On COVID-19, cognitive bias, and open access

Starting out the year 2021 by looking back at the year 2020 might seem like an exercise in masochism, given the horrific loss of life, the untold economic hardships, the resurgence of white supremacy across the country, and the rampant (and at times utterly incomprehensible) political chaos packed into those 12 months. There were, however, many redeeming aspects of the year, not least of which was a panoply of breathtaking scientific achievements. The development of not just one but several remarkably effective vaccines against COVID-19, a hitherto completely unknown virus, in less than 12 months was unprecedented (and was among the reasons Oxford Languages chose “unprecedented” as one of its choices for “word of the year,” https://languages.oup.com/word-of-the-year/). But even beyond assembling, with astonishing rapidity, the body of knowledge required for successful vaccine development, the scientific community across a broad span of disciplines continued to discover and innovate with inspiring resilience. Like so many other journals, PNAS experienced a surge in manuscript submissions—an increase of about 20% for the year, in fact, resulting in the publication of more than 3,600 research articles this year, compared with ∼3,250 last year.

Fig. 1.

Coronavirus illustration. Image credit: Shutterstock/Lightspring.

It should surprise no one that, of the 30 most highly cited of the 3,600+ published papers, 90% were COVID-related. The 10% of highly cited papers that were unrelated to COVID-19, however, represented major advances on other critical fronts. Two studies (1, 2) addressed climate change, a more slow-moving existential threat to human existence that was clearly not forgotten during the pandemic. Also remaining a high priority was an aspect of racial and economic injustice that strikes at the core of the scientific enterprise, with one paper (3) reporting the results of a study aimed at narrowing achievement gaps in underrepresented minority undergraduates in STEM fields.

COVID-19, however, brought with it a seismic shift in the national research profile. At PNAS, we began by closely monitoring COVID-19 manuscripts as of March 6, 2020, in an effort to expedite processing while maintaining quality. Most striking about the submissions was the diversity of disciplines they represented; researchers across the physical, biological, and social sciences pivoted to deploy their specific disciplinary skills to shed light on different dimensions of the pandemic. One study was unexpectedly illuminating for me, in a context broader than just surviving this particular pandemic. Published on June 24, Lammers et al. (4) reported the results of three studies conducted in late March 2020 showing that there existed a widespread misperception that cases of COVID-19 were increasing in a linear, rather than an exponential, fashion; moreover, that misperception was a critical factor contributing to failures to embrace social distancing to stem the spread. Underestimating the threat posed by the epidemic is a manifestation of what has been called exponential growth bias, a phenomenon described decades ago and known for its remarkable robustness, despite its potential to cause individual harm. Lammers et al. (4) was published two weeks after US COVID-19 cases surpassed 2 million. Six months later, by December 21, total cases had exceeded 18 million, and deaths surpassed 310,000 (ourworldindata.com https://bit.ly/3arcigI).

The Elephant in the Room

Over the past 60 years, dozens of forms of cognitive biases have been described in hundreds of studies (5). No human population is immune, and cognitive biases are now widely acknowledged as a threat to the integrity, reliability, and reproducibility of the scientific enterprise. An extensive literature documents the adverse effects of gender bias and race bias in STEM on classroom performance, hiring practices, publication and funding success, tenure and promotion, and academic recognition; confirmation bias influences the interpretation of experimental results; sampling bias, selection bias, and channeling bias distort results of clinical trials; question order bias, interviewer bias, and recall bias can undercut survey research; and publication bias, whereby negative results are far less likely to become part of the scientific literature than statistically significant results, appears to be baked into the contemporary culture of scientific research.

Frankly, it hadn’t occurred to me that statistical biases can undermine how the general public makes sense of scientific data. An exponential growth bias, in particular, is bad news in view of the fact that so many biological phenomena, including a few that are inimical to human survival, increase exponentially. More than 150 years ago, Darwin (6) devoted part of Chapter III of the Origin of Species to the

Geometrical ratio of increase:

there is no exception to the rule that every organic being naturally increases at so high a rate, that, if not destroyed, the earth would soon be covered by the progeny of a single pair. Even slow-breeding man has doubled in twenty-five years, and at this rate, in less than a thousand years, there would literally not be standing room for his progeny… The elephant is reckoned to be the slowest breeder of all known animals, and I have taken some pains to estimate its probable minimum rate of natural increase: it will be under the mark to assume that it [the elephant] breeds when thirty years old, and goes on breeding till ninety years old, bringing forth three pair of young in this interval; if this be so, at the end of the fifth century there would be alive fifteen million elephants, descended from the first pair. [Darwin 1859, p. 64 (6)]

In the ensuing decade, several contemporaries questioned both his assumptions and final calculation, so, by the sixth edition of Origin of Species, Darwin (7) had revised both, changing the paragraph to read,

it will be safest to assume that it begins breeding when thirty years old, and goes on breeding till ninety years old, bringing forth six young in the interval, and surviving till one hundred years old; if this be so, after a period of from 740 to 750 years there would be nearly nineteen million elephants alive, descended from the first pair. [Darwin 1872, p. 51 (7)]

The point of the exercise, however, was that even the estimates made by Darwin’s most punctilious critics, including, for example, “Ponderer” (ca. 1869), whose letter published in Athenaeum put the number at 85,524 after 510 years, still represent a mind-boggling number of elephants (8, 9).

Cognitive biases arise, in part, due to inherent limits on working memory, and exponential growth does push those boundaries. Boggled minds can take shortcuts, falling back on heuristic decision-making limited to familiar domains (10). Graphical representations of exponential growth may even be more difficult to grasp than verbal descriptions, at least in part because, as a species, we haven’t had much time to get used to them. Visual representations of statistical phenomena are a relatively recent innovation; at the turn of the 19th century, political economist William Playfair essentially invented an entire suite of diagrams to incorporate graphical elements into statistics, creating line charts, bar charts, area charts, and pie charts. Initial uptake was certainly not exponential—the word “histogram,” for example, wasn’t even coined for another 90 years (11), a decade after Darwin’s death.

Making Science Accessible

Exponential bias isn’t the only manifestation of cognitive biases associated with data interpretation. A quick check of the scholarly literature identified a natural number bias that leads to a misinterpretation of the relative size of an integer when it is part of a decimal or fractional value (12), as well as a proportional bias in solving missing-value problems (13); there are probably many more. In terms of the total number of cognitive biases, a perfunctory Internet search with the phrase “number of cognitive biases” yielded almost 29 million results, with estimates ranging over a couple of orders of magnitude, leading to the kind of processing overloads that undergird cognitive biases in the first place.

One perceived element shared by many cognitive biases is that they can be difficult to extinguish, possibly because, at some point in evolutionary history, at least some had adaptive value in allowing for more timely decision-making. The failure to grasp the potential impact of exponential growth of a lethal highly contagious disease, however, should be considered exceedingly maladaptive. The scientific literature, with its impenetrable jargon, field-specific methodologies, and assumptions of familiarity with a specific body of knowledge, is, for the most part, written for trained scientist readers. It’s certainly not optimally designed for the casual nonspecialist reader, nor should it be, and correcting biases in occasional readers of the scientific literature has not traditionally been the responsibility of the scientific community. The move toward open access publishing, however, in a way, is making it our responsibility, particularly if the idea is that the general public, having provided the means for federal funding of research, is owed immediate access to the products of that research. Making the scientific literature widely and immediately available probably should bring with it an obligation for making not just the data accessible but every aspect of scientific research more accessible.

If, in fact, such an obligation exists, it’s not at all clear that increasing the accessibility of the more arcane aspects of research is even an achievable goal. Some encouraging evidence that overcoming cognitive biases in understanding statistics is possible is presented in Lammers et al. (4), who demonstrated experimentally, with two studies, that providing just three sentences of instruction or using intermediate steps in estimates of future growth could correct misperceptions about exponential growth to the extent that support for social distancing increases.

One small step toward increasing the accessibility of data might be recognizing and addressing potential cognitive biases in an expanded version of a “significance statement,” an element that is appearing in an increasing number of scientific papers. PNAS began requiring significance statements in 2013 in connection with the introduction of long-format PNAS Plus manuscripts (14). At that time, instructions to authors were minimal: “The Significance Statement should provide enough context for the paper’s implications to be clear to readers. The statement should not contain references and should avoid numbers, measurements, and acronyms unless necessary.” Although PNAS Plus as a publishing option no longer exists (obviated by the transition to entirely digital publishing), the Significance Statement remains, but it has morphed into something different. Today, instructions to authors specify that the Significance Statement should “explain the significance of the research at a level understandable to an undergraduate-educated scientist outside their field of specialty. Include no more than 120 words.” In this form, our Significance Statement resembles those introduced by many other journals in recent years. Aptly, even Cognitive Research: Principles and Implications states, on the journal’s home page, that they “expect that authors will be able to explain in a Significance section how their basic research serves to advance our understanding of the cognitive aspects of a problem with real-world applications.”

As open access publication becomes the norm across the publishing landscape, making data more accessible while at the same time anticipating and making a greater effort to correct potential cognitive biases may be among many tools that the scientific community can use to reduce the likelihood of misperceptions that can lead to widespread rejection of policies and recommendations based on solid scientific evidence, a cultural phenomenon that appears to have grown as exponentially as COVID-19 in the United States in 2020. As 2021 begins with uncertainties about what percentage of the population will actually be willing to receive those “unprecedented” vaccines, it’s probably worth thinking about ways to help the general public contextualize and interpret the data that can be freely accessed.

Finally, speaking of contextualizing, although populations of organisms may have the potential for exponential growth, in reality, density-dependent mortality factors generally prevent any from actually doing so for any length of time. Thus, we’re not really in imminent danger of being buried by elephants. And, speaking of time, if you were wondering, Hardin (15) estimated that it would take 1,484 years for Earth’s surface to be entirely covered by the offspring of just one pair of elephants.