When Beliefs Face Reality: An Integrative Review of Belief Updating in Mental Health and Illness

How beliefs modulate perception and well-being

It is becoming increasingly evident that people’s beliefs ¹ significantly modulate their perception of—and response to—the world, affecting various aspects of life from decision-making to well-being. Evidence for this claim has been provided, for instance, by research into “motivated perception” (Lench, Smallman, Darbor, & Bench, 2014; Voss & Schwieren, 2015). For example, in a well-designed series of five experimental studies that included both explicit and implicit measures of perception, Balcetis and Dunning (2006) found that people’s perception of ambiguous visual stimuli was guided by top-down beliefs, meaning that they had the propensity to perceive a representation of the visual environment that was consistent with their desires. Further substantiating these findings, Leong, Hughes, Wang, and Zaki (2019) investigated the neurocomputational mechanisms of such motivated perception. Using a rewarded visual-categorization task, the authors found that top-down beliefs about expected reward biased perceptual judgments in that ambiguous images were more often labeled as corresponding to a reward-associated category. Beyond perception, performance has also been shown to be affected by beliefs, as demonstrated by research indicating enhanced cognitive abilities such as memory and creativity after sham, placebo-like treatment (Parker, Garry, Einstein, & McDaniel, 2011; Rozenkrantz et al., 2017; Weger & Loughnan, 2013).

Empirical support for the role of beliefs in well-being comes from several studies linking large-scale national health surveys with databases of well-being measures. These studies suggest that holding negative beliefs about aging in young adulthood is associated with up to a threefold higher likelihood of experiencing cardiovascular events 38 years later (Levy, Zonderman, Slade, & Ferrucci, 2009). Further, individuals reporting high levels of stress had a 43% increased risk of premature death, but only if they perceived stress as having a significant impact on health (Keller et al., 2012). In addition, individuals who perceived themselves as less physically active than others had up to a 71% higher mortality risk compared with individuals who believed they were more active than others, even after controlling for actual levels of activity (Zahrt & Crum, 2017). Likewise, Laferton, Fischer, Ebert, Stenzel, & Zimmermann (2020) found that individuals who hold negative beliefs about stress (e.g., believing that stress is uncontrollable) report more negative affect in relation to daily stress.

In addition to this mostly cross-sectional or longitudinal research, Crum, Salovey, and Achor (2013) provided more causal evidence for the influence of beliefs on the stress response. In a series of experimental studies, the authors manipulated the way participants thought about stress. Specifically, they established a “stress is debilitating” mind-set (suggesting detrimental effects of stress on health, performance, and learning/growth) in contrast to a “stress is enhancing” mind-set (suggesting beneficial effects of stress) and found that the latter was related to more adaptive cortisol profiles under acute stress. This implies that beliefs about stress significantly contribute to how people actually respond to acute stress (Crum et al., 2013). Building on this finding, another study demonstrated that a brief psychological intervention aimed at optimizing participants’ beliefs about stress (i.e., enhancing participants’ personal-control expectations) led to lower cortisol levels after acute stress than a gratitude intervention; however, the specificity of this effect was somewhat limited, given that a distraction task also lowered participants’ cortisol levels (Salzmann et al., 2018).

Other interesting insights into the link between beliefs and health have been provided by placebo research, in which expectations for symptom improvement have been found to contribute significantly to clinical outcomes after an inert treatment (i.e., placebo; for reviews, see, e.g., Petrie and Rief, 2019; Schedlowski, Enck, Rief, and Bingel, 2015). Pointing to a causal effect of participants’ expectations, some studies have found that manipulating participants’ treatment expectations experimentally (e.g., expecting to receive an analgesic drug vs. placebo) substantially influenced pain perception (Lund, Vase, Petersen, Jensen, & Finnerup, 2014; Schenk, Sprenger, Geuter, & Büchel, 2014). Intriguingly, another study demonstrated that positive treatment expectations boosted the analgesic effects of an opioid agonist, whereas negative treatment expectations diminished its analgesic effects (Bingel et al., 2011). Furthermore, neuroimaging studies have found that expectations modulate treatment-relevant neuronal processes even before treatment, and these modulations mediate treatment outcomes (Wager et al., 2004; Zilcha-Mano et al., 2019). Expanding our knowledge about the effects of expectations on pain perception, two neuroimaging studies revealed how the expectation of pain relief influenced the perception of pain at the neurobiological level: It was found that higher order brain regions acted on lower order brain regions to suppress the coding of the dissonance between expectation and perception (Jepma, Koban, van Doorn, Jones, & Wager, 2018; Schenk, Sprenger, Onat, Colloca, & Büchel, 2017). In other words, the brain matches its perception to its expectations, much like a self-fulfilling prophecy. Collectively, these findings highlight how human information processing, from sensory perception to decision-making, is significantly shaped by people’s beliefs. However, it should be noted that some of the research cited above used cross-sectional or longitudinal data, limiting the conclusions that can be drawn with respect to causality.

When beliefs make people vulnerable

The importance of beliefs is further emphasized by studies using clinical samples, indicating vulnerabilities in relation to dysfunctional ² beliefs. For instance, research from behavioral medicine has shown that dysfunctional beliefs of patients about their illness can have adverse effects on symptom development and treatment success. In patients with coronary heart disease, for example, negative beliefs about the controllability and consequences of the disease predict worse symptom recovery from myocardial infarction (Petrie, Weinman, Sharpe, & Buckley, 1996) and reduced benefit from cardiac surgery (Juergens, Seekatz, Moosdorf, Petrie, & Rief, 2010; Kube, Meyer, et al., 2020). Even long-term mortality has been found to be increased for patients with dysfunctional illness perceptions (Barefoot et al., 2011). Similar results have been provided for patients with chronic obstructive pulmonary disease (Zoeckler, Kenn, Kuehl, Stenzel, & Rief, 2014), breast cancer (Nestoriuc et al., 2016), diabetes (Griva, Myers, & Newman, 2000; Mann, Ponieman, Leventhal, & Halm, 2009), and chronic pain (Cormier, Lavigne, Choinière, & Rainville, 2016). Beyond this mostly longitudinal research, other studies have used experimental designs to examine the influence of patients’ beliefs on the course of symptoms in a more causal manner. For example, two studies found that interventions aimed at modifying illness beliefs of people who had experienced a myocardial infarction were related to a faster rate of return to work than standard medical care (Broadbent, Ellis, Thomas, Gamble, & Petrie, 2009; Petrie, Cameron, Ellis, Buick, & Weinman, 2002). In line with these results are results from another study that showed that the optimization of preoperative expectations in patients undergoing coronary heart surgery led to less illness-related disability 6 months after the surgery than standard medical care. Because this expectation-optimization intervention was also superior to an unspecific supportive psychological intervention in terms of reducing disability, this study suggests a specific effect of patients’ expectations on symptom improvement (Rief et al., 2017).

Dysfunctional beliefs are also regarded as core features of almost all mental disorders. In major depression, for instance, it has been shown that dysfunctional expectations about one’s own abilities and the personal future predict the course of depressive symptoms and suicidal ideation (Backenstrass et al., 2006; Czyz, Horwitz, & King, 2016; Horwitz, Berona, Czyz, Yeguez, & King, 2017; Strunk, Lopez, & DeRubeis, 2006). More specifically, it has been found that people with depression hold negative global beliefs that elicit dysfunctional situation-specific expectations that then lead to the experience of depressive symptoms (Kube et al., 2018). In posttraumatic stress disorder (PTSD), dysfunctional beliefs in relation to the traumatic experience are assumed to be a core factor that contributes to the development and maintenance of the disorder (Ehlers & Clark, 2000). For instance, Murray (2018) showed that beliefs about being guilty after the survival of a traumatic event is very common among PTSD patients and is associated with higher PTSD symptom severity. Further evidence for the contribution of dysfunctional beliefs to the development and maintenance of mental health problems has been provided for social anxiety (D. M. Clark & Wells, 1995; Rapee & Heimberg, 1997), hypochondriasis (Barsky et al., 2001; Barsky, Coeytaux, Sarnie, & Cleary, 1993), obsessive-compulsive disorders (Hansmeier, Exner, Rief, & Glombiewski, 2016; Salkovskis et al., 2000; Shafran, Thordarson, & Rachman, 1996), eating disorders (Corstorphine, 2006; Konstantakopoulos et al., 2012; Waller, Dickson, & Ohanian, 2002), and schizophrenia (Lincoln, Mehl, et al., 2010; Moritz et al., 2010).

The evidence summarized in this brief introduction suggests that beliefs can influence the way people perceive the world and respond to it; further, in the case of dysfunctional beliefs, beliefs can make people vulnerable and can be associated with mental health problems. However, because of the limited number of experimental studies differentially manipulating beliefs, conclusions about a causal role of beliefs (e.g., with respect to the question of whether dysfunctional beliefs constitute a risk factor for mental disorders) need to be drawn with caution.

A prominent framework in the study of belief updating: predictive processing

The roots of the predictive-processing framework (PPF) go back to von Helmholtz (1867), who introduced the concept of unconscious inference, assuming that the human brain fills in visual information to make sense of a visual scene. In recent years, predictive processing has become a dominating theoretical framework in neuroscience and related areas interested in operating principles of the brain. One basic assumption of the PPF is that perception is inextricably linked to prediction, meaning that prior predictions bias perception, which in turn shapes posterior predictions. In particular, researchers using the PPF assume that the brain does not “passively” process incoming sensory signals; instead, it permanently generates predictions about expected sensory input that is then compared with prior predictions (A. Clark, 2013b; Huang & Rao, 2011; O’Reilly, Jbabdi, & Behrens, 2012). Any mismatches between prior predictions and actual sensory signals are referred to as prediction errors. Prediction errors normally provide corrective feedback and are used to update future predictions (referred to as belief updating). Appealing to the free-energy principle (Friston & Kiebel, 2009; Moran et al., 2013), neuroscientists suggest that the brain aims to adjust predictions such that prediction errors are minimized and incoming sensory information is optimally used (Barrett & Simmons, 2015). With active inference, people can also fulfill their predictions by acting on the world in such a way that prediction errors are minimized (Friston et al., 2015; Pezzulo, Rigoli, & Friston, 2015; Seth & Friston, 2016).

Precision plays a crucial role in active inference. As with all forms of Bayesian inference, the precision of various sources of sensory evidence—or indeed prior beliefs—can have a profound effect on belief updating (Ernst & Banks, 2002). These effects will become prescient later when we consider aberrant precision and the false inferences that accompany psychopathology. Simple examples here include inferring that something is there when it is not (i.e., hallucinations) or that something is not there when it is (i.e., agnosia). Technically, the key determinant of the rate of belief updating or evidence accumulation is the precision afforded to prior beliefs relative to sensory evidence (i.e., sensory prediction errors). In other words, a weakening of prior precision corresponds to a relative increase (or failure to attenuate) sensory precision and vice versa. To keep things simple, we describe this imbalance of precision in terms of prior precision (Pellicano & Burr, 2012). Psychologically, getting the right balance between prior and sensory precision has been associated with attention, such that attending to a particular source of sensory evidence enhances the corresponding precision of prediction errors and the accompanying rate of belief updating (Feldman & Friston, 2010; Kanai, Komura, Shipp, & Friston, 2015; Parr & Friston, 2019; Vossel, Mathys, Stephan, & Friston, 2015). From a neurobiological perspective, this is thought to be encoded by the excitability or postsynaptic gain of neuronal populations—superficial pyramidal cells—encoding prediction error. This aspect becomes important later when we discuss the pathophysiology of conditions such as schizophrenia and Parkinson’s disease, which usually involve an abnormality of neuromodulatory transmitter systems (e.g., dopamine). In short, a pathophysiology of neuromodulation may translate into a failure of belief updating as a result of aberrant precision control (A. Clark, 2013a; Haarsma et al., 2018; C. E. Palmer, Auksztulewicz, Ondobaka, & Kilner, 2019; C. J. Palmer, Seth, & Hohwy, 2015; Rae, Critchley, & Seth, 2019; Van de Cruys et al., 2014; Vuust, Dietz, Witek, & Kringelbach, 2018).

The PPF was originally developed in neuroscience and provided a new understanding of sensory processing and, more recently, infant learning (Köster, Kayhan, Langeloh, & Hoehl, 2020). In recent years, several clinical applications of the PPF have been proposed, for example, for mental disorders such as depression (Barrett, Quigley, & Hamilton, 2016; J. E. Clark, Watson, & Friston, 2018; Kube, Schwarting, Rozenkrantz, Glombiewski, & Rief, 2020), PTSD (Kube, Berg, Kleim, & Herzog, 2020; Linson & Friston, 2019; Wilkinson, Dodgson, & Meares, 2017), and psychosis (Corlett et al., 2019; Sterzer et al., 2018); persistent physical symptoms (Edwards, Adams, Brown, Parees, & Friston, 2012; Henningsen et al., 2018; Kube, Rozenkrantz, Rief, & Barsky, 2020; Van den Bergh, Witthöft, Petersen, & Brown, 2017); pain perception (Wiech, 2016); and placebo analgesia (Büchel, Geuter, Sprenger, & Eippert, 2014; Grahl, Onat, & Büchel, 2018; Ongaro & Kaptchuk, 2019). We refer to some of these clinical applications of the PPF later in this article when discussing certain distortions related to belief updating in clinical populations.

The optimism bias

One particularly striking example of how healthy people integrate new information into their existing beliefs is the optimism bias that Tali Sharot and colleagues have studied extensively. According to this concept, (healthy) people’s beliefs about the future are often unrealistically optimistic and remain so despite disconfirmatory evidence. Remarkably, to maintain optimistic beliefs about the future, people selectively embrace information that favors their optimistic view while discarding information that contradicts it. In other words, people display asymmetric learning from new information according to whether it confirms their existing (biased) beliefs (Lefebvre, Lebreton, Meyniel, Bourgeois-Gironde, & Palminteri, 2017; Sharot, 2011; Sharot & Garrett, 2016). This has been demonstrated in a series of experimental studies (Garrett & Sharot, 2017; Sharot, Guitart-Masip, Korn, Chowdhury, & Dolan, 2012; Sharot, Kanai, et al., 2012; Sharot, Korn, & Dolan, 2011; Sharot, Riccardi, Raio, & Phelps, 2007). In these studies, a trial-by-trial belief-updating design was used: That is, participants were asked to estimate their risk for several adverse life events and then were presented with the real probabilities of encountering these events (i.e., the correct estimation). When then asked to reestimate their risk, participants preferably used this new knowledge when favorable, but not unfavorable, probabilities were presented. For example, an initial estimation of a 30% risk for cancer might be updated to 22% after a base rate of 20% (favorable information; large update), but an initial estimation of a 10% risk might be updated to 13% after the same base rate (unfavorable information; smaller update; Sharot & Garrett, 2016). Notably, this “good news–bad news” effect has also been found in research from behavioral economics, indicating that people do not make decisions solely on the basis of rationality (Eil & Rao, 2011).

A series of neuroimaging studies (Kuzmanovic, Jefferson, & Vogeley, 2016; Sharot et al., 2011; Sharot, Kanai, et al., 2012) elucidated the neural mechanisms of the optimistic bias and showed that undesirable information is coded differently than desirable information, which is thought to underlie the relative neglect of the integration of the former. Thus, overly optimistic beliefs are further reinforced by a selective filter that generates a feedback loop that facilitates the persistence of biased beliefs. A further study found that this optimistic bias in belief updating diminishes when people are under perceived threat: Both in participants with experimentally induced stress and in firefighters on duty, Garrett, González-Garzón, Foulkes, Levita, and Sharot (2018) found that under perceived threat, healthy people are more sensitive to bad news and better integrate it into their beliefs than in the absence of threat. The authors interpreted their findings in light of psychological flexibility, meaning that healthy people can flexibly adjust their otherwise optimistic belief-updating bias when situational circumstances require it.

Cognitive consistency and confirmation bias

According to theories from cognitive and social psychology, people strive for cognitive consistency, meaning that they aim to reduce contradictions between different beliefs. One of the most influential theories in this context is the theory of cognitive dissonance (Festinger, 1962). This theory proposes that holding conflicting cognitions is perceived as being aversive, resulting in the preference to reduce this cognitive dissonance by cognitive or behavioral strategies. To apply this theory to the terminology of belief updating, consider the example of a healthy person holding the belief that, overall, the world is safe. If this person were confronted with disconfirming information, such as reported crimes in the news, a conflicting cognition such as “Some people endanger safety” might come up. Because this cognition contradicts the belief in a safe world, the person may aim to reduce cognitive dissonance, for example, by thinking, “These things only happen to others. My personal world is safe.”

Relatedly, other lines of research from cognitive psychology have also dealt with the phenomenon of confirmation bias, which means that people have the propensity to interpret information in such a way that their preexisting beliefs are confirmed (Nickerson, 1998; Oswald & Grosjean, 2004). According to this concept, people generally assume that their beliefs are true, which is why they use their preexisting beliefs as a heuristic to evaluate new information. Further, people tend to use positive test strategies, meaning that they prefer strategies that are likely to confirm prior beliefs when integrating novel information (Klayman & Ha, 1987).

Note that we have moved beyond perceptual processing and now consider the problem of active inference (i.e., how to optimize belief updating when people are in charge of gathering the sensory evidence on which to base their beliefs). Active inference takes the view that the best way to gather data is to resolve uncertainty in relation to prior beliefs. Implicit in this formulation is a Bayes-optimal optimism bias. In other words, if people are gathering evidence for their models of the world, they will naturally tend to select those data that provide confirmatory evidence (Bruineberg, Kiverstein, & Rietveld, 2018). Crucially, active inference does not just deal with selecting sources of sensory information (e.g., which news channels to watch). There can also be mental actions in terms of deploying precision or attention to various modalities or hierarchical levels of inference (Limanowski & Friston, 2018). In other words, positive test strategies can also be manifested in the things that people attend to—and the things that they ignore—without any overt action.

Self-concept stability

Another bias in belief updating among healthy people has been revealed in social and personality psychology. It is well established in this field that people’s beliefs about themselves remain remarkably stable over time. One aspect contributing to this self-concept stability is that people selectively search for information that confirms their self-concept. In other words, people are inclined to search for information that is consistent with their beliefs about themselves. This inclination has been demonstrated by studying participants’ desire to receive feedback from social-interaction partners about their appraisal of the participants; it was found that people had a strong preference to receive feedback that they believed would be consistent with their self-concept (e.g., regarding self-perceived assertiveness and self-perceived emotionality; Swann & Read, 1981a, 1981b). Conversely, when people receive new information that is inconsistent with their self-concept, they are likely to reject it, which further contributes to self-concept stability (Markus, 1977; Markus & Wurf, 1987; Swann & Hill, 1982). With regard to the latter, one study, for instance, has shown that participants’ expectations for their performance were highly influenced by their preexisting level of self-esteem but not by novel performance feedback received during the task (McFarlin & Blascovich, 1981).

In another study, Korn, Prehn, Park, Walter, and Heekeren (2012) demonstrated that the valence of self-relevant information also influences belief updating. The authors showed that people update their beliefs about themselves after receiving social feedback from peers more toward desirable information than toward undesirable information, suggesting that healthy people are positively biased when integrating social feedback into their beliefs about themselves. Thus, collectively, research on self-concept suggests that belief updating in healthy people is biased toward the integration of (positive) self-concept-congruent information, whereas self-concept-discrepant information is disregarded, thereby maintaining the person’s self-perception.

Truth decay and posttruth: persistence of political attitudes

Insightful examples of how people selectively use novel information to update their beliefs can also be found in the literature on political attitudes. Research in this area suggests that people stick to their attitudes and opinions regardless of disproving facts (Taber, Cann, & Kucsova, 2009; Tappin, van der Leer, & McKay, 2017). This has been referred to as a truth decay, meaning that factual evidence is insufficiently taken into account in political opinion forming and decision-making (e.g., in elections; Kavanagh & Rich, 2018). Authors who assume that this truth decay has increased in recent years even speak of a “posttruth” era (Jasanoff & Simmet, 2017; Lewandowsky, Ecker, & Cook, 2017; Sismondo, 2017) or “postfactual” politics (Sayer, 2017) when characterizing the current (Western) political discourse. Prominent examples from current political discussions for the persistence of political beliefs despite disconfirmatory evidence are climate-change denial (G. T. Farmer & Cook, 2013; Häkkinen & Akrami, 2014; McCright & Dunlap, 2011), the “vaccination confidence gap” (Browne, Thomson, Rockloff, & Pennycook, 2015; Larson, Cooper, Eskola, Katz, & Ratzan, 2011; Tafuri et al., 2014), and the discussion about gun control in the United States (Rogowski & Tucker, 2018).

Research has revealed that the extent to which political beliefs are updated after receiving new information in relation to prior beliefs is influenced by party identification. For example, Swire, Berinsky, Lewandowsky, and Ecker (2017) conducted an experiment in November 2015 in which they used political statements Donald Trump made during his campaign, some of which were objectively false and some factual. When presented to participants, half of the statements were explicitly attributed to Trump, whereas the other statements were presented without attribution. Inaccurate statements were subsequently corrected, and factual statements affirmed, with a brief explanation. The authors found that when initially evaluating the veracity of both false and factual statements, Republican supporters of Trump believed the information more when it was attributed to Trump, whereas the opposite was found for Democrats and Republicans who did not support Trump (albeit to a lesser extent than Democrats). Immediately after being presented with the correction/affirmation of the statements, all participants updated their beliefs; strikingly, however, Trump supporters “rebelieved” in the misinformation a week later and partially forgot that factual information was true. The authors concluded from their findings that even if individuals update their political beliefs temporarily, they become oblivious to disconfirmatory evidence over time and use partisanship and political figures as heuristics to guide their evaluation of what is true or false. These findings are in line with other research that has provided evidence for biased belief updating in the context of political information (Ecker, Lewandowsky, Fenton, & Martin, 2014; Swire, Ecker, & Lewandowsky, 2017).

Now that we have reviewed the basic characteristics of belief updating in healthy people, we next turn to belief updating in clinical populations.

Lack of updating disorder-specific beliefs

The other side: hasty changes in beliefs

Aberrant precision given to prior beliefs

With active inference, it is assumed that the degree of precision that is afforded to prior beliefs significantly influences the extent to which new information (i.e., prediction errors) is used to update beliefs. According to this view, research from computational psychiatry suggests that too much precision given to prior beliefs, accompanied by an attenuation of disconfirming information, might be a mechanism underlying the persistence of dysfunctional beliefs despite disconfirmatory evidence in various mental disorders (Adams, Huys, & Roiser, 2016; Friston, Stephan, Montague, & Dolan, 2014; Paulus, Feinstein, & Khalsa, 2019). Evidence for this strong-prior hypothesis has first been provided for psychosis: For example, Powers, Mathys, and Corlett (2017) demonstrated in an intricate approach that people who hear voices were more susceptible to conditioning-induced hallucinations; using neurocomputational modeling, they showed that this effect was based on strong perceptual priors. This finding was replicated by Benrimoh, Parr, Vincent, Adams, and Friston (2018) for auditory hallucinations. On the basis of these findings, the assumption of strong priors that override disconfirming sensory information has been discussed as a core feature of psychosis (Corlett et al., 2019; Sterzer et al., 2018). Similar to this aberrant-precision account of psychosis, it has been proposed that too much precision afforded to negative prior beliefs accounts for the lack of updating these beliefs despite positive experiences (i.e., prediction errors) in major depression (Barrett et al., 2016; J. E. Clark et al., 2018; Kube, Schwarting, et al., 2020).

Whereas too much precision afforded to priors is thought to contribute to persistent beliefs, we suggest that too little precision given to priors might contribute to the above-described phenomena of hasty changes in beliefs. This is consistent with current thinking in computational neuroscience, assuming that a low degree of precision in priors (i.e., weak priors) leads to a higher relative impact of sensory information (i.e., prediction errors) on the posterior predictions (Barrett & Simmons, 2015; Kanai et al., 2015; Paulus et al., 2019). Some empirical support for this idea can be found in research on autism, as described above. Additional support for the weak-prior hypothesis has been provided by research using dynamic causal modeling (e.g., in schizophrenia; Adams, Bauer, Pinotsis, & Friston, 2016; Bastos-Leite et al., 2014; Fogelson, Litvak, Peled, Fernandez-del-Olmo, & Friston, 2014).

In more general psychological terms, the assumption of weak prior beliefs that are highly susceptible to new information might be regarded as a form of belief fragility. That is, the less firmly people hold their beliefs, the more likely they use new information to update them. This corresponds well to some of the difficulties experienced by people with BPD, as discussed above. For instance, under the assumption that their beliefs about their relationships are fraught with uncertainty, it is understandable that people with BPD are highly sensitive to any new information that they feel is relevant to the stability of their relationship, both negative and positive (such as seeing the boyfriend with another woman vs. receiving a compliment from the boyfriend).

Although aberrant precision has so far been linked to anomalies in belief updating mostly in clinical populations, it is well conceivable that this link may also apply to healthy people under some circumstances. For example, applying the argument of aberrant precision to political attitudes, we suggest that too much precision given to prior beliefs might be an important mechanism underlying the maintenance of particular political beliefs despite disconfirmatory political news. In other words, if people hold political attitudes with a high degree of certainty, they tend to be skeptical about any information that contradicts their beliefs; hence, such information is likely to be ignored or given reduced weight.

Interpretation biases

Another mechanism that may underlie the persistence of dysfunctional beliefs despite disconfirmatory evidence is a biased interpretation of discrepant information. In particular, researchers have found that people with mental disorders have the tendency to interpret novel information through the lenses of their prior beliefs, thereby increasing the likelihood of confirming them. For example, as discussed above, people with major depression tend to interpret ambiguous situations negatively, especially if they contain self-referential information (Everaert, Podina, & Koster, 2017), and once negative interpretations have been established, they fail to use novel positive information to revise their interpretations (Everaert et al., 2018; Liknaitzky et al., 2017). Relatedly, research on affective forecasting has found that people with depression are biased in their prediction of their future affective states, meaning that they overestimate the anticipated presence of negative affect (Hoerger, Quirk, Chapman, & Duberstein, 2012; Marroquín & Nolen-Hoeksema, 2015; Radomsky, Wong, Dussault, Gilchrist, & Tesolin, 2019; Wenze, Gunthert, & German, 2012; Zetsche, Bürkner, & Renneberg, 2019). Similar interpretation biases have been found in patients with social anxiety: People with social anxiety favor negative interpretations of ambiguous social situations and tend to interpret unambiguous but mildly negative social events as catastrophes (Amin, Foa, & Coles, 1998; D. M. Clark & McManus, 2002; Stopa & Clark, 2000).

As with aberrant precision, interpretation biases may not be specific to clinical populations alone. In fact, there is a large amount of research in social and cognitive psychology on biased assimilation, which is conceptually very similar to the clinical research on interpretation biases. Specifically, according to the concept of biased assimilation, people’s interpretations of new information are assimilated into preexisting beliefs, thus sustaining and further “confirming” their beliefs (Lord & Taylor, 2009). For further discussion of persistent political beliefs in social and political sciences, readers may also be referred to the literature on attitude polarization (Boysen & Vogel, 2008; McHoskey, 1995; Munro & Ditto, 1997), resistance to persuasion (Tormala & Petty, 2004; Zuwerink & Devine, 1996), partisan bias (Bartels, 2002; Bullock, Gerber, Hill, & Huber, 2015), motivated reasoning (Redlawsk, 2002; Taber et al., 2009), and metacognitive sensitivity (Rollwage, Dolan, & Fleming, 2018).

Reappraisal

Cognitive reappraisal of disconfirmatory evidence is an additional factor that has been discussed as a contributor to the maintenance of dysfunctional beliefs (Rief & Glombiewski, 2016; Rief et al., 2015; Rief & Joormann, 2019). In particular, it has been suggested that people with mental disorders are prone to devaluing positive information that disconfirms disorder-specific negative beliefs by post hoc questioning of its credibility or considering it to be an exception rather than the rule. This negative reappraisal of disconfirming information resulting in a lack of belief updating has been referred to as “cognitive immunization” against disconfirming information. Experimental research has recently confirmed this hypothesis by indicating that modulating the appraisal of unexpectedly positive performance feedback affected how prior negative performance-related expectations were updated (Kube, Glombiewski, et al., 2019; Kube, Rief, et al., 2019). In these studies, people with depression worked on a performance test in which it is difficult for participants to evaluate whether they solved the tasks correctly. After the initial establishment of negative expectations, all participants received unexpectedly positive feedback for their performance. In people with depression, it was found that promoting the engagement in a post hoc devaluation of positive performance feedback (by informing participants that the test they were working on would not have proven to be valid and reliable) led to reduced expectation updating; in contrast, the inhibition of cognitive-immunization strategies (i.e., increasing the value of positive performance feedback) facilitated the updating of negative expectations in line with positive feedback (Kube, Glombiewski, et al., 2019; Kube, Rief, et al., 2019).

Another study indicated that the promotion of cognitive-immunization strategies in healthy people did not affect their adjustment of initial beliefs; that is, healthy people updated their expectations regardless of a reappraisal manipulation in line with positive performance feedback (Kube & Glombiewski, 2020). These findings suggest that the negative reappraisal of novel positive information is core to depression but is not typical of healthy people. However, it is conceivable that healthy people conversely use cognitive-immunization strategies to maintain their optimistic beliefs when confronted with bad news (Kube, Schwarting, et al., 2020; Sharot & Garrett, 2016).

Relating belief-updating anomalies to different stages of information processing

In the previous section, we compiled evidence of some factors that may contribute to certain anomalies in belief updating. Next, we aim to integrate these factors into the model of belief updating as a process of different stages of information processing. In doing so, we are facing an asymmetry in the depth with which the mechanisms of persistent beliefs versus hasty changes have been investigated so far: Although we were able to invoke a considerable amount of evidence of particular mechanisms underlying the lack of belief updating, little research to date has dealt with factors contributing to hasty changes in beliefs; accordingly, the latter is much less understood. Therefore, we consider these two phenomena separately.

With respect to persisting beliefs despite counterevidence, we identified three potential mechanisms that can be linked to different stages of information processing. First, the idea of hyperprecise prior beliefs, resulting in an increased weight given to them in relation to new information, can be linked to the stage of anticipation of new information. In other words, holding a belief with certainty implies that people anticipate to receive confirmatory rather than disconfirmatory information, or even actively search for information that fits into their beliefs, to put it in active-inference terms. Second, at the level of encoding/interpretation, research into interpretation biases and biased assimilation indicates that beliefs can be resistant to updating if people have a strong bias to reinterpret new information as a confirmation of their prior beliefs. Third, in terms of further processing and appraising new information, people may be inclined to uphold their beliefs if the validity of new information is questioned after the fact, as research on cognitive reappraisal suggests, leading to a disregard of disconfirmatory information so that beliefs are not altered. Figure 3 illustrates how factors underlying the lack of belief updating relate to different stages of information processing (i.e., anticipation, encoding, appraisal).

OPEN IN VIEWER

Fig. 3.

Factors contributing to the lack of belief updating at different stages of information processing. The Bayesian homologues of the belief-updating terminology are shown in italics.

From the discussion of hasty changes in beliefs, it emerged that little is known about particular mechanisms underlying this phenomenon. In terms of the different stages of information processing, past research allows only one cautious suggestion with reference to the PPF: Drawing on this framework, we assume that affording too little precision to prior beliefs leads to the posterior predictions being influenced by new information, whatever it may be, rather than by priors. However, given that little to date is known about mechanisms of hasty changes of beliefs, we refrain from including contributors to hasty changes of beliefs in a mechanistic or conceptual model.

An integrative model

As illustrated in Figure 4, we suggest that the degree of precision afforded to prior beliefs is a lens through which new information is perceived, interpreted, and appraised. In the case of strong priors (i.e., prior beliefs that are afforded high precision), belief updating is biased toward resistance to disconfirmatory information, as new information may be considered less informative or valid. By contrast, low precision afforded to prior predictions may bias perception toward new information such that posterior beliefs are strongly influenced by new information, formalized by the increased weight given to this new information (i.e., prediction error) and pronounced belief updating. Thus, the level of precision afforded to prior beliefs affects the perception of information and its appraisal, resulting in belief updating being tied to the priors (in the case of hyperprecise priors) or being influenced strongly by incoming input (in the case of weak priors).

OPEN IN VIEWER

Fig. 4.

Schematic illustration of how affording high versus low precision to prior beliefs and high versus low precision of new information influence belief updating.

As described previously, the valence of new information (i.e., good news vs. bad news) can critically influence belief updating. Thus, it may also affect the balance of precision given to priors and new information. Specifically, we noted that the computation of posterior beliefs is generally influenced by the relative weight of priors and new information. Valence comes into play here as follows. The precision afforded to new information, on the one hand, may depend on the source of incoming information. For example, one may trust one’s own observations more than others’ descriptions of them, thus resulting in more precision afforded to the former. On the other hand, the precision of new information may be determined according to its content, such as the valence of new information. Differentially weighing new information on the basis of its valence may then lead to some of the phenomena related to belief updating, as discussed above. For example, the optimism bias may be explained by assuming that healthy people, in general, afford high precision to positive information and low precision to negative information. This allows them to integrate new information asymmetrically into their beliefs (i.e., updating beliefs selectively in response to good news). In contrast, people with depression (as well as some other mental disorders) may show an opposite pattern: They afford little precision to unexpectedly positive information and thus only minimally update their initial beliefs.

Aside from valence, precision afforded to incoming information may also be evaluated on the basis of the confirmatory nature of its content (i.e., information supporting my initial beliefs will be weighed stronger than disconfirming information), its relevance, and other factors. As illustrated in Figure 4, we propose that this attribution of precision to new information occurs in a phase we call weighing: That is, people must first assess the information presented to them and categorize it in terms of good/bad, confirming/disconfirming, relevant/irrelevant, and so on, before assigning it a certain weight. This weight of new information is then matched with the precision-weighted priors to compute a posterior belief.

This integrative model is the first to our knowledge to connect the PPF with the belief-updating framework and to synthesize various, formerly unrelated lines of research into one coherent model. Specifically, it bridges neuroscientific research into the role of precision and psychological investigations of several biases in belief updating. Furthermore, this model provides a foundation from which to further investigate differences in belief updating between healthy people and people with certain mental health conditions. In particular, we propose that our model is suitable for explaining the aforementioned biases of belief updating in healthy people by assuming that, for instance, relatively high precision in optimistic future beliefs results in biased perception and appraisal of new information, hence leading to reduced belief updating in response to bad news. On the other hand, the model can explain the formation of pathologies in belief updating in that it suggests that abnormally high (or low) precision given to prior beliefs can result in the persistent lack of belief updating (or hasty updating of beliefs, respectively). Tying in with these ideas, we next discuss the implications of this account for the understanding of mental health and disorders in more detail.

Implications for the understanding of mental health as opposed to disorders

Integrating the phenomena discussed above, we suggest that belief updating in healthy people is quite volatile, meaning that it is biased toward the integration of information that fits into people’s preexisting assumptions in various domains. In particular, according to the literature on optimism bias and self-concept stability, we propose that belief updating in healthy people is driven by the valence of new information as well as by the desire for a stable, positive self-perception. In addition, the literature on attitude change has revealed that belief updating in healthy people is biased toward the integration of information that is consistent with core beliefs to make sense of the world, such as core political attitudes. In other words, if (healthy) people’s beliefs reflect an optimistic view of their personal future, their view of themselves, or basic assumptions about the world, beliefs are likely to be resistant to disconfirmatory information and updating. By contrast, in other domains, belief updating may be more sensitive to new information, especially if the adjustment of a particular belief is considered socially desirable, as suggested by research on social conformity (Asch, 1956; Bond & Smith, 1996; Larsen, 1974). This means that people are inclined to update their beliefs in line with new information if they consider it socially desirable. Collectively, research suggests that mental health is associated with the preferential integration of positive, self-concept-stabilizing information into one’s beliefs and the dismissal of information that is inconsistent with an optimistic view of the future, the self-concept, or core beliefs to make sense of the world.

Some of the biases in updating beliefs that are typical of healthy people (i.e., the optimistic bias) have been shown to be absent in people with clinical disorders, particularly depression and autism (Garrett et al., 2014; Korn et al., 2014; Kuzmanovic et al., 2019). More generally, available evidence suggests that as much as belief updating in healthy people is biased toward the confirmation of core beliefs of them (such as optimistic beliefs about the future and positive beliefs about the self), belief updating in mental disorders is biased toward the integration of information that is consistent with core beliefs of the respective disorder. That is, a belief seems to be particularly immune to updating if it is intertwined with core beliefs of the disorder, such as viewing oneself as being incapable in depression or the overestimation of threat in anxiety disorders. In other words, just as the nature of beliefs differs between people with certain mental disorders and healthy people, so does the nature of the information that is preferentially integrated (vs. disregarded). In both healthy people and people with clinical disorders, the (relative) immunity of core beliefs to disconfirmatory evidence might be accounted for by affording increased precision to prior beliefs or decreased precision to the incoming information (as illustrated in Fig. 4), although the content of the respective beliefs may differ considerably.

In other domains, healthy people and people with mental disorders may differ not only in terms of the contents of their beliefs but also mechanistically in the way new information is integrated. In particular, in cases in which prior beliefs are afforded overly low precision, they become highly susceptible to updating, as indicated, for example, for increased (in fact, more rational) belief updating in response to bad news in people with autism. In general, the phenomenon of hasty changes in beliefs as a consequence of weak priors as discussed for various clinical populations has, to our knowledge, not yet been reported for healthy people. Conversely, whereas in some cases increased precision of prior beliefs may also apply to healthy people, in some clinical conditions prior beliefs are so overly precise that they cause severe perceptual distortions, such as hallucinations in psychosis. Specifically, whereas perception in people who are in good mental health is normally influenced by both prior predictions and sensory evidence, it can be almost exclusively dominated by priors in psychotic hallucinations, resulting in a complete neglect of disconfirming sensory information. Similar accounts have been proposed for the experience of flashbacks in people with PTSD (Linson & Friston, 2019; Wilkinson et al., 2017) and medically unexplained physical symptoms (Edwards et al., 2012; Henningsen et al., 2018; Kube, Rozenkrantz, et al., 2020; Van den Bergh et al., 2017).

There is a compelling convergence of the mechanisms that underwrite attentional filtering (i.e., precision control) in terms of neuromodulation and psychopathology. A key example here is the ubiquitous role of the neuromodulator dopamine in various neurological and psychiatric conditions. For example, the primary deficit in Parkinson’s disease, leading to psychomotor poverty, is a dopaminergic deficit. Conversely, the most enduring synaptic hypothesis of schizophrenia suggests overactive dopamine neurotransmission (that is targeted by antipsychotics). A similar story might be told for oxytocin in autism. It has been noted previously that one of the most challenging computational issues in inference is estimating the precision or uncertainty afforded to (sensory) data. If this assumed role of precision is true, then psychopathologies that arise from aberrant precision control are quite understandable—as the most fragile computational challenge faced by the brain (A. Clark, 2013a; Hohwy, 2013).

Collectively, research into belief updating as reviewed in this article provides a new understanding of cognitive processes in mental health versus mental illness insofar as it suggests that healthy people and people with mental disorders differ not only in the nature of their beliefs but also in the way they integrate new information into their beliefs depending on the particular disorder. The relative precision of prior beliefs and incoming information is assumed to play a key role in this respect. These considerations—and the associated empirical evidence—do not allow any conclusions regarding whether belief-updating anomalies are a vulnerability factor of mental disorders or merely a consequence thereof (in terms of reflecting a clinical problem as a result from the disorder). To address this question, researchers would have to carry out longitudinal studies that allow crossed-lagged panel analyses.

In addition to contributing to the understanding of the psychopathology of some mental disorders, the clinical distortions in belief updating discussed above also have implications for psychological treatment. Specifically, as we highlighted in the introduction, dysfunctional beliefs significantly influence the severity and prognosis of various mental disorders, and so their resistance to updating and learning from new experience is particularly important from a clinical point of view. Accordingly, clinicians may aim to modify dysfunctional patterns of belief updating in their patients through cognitive-behavioral interventions. For example, because cognitive immunization against disconfirmatory evidence is regarded as an important obstacle when attempting to modify dysfunctional beliefs, future clinical research may develop psychological interventions aimed at inhibiting the engagement in cognitive-immunization strategies. A first proof-of-concept study in this context showed that it was possible to prevent people with depression from devaluing expectation-disconfirming positive information through cognitive-immunization-inhibiting strategies, thus facilitating the revision of negative expectations (Kube, Glombiewski, et al., 2019). It has also been shown that an optimistic view of the future can be trained in people with depression through optimism-enhancing interventions (Miranda et al., 2017). Another recent study demonstrated that such optimism training subsequently helped people with mild dysphoria to update their beliefs about future events more toward desirable information, thereby resembling the optimism bias that is typical of healthy people (Yoshimura & Hashimoto, 2020). However, the optimism training had no effects on belief updating in participants with either low or high dysphoria. We encourage researchers to continue this line of research to help people with mental health problems use novel information effectively to alter their dysfunctional beliefs.

To our knowledge, there has not yet been any research aimed at modifying the opposite pattern (i.e., hasty changes in beliefs). Given that affording too little precision to prior beliefs has been discussed as a possible contributor to pathology in the respective conditions, it may be worthwhile to develop and test interventions aimed at increasing confidence in prior beliefs and/or decreasing the precision given to new information.

Societal implications

In our view, the evidence summarized above on the characteristics of belief updating in healthy people also has implications for social and policy issues. For example, whereas the repeatedly referenced optimism bias may be adaptive in terms of mental health, it becomes problematic when people do not take information that threatens their livelihood seriously enough and fail to take necessary action. Sadly, this phenomenon can be observed in the context of climate change: Although scientific evidence is accumulating to document the devastating consequences of climate change, many people still deny the presence of man-made climate change and their responsibility to counteract it. Some evidence suggests that this, in addition to various other factors, may be due to the optimistic bias when confronted with bad news. Specifically, one study using eye tracking found that people with high trait optimism spent significantly less time attending to any information related to climate change, with particularly low fixation times if the arguments about climate change conveyed “bad news” (Beattie, Marselle, McGuire, & Litchfield, 2017). The authors also found that people with high dispositional optimism more likely recalled the arguments in terms of a debate between two opposing positions (as opposed to framing it as clear evidence) and were more confident that the consequences of climate change would not affect them personally (Beattie et al., 2017). Likewise, a study applying the belief-updating framework to information about climate change demonstrated that people who do not believe in the concept of man-made climate change updated their beliefs more toward good news (suggesting that the global rise of the average temperature may not be as dramatic as predicted) than toward bad news (suggesting that the global rise of the average temperature is even more dramatic than predicted; Sunstein, Bobadilla-Suarez, Lazzaro, & Sharot, 2017).

Thus, it seems that people are fleeing into tragic optimism when confronted with bad news about climate change, hoping that it may not become as horrific as science prognosticates or even denying man-made climate change fundamentally. Given that this individual bias—which is understandable from an individualistic point of view—in processing climate-change-related information contributes to collective inaction, it is a serious problem from a societal point of view. Most recent research suggests that this optimistic bias may also contribute to people’s difficulties in handling the current COVID-19 pandemic: A preliminary report from March 19, 2020, found that people from the United Kingdom, Germany, and the United States estimated the probability of getting infected with the virus—and of infecting others if infected themselves—as lower for themselves than for someone similar to them. Further, the authors linked this optimism to the estimated frequency of direct social contacts and to the necessity of—and difficulties in—reducing these contacts (Kuper-Smith, Doppelhofer, Oganian, Rosenblau, & Korn, 2020).

To return to the examples of current divisive and irreconcilable political controversies as discussed in the beginning of this article, another societal implication of biased belief updating might be derived. In particular, the phenomenon of truth decay discussed above must be considered alarming from a political and societal point of view. It is a crucial aspect of democratic systems that people are properly informed before making a political decision (e.g., in terms of voting for a particular party or person). If, however, there is the increasing tendency that people disregard information that does not fit into their view, this is dangerous for the state of democracy. It is inherent to political decisions that their correctness (in terms of, e.g., being more beneficial than harmful) can be confirmed or falsified by new information. We believe that in a healthy democracy new information should be considered thoroughly in examining past decisions, and if new information brings those decisions into question, they should be revised. Thus, from a political and societal perspective, beliefs should be updated if new information is available that disproves previous assumptions. Accordingly, from this perspective, we view the current developments regarding truth decay and “alternative facts” with great concern. Because “filter bubbles,” reductionism, vulnerability to “fake news,” and other features of social networks significantly contribute to this development, we believe that societies should discuss this development and the role of social networks in political decision-making more critically, particularly because more and more people base their political attitudes on information received on social networks. Future research on belief updating might contribute to this discussion by investigating how information can be provided such that people use it to examine the validity of their (political) beliefs critically.

The continuity of disconfirmatory information

As reviewed in this article, new information can be distinguished by whether it confirms or disconfirms previous assumptions. For the sake of simplicity, many researchers tended to regard confirmatory versus disconfirmatory information as binary concepts. However, because information can vary greatly in the extent to which it contradicts previous beliefs, we encourage researchers to develop experimental paradigms that allow taking the continuity of disconfirmatory information into account. An experimental paradigm that already considers this aspect is the one from Tali Sharot’s lab, which was used to study the optimism bias in healthy people as discussed above. A study on pain perception found that the relationship between the magnitude of the prediction error (indicating how large the difference between the predicted and actual outcome is) and the extent to which beliefs are updated is not linear but includes a “tipping point,” meaning that at some point the prediction error is so large that it is considered less reliable and thus entails reduced updating (Hird, Charalambous, El-Deredy, Jones, & Talmi, 2019). Continuing this line of research, researchers may examine the relationship between the magnitude of the prediction error and belief updating in domains beyond pain perception.

Belief updating as a dynamic phenomenon

Another limitation of previous research refers to the fact that belief updating is not a stable trait but varies dynamically on the basis of various factors. For example, Sharot and colleagues, using their optimistic-bias paradigm, found that the extent to which belief updating is biased toward the integration of desirable (vs. undesirable) information changes throughout the life span (Sharot & Garrett, 2016). Whereas the integration of good news remains stable over time, the discarding of bad news appears to be associated with younger age: Adolescents, to a larger degree than young adults, failed to integrate undesirable information, as revealed in a study by Moutsiana et al. (2013). This may contribute to the understanding of risk-taking in adolescence and have implications on educational risk-prevention programs at schools. When examining other age ranges, a U-shape curve is formed, such that the lowest optimistic bias in belief updating is seen in midlife, whereas younger (adolescents and young adults) and elderly individuals display higher resilience to the integration of bad news. Whether this age effect applies to belief updating in general or only to optimistic updating in particular remains an open question.

Beyond age, belief updating fluctuates depending on factors such as hormones and neurotransmitters, particularly dopamine. Administrating exogenous dopamine to healthy participants increased the discarding of bad news but had no significant effect on the integration of good news (Sharot, Guitart-Masip, et al., 2012), much like the effect seen in adolescents. Dopamine is involved in many neural processes, including the reward system, where it mediates the process of learning from reward and punishment (Pessiglione, Seymour, Flandin, Dolan, & Frith, 2006). Likewise, research on people with Parkinson’s disease has demonstrated that dopamine diminished learning from unwanted outcomes when the patients were on dopamine-containing medication (Frank, Seeberger, & O’Reilly, 2004), contributing to its effect on belief updating. Providing a different neurotransmitter, oxytocin (administered intranasally), led not only to the reduced integration of bad news but also to the increased integration of good news. This was explained by oxytocin’s role in promoting adaptation in social situations by modifying cognitive and emotional responses (Ma et al., 2016). In view of these various sources of influences on belief updating, we encourage researchers in the field to consider factors such as age, hormones, and neurotransmitters thoroughly in their studies and test whether this is a by-product of the optimistic bias or a genuine belief-updating trait.

The question of adaptivity

In this article, we presented the examples of persistent beliefs and hasty changes in a fairly descriptive manner. In our view, an underexplored aspect in this context is the more normative question of when it is adaptive for people to update their beliefs. So far, researchers have mostly investigated simply whether or not people updated their beliefs, and some patterns in this regard were considered “adaptive” and hence linked to mental health. This prompts the question of what adaptive means, and we believe that the answer may depend on the (scholarly) perspective of researchers. For instance, from a predictive processing point of view, it is adaptive to update beliefs if there is enough reliable evidence that disconfirms people’s prior beliefs relatively independent of the nature or content of a belief. By contrast, from a mental health perspective, belief updating would be regarded as adaptive when it helps people sustain their mental health. From this perspective, the content of the particular belief is very important in relation to the question of when it is adaptive to update it: There are some beliefs that can reasonably be considered dysfunctional, such as very negative beliefs about oneself (e.g., “I’m worthless”), whereas other types of beliefs can be assumed to be healthy, such as a mildly positive belief in one’s own abilities. From this standpoint, it would be adaptive to update dysfunctional beliefs and to maintain adaptive beliefs, each even on the basis of rather limited evidence. Additional perspectives to consider for the question of adaptivity might be an evolutionary-psychology perspective and a societal perspective.