Dear Donald (and SPM Users),

I should have been more specific. Since I want to compare methods for how each analysis performs at the channel and sample level I am looking to obtain the FWE corrected p-values at the voxel-level for every voxel in the data (the result table only provides the corrected p-value for the peaks in a certain cluster).

I suppose it might be interested to obtain cluster-level corrected statistics and simple assign each channel/sample in that cluster with the cluster-level p-value for comparisons sake.

Apologies for mentioning "uncorrected" t-values. I simply want to say that I could extract the "raw" t-values from the data... but not the corresponding p-values (not even uncorrected, although I do know there are some functions already created to change those t-values into their uncorrected p-values).

Thanks for your help!
Armand


On 2 February 2012 19:27, MCLAREN, Donald <[log in to unmask]> wrote:
See inline comments below:


On Thu, Feb 2, 2012 at 5:12 AM, Armand Mensen <[log in to unmask]> wrote:
Dear SPM User,

For my PhD thesis I am comparing multiple methods of EEG analysis using simulated data. The result structures for all other methods I've tested result in a channel x time matrix of p-values for all data measurements.

Due to the conversion of EEG data into 3D time series images for analysis, SPM does not give such an output and I am trying to convert the resulting structure somehow so that I can run the same ROC analysis to compare the SPM method with other ones I've already tested.

Clearly, being able to compare SPM's results with others would be important in determining a method's validity and superiority... so I don't just want to leave out SPM from the comparison.

Is there any way anyone can think of to extract corrected p-values from the resulting SPM structure for a simple t-contrast?

You need to decide what you mean by corrected. There are numerous potential corrections that could be applied. Some are easier to implement than other. Which one did you want to use? It seems that you would want to use the same type of correction as your other methods used.

 

At the moment I could only extract the corresponding uncorrected t-values from the resulting image by sampling the image at the known channel positions over time. So is there any sort of statistical image which shows the corrected p-values (even if negative log transformed for display purposes). If this existed I could sample from that image to obtain a channel x time matrix.

There is no such thing as a corrected t-statistic or uncorrected t-statistics. The corrections are all based on the p-values. Some are based on the number of tests, some are based on the distribution of tests, and some are based on both.
 

Thanks for any advice at all.

Armand Mensen
PhD Student - University of Zurich