[Eeglablist] problem with PSD unit coversion

[Arnaud Delorme]

Dear Magda,

> If I understand the concept right - the output of spectopo gives me PSD in 10*log10(mv^2/Hz). As I wanted to get rid of log transformation I simply took my matrix and did the math: 10.^(x/10), so that my current data unit is (mv^2/Hz). Am I correct?

This is correct.

> But - something went terribly wrong. My results literally flipped after conversion. Stats look like from different data set…

Bootstrap/permutation and rank stats should be the very similar (in case of rank identical). Normal stats (t-test and Anova) will indeed be very different because of the normality assumption and the fact that the probability distribution of the data is different (my take is that log is much more correct in most cases, because with log the distribution is less skewed and closer to normal). If the result flipped it could be due to outlier so it would be better to use robust statistics (Bootstrap/permutation) if possible with trimmed means for the estimator (Bootstrap/permutation is available in EEGLAB but the trimmed mean version is only available in Limo).

Best wishes,

Arno

> And I'm more than confused as I do not understand why this has happened. I've checked my matlab scripts more than a dozen times and I did not find any mistake. Did I miss something during the conversion? Maybe oversimplified sth?
> 
> I assume that my confusion is caused by the lack of proper knowledge - but I would be more than grateful if any of you could tell me what am I missing?  
> 
> Thanks in advance!
> 
> Greetings,
> Magda
> 
> 
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu