The Longitudinal Interplay Between Attention Bias and Interpretation Bias in Social Anxiety in Adolescents

Social anxiety involves the fear to be negatively evaluated by others, and is often accompanied by the avoidance of social situations (American Psychiatric Association, 2013). Social anxiety has its onset in childhood (Beesdo et al., 2011), but symptoms increase during adolescence with a prevalence of 5–16% (Mesa et al., 2011). Youth with social anxiety experience severe consequences such as the increased risk of peer victimization (Erath et al., 2007), depression, and substance use (Beesdo et al., 2007). Early detection and treatment are therefore warranted, highlighting the necessity of research on factors contributing to the onset and maintenance of social anxiety.

Many cognitive models assume that biased cognitive processing is such a contributing factor (e.g., Beck et al., 2005; Wong & Rapee, 2016). Specifically, negative attention bias (i.e., the attentional preference for negative stimuli) and negative interpretation bias (i.e., the tendency to negatively interpret ambiguous social situations) increase the risk of experiencing social anxiety symptoms in children and adolescents (Dudeney et al., 2015; Stuijfzand et al., 2018). Most studies in children and adolescents so far have been cross-sectional (e.g., Rozenman et al., 2014; Seefeldt et al., 2014; Vassilopoulos et al., 2009; Waters et al., 2011), making it difficult to draw conclusions about the long-term predictive impact of biases on social anxiety. Exceptions are made for two longitudinal studies. One of these studies showed that adolescents with a negative interpretation style are at higher risk for belonging to a high social anxiety trajectory (Miers et al., 2013). The other study provided evidence that social anxiety was able to predict negative interpretation bias two years later. However, interpretation bias could not explain the pathway between shyness and social anxiety over time (Blöte et al., 2019). Longitudinal evidence about the role of attention bias is still lacking. At the same time, most studies in youth investigated the effect of either one of the biases on social anxiety (e.g., Bögels & Zigterman 2000; Miers et al., 2008; Salum et al., 2013), hereby ignoring the fact that biases may be interrelated or interact with each other (Hirsch et al., 2006). To overcome these limitations, the current study focuses on the interplay between attention bias and interpretation bias, and examines the longitudinal predictive ability of cognitive biases on social anxiety across three years in adolescence.

Both attention and interpretation biases may play a role in the existence and development of social anxiety in adolescence. Following the theory of (Piaget, 1971), adolescence in particular, may be a critical period of time for the development of cognitive biases, as a result of socio-cognitive maturation. Adolescents become able to think and reason in an abstract way, and develop metacognitive beliefs (i.e., the ability to reflect on our own thoughts and behaviours; Weil et al., 2013). This cognitive maturation may result in heightened self-consciousness and concerns about self-presentation, and the development of cognitive biases (Rapee & Spence, 2004; Westenberg et al., 2004).

Attention bias is a complex phenomenon consisting of multiple components. On the one hand, it is characterized by an enhanced engagement to threat as (socially) fearful individuals are faster to detect negative stimuli in the environment. On the other hand, it encompasses the delayed disengagement from threat, which is described as the difficulty to withdraw one’s attention from negative cues (Blicher et al., 2020). Most studies on attention bias in anxious youth focus on one specific attention bias component (e.g., enhanced engagement to threat; Shechner et al., 2013) or use a task which is not able to distinguish between the different attention bias components (e.g., the dot-probe task; Clarke et al., 2013). In this study, we aimed to examine both attention bias components by using a visual search task instead in which participants in different trials either have to detect or ignore a social threat. As such, we could investigate whether attention biases in social anxiety are associated with biases in engagement or disengagement (Donnelly et al., 2010).

Despite the theoretical frameworks and the fact that adult studies continuously found evidence for the role of engagement and disengagement attention biases in social anxiety (Liang et al., 2017), the role of attention bias and its specific components in youth is less clear. A review paper showed that some studies failed to detect a link between attention bias and anxiety in children, including social anxiety (Dudeney et al., 2015; Puliafico & Kendall, 2006), but others showed that socially anxious youth are faster at detecting threatening stimuli such as negative emotional faces (Roy et al., 2008; Schmidtendorf et al., 2018; Seefeldt et al., 2014; Waters et al., 2011). Meta-analyses comparing attention bias in youth and adults are also inconclusive: one found attention bias to be similar across different age groups (Bar-Haim et al., 2007), while another concluded that anxious children show attention bias to a lesser degree than adults (Dudeney et al., 2015). Actually, in line with the idea of socio-cognitive maturation (Piaget, 1971), the association between attention bias and anxiety increased with increasing age in children, with adolescence being a critical period for a possible persisting role of attention bias (Dudeney et al., 2015).

Contrary to attention bias, there is ample evidence for the role of interpretation bias in social anxiety. Recent meta-analyses for adults (Chen et al., 2020), as well as children and adolescents (Stuijfzand et al., 2018), concluded that a negative interpretation bias was related to an increased risk of experiencing anxiety symptoms, including social anxiety. Furthermore, socially anxious children and adolescents were more likely to interpret social situations in a threatening manner than non-anxious individuals (Miers et al., 2008; Rozenman et al., 2014). Similar to the findings of attention bias, and in line with the theory of Piaget (1971), the link between interpretation bias and anxiety also seems to increase in strength from childhood to adolescence (Stuijfzand et al., 2018). This highlights the importance of focusing on cognitive biases during adolescence.

Studies regarding cognitive biases in social anxiety varied on whether the sample was unselected (i.e. a community sample), highly trait anxious, or clinically diagnosed with social anxiety disorder. Different meta-analyses on attention bias and interpretation bias in adults, children, and adolescents found that the variation among effect sizes of different studies were not accounted for by the level of social anxiety symptoms as a sample characteristic (Bar-Haim et al., 2007; Chen et al., 2020; Stuijfzand et al., 2018). In other words, the strength of the link of attention bias and interpretation bias with social anxiety was similar for non-clinically as well as clinically anxious individuals. Cognitive biases were comparable across different types of anxious populations. In our study, we focus specifically on a community sample of adolescents to be able to investigate the relative contribution of attention bias and interpretation bias in the development of social anxiety symptoms.

Attention bias and interpretation bias are often studied as two separate constructs, having their own independent impact on social anxiety. However, the Combined Cognitive Bias Hypothesis (Hirsch et al., 2006) assumes that biases do not operate in isolation. The idea of the CCBH comes from Neisser (1967) who suggested that cognitive processes are cyclical, thus mutually reinforcing each other. Instead of acting alone, attention bias and interpretation bias are expected to be interrelated and influencing each other. Different biases may interact in such a way that the combination of attention bias and interpretation bias is expected to have a greater impact on social anxiety than individual cognitive biases alone.

However, most studies in youth so far, investigated the effects of different biases separately (Vassilopoulos & Banerjee, 2008; Waters, Mogg, et al., 2008). Even if attention bias and interpretation bias were investigated simultaneously, the interrelation and interplay between biases were often ignored in studies on youth (Klein et al., 2018; Waters, Wharton, et al., 2008). There are two exceptions of studies showing a positive link between attention and interpretation bias in different types of youth anxiety, including social anxiety (Rozenman et al., 2014; Watts & Weems, 2006). In both studies attention bias was assessed using a dot-probe task (either with pictures of emotional faces, Rozenman et al., 2014; or with threatening words and drawings of angry faces, Watts & Weems 2006). In the study of Rozenman et al. (2014), interpretation bias was measured using a word-sentence association paradigm in which threatening/neutral words appeared, with an ambiguous sentence following the word. Adolescents were prompted to indicate whether the word and sentence were related. Watts & Weems (2006) used an ambiguous vignette task to measure interpretation bias. Although results showed a clear relationship between attention and interpretation bias, both of these studies were cross-sectional, hereby overlooking the predictive magnitude of biases on social anxiety, and did not take into account the interaction between different biases, so did not formally test all aspects of the CCBH.

In sum, while it is generally acknowledged that interpretation bias may play an important role in maintaining, if not causing, social fears, the role of attention bias in social anxiety in adolescents is less clear. At the same time, the combined role of attention bias and interpretation bias during adolescence is only scarcely investigated. The current study is set up to overcome this caveat by investigating how attention bias and interpretation bias relate to each other, and whether they individually and/or mutually predict social anxiety over the course of adolescence. Understanding the interplay between biases may also have clinical implications. While Cognitive Bias Modification (CBM) techniques are prevailing as adjunct treatments to classical psychological treatments, modifying only one bias type without addressing other underlying biases may be insufficient. This might be the reason why current CBM techniques do not unequivocally lead to successful reductions of psychological symptoms in youth (Platt et al., 2017).

Evidence regarding the effectiveness of combined CBM, in which both attention bias and interpretation bias were targeted, are mixed. One study showed that the effect of combined CBM on decreasing social anxiety in adolescents 6 months later, was only marginally significant (Sportel et al., 2013). However, there are also studies showing the utility of combined CBM. For instance, adults receiving such a combined CBM reported reduced social anxiety symptoms and this was also shown in a behavioural speech task (Beard et al., 2011). Similarly, there is some preliminary evidence for socially anxious adolescents to show reductions in social anxiety, negative social behaviour, general anxiety, and depression after receiving both an attention and interpretation training (Lisk et al., 2018). However, both of these studies did not include an active control condition (e.g., a single CBM training), limiting the interpretation of the effects. Thus, more evidence for the combined effect of different cognitive biases on social anxiety is needed to guide clinical practice whether targeting multiple biases is necessary for effective treatment.

The aim of the current study was to investigate how attention biases (both enhanced engagement and delayed disengagement) and interpretation bias predicted social anxiety symptoms during three years in adolescence. Additionally, we tested the Combined Cognitive Bias Hypothesis (Hirsch et al., 2006) by investigating how different biases are related, and whether attention biases and interpretation bias individually predict social anxiety, or whether they interact with each other. Self-esteem and loneliness were included as covariates as they have found to be related to these biases in adults and youth (Lau et al., 2021; Tran et al., 2011), and to social anxiety in youth (Maes et al., 2019; Van Tuijl et al., 2014).

First, we hypothesized that negative attention biases (enhanced engagement and delayed disengagement) and interpretation bias will each independently predict social anxiety one year later. Individuals who are faster to detect negative stimuli (i.e., enhanced engagement), or are slower to withdraw their attention from negative stimuli (i.e., delayed disengagement), or individuals with a higher negative interpretation bias will experience higher levels of social anxiety at a later time point.

Second, regarding the CCBH, we expected that attention biases (enhanced engagement and delayed disengagement) and interpretation bias will be positively and bidirectionally related to each other. That means that individuals who are faster to detect negative stimuli (i.e., enhanced engagement) or are slower to withdraw their attention from negative stimuli (i.e., delayed disengagement) will show higher negative interpretation bias at a later time point. Vice versa, individuals with higher negative interpretation bias will have higher negative attention biases at a later time point (they experience enhanced engagement to or delayed disengagement from negative stimuli).

Third, we expected attention biases (enhanced engagement and delayed disengagement) and interpretation bias to positively interact with each other when predicting social anxiety over the course of adolescence. Thus, the combined effect of attention biases and interpretation bias on social anxiety is expected to be larger than if biases act alone.

Attention biases (i.e., enhanced engagement to threat and delayed disengagement from threat) were not related to social anxiety, thereby adding to the current field of research showing mixed and inconclusive evidence for the role of attention bias in social anxiety in youth (Morren et al., 2004; Puliafico & Kendall, 2006; Roy et al., 2008; Waters et al., 2011). Interpretation bias, on the other hand, did negatively relate to social anxiety, with higher negative interpretation bias predicting an increase in social anxiety. This finding was in line with cognitive models (e.g., Beck et al., 2005; Wong & Rapee, 2016) and previous research in children and adolescents (Stuijfzand et al., 2018). The effect of interpretation bias was only present from grade 7 to grade 8, which was contrary to the suggestion that the link between interpretation bias and social anxiety actually increased with age (Stuijfzand et al., 2018). The absence of an effect from grade 8 to grade 9 could possibly be explained by the fact that adolescents stayed in the same class in grade 7 and 8, but switched classes in grade 9. Entering a new peer context involves establishing new friendships and social ties, and it calls for a new social ranking of the peer group. A new peer context could elevate social fears (as adolescents have to meet new peers which could be threatening), but could also bring positive opportunities to adolescents (e.g., if adolescents are bullied in their previous context, they could form more positive peer relationships in the new context). These social contextual changes may have complicated our findings as it is unsure how the new peer context impacts the relationship between interpretation bias and social anxiety.

The different effects of attention bias versus interpretation bias in relation to social anxiety may also have something to do with the different conceptualization of attention and interpretation biases, and the developmental time frame under investigation. While interpretation bias is found to be a stable trait-like characteristic (Creswell & O’Connor, 2011), researchers argue that attention biases are fluctuating over time (which was also confirmed by our results), being highly context-dependent, and could be better understood as state processes (Zvielli et al., 2015). Indeed, a previous study showed that stressful situations may alter attention biases (Bar-Haim et al., 2010), and in fact, attention biases for threat were already unstable across 7 days (Li et al., 2008). Perhaps, the yearly time interval of the current study was thus too large to find effects of attention biases related to social anxiety. Future research could benefit from measurement burst designs with multiple measurement moments with short time-intervals (Sliwinski, 2008).

In our study, we used an unselected community sample. As expected, the percentage of individuals with clinically social anxiety levels was rather small, ranging between 10.8% and 14.7% across grades (scores of 50 or higher as described by La Greca & Lopez 1998). Re-examining the descriptive statistics for this subsample of clinically anxious adolescents showed that this subgroup had the tendency to negatively interpret situations compared to non-anxious adolescents who showed a positive interpretation bias. However, the attention bias patterns were similar for the entire normative sample compared to the socially anxious subgroup. This suggests that maybe a negative attention bias does not exist for these individuals at all, or that the sample size of this subgroup was too small to find a negative attention bias. To conclude, our results suggest that attention bias does not play an important role in explaining individual differences in social anxiety in a normative sample of adolescents, while interpretation bias did. Replicating this study in a large (sub-)clinical sample of socially anxious adolescents could help to investigate the generalization of findings from analogue samples such as in our study (Chen et al., 2020), and to formally test whether the Combined Cognitive Bias Hypothesis does apply differently to individuals with heightened social fears.

The lack of findings with regards to attention bias in social anxiety could be because of certain other methodological factors. For instance, there is much doubt regarding the adequacy of the psychometric properties of reaction-time based paradigms for measuring attention bias in children (e.g., Brown et al., 2014; Wermes et al., 2017). In contrast, some studies provided evidence that the visual search task was successful for detecting attentional threat biases in youth, and actually better in doing so than other measures of attention bias such as the dot-probe task (de Voogd et al., 2016). Not only the visual search task in itself raises questions, but there is also some discussion about the type of stimuli used in this task. Some argue that especially angry facial expressions may be ecologically valid threat stimuli for adolescents with social anxiety, as it reflects the fear of rejection (Rinck & Becker, 2005). However, others suggested that it would also be interesting to investigate attention to different negative emotional expressions, such as disgust, as that may convey the desire to avoid or reject (Buckner et al., 2010). Future research should formally test the reliability and validity of the task and stimuli.

In our study, the visual search task seemed to be unreliable, seen by the high standard deviations and the weak or non-significant correlations between the attention bias variables across and within grades. This unreliability could be due to two competing explanations. On the one hand, participants could have become fatigued during the attention bias task and may have not responded to the later trials in a concentrated manner. This explanation seems plausible as several adolescents complained that the task was long and repetitive, and that they were easily distracted. However, our results showed that there was no difference between data from the first and the second half of the task (in fact, scores between block 1 and 2 were moderately correlated), and previous studies with adolescents even successfully included more trials (72 compared to 48 trials in our study; De Voogd et al., 2017). On the other hand, we may have not included enough trials in our task, since a higher amount of trials increases the stability of effects. When developing a task there is an important trade-off between the amount of trials necessary to find stable effects and the feasibility of the task for participants (Price et al., 2015). Due to the fact that our data collection procedure consisted of many measures, we decided to keep the attention bias task relatively short. Future research should determine the minimum amount of trials necessary to retrieve reliable attention bias data.

Another methodological explanation comes from the suggestion that attention bias consists of many different subcomponents that unfold from moment-to-moment across time (Rodebaugh et al., 2016; Zvielli et al., 2015). A previous study only found a negative attention bias in anxious individuals during the first 500 milliseconds of stimulus presentation (Gamble & Rapee, 2009). Our visual search task, in which participants were allowed to view the stimuli as long as they needed, did not allow for determining the time course of various attention bias processes. Future studies including for instance eye-tracking would be better able to determine the time frame in which these processes should occur (Armstrong & Olatunji, 2012; Roy et al., 2015), and eye-tracking was found to be a successful method for children and adolescents (In-Albon & Schneider, 2010).

Finally, the difference in the results of attention bias and interpretation bias could also be accounted for by the difference in self-relevance between both tasks. The interpretation bias task specifically instructs participants to imagine themselves being in the situation. However, the attention bias task did not induce any form of self-relevance with the task, but simply asked to detect a face. Previous findings show that negative biases are mostly triggered when faced with stress or in self-relevant situations (Vassilopoulos & Banerjee, 2012). Future research could thus examine whether different attention bias results would be found if a stress induction is used or if the self-relevance with the attention bias task is increased.

Finally, attention biases were not related to interpretation bias, neither directly nor in interaction when predicting social anxiety, showing no support for the Combined Cognitive Bias Hypothesis for social anxiety in adolescents (Hirsch et al., 2006). This is contrary to two previous studies showing a link between attention and interpretation biases in youth anxiety, including social anxiety (Rozenman et al., 2014; Watts & Weems, 2006), and research showing the effectiveness of combined Cognitive Bias Modification (CBM) techniques in reducing social anxiety (Beard et al., 2011; Lisk et al., 2018). There are two possible interpretations of these findings. On the one hand it could be that there is simply no support for the CCBH in adolescence. Attention bias and interpretation bias may just not be related, and rather function as two independent processes. This suggestion could theoretically be supported by neurological models arguing that automatic (i.e., attention bias) and more effortful controlled (i.e., interpretation bias) regulatory processing are managed in a partly different way by the brain (Cunningham et al., 2004). In line with this reasoning, it would then be important to study attention bias and interpretation bias as separate processes in order to understand the development of social anxiety.

On the other hand, it could also be that the different biases were not related to each other because of the methodological difficulties with measuring attention bias in particular (as extensively discussed earlier). In addition, the result that different cognitive biases were not related to each other, could also be attributed to another methodological issue, namely the different modalities used for the attention bias and interpretation bias task. Specifically, in our study, interpretation bias was measured with textual vignettes, while attention bias was assessed using pictures of emotional faces. The use of visual versus verbal stimuli may have complicated the possibility to find a relationship between these different cognitive biases, particularly because the degree of social fears was also assessed with a textual (verbal) measure. Future research should develop comparable cognitive bias tasks to assess both attention bias as well as interpretation bias, ideally in the same modality. An attempt for this has already been made, with two studies integrating attention bias paradigms in a standard textual interpretation task. Specifically, participants completed a scrambled sentence task to see whether they would complete the sentence in a positive or negative way (as an indication of interpretation bias), and tracked the time spent at the positive versus negative words (as an indication of attention bias) (Sanchez et al., 2015; Sanchez-Lopez et al., 2019). Another study also showed the utility of a pictorial task to assess interpretation bias, to ensure that the modality of the task is similar to that of classical attention bias tasks in which visual stimuli are also often used (Henricks et al., 2022). Equalizing the assessment of attention bias and interpretation bias in terms of modality also led to stronger associations between the two biases, with Pearson correlations ranging between r = .30 and r = .32, p < .05 (Sanchez et al., 2015), as compared to the non-significant correlations found in the present study.

This pre-registered, prospective longitudinal study was one of the first to investigate the link between different cognitive biases across three years in adolescence. Including multiple measurement moments across different years allowed for the longitudinal investigation of these cognitive processes during an age period which is critical for social cognitive development (Choudhury et al., 2006) and the rise of psychopathology, including social anxiety (Mesa et al., 2011). Another strength of the study was its relatively large sample size, allowing for well-powered statistical analyses (i.e., in our most complex model we had at least 10 observations per parameter; Kline 2015).

However, there were also some limitations to this study. As discussed extensively earlier, the reliability of the attention bias task needs further improvement. In addition, we investigated two components of attention bias: enhanced engagement to threat (in other studies referred to as early vigilance to threat) and delayed disengagement from threat, which is in line with recent attention bias frameworks (Richards et al., 2014). However, other theories suggest that after detecting threat, socially anxious individuals are characterized by the quick visual avoidance of threat (e.g., the hypervigilance-avoidance model; Mogg & Bradley 1998). It might be that this attentional avoidance is the key process in attention bias and accounts for the maintenance of social fears because it prevents the exposure, and therefore the habituation and objective evaluation of socially threatening information (Amir & Bomyea, 2010; Mogg et al., 1997). Unfortunately, the visual search task used, did not allow for this avoidance process to be captured. It is important for future research to focus on all three attention bias mechanisms to fully understand attention bias as a whole, for instance by using eye-tracking (Cisler & Koster, 2010). Third, in our study we focused specifically on adolescence as this is the age period in which stable social cognitions emerge (Lakdawalla et al., 2007). However, there are also studies supporting the existence of attention biases related to depression in 5-years old children (Kujawa et al., 2011). As recommended by Platt et al. (2017) future studies should concentrate on investigating cognitive biases across a wide developmental period to better understand the emergence of cognitive biases, and their role in the onset and maintenance of psychopathology. Fourth and finally, another important suggestion for future research would be to examine individual differences in our conceptual model. Due to the heterogeneity among socially anxious individuals (Binelli et al., 2015), it could be that the Combined Cognitive Bias Hypothesis only applies to a specific subsample. For instance, concepts such as self-esteem or social competence could play a moderating role, as they are found to be related to cognitive biases and social anxiety (Miers et al., 2013; Tran et al., 2011; Van Tuijl et al., 2014). Other statistical analyses such as Latent Class Growth Modeling and network analyses could be more beneficial to investigate individual differences than the cross-lagged panel models used in the current study.