3.2 Results
Behavioral measures. Recognition accuracy was analyzed in a logit glmer model that included the treatment coded fixed factor Valence (Negative, Neutral and Positive) with random intercepts for Participants and Items. The slope for Valence was not included due to non-convergence of the model when included. The first model with “Neutral” as intercept revealed that neutral words (M=69.4%, sd=18%) significantly differed from both Positive words (M=85.9%, sd=9% ; β = 4.12, se = 0.7321, t = 5.63, p<.001) and Negative words (M=81.3%, sd=14%; β = 2.96, se = 0.7321, t = 4.04, p<.001). The re-leveled model with “Negative” as intercept did not reveal a significant difference between negative and positive words (β = 1.16, se = 0.7321, t = 1.58, p=.119). This pattern of results confirmed that recognition accuracy was higher for emotionally laden target words (whether positive or negative) than neutral words. (cf. Figure 3 )
ERP analyses. To probe the effect of Valence on the processing of printed Mandarin words, a two-tailed permutation t-test (1000 random partitions) was conducted to assess significant differences in ERP amplitude between Neutral and Negative target words, and between Neutral and Positive target words for all selected electrodes (midline: Fz, FCz, Cz, CPz, Pz, frontal central : FC1, FC3, FC5, FC2, FC4, FC6, central-parietal: C1, C3, C5, C2, C4, C6, CP1, CP3, CP5, CP2, CP4, CP6, and parietal: P1, P3, P5, P2, P4, P6), with time points of 5 msec across the entire epoch (0 to 1200 msec after stimulus onset). Only differences that persisted for 10 msec or more were considered statistically significant. Figure 4 demonstrates the results of these tests for 9 representative electrodes.
Based on the results of the permutation tests, the mean voltage amplitude of two time windows, 220-300 msec and 300-500 msec, were chosen. These windows are also in line with the P2 and the N400 components in the literature. Data were modeled using linear mixed effect regression, with the LmerTest package (Kuznetsova & Christensen, 2017) in R program (R Core Team, 2017). Models were performed independently over 3 regions of interest : mid-line sites (Fz, FCz, Cz, CPz, Pz), frontal-central sites (FC1, FC3, FC5, FC2, FC4, FC6, C1, C3, C5, C2, C4, C6) and centro-parietal sites (CP1, CP3, CP5, CP2, CP4, CP6, P1, P3, P5, P2, P4, P6). The first models, summarized in Table 3 , included the treatment coded fixed factor Valence (Neutral, Negative, Positive), with random intercept for Item and Participant. A random slope of Valence for Participant was included. Independent models were performed at all 3 ROIs for the mean voltage amplitude in the 220-300 and 300-500 msec time window. The models revealed significant differences between Negative and Neutral target words at all 3 ROIs in the 220-300 msec time window, and at frontal-central sites in the 300-500 msec time window). The comparison of Positive and Neutral target words revealed a significant difference only at frontal-central sites in the 220 - 300 msec time window.