[Eeglablist] ERSP and ITC measures as revealed in ERPimage plots
smakeig at ucsd.edu
Mon Feb 20 10:15:18 PST 2006
Tineke Grent - 't Jong wrote:
>I have some questions about the output of the ERSP and ITC values in the ERPimage plots that I can't find appropriate answers to. Here are the questions:
>1) What does the number, specified in dB, plotted above the baseline of the ERSPs actually reflects? I understand that it is the output of a log-conversion (10log(power^2) to be precise) of the absolute baseline powerlevel (at least that is what I get from the tutorial). However, how can it reflect an ABSOLUTE value? By convention, dB is used to express a RATIO between two values. Shouldn't it be relative to something else? And what is it that it is compared to?
Absolute means 10*log10(actual_baseline_power_value). It has no
absolute(!) meaning, since power is not here converted to standard units
(e.g. uV^2/Hz), but can be input to a subsequent call to erpimage() to
obtain plots with standardized baselines. It is awkward to place this
value on the plot itself. Perhaps we should only print it on the
commandline and function outputs.
>2) If the ITC is not significant, does this then automatically indicate that all (remaining) activity in the plotted frequency range is non-phaselocked? How do I have to read this? Does an overall non-significant ITC value of 0.2, for example, means that only 20% of the data shows some sort of phase-locking (although seemingly not significantly different between trials), whereas 80% of the data is non-phaselocked? How do I know whether that non-phaselocked activity is actually significant?
ITC is a statistical measure of the randomness of phase at each latency,
across trials. Random data for any finite number of trials will have ITC
> 0. A non-significant ITC value just indicates that more than 'alpha'
of the time, such a value will occur by chance for datasets of this type
(e.g., during the pre-stimulus baseline in ERP paradigms using jittered
ISIs). Here, 'alpha' is your specified significance cutoff.
>3) How can I combine all the information provided by ERPimage in a meaningful way? Say I want to know what percentage of the data (or how much of the data in terms of amplitude values) is non-phaselocked and thus not reflected in the ERPs. Using ERPimage information, how would I get an estimate of this non-phaselocked information, contributing to the effect I am looking for in a particular experiment? Or do I need to adopt a totally different approach to get to that question? If yes, what approach would provide me with an answer to this question?
As above, one cannot interpret an ITC of say 0.3 as meaning that 30% of
the trials are exactly phase locked, while 70% display random phase.
That would be a very odd distribution, one that could be artificially
created but is unlikely to represent a coherent st of actual data trials
(in nearly any paradgim). See my article in TICS, 2004
(http://sccn.ucsd.edu/papers/TICS04.html) for a broader view.
>Any input on any of these questions would be greatly appreciated ;-).
Thanks for your interest. -Scott Makeig
More information about the eeglablist