Dear Tatia,
>Thank you for your reply and information. Further to the question that I
have
>raised, the reviewer wanted a report of the cluster level significance and
>associated values. However, it seems that SPM does not provide cluster level
>significance and associated values for conjunction analysis. I am
wondering if
>we have taken a wrong step in the analysis. If not, is there a reason why
SPM
>does not provide such information?
Yes there is, and an expert reviewer should have known that. The null
distribution of cluster size for minimum T-fields has not yet been derived.
The adjusted [corrected] p-value based on the peak highest should be entirely
sufficient and this is what should be reported (i.e. the equivalent Z
score, the
corrected p-value and the location of the maximum).
The reviewer might have simply meant the adjusted p-vale for the maximum of
each cluster (The p-value based on spatial extent i.e. cluster-level inference
is a different quantity).
With very best wishes,
Karl
>>===== Original Message From Karl Friston <[log in to unmask]> =====
>>Dear Tatia,
>>
>>> I am writing to seek your kind assistance. In a paper that I have recently
>>>submitted to NeuroImage, I reported the use of conjunction analysis and
>>>provided the following description of the procedure.
>>>
>>>"Furthermore, to identify regions of activation common to all three fMRI
>>>studies (Friston et al., 1999), a conjunction analysis was performed by
>>>selecting all the individual contrasts during the result assessment in
>>>SPM99. Conjunction analysis makes it possible to locate the regions of the
>>>brain that were consistently activated, above a fixed threshold, across all
>>>the groups being studied (Friston et al., 1999). Conjunction analysis,
>>>according to Friston et al. (1999), is a net test of random effect.
>>>According to the SPM manual, the p-value (corrected or uncorrected) refers
>>>to the threshold of the conjunction. SPM computes the corresponding
>>>thresholds for individual contrasts (for uncorrected thresholds, the
>>>individual threshold will be p1/n, where p = individual threshold and n =
>>>number of contrasts in the conjunction).
>>>
>>>One of the reviewers commented that we should add further details to our
>>>description of "Conjunction analyses" reported in the manuscript. I am
>>>wondering whether you could help direct me to the source of additional
>>>materials that should be added to better clarify the procedure of
>>>conjunction analysis?
>>
>>I would replace your description with:
>>
>>"Furthermore, to identify regions of consistent activation in all three
>>fMRI studies,
>>a conjunction analysis was performed using an SPM of the minimum t-statistic
>>over the individual contrasts from the three groups (Friston et al., 1999).
>>The p-values of the ensuing regional effects were adjusted for the
>>search volume using random field theory for minimum t-fields. A significant
>>effect (p <0.05) suggests that one or more contrasts were significant. An
>>alternative interpretation is that the probability of the effect being
>>expressed in any contrast, selected at random, is significantly greater
>>than zero.
>>We display our conjunction SPMs at p <0.001 (uncorrected) but only report
and
>>discuss consistent regional effects at p<0.05 (corrected for the [small]
>>search
>>volume)"
>>
>>This interpretation only holds if the contrasts were testing for the same
>>thing.
>>I am not sure what your contrasts were so I have assumed this. I suspect the
>>reviewers just want you to say what thresholds you used.
>>
>>You could cite:
>>
>>Karl J Friston, William D Penny, Daniel E Glaser. Conjunction revisited.
>>NeuroImage,
>>in press.
>>
>>But its not out yet.
>>
>>I hope this helps
>>
>>Karl
>
>
>
|