Introduction
Dysfunctional cognitions about rejection or shame are central in social anxiety disorder. The fear stemming from these convictions leads to a range of behaviours characteristic of social anxiety disorder (Clark and Wells
1995; Rapee and Heimberg
1997). Current cognitive models emphasize the role of information processing biases such as judgmental bias, attentional bias and interpretation bias in maintaining socially anxious convictions (for reviews see Bögels and Mansell
2004; Clark and McManus
2002; Hirsch and Clark
2004). The major focus of current treatment strategies (e.g., Clark and Wells
1995; CPA
2006; Trimbos-instituut
2003) is to somehow challenge these convictions in an attempt to replace dysfunctional and oftentimes irrational beliefs by more rational ones. The alleged crucial role of irrational beliefs in the persistence of complaints points to the vital importance of individuals’ ability to draw adequate conclusions. The inability to draw appropriate conclusions on the basis of available evidence seems a particularly direct way to impede the adjustment of irrational, anxiogenic beliefs. In cognitive psychology, the relatively poor performance in drawing appropriate conclusions when reasoning with materials that are counterintuitive (i.e., have a mismatch between the believability and the logical validity) is known as the belief bias effect. Despite its apparent importance, the ability to evaluate (dysfunctional) beliefs in light of evidence has received little attention in psychopathology research.
Belief bias refers to a bias in deductive reasoning that acts to confirm rather than falsify prior beliefs, which is demonstrated in a tendency to endorse a priori believable conclusions as valid and unbelievable conclusions as invalid, regardless of their actual logical status (Evans et al.
1993a). It is assumed to facilitate the maintenance of a relatively stable belief system from which the world and experiences can be interpreted without great effort, leaving the attentional capacities for more urgent and complex tasks. Therefore, in everyday life some degree of belief bias might be considered functional. Also in potentially dangerous situations, it may well be adaptive to rely on prior beliefs and act on plausible conclusions, rather than to consider whether those conclusions meet the standards of formal logic (e.g., Evans et al.
1993b). If, however, the perceived threat is based on dysfunctional convictions (for instance, ‘If I say something odd, people will ridicule me’), belief bias may become counterproductive. In that case, such a bias in deductive reasoning could impede the disconfirmation of anxiogenic beliefs, which in turn may lead to stable cognitions feeding the anxiety disorder (cf. de Jong et al.
1997). Accordingly, belief bias may play a fairly direct role in the maintenance of fearful preoccupations.
Belief bias can be measured using a linear syllogistic reasoning task (e.g., Smeets and de Jong
2005). In performing this task participants are asked to judge as quickly as possible the logical validity of syllogisms consisting of two statements, the premises, and a conclusion. Logical validity refers to the necessity of a conclusion, assuming that the premises are true. If it is true that ‘A is larger than B’ and that ‘B is larger than C’, it follows that ‘A must be larger than C’. Logical validity would be violated when one concludes that ‘C is larger than A’ based on the given premises. When judging the validity, participants are instructed to ignore the believability of the conclusions. Believability refers to the meaning of the syllogism’s conclusion. An example of a generally believable conclusion would be: ‘An elephant is bigger than a mouse’, whereas ‘A mouse is bigger than an elephant’ represents an example of a generally unbelievable conclusion. A valid yet unbelievable linear syllogism would be as follows:
Premise 1 | A mouse is bigger than a dog |
Premise 2 | A dog is bigger than an elephant |
Conclusion | A mouse is bigger than an elephant |
Thus, participants have to judge whether a syllogism is logically valid, while ignoring its meaning. People are typically faster in reaching a decision about the validity of a syllogism when there is a match than when there is a mismatch between the validity and believability of the conclusion.
Although it seems plausible to apply the belief bias theory to dysfunctional convictions, there is a clear distinction with past research: Past studies have focussed on universal truths and common beliefs for which confrontation with disconfirming evidence is unlikely, whereas the current study focusses on the potential relevance of belief bias for dysfunctional convictions for which disconfirming evidence is oftentimes available.
In a first attempt to explore this relationship, de Jong et al. (
1997) tested spider phobic participants and non-phobic controls for belief bias when reasoning with spider phobia relevant materials. They failed to find a convincing difference between the phobic and the non-phobic group. This might well have been due to methodological problems. Most important, spider phobia relevant beliefs (e.g., as indexed by the Spider Phobia Questionnaire by Arntz et al.
1993) are hard to translate into linear syllogisms, which are based on comparison (e.g., A spider is creepier than a fish, a fish is creepier than a pigeon, hence a spider is creepier than a pigeon). The necessary inclusion of a comparison category decreases the resemblance between the syllogisms’ conclusions and the dysfunctional beliefs, thereby probably decreasing the sensitivity of the task. In addition, it is doubtful whether spider phobia is the optimal candidate for testing this hypothesis. Although there is evidence that spider phobic individuals do report high believability ratings for irrational spider related beliefs (e.g., ‘the spider will kill me’; Arntz et al.
1993; Thorpe and Salkovskis
1995), it is still a matter of dispute whether dysfunctional beliefs indeed play a crucial role in the aetiology and maintenance of the phobic complaints. Some authors described spider phobia as a prototypical “non-cognitive” (evolutionary prepared) fear (e.g., Seligman
1971). Accordingly, spider fearful individuals find it extremely difficult to articulate what they actually fear (e.g., Davey
1992).
Therefore, the present study focussed on social anxiety (rather than spider phobia) to test further the potential role of belief bias in anxiety disorders. Dysfunctional beliefs are generally assumed to be central to social anxiety disorder (e.g., Clark and Wells
1995), and a striking feature of these beliefs is their persistence in the face of incompatible data. That is, because socially anxious individuals cannot so easily avoid the situations they strongly fear (as spider phobic individuals can), most socially anxious individuals will have been involved in many social situations that contradicted their fearful convictions (e.g., situations in which they are not ridiculed for saying something odd). Moreover, social anxiety beliefs often imply social comparison, making social anxiety convictions more suitable for translation into linear syllogisms (e.g., ‘I am not likeable’ translates into ‘I am less likeable than others’ or into a linear syllogism such as ‘I am less likeable than Jane and Jane is less likeable than John’). The main aim of the present study was thus to test the hypothesis that socially anxious individuals are characterized by belief bias when reasoning about social anxiety themes. Therefore, a group of individuals varying in their level of fear of negative evaluation (one of the central cognitive concepts within social anxiety, e.g., Clark and Wells
1995) was presented with a series of linear syllogisms concerning themes relevant to social anxiety.
If enhanced belief bias is only evident for dysfunctional convictions, this would be consistent with the idea that the rigidity of anxiogenic beliefs may not itself result from a reasoning abnormality, but may represent a normal tenacity of important and strongly held beliefs (cf. Garety and Hemsley
1997). Yet, research in the context of spider phobia (de Jong et al.
1997) provided preliminary evidence to suggest that psychopathology patients show a generally enhanced belief bias (i.e., not restricted to the domain of the psychopathological concerns). This raises the possibility that this reasoning bias reflects a trait-like information processing bias that acts as a diathesis in the development of psychopathological disorders in general (cf. Arntz et al.
1995). As a subsidiary issue it was therefore tested whether socially anxious individuals are (also) characterized by enhanced belief bias for factual information that is irrelevant for their social anxiety concerns.
To summarize, content interferes with logical reasoning when reasoning with highly believable materials. As socially anxious people hold strong social anxiety convictions, one can expect to find a belief bias effect concerning social anxiety related materials for the high social anxiety group and not for the low social anxiety group. In addition to this content-specificity hypothesis, it is explored, based on the earlier finding by de Jong et al. (
1997), whether high socially anxious people have a general tendency to apply belief bias more often compared to low anxious people. Therefore, we also tested whether socially anxious individuals will show a relatively strong belief bias when reasoning with neutral, generally believable, materials.
Discussion
This study investigated the relationship between belief bias and social anxiety. Although the study was initially designed to compare a high and low anxiety group, a shift in design had to be made due to the change in BFNE scores after preselection. Instead of making group comparisons, it was tested whether belief bias increased with increasing BFNE scores. Furthermore, the analyses were restricted to the RT data, as the error rates and the dispersion were too low to be analysed. The low error rates indicate that participants confirmed to the task and did not show response biases. The main results can be summarized as follows. First, for the social anxiety relevant materials, results indicated that the higher participants’ fear of negative evaluation, the stronger the belief bias effect. Second, for the neutral common knowledge syllogisms, there was an overall belief bias effect that was independent of participants’ fear of negative evaluation.
According to contemporary cognitive models of anxiety disorders, persistent dysfunctional cognitions (such as ‘If I make a mistake, people will make fun of me’) play a vital role in the maintenance of complaints (e.g., Beck et al.
1985; Clark and Wells
1995; Rapee and Heimberg
1997). One obvious explanation for the refractoriness of this type of anxiogenic convictions is that socially anxious individuals are actually evaluated less positively than non-anxious individuals, for example because they behave less skilful in social situations. In line with this, there are indications that in some situations people suffering from social anxiety may indeed perform less well than non-anxious controls (e.g., Stopa and Clark
1993; Voncken and Bögels
2008). This does not however imply that the convictions of social anxiety patients are necessarily true, as these oftentimes concern blunt negative appraisal or rejection by others. Another mechanism that may play a fairly direct role in the persistence of these anxiogenic convictions concerns individuals’ difficulty to correct their dysfunctional convictions when confronted with disconfirming evidence. Correcting erroneous convictions requires the ability to accurately deduce the logical implications of empirical evidence for certain convictions. For instance, not being made fun of after having made a public mistake should lead to correction of the dysfunctional belief ‘If I make a mistake, people will make fun of me’, since it proves that the cognition is invalid. In support of the hypothesis that belief bias may be involved in social anxiety, the results for the RT data showed that individuals high in fear of negative evaluation have relative difficulty in judging anxiogenic (i.e., social anxiety congruent) information as false and reassuring non-congruent information as true. Such a belief bias effect for social anxiety convictions logically prevents dysfunctional cognitions from being corrected, thereby sustaining phobic fear.
It should be acknowledged that belief bias theory concerns errors in reasoning. In the present study we used linear syllogisms that are known to be relatively easy and to produce little errors (Huttenlocher
1968). Indeed, in line with previous research using this type of syllogisms (e.g., de Jong et al.
1997), participants in this study made only few errors. This implies that the participants actually reasoned analytically when performing the task. In this study, a belief bias effect for RTs was found in a single-task situation where all resources could be employed to the task. With all resources available, the participants needed more time to answer the mismatched syllogisms, indicating that it took more effort and/or resources to answer these syllogisms. It seems safe to assume that when reasoning takes more effort in a lab, it will result in faulty reasoning when sufficient cognitive resources and/or the motivation to reflect on the validity of their initial convictions are lacking, which is likely to be the case in most real life situations (e.g., Beevers
2005; Evans and Curtis-Holmes
2005). Obviously, further research manipulating the availability of cognitive resources is necessary to arrive at more final conclusions in this respect.
The absence of a relationship between belief bias for neutral common knowledge and fear of negative evaluation indicates that anxious individuals are not characterized by a reasoning abnormality and that the belief bias for social anxiety convictions that was found in the present study reflects a normal tendency to reason in a belief biased manner with respect to strongly held convictions.
7 This belief bias for social anxiety convictions is merely problematic because it logically acts to maintain convictions that are
dysfunctional.
The finding of complaint-related belief bias for individuals who are fearful of negative evaluation is an important first step in determining whether belief bias may indeed be involved in the maintenance of social anxiety disorder. Meanwhile, it should be acknowledged that on the basis of the present study it cannot be ruled out that this belief bias for social anxiety convictions is a mere symptom of social anxiety rather than a mechanism that reciprocally strengthens the dysfunctional convictions. While causality problems of the present type are hard to solve, they are theoretically important. As a next step it would be worthwhile investigating whether post-treatment belief bias is predictive of relapse after successful treatment (cf. de Jong et al.
1995). If not, causality seems highly unlikely. A more direct and rigorous way to test the causal properties of belief bias would be to specifically reduce belief bias and to test whether this results in a reduction of dysfunctional beliefs and symptoms of social anxiety (cf. MacLeod et al.
2002). Perhaps most relevant to the clinical context is the question whether enhanced belief bias present after successful treatment of the social anxiety disorder can predict relapse. If the complaints have disappeared, but social anxiety related belief bias is still present, this belief bias potentially indicates that the patient still holds social anxiety related convictions. As such, the belief bias task may serve as an implicit measure to detect such (potentially unreported) remaining beliefs. Of course, further research is required to actually test these notions.
It is a well-established fact that the belief bias theory holds for common knowledge and commonly shared prejudices (e.g., Evans et al.
1993a). The current study illustrates that belief bias effects can also be found for irrational convictions for which disconfirming evidence is available. The finding that correct information does not necessarily result in disconfirmation of irrational convictions emphasizes the difficulty for people to reason following logical rules. This underscores the importance of explicitly discussing the arguments for and against dysfunctional convictions in the context of behavioural experiments as a way to help patients to detect the relevant premises or arguments for their dysfunctional conclusion.
Limitations
Although the correlation between the believability check and the BFNE was significant and supports the validity of the stimulus materials that were used, the modest strength of the association suggests that there is also still room to further improve the validity of the stimulus material and thereby the sensitivity of the present belief bias task. It should be acknowledged that global social anxiety themes were used. The validity of the task may be enhanced by adjusting the syllogisms to individuals’ core beliefs. In addition, the construction of linear syllogisms required the inclusion of abstract contrasts (e.g., I am less socially skilled than person A and person A is less socially skilled than person 1) which might have resulted in a suboptimal reflection of the individual’s actual convictions. Future research may need to search for different paradigms to measure belief bias which allow for a better match of the materials with the actual convictions.
There was a discrepancy between the BFNE scores during the mass-screening and during the experiment proper. This could raise some doubts concerning the reliability and validity of our screening instrument. Yet, the reliability scores of both test administrations were high. Hence, there is reason to suspect that the changes in scores reflect real changes in social anxiety rather than a statistical artifact (cf. Dijk and de Jong,
in press) or unreliability of the BFNE. Ample new social experiences associated with starting a new life as a student could potentially explain the unexpected deviance in FNE scores between the mass-screening and the actual experiment. These change in BFNE scores interfered with our planned factorial approach. Fortunately, the range and distribution of BFNE scores during the actual experiment allowed us to test our hypotheses while maintaining the continuity of our data, resulting in a relatively powerful design.
Another point of attention lies in the use of the BFNE as a measure of social anxiety. There have been some concerns with the use of BFNE as a measure of social anxiety, given that it only measures beliefs and not behaviours (Wilson and Rapee
2005). On the other hand, Collins et al. (
2005) and Weeks et al. (
2005) have found that the BFNE is a valid measure for clinical social anxiety groups. In addition, Stopa and Clark (
2001) showed that for psychological process studies, an analogue design based on BFNE-scores produces findings that are essentially the same as those found in studies using social anxiety disordered patients and non-clinical controls. The results of the current study can be potentially relevant to other patient groups as well: Studies using different analogue or patient groups such as eating disorders have found correlations between the BFNE and self-reported eating disorder and depressive complaints (e.g., Gilbert and Meyer
2003; Hinrichsen et al.
2003). On the other hand, both eating disorder and depression self-report questionnaires are known to correlate with other measures of social anxiety complaints as well (e.g., Gibb et al.
2005; Hinrichsen et al.
2004), and both disorders are found to have high comorbidity with social anxiety disorder (e.g., Kessler
1995; Pallister and Waller
2008). Whether the results of the current study can be generalized to disorders such as depression and eating disorder remains to be seen.
The order of the BFNE and the syllogistic reasoning task was not counterbalanced over participants. The BFNE was always administered after completion of the reasoning task. This was done to avoid potential priming effects of the BFNE on the reasoning task (cf. Bosson et al.
2000), however this procedure may have enhanced existing individual differences in BFNE scores.
A final remark concerns the generalisation of the current findings. It remains to be seen whether similar findings will be obtained in a more male/female balanced group, as well as in less highly educated groups. In addition, the present study relied on an analogue sample, and it remains therefore to be seen whether similar findings will be obtained in treatment seeking individuals suffering from a clinically diagnosed social anxiety disorder.