Monday, February 28, 2005 1:21 PM
>> Subject: Re: [SPM] Negative T Values and Conjunction Analysis
>>
>>
>>> Hi,
>>>
>>>> I'm writing in hopes of raising a discussion on the phenomenon of
>>>> negative t
>>>> values with conjunction analysis.
>>>
>>>
>>> Have you read the material at:
>>>
>>> http://www.sph.umich.edu/~nichols/Conj/
>>>
>>> The paper and poster deal with examples of negative t values using the
>>> standard SPM2 (global null) procedure.
>>>
>>> Best,
>>>
>>> Matthew
>>
>>
>
>
|