[Eeglablist] time-frequency contiguity threshold?

[Kieffaber, Paul]

I've been looking (so far in vein) for some precedent or algorithm to determine a reasonable contiguity threshold for time-frequency analysis.  My understanding is that the zero-masking done by newtimef() is on a pixel (resel?) by pixel basis, showing only those pixels where the deviation from baseline is statistically significant.  I'm wondering if anyone is using the added constraint of a contiguity threshold in order to mask even those pixels with statistically significant deviations by requiring that there be multiple consecutive (in time and/or frequency) pixels with significant deviations before that time-frequency perturbation is considered "significant."  Two alternatives I'm imagining are: (1)a theoretically determined threshold in terms of time and frequency (e.g., perturbation must span 50ms and/or 5Hz) that is independent of the temporal/frequency resolution and (2) an algorithmically determined threshold that is based on the dimensions of the time-frequency  matrix (e.g., the joint probability of N consecutive time-bins and N consecutive frequency-bins is less than alpha).  Any comments,  recommendations and/or references would be greatly appreciated.

Thanks!
Paul

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20080620/46124d51/attachment.html>