2.4 Data recording and analyses
Electrophysiological data were recorded using 64-channel caps with
Ag/AgCI impedance-optimized active electrodes. Standard electrodes were
placed according to the extended 10-20 positioning system, and
impedances were kept below 5 kΩ. The continuous signal was recorded with
a sampling rate of 1000 Hz and referenced online to the electrode placed
over the Cz. Using EEGLAB (Brunner et al., 2013; Delorme & Makeig,
2004; Makeig et al., 1995) for data processing, EEG data was
down-sampled to 500 Hz offline and filtered online within a high-pass
filter for .1 Hz and a low-pass filter for 30 Hz. All electrode sites
were rereferenced offline to the average of the left and right mastoids.
Reduction of ocular movement artifacts was performed through Independent
Component Analysis (ICA) rejection (Makeig et al., 1995; Delorme &
Makeig, 2004). The mean number of ICs rejected was 1.66 ± .91 per
participant (natural: 1.65 ± .71; forced: 1.79 ± .89; voluntary: 1.56 ±
1.11). The continuous recordings were cut into epochs ranging from -200
to 800 ms relative to the onset of each flanker trial. Baseline
correction was performed in reference to pre-stimulus activity (-200 to
0 ms). We subjected all epochs to a rejection procedure in which epochs
containing deviations larger than 100 μV were discarded. This procedure
rejected 670 epochs (rejection rate: .17%), and the mean rejection rate
per participant was 1.73%.
ERP components were defined based on previous literature and visual
inspection of wave-forms. Two time-windows were used to explore the
components of interest. Firstly, a 120-220 ms time-window classically
defined as N1 was selected. Secondly, a 350-450 ms time-window was
chosen to analyze the P3 component. In accordance with prior studies
(Christoffels et al., 2007; Jackson et al., 2001; Martin et al., 2013;
Verhoef et al., 2010), we focused our analyses on two regions of
interest (ROIs): central-parietal: CP3, CP1, CPz, CP2, CP4; and
parietal: P3, P1, Pz, P2, P4. The N1 effect is assumed to be indicative
of persistent covert visual attention control and sensitive to the level
of visual attention intensity on the target stimuli (Di Russo). The
greater N1 amplitude for incongruent relative to congruent trials implys
more attention allocated to a given stimuli (Luck et al., 2000).
Traditionally, a parietal P3 component was thought to index context
updating operations, demonstrating the late, top-down controlled process
associated with the allocation of attentional resources and updating of
working memory (Donchin & Coles, 1998; Duncan-Johnson & Donchin, 1977;
Polich, 2007; Liu et al., 2023b). In flanker tasks, the P3 effect has
often been found to be enhanced and delayed on incongruent trials
compared to congruent trials (Folstein et al., 2008).
Response times (RTs) and accuracy on each flanker trial were recorded.
We excluded the first two warm-up trials of each block and RTs ±2.5SD (.64% of the total) from the analyses. In addition, since
accuracy rates of flanker trials was 97.88% (SD = .14), and
there was only a significant main effect for trial type, we did not
conduct further analyses on accuracy rates. In total, the remaining
96.07% trials were included in the RT analyses. RTs were
log-transformed to better approximate a normal distribution.
Statistical analyses were conducted in R using the lme4 package (Bates
et al., 2015). The RT values and single-trial mean amplitude were
submitted separately to 3 × 2 × 2 linear mixed-effects models. The fixed
effects included “context” (natural vs. forced vs. voluntary), “trial
type” (switch vs. non-switch), and “flanker” (congruent vs.
incongruent), and interactions between these factors; and random effects
in the RT model included participants and items, whereas only
participants were added as a random effect of the model on single-trial
mean amplitude. We started with a full model including the structure of
maximal random effects (Barr et al., 2013), and added the three factors
and their interactions as random slopes. If the model failed to
converge, we used a backwards-stepping procedure until the model fit.
The decision to include or exclude random slope effects of each model
was based on the smallest Akaike Information Criterion (AIC) value, and
the model with the smallest AIC was selected as the best-fitting model.
Because “context” in the model involved three levels, any significant
main effects or interactions were further analyzed by comparing the
estimated means using the “emmeans” package. Moreover, we performed
pairwise comparisons between flanker trials and each context to identify
any differences. The p -values obtained were adjusted for multiple
comparisons using false discovery rate (FDR) correction. Given the aims
of the study, we only present the effects involving the factors
“flanker” and/or “context.”