Statistical Analysis

A 350–500 ms time window was chosen for the N400 component analysis after visual inspection of the waveforms. Mean amplitudes were calculated for each participant in each condition, using custom-made MATLAB (2012b, The MathWorks, Natick, MA, United States) scripts. Visual inspection of the scalp topography and waveforms showed that the N400 component was largest at mid-central and posterior areas, and nine midline electrodes, in which the signal was most prominent, were chosen for further statistical analysis: CP3, P3, PO3, CPz, Pz, POz, CP4, P4, PO4. For the non-switched dataset, a four-way mixed model ANOVA was conducted to test the effects of Congruency (two levels; congruent/incongruent), Hemisphere (HS, three levels: left, midline, right), and Anterior–Posterior division (AP, three levels: anterior, central, and posterior) on the mean amplitudes for the three groups (between-subjects factor Group: three levels).

To test modulation of the N400 during language switches, a separate four-way mixed model ANOVA was conducted to compare the effects of Congruency (two levels; congruent/incongruent), Switch (two levels; code switch/no code switch), Hemisphere (HS, three levels: left, midline, right) and Anterior–Posterior division (AP, three levels: anterior, central, posterior) on the mean amplitudes in the previously mentioned time window.

To further examine the difference between switched and non-switched stimuli, another four-way ANOVA was performed on a different set of nine electrodes, which best represented the distribution of the observed switch-specific effects: F5, FC5, C5, F3, FC3, C3, F1, FC1, C1. Effects of Congruency, Switch, AP and HS were tested in both proficiency groups during the time windows of 200–300 ms and 300–450 ms. A time window of 520–670 ms was further chosen to investigate late switching effects, on a set of nine right-lateralised parietal electrodes: CP2, P2, PO2, CP4, P4, PO4, CP6, P6, and PO8. To all mixed model analyses of variance, Greenhouse-Geisser corrections were applied wherever appropriate and p-values after correction are reported in the results. Bonferroni corrections were applied to Post hoc analyses.