Material and Method
Participants
Participants were 30 native Thai undergraduates (mean age = 20.97; 16 females; 1 left-handed) from King Mongkut’s University of Technology Thonburi (KMUTT). Each participant had normal or corrected-to-normal vision, and each received financial compensation for participation. Informed consent forms were signed prior to the experiment. The study was approved by the KMUTT Institutional Review Board and conducted in accordance with relevant guidelines and regulations.
The number of participants for both experiments was estimated based on a previous study using a behavioral Eriksen Flanker task to assess the impacts of emotion on attention [6]. Since the effect size was not directly mentioned in their study, was estimated from the reported ANOVA outcome and converted to Cohen’s f to facilitate the power analysis. The estimated effect size (Cohen’s f) of Experiment 1A in their study was 0.81. Consequently, a minimal sample size of 20 was required to achieve a statistical power of 0.95 while maintaining a significance level of 0.05 ('pwr' package; R software[22]). Thirty participants were recruited for both experiments in our study. In Experiment 1, data from one subject were excluded from the analysis due to excessively low accuracy (>3SD below the mean). As a result, the final dataset included twenty-nine subjects (mean age = 21.03 years; 15 females; 1 left-handed; total number of trials = 25,056).
Stimuli
Emotional cues consisted of seventy-two face images selected from the Nimstim Face Database [23]. To ensure that these images elicited the desired emotional reactions from our Thai participants, a separate mini-experiment was conducted. In this experiment, an independent group of 30 Thai observers (mean age: 20.20; 14 females; 1 left-handed) who did not take part in the main experiment provided ratings of the emotional content for each image in the Nimstim database. Three types of ratings were acquired: (1) emotional valence, or the perceived level of positivity or negativity of each facial expression, on a 0-9 Likert scale; (2) emotional intensity, or the degree of emotion expressed in the image, on a 0-9 Likert scale; and (3) emotional category, or the interpreted emotional content of each image (i.e., happy, sad, fear, angry, surprised, disgusted, calm, neutral). Data from this mini-experiment were used to further screen face images for our study.
From the total of 673 images in the original database, 72 face images were selected from 24 actors, each expressing happy, angry, and neutral emotions (i.e., 3 images per actor). Each selected image achieved a categorization accuracy beyond 70%, and the mean accuracies and emotional intensity ratings did not significantly differ across categories (mean accuracies: 90.83%, 87.78%, 90.00% for happy, angry, and neutral expressions, respectively). To maximize the efficacy of the selected images in evoking the desired emotional response, happy and angry expressions were chosen to represent positive and negative emotions, respectively, due to the highest accuracy scores observed for these emotional categories compared to others. Only images of faces with closed mouths were chosen because, as noted by the original authors [23], these images were noticeably perceptually dissimilar from the open-mouth face images due to the visibility of teeth or the shape of the mouth. Each image was converted to grayscale and manually cropped to remove hair, ears, and other non-essential details prior to the experiment.
Experimental Procedure
Each participant was seated in a comfortable chair approximately 57 cm away from the computer screen. In each trial, participants observed a 2000-ms central fixation, followed by a 75-ms face cue (width: 4.4˚, height: 6.4˚), which appeared at one of twelve possible locations around the central fixation (eccentricity: 9.8˚; see Figure 1). We presented the cue and stimuli at different eccentricities to prevent potential masking effects. After a brief period with a randomly-varied cue-stimulus interval (CSI; 125, 250, or 500 ms), 12 basic shape stimuli appeared on the screen for 1,550 ms (eccentricity: 10.6˚; diameter: 3.6˚). Participants were instructed to discriminate the shape of the target (bowtie vs. diamond) while ignoring the presence of surrounding distractors. Stimulus congruency was manipulated by altering the shapes of two flanking distractors on either side of the target. Participants were given three seconds to respond by pressing either ‘j’ or ‘k’ using their right index and middle fingers, respectively. Visual feedback was then provided, indicating whether the response was correct, incorrect, or too slow (i.e., if the response took longer than 3,000 ms). The next trial began after an inter-trial interval of 1,000 ± 50 ms.
The trial structures were designed to achieve an even distribution of stimulus conditions. Images depicting three facial emotions (happy, angry, neutral) were presented in both upright and inverted orientations at one of twelve possible screen locations. These stimuli conditions were paired with three cue-stimulus intervals (125 ms, 250 ms, and 500 ms) and two congruency conditions (congruent, incongruent), creating a total of 432 stimulus combinations (3 emotions × 2 orientations × 12 locations × 3 CSIs × 2 congruency conditions). For each participant, these combinations were duplicated to create a total of 864 trials. The trial order was randomly shuffled and distributed across 16 blocks, with each block consisting of 54 trials. The entire experiment lasted approximately 2.5 hours.
Each participant received a brief practice session prior to the experiment. Additionally, the short version of the Depression, Anxiety, and Stress Scale (DASS-21) [21] was administered to assess current and recent levels of psychological distress. Participants were required to rate their experiences of negative affect over the previous week on a 4-point scale. The questionnaire consisted of 21 items, evenly divided into three subscales (depression, anxiety, and stress; 7 items per scale). The Thai translation of the questionnaire, which has been extensively validated in previous studies to ensure translation accuracy and validity[24,25], was used. Instructions were given in Thai throughout the experiment.
Stimuli presentation was carried out in a dark room on a 24-inch LG monitor with a 144 Hz refresh rate. MATLAB software (R2020b version) with the Psychophysics Toolbox-3 package [26] was used to execute the display sequences and collect response data.
Data Analysis
Data from subjects whose average accuracy scores were below three standards deviations from the mean were excluded from the analysis. Planned analyses involved using repeated-measures Analysis of Variance (ANOVA) implemented in R [22]. An omnibus four-way ANOVA was initially performed to analyze the effects of facial emotion (happy, angry, neutral), facial orientation (upright, inverted), stimulus congruency (congruent, incongruent), and cue-stimulus interval (CSI; 125, 250, and 500 ms) on trial accuracies. Following the identification of a significant three-way interaction (emotion x congruency x orientation), separate two-way ANOVAs were conducted to probe the interaction between facial emotion and stimulus congruency and the nature of this interaction. Simple main effects tests with Bonferroni-corrected p-values were subsequently performed to examine the direction of the attention modulation effect for each emotional category. Effect sizes in this study were calculated using generalized eta squared (ges), which quantifies the proportion of total variance in the dependent variable that is attributable to each factor.
Analysis for DASS score correlation was conducted by developing a measure to quantify the magnitude of emotion-dependent attention modulation. A score reflecting the size of the incongruency effect evoked by positive versus negative faces was first computed. For each participant, the mean accuracy difference between congruent and incongruent trials (congruent-incongruent) for each emotional category and facial orientation was calculated. These accuracy differences were then subtracted across emotional categories (happy-angry) and facial orientations (upright-inverted) to derive a score of emotion-dependent attention modulation, normalized according to inverted faces. These scores were subsequently correlated with data from the DASS questionnaires to examine the relationships between attention modulation and emotional well-being.
Results
The mean accuracy score (and standard deviation) in the Flanker task was 96.28% (0.19%). The average scores (SDs) from the DASS-21 questionnaire were 9.10 (6.90) on the depression scale, 12.14 (8.93) on the anxiety scale, and 13.10 (7.63) on the stress scale. According to the standard DASS scoring guidelines [21], these scores suggest a normal level of depression, a moderate level of anxiety, and a normal level of stress among our subject population.
Valence-Induced Attention Modulation
A four-way repeated-measures ANOVA was performed to investigate the impact of facial emotion (happy, angry, neutral), facial orientation (upright, inverted), stimulus congruency (congruent, incongruent), and cue-stimulus interval (CSI; 125, 250, and 500 ms) on trial accuracy. As expected, the analysis yielded a significant main effect of stimulus congruency (F(1, 28) = 35.81, p < 0.001, ges=0.07), with incongruent trials eliciting lower average accuracy scores than congruent trials. Importantly, a significant three-way interaction was also observed among stimulus congruency, facial emotion, and facial orientation (F(2, 56) = 5.36, p < 0.001, ges=0.007). This interaction suggests that the ability to selectively attend to the target stimulus varied depending on the emotional valence and the orientation of the preceding face cue. The analysis revealed no significant main effects of emotion, orientation, or CSI (p > 0.1), and no significant interactions between CSI and other experimental variables were observed (p > 0.1). Consequently, subsequent analyses combined data across different cue-target time intervals.
To interpret the three-way interaction, data from upright and inverted face trials were submitted to separate repeated-measures ANOVAs that included facial emotion and stimulus congruency as within-subjects variables. For upright faces, the analysis revealed a significant interaction between facial emotion and stimulus congruency (F(2, 56) = 6.45, p < 0.001, ges=0.02; see Figure 2), suggesting that the presentation of upright emotional face cues indeed alter the spatial scope of visual selective attention. For inverted faces, no significant interaction between emotion and congruency was observed (F(2, 56) = 0.30, p > 0.1, ges=0.0008). This suggests that emotion-dependent attention modulation was evident only when the face cues were presented in their canonical orientation, where their semantic content and identities were preserved.
Post-hoc analyses were conducted to examine the direction of the incongruency effect across facial emotions. For positive faces displayed in the upright orientation, simple main effects tests revealed higher accuracy on congruent trials than on incongruent trials (mean difference = 4.0%, p_Bonf < 0.001, ges=0.22). However, this accuracy difference disappeared for trials with upright negative faces (mean difference = 0.9%, p_Bonf > 0.1, ges=0.03). As expected, a significant incongruency effect was observed for upright neutral faces (mean difference = 2.5%, p_Bonf < 0.01, ges=0.14). For faces presented in the inverted orientation, a significant incongruency effect was observed regardless of emotion type: positive (mean difference = 1.9%, p_Bonf < 0.05, ges=0.09), negative (mean difference = 2.4%, p_Bonf = 0.01, ges=0.13), and neutral (mean difference = 2.4%, p_Bonf < 0.01, ges=0.12). Together, these results suggest that the observed valence-induced attention modulation occurred in the expected direction, with a broadening of attention scope in response to positive faces and a narrowing in response to negative faces.
Impacts of Psychological Distress
To investigate whether the observed attention modulation was influenced by inter-subject variability in levels of depression, anxiety, or stress, we analyzed the correlation between the extent of attention modulation and data from the DASS questionnaire. A score reflecting the magnitude of valence-induced attention modulation was computed (see Materials and Methods). Data from one participant were excluded from the correlation analysis due to an excessively high attention modulation score (>3 SD from the mean). Scatterplots in Figure 3 depict the relationships between attention modulation and DASS scores for the remaining participants. Our analysis revealed a significant negative correlation between attention modulation and emotional distress for the Depression scale (Pearson’s r = -0.47, p = 0.01) and the Stress scale (Pearson’s r = -0.41, p = 0.03), but not for the Anxiety scale (Pearson’s r = -0.27, p > 0.1). An additional analysis performed by accumulating scores from all three DASS scales indicated a significant negative correlation (Pearson’s r = -0.42, p = 0.03). This suggests that valence-induced attention modulation is indeed influenced by inter-subject variability in emotional distress, with participants reporting lower levels of depression or stress on the DASS showing a more pronounced pattern of attention modulation in response to positive versus negative facial expressions.
To further analyze whether the observed correlation was driven by attention broadening induced by positive emotion, attention narrowing induced by negative emotion, or both; we made a slight modification to the same analysis. Rather than computing the magnitude of attention modulation by directly subtracting the incongruency effect across positive and negative emotions (see Materials and Methods), neutral emotion was used as the baseline for subtraction. Figure 4 shows correlation plots for attention modulation measured for positive relative to neutral emotions (positive-neutral) and negative relative to neutral emotions (negative-neutral). Our analysis revealed a significant negative correlation for positive-induced attention modulation and stress scores (Pearson’s r=-0.4, p=0.03; Fig.4A right panel). In other words, participants who self-reported higher stress levels exhibited less attention broadening after exposures to positive faces. The analysis also revealed a marginally significant positive correlation between negative-induced attention modulation and depression scores (Pearson’s r=0.38, p=0.05; Fig.4B left panel). This suggested that high-depression participants displayed a tendency towards increased attention narrowing in response to negative faces compared to neutral faces. Overall, the data suggest that psychological states of participants played roles in mediating the magnitudes of attention broadening and narrowing observed in the current experiment.