Setting the Context

To begin, a clarification: we are in no way denying the value of accurate information or the cost of false information. Both can and do tend to have the effects one would anticipate, in that accurate information tends to foster accurate beliefs, and false information tends to undermine them. Efforts to reduce the spread of misinformation (Bak-Coleman et al. Reference Bak-Coleman, Kennedy and West2022; Kim et al. Reference Kim, Tabibian, Oh, Schölkopf and Gomez-Rodriguez2018), to identify and correct false information (Rubin et al. Reference Rubin, Brogly, Conroy, Chen, Cornwell and Asubiaro2019; Vraga and Bode Reference Vraga and Bode2017), and to assist information consumers to do the same (Sharon and Baram-Tsabari Reference Sharon and Baram‐Tsabari2020; Vraga, Tully, and Bode 2020) are now and will remain tremendously important.

Nevertheless, we contend that even if these goals are perfectly achieved, misapprehensions will persist – because people are not simply passive receivers of information. The act of [mis]informing does not end with [mis]information – a great deal happens in the mind after information is encountered. When someone receives information, it is filtered through cognitive processes, including strategies and heuristics that can introduce bias, and incorporated into existing belief structures. In other words, the information is used. Our focus is on those moments, and processes, of information use – what Savolainen (Reference Savolainen2009) terms the typically “unspecified ‘appendix’ of information seeking” (116) – and the impact that these have on beliefs.

Thus, in this chapter, we examine the impact of information use in developing and sustaining misapprehensions – even when the information being processed is not itself misleading or inaccurate. Examples of this type of false belief abound. For example, many people incorrectly estimate the frequency of murder to be much higher than that of suicide (Fischoff, Slovic, and Lichtenstein Reference Fischhoff, Slovic and Lichtenstein1977), and travelers often express greater concern about flying as opposed to driving. There are, however, no false news reports or other statements that assign a higher probability to murder as compared to suicide, or a greater risk to flying as compared to driving. The misapprehensions arise instead from the ways in which decision makers cope with the limitations of what Herbert Simon termed “bounded rationality” (Wheeler Reference Wheeler2018). These limitations lead us to rely on mental shortcuts, or heuristics, that support the best possible decisions with the limited resources – attention, memory, and information – that we have at our disposal. Although these heuristics generally work well, they lead to severe and systematic errors in thinking, or cognitive biases (Tversky and Kahneman Reference Tversky and Kahneman1974, 1124). These biases, in turn, can lead us to develop or persist in incorrect conclusions and false beliefs.

We explore these issues in the specific context of the COVID-19 pandemic, but the arguments, and conclusions, have wide application. We structure the chapter around specific misconceptions, exploring how cognitive biases and heuristics can influence these misperceptions, and how careful information design and education can ameliorate those effects.

Misconception 1: “COVID-19 Isn’t Really a Problem – There Are Only a Few Cases”

Even in the earliest stages of the COVID-19 crisis, there was widespread understanding that case counts were growing – but there was also widespread and severe underestimation of the rate of growth, and thus severe underestimation of the number of future COVID-19 cases (Banerjee and Majumdar Reference Banerjee and Majumdar2020; Villanova Reference Villanova2022). Part of the explanation lies in exponential growth bias: a general tendency to underestimate the impact of exponential growth. This bias is well-demonstrated in a children’s book by Demi entitled One Grain of Rice. According to the story, an Indian peasant girl was offered a reward of her choice for service to the king. Her request, which appeared modest to the sovereign, was for a single grain of rice on the first day, and for each of the following thirty days an additional amount of double the number of grains received the day before. The greedy king readily agreed, thinking he had struck a very good bargain. On the last day of the month, however, the girl received four storehouses full of rice in addition to all that had been provided up to that time. In the story, the king has fallen prey to exponential growth bias, and the young girl has used his misconception to her advantage to feed her entire community.

Unfortunately, the impact of exponential growth bias is not as positive when it comes to responding to the COVID-19 crisis. Especially in the initial stages, the spread of infectious diseases follows an exponential function, with a few positive cases exploding into a widespread pandemic if the disease is, as with COVID-19, sufficiently transmittable (Banerjee, Bhattacharya, and Majumdar Reference Banerjee, Bhattacharya and Majumdar2021). Initially, the number of reported COVID-19 infections was low, and a focus on this low number, coupled with a predictable misapprehension of the impact of exponential growth, led many to underestimate the severity of the situation (Lammers, Crusius, and Gast Reference Lammers, Crusius and Gast2020). During the outbreak of March 2020, for example, the number of coronavirus patients in the United States doubled about every three days; however, a study conducted during this period showed that American participants mistakenly perceived the growth of cases of the virus as linear (Lammers, Crusius, and Gast Reference Lammers, Crusius and Gast2020). This misperception could, in turn, hinder the adoption of measures to fight and contain the pandemic. Research has shown that, in the comprehension of COVID-19 disease data, the exponential growth bias can predict noncompliance with safety measures such as handwashing, mask-wearing, and the use of sanitizers (Banerjee, Bhattacharya, and Majumdar Reference Banerjee, Bhattacharya and Majumdar2021). Moreover, interventions that correct the misperception of exponential coronavirus growth have been shown to significantly increase support for social distancing (Lammers, Crusius, and Gast Reference Lammers, Crusius and Gast2020).

Some instructional or educational strategies appear to reduce the effect of exponential growth bias. Lammers, Crusius, and Gast (Reference Lammers, Crusius and Gast2020) found that exponential growth bias could be reduced by asking audiences to read a few sentences that explained the bias (e.g., “keep in mind that many people forget that the speed by which the corona virus spreads, increases each day …”). Even greater reduction in the bias was observed in an experimental context where participants were encouraged to “step through” case number increases when asked to estimate a total number at a specific future date given an initial case count and doubling time (e.g., estimate cases at day fifteen given five initial cases and a doubling time of three days).

Exponential growth bias can also be reduced by careful information design. There is some evidence that presenting growth information in terms of doubling times rather than growth rates (i.e., “the number of cases is expected to double every fifteen days” as compared to “cases are expected to grow at a rate of 5 percent per day”) can reduce exponential growth bias (Schonger and Sele Reference Schonger and Sele2021). Graphical representations of exponential growth are especially difficult to grasp (Berenbaum Reference Berenbaum2021), and the dominant mode of presenting COVID-19 case numbers in the print and online media has been graphical (Banerjee, Bhattacharya, and Majumdar Reference Banerjee, Bhattacharya and Majumdar2021). Banerjee, Bhattacharya, and Majumdar (Reference Banerjee, Bhattacharya and Majumdar2021) experimentally demonstrated that the representation of past COVID-19 case numbers in numerical form (see Table 3.1) significantly decreased the exponential growth bias relative to graphical representation (see Figure 3.2). From this, they suggest that data should be shown via raw numbers (see Table 3.1) and presented “alongside familiar ‘flatten-the-curve’ style graphics” (8).

Figure 3.2 Hypothetical growth of COVID-19 cases over fifteen days with a doubling time of three days, presented graphically.

Misconception 2: “COVID-19 Won’t Happen To Me”

Epidemiologists raised the threat of a global pandemic as early as January 2020 and “announced that more than 40–70 % of the world population could be infected within the end of the year” (Bottemanne et al. Reference Bottemanne, Morlaàs, Fossati and Schmidt2020, 2); in the same month, the Chinese city of Wuhan, with a population of 11 million people, went into full quarantine (Cai Reference Cai2020). Despite this and other strong and mounting objective evidence of widespread infection, people underestimated their personal risk of contracting and/or transmitting COVID-19. Survey data collected in several European countries in February 2020, for example, revealed that a large majority of participants believed that their risk of contracting the coronavirus was around 1 percent (Raude et al. Reference Raude, Debin and Bonmarin2020). (By comparison, as of February 2022, approximately 58 percent of the United States population – and 75 percent of US children – had contracted COVID-19, according to clinical testing of blood samples for SARS-CoV-2 antibodies (Ducharme Reference Ducharme2022).) Many also viewed themselves as less likely than average, and less likely than others, to contract the disease. During the early phases of the outbreak in March 2020, German adults who had not yet tested positive for COVID-19 perceived family members and friends to be at a higher risk of infection than themselves (Gerhold Reference Gerhold2020), and Polish university students estimated their likelihood of contracting the coronavirus as lower than that of their peers (Dolinski et al. Reference Dolinski, Dolinska, Zmaczynska-Witek, Banach and Kulesza2020). Meanwhile, one respondent in an American poll claimed that they were “cautiously optimistic” the United States would “nip [the virus] in the bud” before it could spread as it had in Europe (Allyn and Sprunt Reference Allyn and Sprunt2020).

This unwarranted and elevated perception of personal safety is consistent with the optimism bias (Weinstein Reference Weinstein1983). Optimism bias leads us to think ourselves more likely than average to experience positive outcomes, and less likely than average to experience negative outcomes. This bias appears to arise from an inappropriate comparison group – when thinking about our risk of serious consequences of disease, for example, we implicitly compare ourselves to others who we believe to be less healthy and more vulnerable than ourselves, rather than to a more representative “average” other – and by comparison, we estimate ourselves as less likely to experience the negative effects (Weinstein Reference Weinstein1983).

The misapprehension of reduced personal risk could lead people to ignore public health recommendations and refrain from personal hygiene practices and precautions (Pascual-Leone et al. Reference Pascual‐Leone, Cattaneo, Macià, Solana, Tormos and Bartrés‐Faz2021; see Wise et al. Reference Wise, Zbozinek, Michelini, Hagan and Mobbs2020 for a discussion of the relationship between risk perception and protective behaviors). Indeed, research shows that individuals with high optimism bias engaged less in protective behavioral changes during the COVID-19 pandemic in 2020 (Fragkaki et al. Reference Fragkaki, Maciejewski, Weijman, Feltes and Cima2021).

The delivery of objectively accurate risk information, even by trusted messengers, will not correct this bias (Felgendreff et al. Reference Felgendreff, Korn, Sprengholz, Eitze, Siegers and Betsch2021); however, early research on the optimism bias indicates that personalized risk comparators on a group (Weinstein Reference Weinstein1983) or individual (Alicke et al. Reference Alicke, Klotz, Breitenbecher, Yurak and Vredenburg1995) basis can help to correct risk perceptions. This targeted communication approach, with messages designed for specific individuals or subpopulations, will be effective in reducing optimism bias when audiences can be segregated. For general (untargeted) communications, there is specific research in the COVID-19 context, consistent with prior research in other domains, showing that unrealistic optimism is reduced by communications (video or text) that emphasize the risk-reducing activities of others (e.g., compliance with medical recommendations for social distancing; Dolinski et al. Reference Dolinski, Kulesza, Muniak, Dolinska, Węgrzyn and Izydorczak2022). In general, interventions that explicitly or implicitly provide an appropriate comparison group for personal risk estimation will help to mitigate optimism bias.

Misconception 3: “Vaccines Don’t Work”

Research shows that, in real-world settings, COVID-19 vaccines offer a high degree of protection against SARS-CoV-2-related diseases (Zheng et al. Reference Zheng, Shao, Chen, Zhang, Wang and Zhang2022), and the efficacy of these vaccines has been reported extensively and positively in the mainstream media (e.g., Hayes Reference Hayes2021; Thomas and Hanna Reference Thomas and Hanna2021). Nonetheless, many remain “vaccine hesitant” (Kirzinger et al. Reference Kirzinger, Sparks and Brodie2021), choosing not to take the vaccination. One common argument among the vaccine hesitant is that “vaccines don’t work” – and the cited evidence is that the vaccinated continue to make up “most people admitted to hospital with Covid-19” (Benedictus Reference Benedictus2021),Footnote ¹ and a large proportion of those dying from the disease (latest data from Canada indicates that from December 2020 to June 2022 there have been 10,385 COVID-related deaths among the unvaccinated, and 7,423 COVID-related deaths among the vaccinated: Statista 2022). If vaccines work, the reasoning goes, those who are vaccinated should not be contracting COVID-19, should not be hospitalized with COVID-19, and should not die from the disease. This is, of course, a misunderstanding of the math and science. No vaccine is 100 percent effective, but an effective vaccine reduces the likelihood of contracting a disease, and can also reduce the severity of the disease if it is contracted. According to the best available data, COVID-19 vaccines provide both types of protection (Centers for Disease Control and Prevention 2022). So, why the persistent belief that the vaccines don’t work?

This misperception arises at least in part from a failure to take into account the vaccination base rate, or the proportion of people in the entire population who are vaccinated compared to those who are unvaccinated. Imagine a virus that will, without any protection, affect 30 percent of the population. Imagine also that 95 percent of the population have received a vaccine for the virus. If the vaccine provides no protection, then 30 percent of the unvaccinated and 30 percent of the vaccinated will be infected, and if we consider the positive cases only, 95 percent of those will be individuals who have received the vaccine (because the vaccine offers no protection, and 95 percent of the population has received the vaccine: Egger and Egger Reference Egger and Egger2022). If the vaccine provides perfect protection (which never happens), then all cases will be among the unvaccinated. In general, however, reality falls somewhere in between. Say, for example, that the vaccine is 80 percent protective against disease; in other words, only 20 percent of those who are vaccinated and who would otherwise have been infected will actually get the virus. In this case, 30 percent of the unvaccinated population will still get the disease, but the rate among the vaccinated will fall to 6 percent (the 20 percent for whom the vaccine is ineffective, out of the 30 percent who would otherwise be expected to become infected).

So far, so good – but here comes the surprising part: under these circumstances, the large majority of infections will still occur in people who are vaccinated. In fact, almost 80 percent of the people who get the condition will have received the vaccination. Put another way, of 1,000 randomly selected people in this population, 15 unvaccinated are expected to contract COVID-19 (30 percent of the 5 percent of the population who are unvaccinated), and 57 among the vaccinated are expected to get the disease (20 percent of the 30 percent who would otherwise get COVID-19, among the 95 percent of the population who are vaccinated). In general, if the proportion of those who have a condition is anything less than the proportion of the population vaccinated against the disease, then the vaccine is providing protection – but that doesn’t seem right when the large majority of those we see with the condition have been vaccinated against it.

Base rates – in this case, the rate of vaccination in the population – have a surprisingly large and counterintuitive impact on outcomes, leading to inaccurate intuitions in a wide variety of circumstances. Thus, for example, we tend to over-interpret the results of accurate screening tests for rare conditions (Burkell Reference Burkell2004), incorrectly assuming that a positive screening test is a strong indicator that the condition is present. Screening tests, by definition, test for relatively rare conditions with a fast and low-cost initial test designed to capture the large majority of true positive cases. These tests are by definition imperfect, and are focused on avoiding false negative results (saying that someone does not have the condition when in fact they do) at the expense of an increased risk of false positive results (saying that someone has the condition when in fact they do not) that can only be resolved through further testing. As a consequence, and as illustrated in Table 3.2, a large proportion of positive screening test results are actually false (in the case presented, 95 of 150 positive results, or 66 percent) – that is, the individual does not in fact have the identified condition. This situation arises precisely because there are so many people without the condition – or, to put it another way, because the base rate of the condition in the population is low, with the result that even a small tendency to give a false positive result creates a large absolute number of such results. Base rate neglect may also play a role in the alarm recently expressed by bicycling advocacy groups over an increase in bicycling deaths (Advocacy Advance 2021).Footnote ² While there are undoubtedly many explanations for this shift, one that is not regularly acknowledged is the increase in the number of cyclists on the road (Mazerolle Reference Mazerolle2021).Footnote ³ This represents an increase in the base rate of cycling as a mode of transportation, which would increase the number of observed cycling accidents even if no other risks to cyclists were to change.

Base rate neglect may explain, at least in part, why a sizable segment of unvaccinated people perceive COVID-19 vaccines to be ineffective (Kirzinger et al. Reference Kirzinger, Sparks and Brodie2021) despite credible proof to the contrary.Footnote ⁴ Evidence of this misperception can be easily observed, even among healthcare professionals (Kampf Reference Kampf2021). Governmental data showing a higher proportion of vaccinated than unvaccinated people in COVID-19 deaths has been used to argue for the inefficacy of vaccines online (e.g., The Exposé 2022).Footnote ⁵ Base rate neglect was also at the root of the February 2022 pronouncement by a high-ranking Canadian politician that the COVID-19 vaccine was no longer effective (Quon Reference Quon2022).Footnote ⁶ In that case, data showing that the number of new COVID-19 cases in the province of Saskatchewan was “about the same in [sic] vaccinated and unvaccinated people” was presented as evidence of vaccine inefficacy. This conclusion ignored the critical fact that nearly 80 percent of the province’s eligible population had received at least two vaccine doses, and even a small proportion of breakthrough cases would result in a relatively large number of vaccinated individuals testing positive for COVID-19.

At the same time, many recognize that these perceptions arise from base rate neglect, and efforts to correct this misperception have appeared in a variety of sources including news articles (e.g., Devis Reference Devis2021; Ferreira Reference Ferreira2022)Footnote ⁷Footnote ⁸ and social media posts (e.g., Rumilly Reference Rumilly2021).Footnote ⁹ Approaches to addressing base rate neglect include graphics that “zoom out” to show the base rate of vaccinated and unvaccinated groups in a hypothetical population (Rumilly Reference Rumilly2021), allowing viewers to visualize how the unvaccinated population is disproportionately hospitalized with COVID-19, even if more hospitalized patients are vaccinated than unvaccinated. Animations demonstrating changes in the percent of vaccinated (as opposed to unvaccinated) COVID-19 hospitalizations with changes in the base rate of vaccination in the population have also been used to achieve this outcome (Panthagani Reference Panthagani2022).Footnote ¹⁰

Many of these alternative visualizations are presented in the context of “corrective” or educational information (e.g., in a blog intended to “help you stay on the frontline of health information”: Panthagani Reference Panthagani2022). These are effective tools for motivated audiences focused on the accuracy of their own perceptions, but it is important also to ensure that information about case counts is initially presented in ways that minimize base rate neglect and thereby minimize the inaccurate interpretation that vaccines do not provide protection. Case counts presented with vaccination status but without drawing attention to the base rate of vaccination in the population invite misinterpretation arising from base rate neglect. At the very least, therefore, media and other public reports should include the relevant base rate information along with counts. A better approach is to take advantage of presentation formats that have been demonstrated to lead to better understanding. Specifically, frequency histograms (Burkell Reference Burkell2004) that show infected and uninfected members of vaccinated and unvaccinated populations have been demonstrated to reduce base rate neglect.

Misconception 4: “Vaccines Aren’t Safe”

In general, vaccines are safe and effective at reducing disease, having undergone rigorous testing for both efficacy and negative side effects before approval. Although the COVID-19 vaccines were developed in a relatively short time in response to an acute health crisis, they are no exception to these general rules, and have been widely demonstrated to reduce incidence and severity of COVID-19 infection with few if any negative health consequences (Centers for Disease Control and Prevention 2022; Henry and Glasziou Reference Henry and Glasziou2021). Nonetheless, and consistent with public response to other vaccines, there remains a portion of the population who refuse the vaccine, citing safety concerns (King et al. Reference King, Rubinstein, Reinhart and Mejia2021; Monte Reference Monte2021).Footnote ¹¹ The issue of vaccine safety is complex, and indeed there are questions about the long-term impact, including unintended consequences, of the mRNA vaccines (Seneff and Nigh Reference Seneff and Nigh2021). In many cases, however, hesitancy is linked to an overestimation of the incidence of vaccine-related adverse effects, including among unvaccinated health professionals (Ehrenstein et al. Reference Ehrenstein, Hanses, Blaas, Mandraka, Audebert and Salzberger2010).

This inaccurate risk perception can be attributed at least in part to the availability bias, which is a tendency to base frequency or probability estimates on the ease with which specific instances can be recalled or imagined (Tversky and Kahneman Reference Tversky and Kahneman1973). When less likely outcomes are more salient – for example, by virtue of the frequency with which they are reported – the probability of these events can be overestimated. Thus, for example, we will tend to overestimate the likelihood of an automobile accident immediately after witnessing one, and we will overestimate the likelihood of an airplane crash if one has recently been covered in the news. The availability bias is stronger for negative than for positive events, perhaps because those events are more salient (Stapel, Reicher, and Spears Reference Stapel, Reicher and Spears1995). Media coverage is a significant factor in the availability bias, and has been shown to influence health concerns among the general public (Brezis, Halpern-Reichert, and Schwaber Reference Brezis, Halpern-Reichert and Schwaber2004) and even physician diagnostic decisions (Schmidt et al. Reference Schmidt, Mamede, Van Den Berge, Van Gog, Van Saase and Rikers2014).

A rare serious adverse effect of a vaccine covered in the media provides “a vivid and emotionally compelling anti-vaccination message, likely to be recalled during decision making” that could cause one to overestimate the probability of an adverse effect following immunization (Azarpanah et al. Reference Azarpanah, Farhadloo, Vahidov and Pilote2021, 8). In March 2021, for example, reports emerged in the mainstream media of a possible link between AstraZeneca’s COVID-19 vaccine and rare but potentially fatal blood clots (e.g., Gronholt-Pedersen Reference Gronholt-Pedersen2021; Reuters staff 2021).Footnote ^12, Footnote ¹³ Some of these reports included descriptions of individual cases, such as that of a sixty-year-old Danish woman who developed a blood clot and died after vaccination (Gronholt-Pedersen Reference Gronholt-Pedersen2021). In the wake of this media coverage, one March 2021 poll showed that Canadians were much warier of the AstraZeneca vaccine than other vaccines approved for use in Canada (Bryden Reference Bryden2021).Footnote ¹⁴ Although blood clots among those with low blood platelet counts were eventually confirmed to be a side effect of the AstraZeneca vaccine, the European Medicines Agency (2021) stressed that the complication was “very rare” and that the vaccine’s benefits in preventing COVID-19 outweighed the risks of side effects.

In a context where low-probability events, such as vaccine side effects, are more likely to be reported than higher-probability outcomes, such as vaccination without incident, the availability bias will lead audiences to overestimate the risk of the low-probability event. Counteracting this bias requires that the alternative – safe and incident-free vaccinations – are made more salient and more easily recalled. Strategies to achieve these outcomes include multiple repetitions of the items judged less frequent, and/or increasing the memorability of the reports of incident-free vaccinations (Lewandowsky and Smith Reference Lewandowsky and Smith1983). The latter could be supported, for example, by the intuitively appealing strategy of reporting stories of high-profile individuals who have received vaccination and not experienced side effects. Some strategies to reduce availability bias focus on instruction and training (e.g., Mamede et al. Reference Mamede, de Carvalho-Filho and Schmidt2020). These approaches are most useful when specific individuals are engaged in repeated risk estimation tasks (e.g., physicians making a diagnosis), and often involve guided reflection on decision-making processes along with specific stepwise strategies including the consideration of alternative conclusions (Prakash, Sladek, and Schuwirth Reference Prakash, Sladek and Schuwirth2019).

Misconception 5: “We’ve Had the Pandemic – So We’re Safe for a Long Time”

No one knows when the next pandemic will occur, and our expectations on this point matter. Belief that another pandemic is likely in the near future will spur prevention and resilience activities, while the assumption that it will be a long time before we have another pandemic will foster complacency and a disinterest in taking up risk mitigation activities. Like deciding whether to take actions to curb the spread of COVID-19, choosing to engage in or avoid measures that could prevent the next pandemic can be conceptualized as a distinct action arena. Although there is no “correct” prediction regarding when we will next see a pandemic, it is important that perceptions are unbiased and appropriately calibrated to objective and current information about risk. These perceptions, however, are influenced by biases and heuristics, and it is crucial that we understand what these are, how they operate, and how they will influence the perceived risk of a new pandemic.

There are numerous cognitive biases that could and likely do influence our expectations about when another pandemic will occur. One of these is the gambler’s fallacy, well demonstrated in the history of the “Lion’s Share” slot machine at the MGM Grand hotel-casino. In 2014, this slot machine paid out a progressive jackpot of $2.4 million after collecting money for twenty years from unsuccessful gamblers (CNN Wire 2014). Some, like the author of one blog post, believed that the long “dry” period meant the slot machine was “due for a win” (Best US Casinos, n.d.)Footnote ¹⁵ Assuming a fair machine with outcomes determined randomly, this intuition is unequivocally wrong. Players at that particular machine were no more likely than those at any other machine with the same odds to have a “winning” spin – because like all independent random events, slot machine spins have no “memory.” For each new instance, the probability of winning is exactly the same. The contrary belief – essentially the belief that luck (good or bad) must change – is called the gambler’s fallacy (Tversky and Kahneman Reference Tversky and Kahneman1974).

The gambler’s fallacy isn’t limited to the casino. The fallacy leads people to believe that improbable events operate on an implicit “schedule,” and thus become increasingly likely as time passes without an event (in Eastern Canada, after a few years without a big snowstorm you’ll hear the refrain “we’re due for a big one”) and, once they occur, are unlikely to be repeated, at least in the near future (think about the familiar adage “lightning never strikes twice”). Soccer goalkeepers fall prey to the gambler’s fallacy when faced with penalty kicks (Wogan Reference Wogan2014), and asylum judges, loan officers, and baseball umpires show signs of the gambler’s fallacy when making multiple decisions (Chen, Moskowitz, and Shue Reference Chen, Moskowitz and Shue2016).

The gambler’s fallacy appears to be operating in at least some expectations regarding pandemics. COVID-19 has been described as a “once-in-a-lifetime” (e.g., Guterres, n.d.)Footnote ¹⁶ or “once-in-a-century” (e.g., Gates Reference Gates2020) event.Footnote ¹⁷ Even World Health Organization (WHO) Director General Tedros Adhanom Ghebreyesus characterized the pandemic as a “once-in-a-century health crisis” during a 2020 meeting of the WHO’s emergency committee (Reuters staff 2020).Footnote ¹⁸ This belief might partially arise from the fact that the Spanish flu – which has attracted “unprecedented interest” due to COVID-19 and has clear parallels to the COVID-19 pandemic itself (Simonetti, Martini, and Armocida Reference Simonetti, Martini and Armocida2021, E613) – broke out approximately 100 years prior, in 1918. According to some:

Even if pandemics were completely random events – and they are not – this reasoning would be flawed: the fact that we had a pandemic 100 years ago and another one just recently does not suggest that the next is due in another 100 years. Indeed, research that models pandemic risk “shows that the frequency and severity of spillover infectious disease – directly from wildlife host to humans – is steadily increasing” (Smitham and Glassman Reference Smitham and Glassman2021). Contrary to the assumption that it will be a long time before another severe pandemic, one research team estimates “the annual probability of a pandemic on the scale of COVID-19 in any given year to be between 2.5 and 3.3 percent, which means a 47–57 percent chance of another global pandemic as deadly as COVID in the next 25 years” (Smitham and Glassman Reference Smitham and Glassman2021). Moreover, and contrary to the gambler’s fallacy, barring factors that increase the annual risk (e.g., increased zoonotic spillover; see Birch Reference Birch2021), the chances of a new pandemic in each of the next 100 years is identical, and a new pandemic just as likely after one virus-free year as it is after 99 virus-free years.

The gambler’s fallacy (also known as belief in the law of small numbers; Tversky and Kahneman Reference Tversky and Kahneman1971) is a remarkably persistent cognitive bias (see, e.g., Bishop, Thompson, and Parker Reference Bishop, Thompson and Parker2022). Basic training regarding random events does not appear to reduce the bias (Beach and Swensson Reference Beach and Swensson1967). There is some evidence, however, that the perceptual grouping of events can influence the gambler’s fallacy. In particular, if an event (say, for example, the upcoming year) is seen as grouped with or a continuation of a prior series of events (e.g., the past 100 years), then the gambler’s fallacy is evident. If, however, the event is viewed as the beginning of a new series, the gambler’s fallacy is reduced or even eliminated (Roney and Trick Reference Roney and Trick2003). In the COVID-19 context, this would suggest that the underestimation of the risk of a new pandemic could be reduced if risk discussions were forward looking (e.g., “over the next 100 years”), rather than backward looking (“in the past 100 years”).

Confirmation Bias

Confirmation bias describes the tendency to selectively seek out and attend to information that is consistent with previously held beliefs or attitudes (Jones and Sugden Reference Jones and Sugden2001; Wason Reference Wason1960); evaluation bias, which is associated, is the tendency to more positively evaluate information that is consistent with prior beliefs and attitudes. The effect of these two biases is to minimize cognitive dissonance (Festinger Reference Festinger1957), or the uncomfortable state of internal cognitive inconsistency that can arise when we are confronted with something inconsistent with our beliefs and/or attitudes. The effects of confirmation bias are exacerbated in the online social environment, where past information consumption choices influence future offerings, not only to the individual but also to others who are identified as “similar” based on complex algorithmic processing (Ling Reference Ling2020). Confirmation bias contributes to opinion polarization, through a feedback loop wherein even a slight bias or tendency in thinking (e.g., belief that vaccines are unsafe) is exaggerated by selective attention to supporting information, which in turn creates an even stronger tendency to ignore inconsistent messages (Modgil et al. Reference Modgil, Singh, Gupta and Dennehy2021; Xu et al. Reference Xu, Coman, Yamamoto and Jimenez Najera2022). Surprisingly, there is some evidence that the tendency to select confirmatory messages and to perceive them as more convincing is higher among individuals with higher literacy levels (in the case of the research, health literacy levels: Meppelink et al. Reference Meppelink, Smit, Fransen and Diviani2019). Evaluation biases similarly lead to discounting of disconfirming evidence. Information that is strongly inconsistent with previously held beliefs is evaluated as less credible, and is therefore less likely to be accepted, than information consistent with those beliefs (Christensen Reference Christensen2021).

Confirmation bias can be reduced (somewhat paradoxically) by making the corrective information more difficult to process – creating what some researchers have termed “disfluency” in information processing (Hernandez and Preston Reference Hernandez and Preston2013; see also Rajsic, Wilson, and Pratt Reference Rajsic, Wilson and Pratt2018). Disfluency is affected by features such as the visual clarity of text and is believed to promote a critical and analytical mindset (Hernandez and Preston Reference Hernandez and Preston2013). Research has shown, for instance, that high school students score higher on exams when learning materials are presented in hard-to-read fonts (Diemand-Yauman, Oppenheimer, and Vaughan Reference Diemand-Yauman, Oppenheimer and Vaughan2011). In two studies, Hernandez and Preston (Reference Hernandez and Preston2013) found that participants’ preexisting attitudes toward certain issues became less extreme when they read arguments on the issues in a disfluent format (e.g., light gray, bolded and italicized Haettenschweiler font vs. standard 12-point Times New Roman). The second study further determined that participants were only able to disconfirm their prior biases when they were not under “cognitive load” (i.e., distraction or time pressure) (Hernandez and Preston Reference Hernandez and Preston2013).

In some cases, simply increasing the salience of corrective information can ameliorate false beliefs. Schwind and colleagues (Schwind et al. Reference Schwind, Buder, Cress and Hesse2012; Schwind and Buder Reference Schwind and Buder2012) suggest that explicitly recommending information that is inconsistent with prior preferences or beliefs (e.g., by highlighting the results in an online search) can be effective in reducing or correcting those misapprehensions.

Confirmation bias may also play a role in situations where efforts to deliver corrective information “backfire.” Worldview backfire effects occur when corrections that challenge people’s worldviews bolster their belief in the original misinformation (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022; Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). This phenomenon was originally identified by Nyhan and Reifler (Reference Nyhan and Reifler2010), who found that corrections that contradicted individuals’ political beliefs increased their prior misperceptions.

Worldview backfire effects have been conceived of as a product of directionally motivated reasoning (Holman and Lay Reference Holman and Lay2019; Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). In short, individuals may be motivated to arrive at either an accurate conclusion or a particular, directional conclusion (Kunda Reference Kunda1990), and “worldview backfire effects transpire when directional motivations take precedence over accuracy goals” (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020, 170). These effects are believed to be tied to both confirmation bias and disconfirmation bias, the latter of which involves calling to mind opposing arguments, or “counterarguing,” when faced with ideologically dissonant information (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). In essence, when a correction challenges false beliefs that are central to one’s identity, they may discount the accurate information and generate counterarguments to it (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022; Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). For instance, someone identifying as an antivaxxer might perceive information proving that the risks of a vaccine do not outweigh the risks of a disease to be an identity threat, and subsequently ignore this worldview-inconsistent evidence while selectively focusing on evidence that supports their own position (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022).

To avoid worldview backfire effects, corrections “should be tailored to their target audience: the subset of people for whom these corrections would feel the most threatening” (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020, 171). Specifically, corrections should be framed to be consonant with their target audience’s values and worldviews (Lewandowsky et al. Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012; Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). When attempting to convince someone with an “ecocentric” outlook that nanotechnology is safe, for example, the target may be more receptive to evidence of the technology’s safety if its use is portrayed as part of an effort to protect the environment; here, potential benefits and opportunities are highlighted rather than risks and threats (Lewandowsky et al. Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012). Corrections that threaten a target’s worldview can also be made more palatable by pairing them with a self-affirmation or identity affirmation, which involves a message or task that affirms one’s basic values (Lewandowsky et al. Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012) or highlights sources of self-worth (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022).

The Continued Influence Effect (CIE)

The fact that misinformation may continue to influence people’s thinking – even after a retraction has been acknowledged and recalled – is known as the continued influence effect (CIE) (Johnson and Seifert Reference Johnson and Seifert1994; Seifert Reference Seifert2002). The CIE has been explained by dual-process theory and mental model theory (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). Dual-process theory assumes that memory retrieval can be automatic (fast and unconscious) or strategic (deliberate and effortful) (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). While automatic processing is “relatively acontextual, distilling information down to only its most essential properties,” strategic processing is needed “to retrieve specific details about a piece of information” (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020, 174). From this perspective, it is possible that individuals could automatically retrieve a piece of misinformation and fail to recall relevant features, such as its source or perceived accuracy (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020).

On the other hand, mental model theory suggests that people “construct models of external events in their heads, which they continuously update as new information becomes available” (Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020, 174). Retractions may create a gap in an individual’s model of an event, motivating them to continue to invoke the original misinformation (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022; Lewandowsky et al. Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012; Wittenberg and Berinsky Reference Wittenberg, Chloe, Berinsky, Persily and Tucker2020). This effect can be reduced by providing an alternative causal explanation for an event that fills the gap left behind by the retracted misinformation (Ecker et al. Reference Ecker, Lewandowsky and Tang2010; Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022; Lewandowsky et al. Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012; Seifert Reference Seifert2002). For instance, to correct the misperception that a fire was caused by negligence, a causal explanation (“there is evidence for arson”) is more effective than a plain retraction (“there was no negligence”) (Ecker et al. Reference Ecker, Lewandowsky and Amazeen2022, 21). One study found that a specific warning describing the CIE reduced, but did not eliminate, participants’ continued reliance on outdated information (Ecker et al. Reference Ecker, Lewandowsky and Tang2010). When the warning was combined with a plausible alternative explanation for the retracted information, the CIE was further reduced – though still not fully eliminated (Ecker et al. Reference Ecker, Lewandowsky and Tang2010).

Moving Beyond Panaceas

At the outset of this section, we asked the question of whether information is a “panacea” for the misinformed. Not surprisingly, the answer is a resounding “no.” Complex problems such as the issue of mis/disinformation require complex, multifaceted, and multitiered solutions (Ostrom and Cox Reference Ostrom and Cox2010), and are resistant to simple, single-strategy approaches. Addressing the pressing issue of false beliefs – and false information – will require interventions with information consumers, information providers and producers, and information hosting platforms, and regulatory bodies including governments at the very least. Moreover, and critically, while interventions at the individual level including training and appropriate information design will be essential to reducing the impact of mis/disinformation, interventions with hosting platforms including regulation by government and other bodies will be critical to an effective and appropriate governance response.