To help with improving the transparency of what we do in our laboratory in terms of data analysis, and in order to help others either interpret our results, we are providing here a step by step "tree" of our analysis pipeline. Here, we will cover the analyses we use in Brain Vision Analyzer 2.
The Krigolson Lab Analysis Tree (in Brain Vision Analyzer 2.0)
1. Topographic Interpolation
We interpolate bad channels here using a spherical spline technique. Sometimes this is done based on visual inspection of the data or we restart the analysis after artifact rejection (see below).
2. Filter
Low Pass 0.1 Hz, High Pass 30 Hz, both with a 24 db/oct cutoff.
Notch Filter 60 Hz.
3. Re Reference
All data is re-referenced to an average mastoid reference.
4. Segmentation
Epochs of data (typically -200 to 800 ms) are selected for all of the markers of interest.
5. ICA
An ICA analysis is run to identify components associated with ocular movements. Note, we are running the ICA analysis on data that has been filtered, re-references, and segmented.
6. Inverse ICA
Inverse ICA is then used to remove these components from the data. We use visual inspection to verify the success / failure of components selection.
7. DC Detrend
We next use a DC de-trend (first 100 ms, last 100 ms) to remove constant drift from the data.
8. Baseline Correction
We baseline correct the data using the first 200 ms (-200 to 0) of the data unless otherwise specified. For example, sometimes a different baseline is used when looking at response locked motor potentials.
9. Artifact Rejection
We remove all artifacts that break a gradient criteria of 10 uV/ms or a max-min criteria of 100 uV over the entire epoch.
10. Segmentation
We re-segment the data at this point into the separate conditions of interest.
11. Average
Average waveforms are generated for each condition.
12. Difference Waveforms
We finally create difference waveforms to isolate effects of interest (see Luck, 2005). Given our interest in the ERN/FRN we typically are subtracting win/correct waveforms from loss/incorrect waveforms.
Statistical Analysis
In the past we have quantified the FRN as either the maximal negativity on the difference waveform in a window centred around the peak of the grand average difference waveform (e.g., between 200 and 400 ms) or as the mean over a shorter window (peak +/- 20 ms) also centred on the peak of the grand average difference waveform. In terms of topographies, we typically used the "peak topography" generated from the peak detection process as opposed to the topography of the peak difference of the grand average difference waveform.
In the future, we hope to move away from this and instead use and promote the use of confidence intervals plotted for each time point of the grand average difference waveform for each effect of interest.