Social Anxiety is Related to Worse Recognition Memory of Angry Faces

Cognitive biases are major factors underlying social anxiety disorder (SAD; Hirsch & Clark, 2004). These biases include heightened attention to social threats (Kim et al., 2018), perception of negative evaluation from others (Chen et al., 2020), and negatively biased memory of social experiences (Krans et al., 2014). Theories propose that cognitive biases are important in constructing expectations that rejection is likely, thereby maintaining social anxiety (Clark & Wells, 1995). Research has focused on attention and interpretation biases, showing that socially anxious individuals preferentially attend to perceived sources of social threat, such as angry faces (Kim et al., 2018), and interpret neutral or ambiguous social cues as negative (Chen et al., 2020). Fewer studies have investigated biased memory related to social anxiety, and those studies have achieved conflicting results. The relative lack of research on memory biases has led to a gap in understanding their potential role in social anxiety.

Most studies of memory biases have examined autobiographical memory (Krans et al., 2014), memory of social feedback (Caouette et al., 2015) or prose passages (Wenzel & Holt, 2002), or working memory for threatening words or faces (Segal et al., 2015). Working memory research suggests that socially anxious individuals have better working memory capacity for social threat words than neutral words (Amir & Bomyea, 2011; Yoon et al., 2017) and may have biased working memory maintenance of socially threatening faces (Judah et al., 2016). Among the few studies that have tested recognition memory bias for facial expressions, findings have been equivocal. Foa and colleagues (2000) found that patients with SAD displayed better recognition memory for angry and disgust expressions compared to happy and neutral expressions. In contrast, LeMoult and Joormann (2012) found that people with SAD had worse recognition memory of angry faces, but not other emotional faces, compared to healthy controls. Similarly, Leung and colleagues (2022) found that social anxiety was correlated with worse memory for angry faces compared to happy faces. Another study did not find any difference between people with SAD and controls in recognition of angry faces, despite evidence of an attentional bias to angry faces among participants with SAD (Hagemann et al., 2016). Discrepant memory bias findings suggest a need to examine potential moderators. Based on the dependence of memory on attention (Chun & Turk-Browne, 2007), it is possible that biased memory of socially threatening stimuli depends on whether they are the focus of attention.

As is the case with memory bias research, findings regarding attention to socially threatening stimuli are mixed (Bantin et al., 2016; Bockstaele et al., 2018). Studies typically examine attention to visual images of faces with socially threatening expressions, such as angry faces, in comparison to neutral stimuli, such as expressionless faces (Buckner et al., 2010). Some studies suggest that socially anxious individuals attend more to threatening social stimuli, such as angry facial expressions (Heeren & McNally, 2016; Reutter et al., 2017; Wieser et al., 2018), but others suggest avoidance (Mansell et al., 1999; Wermes et al., 2018). This may be due to variations in attention over time (Dong et al., 2020), symptom severity (McTeague et al., 2018), task characteristics and conditions (Bantin et al., 2016; Judah et al., 2013, 2016), comorbid symptoms (Kishimoto et al., 2021; Singh et al., 2015), and the ability to filter out processing of social threat that is not task-relevant (Yuan et al., 2021). Given the dependence of memory on attention and the diverse and complex results in the attention bias literature, it is possible that the mixed findings in memory bias research are due to lack of control over attention in past studies. Although it is not possible to fully manipulate attention, memory biases to threatening faces can be examined in conditions where the faces are task-relevant compared to when they are not. Given past research suggesting that socially anxious individuals attend to task-irrelevant information associated with social threat (Yuan et al., 2021), there is a need to test how memory biases may depend on the relevance of stimuli to the task in order to improve understanding of memory biases related to social anxiety and how they may be dependent on attention.

The present study tested the association of social anxiety with memory bias towards socially threatening stimuli in conditions where the stimuli were task-relevant and task-irrelevant. Participants completed a two-part Face-House task, in which they viewed a series of images comprised of a semi-transparent face spatially overlapped with a semi-transparent house. Participants were cued to make a judgement about either the gender of the face or whether the house had a porch. This design controlled which stimulus was task-relevant (i.e., cued images were task-relevant stimuli, and uncued images constituted task-irrelevant). 50% of the faces expressed an emotion of anger, and 50% were neutral faces. Afterwards, participants completed a surprise recognition memory task of the stimuli. Participants also completed a measure of social anxiety and a measure of anxiety.

Due to the conflicting nature of the previous research, two opposing sets of hypotheses were tested. Based on previous research showing that individuals with SAD tend to fixate on social stimuli, even when irrelevant to the ongoing task (Buckner et al., 2010), it was expected that, overall, social anxiety would be associated with better memory for faces (vs. houses). It was also expected that social anxiety would be linked with better memory for angry (vs. neutral) faces on the Face-House task, on account of the research showing that individuals with SAD have a more difficult time disengaging from threatening social stimuli (Kim et al., 2018). Conversely, when considering the previous work showing that individuals with SAD tend to exhibit avoidance of attending to eyes (Günther et al., 2021), it was expected that social anxiety would be associated with worse memory for faces (vs. houses) on the Face-House task. Furthermore, it was also expected that social anxiety would be linked with worse memory for angry (vs. neutral) faces on the Face-House task, because of a tendency to avoid paying attention to emotional faces (Mansell et al., 1999). Analyses were also conducted with anxiety in place of social anxiety to determine whether any findings were specific to social anxiety.

In the social anxiety model, the main effect of image focus was significant, t(471) = -2.00, p = .046. Recognition memory was better for task-relevant faces (M = 0.55, 95% CI [0.49, 0.61]) than task-relevant houses (M = 0.47, 95% CI [0.39, 0.56]). There was also a significant main effect of facial emotion, t(471) = -2.51, p = .013. Memory for neutral faces (M = 0.61, 95% CI [0.53, 0.70]) was better than memory for angry faces (M = 0.55, 95% CI [0.49, 0.61]). These main effects were qualified by an interaction between image focus and facial emotion, t(471) = -2.49, p = .013 (see Fig. 2). Memory for neutral faces (M = 0.61, 0.95% CI [0.53, 0.71]) was better than memory for angry faces (M = 0 0.55, 95% CI [0.49, 0.61]) in the face-focused condition. Conversely, memory for angry faces (M = 0.47, 95% CI [0.39, 0.56]) was better than memory for neutral faces (M = 0.31, 95% CI [0.21, 0.45]) in the house-focused condition. Facial emotion also significantly interacted with social anxiety, t(471) = 2.22, p = .027, such that memory of angry faces was worse at high (+ 1 SD) levels of social anxiety (M = 0.44, 95% CI [0.33, 0.54]) than it was for low (-1 SD) levels of social anxiety (M = 0.59, 95% CI [0.49, 0.69]). Memory of neutral faces was similar at high (+ 1 SD) levels of social anxiety (M = 0.48, 95% CI [0.38, 0.59]) and low (-1 SD) levels of social anxiety (M = 0.44, 95% CI [0.33, 0.54]; see Fig. 3). The three-way interaction between image focus, facial emotion, and social anxiety was not significant, t(471) = 1.42, p = .157, nor was the main effect of social anxiety, t(157) = -0.92, p = .360.

The model was tested again with anxiety in place of social anxiety. The main effect of image focus was significant, t(471) = -3.12, p = .002, and there was a marginally significant interaction between facial emotion and image focus, t(471) = -1.95, p = .051, both effects resembling those of the social anxiety model. The main effect of facial emotion was not significant, t(471) = -0.87, p = .386, nor was the interaction of facial emotion with anxiety, t(471) = 0.20, p = .845. The three-way interaction between image focus, facial emotion, and anxiety was not significant, t(471) = 0.51, p = .608, nor was the main effect of anxiety, t(157) = 0.03, p = .977.

This study adds to the understanding of memory biases related to social anxiety, a topic that has been investigated much less than other information processing biases related to social anxiety. We found that social anxiety was associated with worse recognition memory for angry faces. This effect was specific to social anxiety, as indicated by the lack of an effect when anxiety was tested in place of social anxiety. The effect was not dependent on whether the faces were the focus of the task, supporting past findings indicating reduced processing or avoidance of social threat in socially anxious individuals (Wermes et al., 2018). The result is contrary to the view that social anxiety is associated with information processing biases toward stimuli representing social threat (Morrison & Heimberg, 2013) and a study that suggested better recognition memory for angry and disgust faces in socially anxious patients (Foa et al., 2000). Instead, the results are in accord with findings suggesting that social anxiety is related to avoidance of processing social threat (Mansell et al., 1999). Among memory bias studies, the findings align those that have found social anxiety to be related to worse memory for angry faces (LeMoult & Joormann, 2012; Leung et al., 2022). Our findings build upon these studies by suggesting that the effect is not dependent upon the task-relevance of the faces. The evaluation of the simple slopes suggests the effect is due to social anxiety being related to reduced memory of angry faces rather than better memory for neutral faces. Although some studies have suggested attentional avoidance of neutral faces may be related to social anxiety (Stevens et al., 2014), we did not see evidence of this in recognition memory.

The present results support the idea that memory biases may be a contributing factor in the perpetuation of social anxiety symptoms. As noted earlier, previous research has revealed that social anxiety may be associated with reduced processing of threatening social stimuli (Mansell et al., 1999) and reduced memory recall of socially threatening material (Wenzel & Holt, 2002). Together with this literature, our study supports social anxiety being related to reduced processing of socially threatening faces. Though avoidance of processing threatening stimuli may temporarily relieve some symptoms of anxiety, attending less to social cues reduces opportunities to correct one’s perceptions of others and to see a social interaction as less threatening. Thus, individuals may benefit by interventions that develop the ability to maintain attention to perceived threats to allow for reappraisal (Bögels & Mansell, 2004).

In addition to our findings involving social anxiety, we observed that memory of neutral faces, but not angry faces, was worse when the task required attention to be focused on houses. This is consistent with evidence that threatening stimuli are more difficult to ignore (Fox et al., 2001). However, the findings are somewhat inconsistent with studies using a similar paradigm (i.e., spatially overlapping faces and scenes) that found no evidence of recognition above chance levels for task-irrelevant stimuli (Yi & Chun, 2005; Yi et al., 2006). Although our paradigm was similar, our study differed in that d′ was used rather than hit rate or ratings of confidence in having seen the image previously. The confidence intervals of our d′ values suggest at least some memory storage of task-irrelevant stimuli, both houses and faces, in our task. Further research is needed to better understand how task-irrelevant stimuli are processed.

Our conclusions are limited by several aspects of the study design. Using a non-clinical sample has the advantage of studying social anxiety continuously, which aligns with dimensional perspectives of psychopathology. But it limits what conclusions can be drawn about the general population or clinical populations. The cross-sectional design precludes conclusions about causality. Additionally, the administration of the study online, due to the circumstances of the COVID-19 pandemic, is less controlled than if the study had been administered in a laboratory. There is a need for future research to address these limitations and to further this line of research. For example, including positive social cues, such as happy facial expressions, may be important for understanding memory of social cues more broadly.

In summary, our findings advance the understanding of memory biases related to social anxiety, suggesting that social anxiety is associated with worse recognition memory of angry facial expressions than neutral facial expressions. The bias was not dependent on the relevance of the face to the task, suggesting that the bias is not dependent on goal-oriented attention. Additional research is needed to better understand memory biases related to social anxiety and how they relate to other cognitive biases (Hirsch et al., 2006).