[FieldTrip] Discrepancy between trialwise and average data using ft_freqanalysis

[Eelke Spaak]

Dear Dan,

Note that db = 10 * log10(stim/bsl), so a non-linear operation.
Therefore the order of steps matters. Let mean(x) denote mean over
trials for x. Then the keeptrials=yes approach results in mean(10 *
log10(stim/bsl)), whereas the keeptrials=no approach results in 10 *
log10(mean(stim) / mean(bsl)). These are in general not the same.

Cheers,
Eelke

On Wed, 4 Sep 2019 at 23:34, Dan McCarthy <mccarthyd at mail.wou.edu> wrote:
>
> Dear list,
>
> I am conducting a trialwise LMM analysis on time-frequency data and compared my grand averages between two ft_freqanalysis pipelines for sanity checking. Though the data look qualitatively similar, I noticed the values are baselined roughly 1.5 dB lower (i.e., values around zero vs. -1.5) for the grand averages using the cfg.keeptrials = 'yes' option compared to the cfg.keeptrials = 'no' option.
>
> Here are the processing steps for the keeptrials = 'yes' dataset:
>
> nSteps = 90;
>
> fRange = [1 30]; % Range of frequencies to analyze
>
> nCyc = [2 12]; % range of cycle widths for each freq
>
>
> cfg = [];
>
> cfg.output = 'pow'; % return power spectra
>
> cfg.channel = 'all';
>
> cfg.method = 'wavelet'; % multipaper method
>
> cfg.foi = linspace(fRange(1),fRange(2),nSteps); % 1 to 30 Hz in 90 steps
>
> cfg.width = linspace(nCyc(1),nCyc(2),nSteps); % width of  wavelets in number of cycles
>
> cfg.toi = [-3 3]; % time windows of interest
>
>
> cfg.keeptrials = 'yes';
>
> cfg.baseline = [-.3 0];
>
> cfg.baselinetype = 'db';
>
>
> for i = 1:nSub
>
>
>      tfDataSLLoCohQ1{i} = ft_freqanalysis(cfg, eeg{bin1,a});
>
>
>
>      tfNormDataSLLoCohQ1{i} = ft_freqbaseline(cfg, tfDataSLLoCohQ1{i});
>
>
> end
>
>
>
> cfg = [];
>
>
>
> % Get subject averages
>
> for j = 1:length(tfNormDataSLLoCohQ1)
>
>
>
>     tfNormAvgQ1{j} = ft_freqdescriptives(cfg,tfNormDataSLLoCohQ1{j});
>
>
>
> end
>
>
>
> % Get grand average
>
> tfNormGrandAvgQ1 = ft_freqgrandaverage(cfg,tfNormAvgQ1{:});
>
>
> % Get theta band for frontocentral sites
>
> cfg.channel = {'Cz','FC1','FC2'};
>
> cfg.avgoverchan = 'yes';
>
> cfg.frequency = [4 8];
>
> cfg.avgoverfreq = 'yes';
>
>
>
> singleThetaQ1 = ft_selectdata(cfg,tfNormGrandAvgQ1);
>
>
> In the cfg.keeptrials = 'no' pipeline, all is the same except the option is switched off and the extra step to get descriptives (ft_freqdescriptives) is removed before ft_freqgrandaverage as the trials were discarded so the extra step isn't necessary.
>
> When I plot the theta time course, all the values in the cfg.keeptrials = 'yes' data are shifted roughly 1.5 units down on the y-axis compared to the cfg.keeptrials = 'no' data.
>
> Has anyone else experienced a similar issue? Any insight into why these two approaches end up with different results?
>
> Any help is greatly appreciated. Thank you!
>
> Best,
>
> Dan McCarthy
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://doi.org/10.1371/journal.pcbi.1002202