Greetings-<br><br>I am using newtimef to calculate the ersp for a set of
data where the baseline will be taken from a temporally non-adjacent
time window, and I had a couple of questions about the appropriate
settings to use, and about the outputs of newtimef using different
settings. Our experiment involves the sequential presentation of 4
to-be-remembered objects, followed by a 3-sec memory delay, the
appearance of a probe object, and a short (1 sec) ITI. We are interested
in exploring sustained neural oscillations during the delay interval,
and are using a portion of the ITI as our baseline. Because the ITI
begins ~4200 ms before the delay period starts, we are calculating the
ersp for the delay period (+-500 ms) and the ITI separately, and doing
the baseline subtraction as a separate step, following the ersp
calculation. This raises a few questions:<br>
<br>First, a clarification question: If I leave 'baseline' unspecified,
does newtimef use
the mean across the entire time window for each frequency as the
baseline? If this is correct, it seems like this method of baseline
correction would tend to mask the presence of sustained power increases
that may be present throughout the interval in question. For
instance, if I have a steady elevation in power of 12 mV^2/Hz throughout the
delay in some frequency band, the mean across the time window will be
~12 as well, and when I do the subtraction, the resulting value will be
~0.<br>
<br>If I have this right, it seems like the appropriate thing
to do would be to set 'baseline' to 'NaN' for both the delay and the
ITI, and then do the subtraction using these "raw" power values.
However, when I do this, I notice that all the power values (the values
in 'ersp') I get for either the delay period or ITI (before doing the
baseline subtraction) are negative, although the overall pattern looks
very similar to what I get when I use comparable settings using the
ft_freqanalysis routines in fieldtrip, just shifted down by some amount. Is it normal for the output to be negative, or is this just something weird happening with my data or analysis pipeline?<br><br>In case it's relevant, here are samples of the code I'm using to do the time-frequency analysis in EEGLab:<br>
<br>
%%<br><br>
[ersp,itc,powbase,times,freqs,<div id=":1y6">tfdata] = pop_newtimef(EEG, 1, chan, [-500 3500], 0, ...<br>
'type','phasecoher','padratio',4,
'timesout',100,'plotphase','off','plotersp','off','plotitc','off','freqs',[5
30],'baseline',NaN);<br><br>
%%<br>
<br>
And what I think is the approximately equivalent procedure in Fieldtrip
(some specifics are likely different, such as the number of cycles used
to estimate power for each frequency--I'm not sure exactly how to
specify this when using Hanning windows in eeglab--and the time steps):<br>
<br>%%<br>
cfg = [];<br>
cfg.output = 'pow';<br>
cfg.channel = 'all';<br>
cfg.method = 'mtmconvol';<br>
cfg.taper = 'hanning';<br>
cfg.foi = 5:30;<br>
cfg.t_ftimwin = 3./cfg.foi; %3 cycles per freq<br>
cfg.toi = -.5:0.05:3.5; % 50 ms window<br>
cfg.keeptrials = 'no';<br><br> TFRHann = ft_freqanalysis(cfg,data);<br><br>%%<br><br>I think this about covers it. Any help would be greatly appreciated.<br><br>Thanks,<br><font color="#888888">Jeff</font></div>
<br clear="all"><br>-- <br><br>Jeffrey S. Johnson, Ph.D.<br>
Postdoctoral Fellow<br>
Department of Psychiatry<br>
University of Wisconsin-Madison<br>
6001 Research Park Blvd.<br>
Madison, WI 53719<br>
<br>
Email: <a href="mailto:jsjohnson3@wisc.edu" target="_blank">jsjohnson3@wisc.edu</a><br>
Office phone: 608-265-8961<br>