2.3 EEG acquisition and ERP quantification
Continuous EEGs were recorded from 32 scalp sites using Ag/AgCl
electrodes embedded in elastic caps (Brain Products, Germany) according
to the international 10–20 system. Online recordings were referenced to
the right mastoid. Electrode impedance were kept below 10 kΩ. The
sampling rate was set to 1000 Hz.
For offline analyses, all data were processed with BrainVision Analyzer
software (Brain Products, Germany). Continuous EEGs were filtered with a
30Hz low-pass filter and a 0.1Hz high-pass filter. Ocular Correction ICA
was used to remove eye movements. Continuous EEGs were then segmented
into 1,200 ms epochs and were time-locked to either the probe/irrelevant
stimuli or to the positive/negative feedback stimuli. The 1,200 ms epoch
contained a 200 ms pre-stimulus baseline, and a 1,000 ms time window
after stimulus onset. The irrelevant ERPs were averaged across all four
irrelevant items. Trials with signals exceeding 100 μV were defined as
artifacts and excluded from averaging.
For ERPs time-locked to CIT stimuli (i.e., probe vs. irrelevant), we
focused on the P300 component at Pz which often occurs between 350 - 550
ms (Matsuda & Nittono, 2018; Matsuda, Nittono, Hirota, Ogawa, &
Takasawa, 2009), the P200 component at FCz which often occurs (between
150 - 250 ms) (Tacikowski, Cygan, & Nowicka, 2014; Xu et al., 2011),
and the N200 component at FCz which often occurs between 250 - 350 ms
(Ganis, Bridges, Hsu, & Schendan, 2016; Hu, Wu, & Fu, 2011). These
recognition-related components were quantified using temporal PCA, a
technique that extracts linear combinations of data points meeting
certain criteria which tend to distinguish between consistent patterns
of electrocortical activity (Dien and Frishkoff, 2005). This analysis
was conducted using the ERP PCA Toolkit, version 2.86 (Dien, 2010). This
PCA used 1200 time points (200 ms before CIT-stimuli onset served as
baseline) from each participant’s averaged ERP as variables, with
participants and conditions as observations. Promax rotation was used
(Dien, Khoe, & Mangun, 2007), and eleven temporal factors were
extracted based on the resulting scree plot (Cattell, 1966). Of these
factors, nine factors accounted for more than 1% of the total variance
in the data. We selected the positive component peaking at 384 ms as the
recognition-P300, the positivity component at 206 ms as the P200, and
the negativity component at 316 ms as the N200 for further calculation
(see Table 1).
For feedback-locked ERPs, we concentrated on the FRN at Fz which occurs
250-350 ms and the feedback-P300 at Pz which occurs 350-550ms (Foti,
Weinberg, Dien, & Hajcak, 2011; Sai et al., 2016 ). Since the
FRN and the feedback-P300 are largely overlapped, we also conducted a
temporal PCA to analyze them. This PCA was performed with promax
rotation, using 1200 time points (200 ms before feedback onset served as
the baseline) from each participant’s averaged ERP as variables, with
participants and conditions as observations. Based on scree plots, 13
factors were extracted and 6 factors accounted for more than 1% of the
total variance in the data. We selected the positive component peaking
at 282 ms as the FRN and the positivity component at 424 ms as the
feedback P300 for further calculation (see Table 1 and Table S1). The
waveforms for each factor were reconstructed (i.e., converted to
microvolts) by multiplying the factor pattern matrix with the standard
deviations. These factors were scored using the peak values (Foti et
al., 2011), which were applied to subsequent analyses, including
analyses of variance (ANOVA), and ROC analyses.
All analyses were conducted using SPSS 20.0. For ANOVA, the
Greenhouse-Geisser correction was applied when the assumption of
sphericity was violated. Post-hoc comparisons were computed with
Fisher’s Protected Least Significant Difference. Effect size was
estimated by the partial eta squared valueηp2 .
In addition, Bayes factors (JZS BFs, with scaled r = 0.707, as
in Rouder,
Speckman, Sun, Morey, & Iverson (2009) were reported as a complement
to classical statistics. The BFs serve as a method of quantifying the
ratio of the likelihood of the null hypothesis to the likelihood of the
alternative hypothesis. The ratio will be stated as favoring either the
null hypothesis (no difference) or the alternative hypothesis
(Jeffreys, 1998;
Kass &
Raftery, 1995).
For t -tests, the BF10 (favoring the alternative
hypothesis) or the BF01 (favoring the null hypothesis)
is reported. For ANOVA effects, the
BFInclusion (favoring the alternative hypothesis) or
BFExclusion (favoring the null hypothesis) is reported,
reflecting a comparison of all models containing a particular effect to
those without the effect (also see Klein Selle, Gueta, Harpaz, Deouell,
&
Ben-Shakhar, 2021).
A BF value of ≥ 3 was regarded as moderate evidence for the respective
hypothesis (Kass &
Raftery, 1995).
BFs were computed using JASP (Version
0.14.1, https://jasp-stats.org/).