Interpretation bias in social anxiety: A systematic review and meta-analysis

Highlights

• Examined the relationship between interpretation bias (IB) and social anxiety.

• Investigated five moderators of the relationship between IB and social anxiety.

• The relationship between IB and social anxiety yielded a large effect size.

• Subjective measures and verbal stimuli emerged as significant moderators.

Abstract

Background:

Interpretation bias, which involves interpreting ambiguous social events negatively and catastrophising even mildly negative social events, has been suggested as a key maintenance factor of Social Anxiety Disorder. Although some individual studies and narrative reviews have demonstrated a role for negative interpretation bias in social anxiety (disorder), findings have been mixed. Given the lack of a quantitative synthesis of the evidence, the current systematic review and meta-analysis examined the strength of the relationship between interpretation bias and social anxiety. It also investigated potential moderators of this relationship (i.e., types of measures and stimuli, samples, and study designs).

Methods:

Five databases (PsycINFO, PubMed, Medline, Scopus, and CINAHL) were searched. Of 46 studies identified, 44 were suitable for meta-analysis (N = 3859).

Results:

There was a large effect for the relationship between social anxiety and interpretation bias (g = 0.83). Types of measures (subjective versus objective) and stimuli (verbal versus visual) were identified as significant moderators, with subjective measures and verbal stimuli particularly adept at capturing interpretation bias in socially anxious individuals.

Limitations:

The effect sizes displayed significant heterogeneity between studies, which likely reflects some publication bias, and thus, the overall effect size may be inflated.

Conclusion:

Findings may help to refine clinical models and interventions for Social Anxiety Disorder, which in turn may maximise evidence-based interventions that target negative interpretation bias in this disorder.

Introduction

Social Anxiety Disorder is characterised by a disproportionate fear and anxiety of social performance and social situations. Socially anxious individuals have a core fear of negative evaluation and actively avoid social situations where potential scrutiny, humiliation, or embarrassment may occur, leading to marked distress and significant disturbances in functioning (American Psychiatric Association, 2013). Cognitive behavioural models of Social Anxiety Disorder (e.g., Clark, 2001; Heimberg et al., 2014) have proposed that information processing biases play a central role in the development and maintenance of the disorder. Thus, a comprehensive understanding of such biased cognitive processing may inform future treatment strategies.

One cognitive processing bias is interpretation bias, which involves interpreting ambiguous social events negatively and catastrophising even mildly negative social events (Clark, 2001; Heimberg et al., 2014). This bias has also been described as judgmental bias or inferential bias (Heinrichs and Hofmann, 2001; Hirsch et al., 2016). According to Clark's (2001) cognitive model of Social Anxiety Disorder, socially anxious individuals tend to interpret social events or situations as threatening (i.e., perceived criticism or rejection by others), which is associated with negative social costs. To keep themselves ‘safe’ from perceived adverse social outcomes, socially anxious individuals avoid social engagements, which prevent them from having positive social experiences. Such negative interpretations are also carried into post-event processing and the cycle repeats and thereby compounds the lack of opportunity for benign disambiguation of social information. Thus, a negative interpretation bias for social information is a significant maintenance factor in Social Anxiety Disorder (Clark, 2001; Heimberg et al., 2014).

In support of this cognitive theoretical framework, empirical evidence and some early narrative reviews have demonstrated that socially anxious individuals hold a negative or threat bias for ambiguous social situations (Beard and Amir, 2009; Brendle and Wenzel, 2004; Clark and McManus, 2002; Franklin et al., 2005; Gutierrez-Garcia and Calvo, 2017; Heinrichs and Hofmann, 2001; Mohlman et al., 2007) and report catastrophic interpretations of mildly negative social situations (Amir et al., 2005; Stopa and Clark, 2000). Socially anxious individuals also interpret neutral and positive social information negatively (Campbell et al., 2009; Voncken et al., 2003; Yoon and Zinbarg, 2007) and show a lack of positive interpretations of social events (Amir et al., 2012; Constans et al., 1999; Maoz et al., 2016). However, not all studies nor more recent narrative reviews have found that socially anxious individuals demonstrate a negative bias in interpreting social information, with some showing a lack of nonthreat or positive interpretation bias and others showing no impact of social anxiety on interpretation bias (Garner et al., 2009; Hirsch and Clark, 2004; Hirsch et al., 2016; Jusyte and Schonenberg, 2014; Qiu et al., 2018; Schofield et al., 2007). Given the inconsistencies in findings, and the high risk of biases (selection, assessment, and publication biases) inherent in narrative reviews (Foroutan et al., 2018), a quantitative synthesis by way of a systematic review and meta-analysis is warranted to ensure quality of evidence. Thus, the first aim of the present study was to estimate the size of the relationship between interpretation bias and social anxiety via a systematic review and meta-analysis.

Researchers have suggested that the inconsistencies in findings may be due to methodological differences, such as the use of subjective versus objective measures, or the type of stimuli (e.g., written scenarios versus faces) used to assess interpretation bias (Hirsch and Clark, 2004; Hirsch et al., 2016; Staugaard, 2010). Some researchers have further suggested that the examination of interpretation bias in social anxiety should take a dimensional perspective and address not only negative but also positive interpretation biases (Huppert et al., 2003). Thus, the second aim of the present study was to examine these and other potential factors that may moderate the relationship between interpretation bias and social anxiety, and thus may have contributed to the heterogeneity in previously reported studies and narrative reviews. In addition to examining objective versus subjective interpretation bias measures, positive versus negative stimulus valence, and different types of stimuli as moderators, we also investigated the potential roles of study population and study design.

In their narrative review, Clark and McManus (2002) recommended including objective measures (i.e., on-line paradigms) to determine the existence of biased interpretation as has been found with subjective measures (i.e., off-line procedures). Unlike subjective measures, which capture interpretation processing by self-report methods such as rating scales, objective measures capture this processing by way of response latencies to stimuli, and are thus less subject to demand, selection, or response bias effects (Hirsch et al., 2016). Accordingly, the current study sought to meta-analyse differences between subjective and objective measures of interpretation bias in socially anxious individuals.

Some researchers have argued that negative versus positive interpretation biases are not opposites of a single dimension (Amir et al., 2012; Huppert et al., 2003; Steinman et al., 2020). However, many paradigms do not allow for an independent assessment of these two types of interpretation bias, making it difficult to distinguish them. A more recent meta-analysis by Kivity and Huppert (2016) examined whether socially anxious individuals respond differently to emotional expressions of threatening, neutral, and positive valence than controls. They found that socially anxious individuals rated people as less approachable irrespective of the valence of expressions and reported higher arousal toward neutral expressions. These findings highlight the impact of stimulus valence on the judgement of emotional reactions in socially anxious individuals. The current meta-analysis therefore examined the role of stimulus valence (i.e., negative, positive, neutral, and ambiguous) in impacting interpretation bias. Specifically, we investigated whether social anxiety is characterised not only by the presence of a threat bias but also by the absence of a positive interpretation bias.

Previous narrative reviews have highlighted the importance of type of stimulus materials when investigating interpretation bias in Social Anxiety Disorder (Heinrichs and Hofmann, 2001; Staugaard, 2010). Two types of stimuli, namely verbal and visual stimuli are commonly used to assess interpretation bias in social anxiety (disorder). Verbal stimuli have included those in a written format such as words, sentences, vignettes, and scenarios whereas visual stimuli have consisted of facial expressions (photographs or video clips) and videotaped social situations or behaviours. Heinrichs and Hofmann (2001) argued that verbal stimuli are more descriptive, whereas other researchers have argued that verbal stimuli are not especially socially threatening (Amir et al., 1998; Hertel et al., 2008; Stopa and Clark, 2000). Yet others have argued that visual stimuli, such as faces are more ecologically valid, as a face can more readily present a social threat and convey negative evaluation through a change in expression (Heuer et al., 2010).

Previous narrative reviews have concluded that studies that have used verbal stimuli have generally found an interpretation bias towards socially threatening information in socially anxious individuals (Heinrichs and Hofmann, 2001; Hirsch et al., 2016). However, studies that have used visual stimuli (i.e., threatening faces) have reported mixed results, with some studies, but not others, demonstrating a negative interpretation bias in socially anxious individuals (Staugaard, 2010). Staugaard (2010) explained the inconsistencies by suggesting that the particular stimuli used across studies may differentially affect the perceptual processing of people with social anxiety. This highlights the importance of examining the impact of different types of stimuli in the assessment of interpretation bias in socially anxious individuals. The current study aimed to address this gap by meta-analysing the impact of the two types of stimuli (i.e., verbal and visual) and their subtypes (i.e., sentences, scenarios, facial expressions, and videos) on interpretation bias in socially anxious individuals. Specifically, we examined whether there are differences in the observation of interpretation bias for different types of stimuli, and whether this bias exists consistently across studies that used a diversity of stimuli.

A recent systematic review and meta-analysis which examined the relationship between negative interpretations and anxiety in children and adolescents, categorised moderators of this relationship into two groups, namely “Procedural” (e.g., design and assessment) and “Population” (e.g., clinical or sub-clinical sample) (Stuijfzand et al., 2018). The current study therefore further extended the examination of the potential moderators that could contribute to the size of the relationship between biased interpretation and social anxiety by including sample population and study design as moderators.

In terms of sample population, previous research has suggested that social anxiety may be better understood as a severity continuum, from mild presentations such as shyness, to severe and impairing forms, namely Social Anxiety Disorder (McNeil, 2001; Ruscio, 2010). Populations with high levels of social anxiety, but sub-threshold for diagnosis, are comparable to populations with a clinical diagnosis, because of the strong fear of social situations and associated severe impairment in functioning (Wittchen and Fehm, 2003). Thus, it has been suggested that studying individuals with social anxiety at a sub-diagnostic level provides empirical and clinical information on Social Anxiety Disorder (McNeil, 2001; Rapee and Heimberg, 1997; Ruscio, 2010). However, research has also acknowledged that the nature of the symptoms in analogue samples is not necessarily comparable to patients with social phobia. As such, researchers have recommended investigating the generalisation of findings from analogue populations to clinical populations (Badra et al., 2017; Chen et al., 2019; Schofield et al., 2007). Accordingly, the current study sought to examine whether the size of the relationship between interpretation bias and social anxiety would be different in clinical and non-clinical populations.

Determining the impact of the study design on outcomes is common practice in systematic reviews and meta-analyses (de Lijster et al., 2017; Haines et al., 2007; Varese et al., 2012). Designs vary in the degree of confidence they instil regarding a causal relationship between two factors. To reduce the likelihood of over optimising conclusions about the results due to study designs and ensure the validity of conclusions drawn (Haines et al., 2007), the current study examined whether the size of the relationship between interpretation bias and social anxiety would differ based on the design of the included studies.

The current systematic review and meta-analysis of the relationship between interpretation bias and social anxiety is both timely and important. Converging or diverging evidence from different methodologies, sample populations, and study designs would provide a comprehensive understanding of how interpretation bias operates in social anxiety. The findings will help to consolidate the evidence on the role of biased interpretations in socially anxious individuals. Such information is critical for the development of individually tailored interventions that target interpretation bias in Social Anxiety Disorder.