Evaluation of the Factor Structure and Content Specificity of the Interpretation Bias Task (IBT)

Theoretical models suggest that negative interpretation bias, the tendency to interpret ambiguous information in a negative way, underlie the course and severity of a range of mental disorders including depression, anxiety, and psychosis, and can also influence the treatment of these disorders (Blanchette and Richards 2010; Everaert et al. 2017; Hirsch et al. 2016; Leonidou and Panayiotou 2018; Mobini et al. 2013; Savulich et al. 2012; Stuijfzand et al. 2018). To date, evidence of interpretation bias in relation to different types of anxiety disorders have revealed reasonably consistent results regarding the direction and intensity of these biases (Hirsch et al. 2016; Leonidou and Panayiotou 2018; Mobini et al. 2013; Schoth and Liossi 2017; Stuijfzand et al. 2018). For instance, people with health anxiety might endorse more pain-/illness-related interpretations for ambiguous health-related scenarios (Leonidou and Panayiotou 2018), whereas people with social anxiety might endorse more negative interpretations for situations related to interpersonal interaction or social evaluation, compared to non-anxious people (Hirsch et al. 2016). These negative interpretation biases can contribute to an exaggerated perception of physical/social threat, which might not only influence how people attend to threat-related information but might also encourage threat avoidance that limits the amount of additional and contradictory information that a person might encounter (Hirsch et al. 2016).

Numerous paradigms that use ambiguous words (e.g., die/dye, pain/pane, bury/berry, etc.), pictures (e.g., emotional and neutral facial expressions), or scenarios (e.g., “You have visitors round for a meal and they leave sooner than expected”) as stimuli have been developed to measure interpretation biases, among which ambiguous scenarios have been most frequently adopted (Schoth and Liossi 2017). One of the advantages of ambiguous scenarios over word and pictorial stimuli is that they contain contextual information that approximate real-world situations and therefore may be more ecologically valid. For example, in a similarity rating task, participants are first presented with ambiguous scenarios that can be interpreted in either a negative or a benign way (Mathews and Mackintosh 2000; Schoth and Liossi 2017). In a subsequent recognition phase, participants are shown disambiguated threatening or non-threatening sentences and they need to rate the similarity of each of these sentences to the scenarios formerly presented (Mathews and Mackintosh 2000; Schoth and Liossi 2017). Higher similarity ratings for threatening sentences indicate more negative interpretation biases.

Despite the wide variety of tasks available for measuring interpretation biases, research findings regarding the content specificity of this bias are not yet conclusive. Content specificity refers to the extent to which the contents of the observed biases specifically match the psychopathology that a person experiences (Beck 1976). For example, patients with depression may endorse more negative interpretations for depression-/rumination-specific information (Hirsch et al. 2016), whereas psychosis-related paranoia may be characterised by an interpretation bias specifically for paranoid information (Savulich et al. 2015, 2017). Similar findings have been revealed in the anxiety literature, where negative interpretation biases and anxiety are associated more strongly when the contents of interpretation correspond to the subtype of anxiety (Blanchette and Richards 2010; Stuijfzand et al. 2018). However, almost all existing evidence regarding the content specificity of interpretation biases in anxiety disorders has only compared social and non-social scenarios, broadly defined, within the context of social anxiety (Blanchette and Richards 2010; Stuijfzand et al. 2018).

Paralleling the psychopathology literature, interpretation biases have also been studied in the context of physical health problems such as chronic pain. In particular, patients with chronic pain have been found to endorse more threatening interpretations for pain-/illness-related information (Schoth and Liossi 2016). This interpretation bias has also been suggested to contribute to heightened levels of pain-related fear, which may then alter pain expectations and exacerbate the intensity and interference of pain in this population (Vlaeyen and Linton 2000). One novel paradigm that emerged from this literature is the Adolescent Interpretations of Bodily Threat (AIBT) task which contains 16 vignettes that describe ambiguous situations in two different domains of daily life (i.e., eight situations related to bodily threat and eight situations related to social evaluation; Heathcote et al. 2016). Each ambiguous situation contains one benign and one negative possible consequence. Participants are asked to first imagine themselves in these situations and then (1) rate whether each consequence is likely to pop into their mind, and (2) rate their belief that each interpretation would actually happen in reality. Using this task, Heathcote et al. (2017) found that adolescent chronic pain patients endorsed fewer benign interpretations for bodily threat scenarios compared to healthy controls, but this group difference did not extend to social situations, suggesting that the AIBT task is a suitable measure of interpretation biases while also providing evidence for content specificity in their patient sample.

More recently, this task has been adapted to include a wider array of scenarios and validated by factor analyses in adolescents with and without persistent and impairing pain (Lau et al. 2020). The original bodily threat domain has been expanded to include eight scenarios describing immediate bodily injury (e.g., cut) and eight scenarios describing long-term illness (e.g., tumour). Similarly, the original social evaluation domain has been expanded to include eight scenarios describing social rejection and eight describing performance failure (Lau et al. 2020). In this expanded AIBT, an example of an immediate bodily injury scenario is: ‘Someone kicks a ball and it hits you in the face. In the mirror you see your face is covered in …’. Participants are first presented with each ambiguous situation and then are offered words that resolve the situation in a negative or benign manner, for example, ‘mud’ or ‘blood’. They are then asked to rate the likelihood that each consequence would actually happen on a 5-point scale. Higher ratings for the word ‘blood’ reflect a more negative interpretation while higher ratings for the word ‘mud’ indicate a more benign interpretation. Similarly, an example of a long-term illness situation is ‘You take a pill every morning at breakfast. The pill is a …’ followed by ‘vitamin’ and ‘medicine’. An example of a social rejection situation is ‘You are walking and as you approach the people in front of you, you hear them say your name. They are saying something …’ followed by ‘positive’ and ‘negative’. Finally, an example of a performance failure situation is ‘Your teacher has decided to give a surprise test. You are sure that you will do …’ followed by ‘well’ and ‘badly’. In their study, Lau et al. (2020) collapsed the immediate bodily injury and long-term illness domains into one subscale and the social rejection and performance failure domains into another subscale following model-fitting of item coherence. However, contrary to the content specificity hypothesis, their results showed that adolescents with moderate-to-high pain interference endorsed a more negative and less benign interpretation bias across all domains compared to adolescents without interfering pain (Lau et al. 2020).

Despite the novelty of the original and adapted AIBT tasks, several research gaps are evident. First, this task has only been used in Western adolescent samples with and without pain problems (Heathcote et al. 2016, 2017; Lau et al. 2020) and therefore its suitability for adult samples and for people from non-Western cultures remains unknown. Second, although a significant body of research has examined the role of interpretation biases in anxiety disorders (Blanchette and Richards 2010; Hirsch et al. 2016; Mobini et al. 2013; Stuijfzand et al. 2018), no study has yet tested the associations between responses to the AIBT task and different subtypes of anxiety symptoms, such as health and social anxiety. Although the AIBT task was originally developed for pain research, scenarios in the immediate bodily injury and long-term illness domains also appear highly relevant to health anxiety, which is characterised by the catastrophisation of bodily sensations and constant worry about having or acquiring a disease (Rachman 2012). In comparison, scenarios in social rejection and performance failure domains appear more relevant to social anxiety, which is characterised by the fear of negative social feedback and perfectionistic self-presentation (Nepon et al. 2011). Additionally, previous studies using the AIBT task revealed mixed results regarding its content specificity (Heathcote et al. 2016, 2017; Lau et al. 2020). This inconsistency highlights the need for replication in larger and more varied samples of participants.

The current article presents two studies to fill these research gaps. We adopted the expanded AIBT task from Lau et al.’s study (2020) that includes four domains of scenarios (i.e., immediate bodily injury, long-term illness, social rejection, and performance failure) and then examined its factor structure using two adult samples. Since the acronym “AIBT” stands for Adolescent Interpretations of Bodily Threat, to avoid ambiguity, and to highlight the inclusion of social scenarios within this measure, we renamed this task as the Interpretation Bias Task (IBT). Unlike previous studies, we instructed participants in the current study to rate how likely each consequence would actually happen on a scale from 1 (not at all likely) to 100 (extremely likely) since we hypothesised that individual differences in interpretation biases would be more precisely captured by measuring the perceived likelihood percentage of potential consequences rather than the 5-point Likert scales used previously (Heathcote et al. 2016, 2017; Lau et al. 2020). We expected that each domain of the IBT could be represented by two unique factors, one for benign resolutions and one for negative resolutions. We also expected that the four domains could be represented by distinct factors. In addition, to determine the content specificity of interpretation biases, the associations between responses to this task and health and social anxiety symptoms were investigated. In particular, we expected that people with elevated health anxiety symptoms might endorse more negative and fewer benign interpretations for immediate bodily injury and long-term illness scenarios, whereas people with higher social anxiety symptoms might endorse more negative and fewer benign interpretations for social rejection and performance failure scenarios.

To confirm the factor structure of the IBT derived from Study 1, and to examine whether the correlational findings between anxiety symptoms and interpretation biases could be replicated, we conducted a second study in a larger community sample (not limited to university students) with a wider age range. Specifically, we collected responses from an Asian adult sample again and conducted CFAs with only the remaining scenarios in the IBT (i.e., immediate bodily injury: scenarios 3–6 and 8; long-term illness: scenarios 1–5 and 7; social rejection: scenarios 2, 5, 7 and 8; performance failure: scenarios 1–8). We also performed similar correlation tests between responses to the IBT, HAI and LSAS.

The current article presented two studies that evaluated the psychometric properties of the IBT, an adapted version of the AIBT task (Heathcote et al. 2016), a computerised task that measures interpretation biases of ambiguous scenarios that could happen in different domains of daily life. Results of the factorial analyses revealed a clear structure of the IBT with eight distinct factors (4 domains × 2 valences). More specifically, the four domains of the IBT (i.e., immediate bodily injury, long-term illness, social rejection and performance failure) were separable and each could be represented by a two-factor structure (i.e., benign and negative). The presence of these eight factors indicated that interpretation biases vary across domains and across valences. Results demonstrated the suitability of the IBT in assessing interpretation biases in different contexts among young Asian adults.

Regarding the correlations between health anxiety symptoms and interpretation biases, the two studies showed inconsistent results. In particular, in Study 1, we found that those with elevated health anxiety symptoms had fewer benign interpretations for immediate bodily injury and long-term illness scenarios, but these effects did not extend to negative interpretations or to other social scenarios (after controlling for social anxiety symptoms). In contrast, in Study 2, health anxiety symptoms were correlated with all IBT scores whether or not social anxiety symptoms were added as a covariate. Here we suggest two possible explanations for this inconsistency. First, Study 1 was limited to a university student sample, whereas Study 2 recruited adults who aged from 17 to 64 years. It is possible that interpretive processes and their association with anxiety symptoms vary across age cohorts (e.g., younger vs. older adults) and across different samples (e.g., university vs. community samples). Second, Study 2 had a sample size that is four times larger than that in Study 1; the overall significant correlations between health anxiety symptoms and IBT scores in Study 2 may be a partial function of the sample size. Despite the inconsistent findings, both studies suggest that injury-/illness-related interpretation biases may be important cognitive factors in the context of health anxiety. In particular, the decreased tendency to interpret injury-/illness-related information in a benign manner might influence the way people attend to, encode and remember this information, which might then result in an overestimation of the likelihood and negative consequences of potential illnesses, as well as maladaptive behaviours such as repeatedly checking one’s body for signs of illness (Leonidou and Panayiotou 2018; Rachman 2012).

Both studies found that people with higher social anxiety symptoms endorsed more negative and fewer benign interpretations for social rejection and performance failure situations, but not for immediate bodily injury and long-term illness situations (after controlling for health anxiety symptoms). These results are in line with recent reviews that confirmed the associations between social anxiety and the tendency to interpret ambiguous social information in a catastrophic fashion in both adults and adolescents (Blanchette and Richards 2010; Hirsch et al. 2016; Stuijfzand et al. 2018). These findings also add to current evidence regarding the role of interpretation biases in the maintenance and exacerbation of social anxiety. In particular, the tendency to interpret social information negatively might result in an exaggerated estimation of the negative consequences of social interactions, which might then contribute to behaviours such as avoidance of eye contact or of social interaction altogether (Heimberg et al. 2014; Nepon et al. 2011).

The findings of the current study only provided partial support for the content specificity of the IBT and interpretation biases. Put otherwise, the content specificity hypothesis only held for correlations between interpretation biases and social anxiety symptoms, but not when health anxiety symptoms were examined. It is possible that social anxiety is characterised by a negative interpretation bias for ambiguous social scenarios only, while health anxiety is characterised by an overall interpretation bias across domains. Although this contention warrants further investigation, this study is not the first that provided evidence against the content specificity of the IBT. Heathcote et al. (2016) found that adolescents with more pain issues in the past 3 months endorsed more negative and fewer benign interpretations for ambiguous scenarios across bodily threat and social domains. Lau et al. (2020) found that adolescents with moderate-to-high pain interference had a more negative and less benign interpretation bias across all four domains of the expanded AIBT task compared to those without interfering pain. It is important for future studies to test whether the domain specificity of interpretation biases varies between anxiety subtypes. It is also important to determine which aspect of interpretation biases (i.e., domain and valence) underlies the maintenance of anxiety symptoms using longitudinal designs.

One might be concerned that some resolutions in the IBT scenarios are seemingly two ends of a single dimension. For example, the scenario “You have made an appointment to see your doctor to discuss your test results. You think the results will probably show you are …” contains two resolutions, “ill” and “fine”, that can be considered as two extreme cases on a single dimension. However, the two resolutions in each scenario do not necessarily have to be mutually exclusive. In both of our studies, we instructed the participants to give ratings to the benign and negative resolutions that do not need to add up to 100, which means that they could give high/low ratings to both resolutions at the same time. Heathcote et al. (2017) found that adolescents with chronic pain had fewer benign interpretations for bodily threat scenarios than healthy controls, but this difference was not evident for the negative interpretation score. Therefore, being less likely to interpret ambiguous scenarios in a benign manner might not always mean a person is more likely to interpret these scenarios negatively. This suggests that the benign and negative resolutions in each domain are independent and should not be condensed into a single dimension.

Several limitations are evident. First, both studies were based on participants’ self-report of their anxiety symptoms. Although recent reviews have confirmed that illness- and social-related interpretation biases are evident in patients with clinical health and social anxiety disorders (Hirsch et al. 2016; Leonidou and Panayiotou 2018), future studies should assess the IBT in clinical populations so that the psychometric properties of this task can be further established. Relatedly, in both studies we did not assess participants’ current status or history of mental disorders. Therefore, we were not able to control for the possible confounding effects of concurrent, or historical, psychiatric diagnoses. Also, we did not control for the effect of generalised anxiety symptoms in both studies. Recent reviews have confirmed that people with generalised anxiety disorder have a nonspecific negative and threat-oriented interpretation bias for all sorts of ambiguous information, which may be reflective of the wide range of worry topics evident in this population (Hirsch et al. 2016; Stuijfzand et al. 2018). Participants with elevated generalised anxiety symptoms may also have high health anxiety symptoms, and may endorse more negative interpretations across all domains of the IBT. Therefore, the significant correlations between health anxiety symptoms and all IBT scores in Study 2 might be driven by generalised anxiety symptoms in our participants.

Moreover, scenarios in the original version of our task (i.e., the AIBT task) were generated based on two adolescent samples (aged 16–18 years) recruited from secondary schools in the UK (Heathcote et al. 2016). The expanded version of the AIBT task was also assessed in British adolescents aged 16–19 years (Lau et al. 2020). Therefore, it may be that the current set of IBT scenarios are more concern-relevant to British adolescents than to Asian adults. For example, consider one scenario in the performance failure domain: “Your teacher has decided to give a surprise test. You are sure that you will do …” followed by “well” (benign) and “badly” (negative). This scenario appears more suitable for student samples than for older adults who no longer need to take courses and exams, and therefore may not accurately capture interpretation biases in non-student adults in our second sample. These items were nonetheless considered appropriate for inclusion in the IBT because most individuals can probably recall such a scenario even if they are not currently experiencing it. Also, a recent study found that Hong Kong residents were more positively biased compared to people living in the UK when tested on culturally-validated tasks measuring interpretation biases (Yiend et al. 2019). As such, future studies may need to examine age and cultural differences in participants’ responses to the IBT, and determine whether inclusion of more age- or culture-specific scenarios might improve the reliability and validity of the IBT.

Finally, in our IBT we explicitly instructed participants to imagine themselves in the ambiguous scenarios, which may involve mental imagery in addition to interpretation biases. In the depression literature, theoretical models and evidence have emphasised the critical role of mental imagery in interpretation biases such that the use of mental imagery may amplify the impact of interpretation biases on depressive symptoms (Everaert et al. 2017; Holmes et al. 2009). Future studies should examine whether responses to the IBT differ between imagery perspectives (i.e., first-person vs. third-person) and vary as a function of the extent to which mental imagery is involved during the task.

In summary, the two studies in the current article evaluated the factor structure of the IBT. Results showed that interpretation biases can be differentiated across domains (i.e., immediate bodily injury, long-term illness, social rejection and performance failure) and between valences (i.e., benign vs. negative). The IBT is a suitable measure of interpretation biases for healthy Asian adults including those with elevated levels of health/social anxiety symptoms. Correlational analyses showed that social anxiety symptoms were associated with interpretation biases for social rejection and performance failure scenarios, but not for injury-/illness-related scenarios. However, it remains unclear whether interpretation biases in relation to health anxiety symptoms are content-specific. Future testing with clinical samples, comparisons between different populations, and considerations of mental imagery as well as other important moderating factors, is warranted.

This line of research will also serve as a foundation for future studies that modify and advance existing treatments that specifically target this cognitive mechanism. In particular, interpretation bias modification programs encouraging stronger beliefs in benign interpretations of ambiguous information have already been developed for populations with health (Antognelli et al. 2020) and social anxiety (Mobini et al. 2013). Although most bias modification studies have been conducted in Western, English-speaking samples, here the psychometric properties of the IBT have been evaluated in different languages (i.e., English and Chinese) and in different age groups (i.e., adolescents and adults). As such, the IBT may now be incorporated as a useful outcome measure in future bias modification studies to allow for more direct comparisons across study populations.