Tag: COVID-19

Special Issue, Uncategorized

Should I Do as I’m Told? Trust, Experts, and COVID-19

Matthew Bennett

[This is an advance copy of an article that will appear in print in September 2020 as part of the KIEJ’s special double issue on Ethics, Pandemics, and COVID-19.]

ABSTRACT. The success of public health responses to the COVID-19 pandemic is sensitive to public trust in experts. Despite a great deal of attention to attitudes towards experts in the context of such crises, one significant feature of public trust remains underexamined. When public policy claims to follow the science, citizens are asked not just to believe what they are told by experts, but to follow expert recommendations. I argue that this requires a more demanding form of trust, which I call recommendation trust. I argue for three claims about recommendation trust: recommendation trust is different from both epistemic and practical trust; the conditions for well-placed recommendation trust are more demanding than the conditions for well-placed epistemic trust; and many measures that have been proposed to cultivate trust in experts do not give the public good reasons to trust in expert-led policy.

1. Introduction

In 2019 public trust in politicians was at an all-time low in many parts of the world. Then again, it had been low for quite some time.[1] Public trust in epistemic authority is a more complicated matter. The success of many right-wing politicians—Trump, Johnson, Bolsonaro, etc.—is partly due to their efforts to discredit traditional sources of information, including the “mainstream media” and whichever scientific institutions prove inconvenient to their political interests. But worries about the erosion of epistemic standards in public discourse are easily overstated, and some polling data have shown that publics are not as distrustful of experts as we might fear.[2]

Nonetheless, there are some political and social crises in which even a small degree of suspicion of scientific authorities can have disastrous consequences. One such crisis is the climate change emergency. It is often with a focus on the climate crisis that social epistemologists and philosophers of science have tackled a range of questions around trust in expertise, including how we can defend the rationality of trusting experts and what measures we can take to cultivate such trust (see e.g., Anderson 2011; John 2016). Public health crises are also highly sensitive to public trust in experts, and the COVID-19 pandemic is no exception to the principle that when epidemics pose serious threats to the health of whole nations, distrust of the relevant public health experts can have serious consequences.

Despite a great deal of attention to attitudes towards experts in the context of such crises, one significant factor in public trust in experts remains underexamined. The vast majority of philosophical work on public trust in experts has addressed questions about epistemic trust: forming beliefs on the basis of the testimony, broadly construed, of scientists and scientific communities. But as the COVID-19 pandemic shows us, a distinct kind of trust in experts is needed when experts not only provide information and data, but also play an instrumental role in the formation of science-led policy. When public policy claims to follow the science, citizens are asked not just to believe what they are told, but to follow expert recommendations. I argue that this requires a distinct form of trust, which I call recommendation trust.

I will argue for three claims about recommendation trust. The first is that recommendation trust is different from both epistemic trust and a third form of trust, also popular among philosophers, which I call practical trust (Section 3). The second is that the conditions in which recommendation trust is well-placed are more demanding than the conditions for well-placed epistemic trust (also Section 3). This is because to have good reason to follow an expert recommendation I must not only believe that the expert is sincere and competent in their field, but also that the action they recommend is in my interest. This leads me to argue for my third claim (Sections 4 and 5): because the norm that governs the rationality of recommendation trust is more demanding than that which governs epistemic trust, many measures that have been proposed to cultivate trust in experts do not help to cultivate recommendation trust in science-led policy. In short, experts have a higher bar to clear when we are asked to follow their recommendations. If we are to ask the public to trust the recommendations of scientists, we must acknowledge that this is different from asking novices to accept facts. When science leads policy, it must work harder to merit public trust


My account depends foremost on the assumption that public trust in experts is both necessary and desirable for an effective public health response to the coronavirus pandemic.[3] One reason to think trust in experts is particularly important in the context of the pandemic is that most current public health policies in response to the crisis depend on a significant degree of voluntary compliance. Many countries have taken social distancing measures to reduce infection rates to a manageable level. This usually involves radical lifestyle changes for citizens, including the closure of businesses, workplaces, and many public spaces, bans on travel for all but essential journeys, and complete self-isolation for those at greater risk of serious illness. It is fair to assume that it would be wholly unpractical to apply social distancing solely through comprehensive policing and direct enforcement. And I am confident that even if compliance with social distancing could be guaranteed wholly through coercion, this would be undesirable.

A state could coerce compliance with social distancing in ways that are less direct than totalitarian policing. The simple threat of such measures might be enough to persuade the public to comply. I am also taking for granted that we would prefer voluntary compliance with a public health policy to compliance secured solely through public fear of what the state might do to them if they do not comply. But note that this is not an all-or-nothing test of the desirability of a public health policy, as if such a policy either depends on states applying coercive power to all citizens, or it does not depend on such power at all. Any state-enforced policy, even where it is granted democratic legitimacy through public trust, will involve exercising coercive power over the few who refuse to comply. In this respect, trust in government secures not just a sufficient level of compliance with government policy, but also sufficient public support for the policing of that policy. It would be wrong to think that a public health response secured through trust would involve no coercive state power at all.

Though I am taking for granted that public trust is a necessary part of a desirable response to the pandemic, I do not assume it is sufficient. There are many reasons for this. One is that citizens may be willing but unable to comply with government policy. Emerging research on public behavior during the pandemic in the UK indicates that although willingness to follow social distancing measures is generally high, the ability to self-isolate is not evenly distributed, with the most economically disadvantaged unable to comply (Atchison et al. 2020). Thus, some citizens who trust policymakers enough to be willing to follow policy might not be able to do so.

It is also possible for citizens to trust policy that does not merit trust, and I will be assuming that the kind of public trust that features in a desirable response to the pandemic is specifically trust that is well-placed. Here it helps to invoke a distinction sometimes made by social epistemologists between trustworthiness and credibility (Rolin 2002). An authority is credible when it is likely to be believed. Credibility is vulnerable to faulty reasoning by individuals when making judgements about whom to believe, but the more insidious threat to well-placed trust is posed by structural injustices that grant illegitimate authority to privileged social groups (Fricker 1998). By contrast, an authority is trustworthy when its claims should be given credence. Following Onora O’Neill’s work on public trust (see e.g., O’Neill 2020), I will assume that desirable public trust must be trust in the trustworthy, and it is only accidentally related to credibility. However, one of the tasks of good science-led policy is to bridge any extant gaps between credibility and trustworthiness. I will return to this challenge in Sections 4 and 5.

My framing of the issues so far might be accused of confusing two very different matters: public trust in scientific experts and public willingness to comply with government policy. If some readers worry about this, then all the better for the purposes of this paper, for the distinction I wish to argue for is very similar to the distinction between trust in science per se and compliance with scientifically-informed policy. But before proceeding to put some conceptual distance between trust in science and trust in technocracy, it is worth noting some observations about the tandem pairing of government policy and scientific expertise that we see in many (though not all) government responses to the coronavirus crisis. What we find in the pandemic is that when science has a significant role to play in policy, trust in experts and compliance with government go hand in hand.

In the UK, where I write this paper, the government’s response to the pandemic has been consistently presented to the public as “following the science.” Daily press conferences began on March 12, 2020 with a briefing delivered by the prime minister flanked either side by the government’s chief medical officer, Chris Whitty, and chief science officer, Patrick Vallance. Vallance and Whitty have since played a prominent role in communicating the government’s policy on the coronavirus, standing alongside government ministers in many more press conferences and appearing on television, on radio, and in print to explain the UK’s public health strategy. Between March and May, Whitty and Vallance have been joined in this communications effort by inter alia Angela McLean, deputy chief scientific adviser, Jenny Harries, deputy chief medical officer, and NHS England’s medical director, Stephen Powis.

Each of these government spokespersons are scientists—epidemiologists, clinical pharmacologists, mathematicians, public health physicians, and medical professors. Their most evident role in the current emergency appears to be to amplify the message that there is authoritative scientific reasoning that supports the social distancing measures the public are being asked to adopt. This “following the science” message depends on the public believing not just that these scientists are part of the communication of the policy, but also that they are involved in deliberations that result in that policy. That the UK government is in fact following the science is very much open to question, but it is evident that the UK’s approach to communicating its message to its citizens is to heavily emphasize its (putative) technocratic nature.

Most other national government responses lie somewhere between two extremes: namely, policy that is directly contradicting the recommendations of most scientists and policy that is designed and implemented by scientists. These extremes are somewhat idealized insofar as the question of whether a given national policy denies or follows “the science” ignores the fact that there is no single authoritative scientific perspective on what should be done. Nonetheless, some notable examples approximate these extremes. In Brazil, Jair Bolsonaro has argued publicly with regional governors and the national health ministry about the wisdom of advice to all citizens to stay home. In the US, Donald Trump has made a series of dangerously ill-informed claims about possible treatments for COVID-19, from hydroxychloroquine to disinfectant, prompting multiple high-profile corrections from Anthony Fauci, director of the US National Institute of Allergy and Infectious Diseases. By stark contrast, the Swedish government has in large part handed over their public health response to state epidemiologist Anders Tegnell, who has become the public face of the Swedish government during the pandemic and leads a state science agency with a significant degree of autonomy from the government. Though it would be easy to overstate the level of political authority granted Swedish public health agencies (responsibility for policy still formally lies with government ministers), the Swedish case is perhaps the closest thing we have to a purely technocratic response to the pandemic.

For those governments that are not actively discrediting the recommendations of scientific experts, public trust in experts and public compliance with government policy go hand in hand. The significance of public trust in experts increases with the degree to which government policy is presented as technocratic. Note that this does not mean that a government’s strategy for dealing with the pandemic must in fact be technocratic in order for public trust in experts to be a significant factor in securing public support. Public trust in experts can support policy implementation by ensuring that citizens are willing to comply with government public health responses that are perceived to be led by science, regardless of whether they are in fact led by science. Public trust in experts is also a significant factor in public support for genuinely technocratic policy, provided we wish to avoid measures to secure public endorsement that involve coercion or dishonesty.[4]


Limited though they are, there are lessons to be learned from the coronavirus pandemic about the nature of the public trust that supports acceptance of (at least putatively) science-led policy. Before outlining those lessons directly, it will help to be more precise about the kinds of trust that are relevant to the pandemic.

I will be treating trust in all of its forms as a tripartite relation: one person trusts another with regard to a particular object of trust.[5] What I call practical trust is to stake something of importance on a particular action or range of actions that I expect another to perform. Thus, what I mean when I say ‘I trust the babysitter’ is that I place the wellbeing of a child in the hands of the babysitter, and I am confident in doing this because I expect the babysitter will take good care of the child. Moreover, the domain of my trust in the babysitter is restricted, insofar as I do not trust them to do just anything. Epistemic trust is to believe something because another person has told us it is true. Epistemic trust requires that the truster ascribes sincerity to the trusted, but also competence in the matter about which the trusted is providing testimony. A similar domain restriction applies. I might trust a classicist to help me with Greek etymology but not trust them on matters of bicycle repair, even if I expect them to be sincere when answering questions in either area.[6] Both epistemic and practical trust differ from a third form of trust that I will call recommendation trust. I recommendation-trust someone when I believe I should do something because they have told me I should. The domain restriction also applies to the act recommended by the person I recommendation-trust: when I trust the recommendation of another it is because there is a particular area in which I take that person’s recommendations to be trustworthy.

As I am primarily interested in differences between recommendation trust and epistemic trust, I will stop short of a full account of these three forms of trust. A comprehensive definition of recommendation trust would need to do more to distinguish it from other, deceptively similar ways in which we form beliefs about what we should do. Say I am taking a trip by car with a friend. The friend recommends I avoid the M25 this afternoon because of the likely traffic. But I have good reason to think that taking a detour route will take just as long, despite the heavy traffic on the M25. There are a number of ways I might do as my friend suggests, in response to their suggestion, without thereby following their recommendation. Perhaps I avoid the M25 to humour my friend (I know they tend to be insecure about others not taking their travel advice). Perhaps I avoid the M25 because I know that my friend has a phobia of traffic jams and is too proud to cite this as the reason for asking that we take a detour. Perhaps my friend is an intimidating person, and I fear the consequences of questioning their advice. Doing as my friend recommends for any of these reasons is not trusting their recommendation.

Note also that the same principle applies to public trust in the recommendations of politicians and experts. If members of the public accord with the recommendations of government because they fear what the government would do were they not to follow its recommendations, then the public is not acting on trust, but on fear. Policy that depends on this form of compliance is precisely the coercion-based policy that earlier I suggested we would prefer to avoid. Moreover, compliance with public health recommendation is significantly different from compliance with law. I will restrict my focus to the question of when we have good reason to follow public health recommendations issued by experts.

One very common concern about epistemic trust is that to believe what we are told by experts we must believe without the warrant available to other kinds of belief (Burge 1995; Hardwig 1985, 1991; John 2018; Kukla 2007). Experts know about things I do not know about and sometimes could not know about. But many areas of expertise are areas in which it is important for non-experts to form true beliefs. Such beliefs must be formed without access to all of the evidence and without expert ability to understand the evidence. This threatens the possibility of both knowledge and rational belief about a wide array of basic facts. Do I know whether the Earth is flat without travelling? Should I believe that penicillin can be used to treat infection without first studying biochemistry? Knowledge of such matters depends on the rationality of beliefs grounded in the testimony of others, and the rationality of such beliefs is questionable in ways that the rationality of other beliefs is not.

John Hardwig’s influential solution to this problem was to argue for the following principle:

(T) If person A has good reasons to believe that person B has good reasons to believe that P, then A has good reasons to believe that P. (1991, 697–98)

Hardwig suggests that if such principles can be demonstrated, we can rescue testimony-based beliefs from scepticism. If (T) holds, Hardwig argues, then we can be epistemically justified in believing what others tell us is the case without having to investigate the evidence ourselves. If Hardwig is right, the principle also gets us part of the way to understanding how epistemic trust in experts could be rational. Say that I believe my doctor to be competent, well-informed, and sincere. This same doctor tells me that if a person’s heart stops outside of hospital there is a 20 percent chance that CPR will save their life. Provided I have good reason to believe the doctor to be competent, well-informed, and sincere, I have good reason to believe that if a person’s heart stops outside of hospital there is a 20 percent chance that CPR will save their life.

With much greater complexity, a novice can adopt a comparable relation of epistemic trust in public experts communicating facts about COVID-19. The greater complexity comes partly from the fact that it is very unlikely that most non-experts would have direct access to the testimony of experts. Most of us will read about what experts have to say about the pandemic as it is reported by journalists. Thus, for example, the question of whether I should (epistemic) trust UK chief medical officer Chris Whitty is made more difficult if, rather than watching the press briefings led by Whitty, I instead read reports based on the briefings in a national newspaper. Most obviously the problem of the rationality of my trust in this situation is compounded by the additional question of whether I can trust the newspaper; I must have reason not just to believe that Whitty is competent and sincere, but also to believe that the journalists working at the newspaper are competent and sincere.

Hardwig’s principle does not render epistemic trust easily won; there are many obstacles to rational epistemic trust.[7] It nonetheless does give us an account of how epistemic trust can be rational, and it is not implausible that its conditions for rational trust might sometimes be met. But even if the conditions set by Hardwig’s principle for rational trust are met, we have not thereby met conditions for rational recommendation trust. This much is evident when we attempt to adjust Hardwig’s principle to apply to recommendation trust.[8] When a person provides factual testimony, they communicate their belief that something is the case, and the question is whether this gives the receiver of testimony good reason to believe the same. When a person makes a recommendation, they communicate their belief that the receiver of the recommendation should do something. Accordingly, the analogous question is whether the recommendation gives the receiver good reason to agree that they should act as recommended. If we adapt Hardwig’s principle to apply to recommendations, we arrive at the following:

(R) if A has good reasons to believe B has good reasons to believe that A should act a certain way, then A has good reasons to believe that they should act that way.

But (R) is not true.[9] Consider again the doctor who has told me about the likelihood of a person being revived through CPR. Imagine that this same doctor follows up with a recommendation: given the low likelihood of resuscitation and the potential for distress if CPR is attempted unsuccessfully, it would be best if I signed a Do Not Attempt CPR (DNACPR) form. Grant that the doctor has good reasons to believe I should sign the DNACPR and that I still think my doctor competent and sincere. And set aside for the sake of argument the significant ethics violations involved in a physician making a recommendation on the decision to sign a DNACPR. Do I have good reason to sign?

Not necessarily. If the doctor’s good reason to think I should sign is that their employer has imposed perverse targets for DNACPR signatures, then this is not likely to be a good reason for me. Perhaps that is stretching what counts as a good reason for the doctor to recommend signing. But say instead that the doctor recommends signing because they have seen too many patients and families suffer the indignity of a failed CPR attempt at the end of life or that they think it is likely that patients whose lives are saved by CPR would only live for a few days more with broken ribs and severe pain. Though these are still questionable reasons for the doctor, they are not evidently bad reasons for them to believe I should sign the DNACPR. But even if they are good reasons for the doctor to think I should sign, they are not necessarily good reasons for me to think this. For I may, for example, have a deeply held conviction in the value of life or in the purpose of medicine to preserve life at all costs. The doctor’s good reasons are not necessarily even reasons that I accept but decide are overridden by other reasons (“I agree, but I consider other things more important”), for I may altogether repudiate the value of a dignified end of life. Thus, even if I am in a position, given the sincerity and competence of the doctor, to believe that they have good reasons to think I should sign the DNACPR, I do not thereby have reason to sign.

An additional condition must hold for it to be rational to trust in the recommendation of an expert. That additional condition is that the person has good reason to believe that the expert bases their recommendation on values that are held by the recipient of the recommendation. To be more specific, a recommendation is based on the values of another when B recommends an action to A because B believes that action is in A’s interests. The ambiguity of “interest” here helps to keep this condition flexible enough to cover actions that are valuable for a person in virtue of their desires (“Given you wish to lift heavier weights, I recommend you work on your shoulder mobility.”) and actions that are valuable for a person despite their occurrent desires (“Smoking is bad for you; stop it.”). This flexibility is important because without it we would be committed to thinking that one can never rationally trust a person who is trying to teach us to change our values. Sometimes we turn to experts not just because they can show us the means to our ends, but because we want their advice on the ends themselves.

We thus reach a more demanding version of a principle for recommendation trust:

(R*) if A has good reasons to believe B has good reasons to believe that a certain action is in A’s interest, then A has good reasons to believe that they should perform that action.

One way in which this is a more demanding principle than that which governs epistemic trust is that it requires that we have good reason to think that the expert issuing the recommendation understands what is in our interest. In some cases, this condition might be easily met. Most of the time when we ask a doctor for a recommendation we expect them to tell us what is good for our health, and we will be happy to defer on precisely what health means and what we should do to preserve it. But even in medical cases we will sometimes encounter situations in which we cannot take for granted that a competent doctor’s recommendations will be in our interest. The potential divergence of values in a decision about a DNACPR is one example of this.


The difference between epistemic trust and recommendation trust has ramifications for the measures appropriate to cultivating trust in experts. Much of the literature theorising ways in which we could help to build trust in science focuses on the sceptical problem posed by the decline of the value-free ideal of science (see, e.g., Anderson 2011; John 2018; Schroeder 2019). In doing so, this work focuses on how we can cultivate epistemic trust. There is no one answer to this challenge for epistemic trust in science, nor is this the only challenge.[10] To keep things manageable, I will focus only on the problem that values in science poses for trust, and to a selection of solutions to this problem. The question I raise is whether any proposed measures to cultivate epistemic trust in experts could also build recommendation trust. The purposes of this are twofold: to demonstrate that not all measures that help to cultivate epistemic trust help to build recommendation trust; and to arrive at a positive—though provisional—proposal for two measures to cultivate recommendation trust in experts.[11]

Consider first a popular argument against the value-free ideal of science, namely the argument from inductive risk (Douglas 2000; John 2016). Evidence gathered to test a scientific hypothesis typically underdetermines whether we should accept or reject the hypothesis because it fails to support deductive inferences either way. This introduces an inductive risk of either false positives or false negatives, depending on the level of caution we exercise in rejecting or accepting hypotheses. The level of caution we exercise cannot be grounded in epistemic reasons, and so we must appeal to non-epistemic reasons—primarily ethical values—to decide how cautious we will be.

Such judgements about the inductive risk we are willing to accept are particularly prevalent in our current emergency, where decisions need to be made quickly and data are scant. The parameters of our epidemiological modelling, for example, will partly be determined by non-epistemic reasons for whether we are more willing to risk overestimates or underestimates in infection and mortality rates. These decisions are sometimes said to be based on a trade-off between COVID-19 deaths now and deaths or decreased wellbeing in the future as a result of the economic impact of social distancing. However, the role of values is likely to be much more complex than this. Our values will affect, among other things, what we consider to be possible responses to a pandemic-generated recession, and political decisions will partially determine the outcomes of a recession, including who bears the brunt of its negative effects. Even setting the terms of the trade-off is influenced by our politics.

In response to such reasons to think science cannot be value free, some have called for greater transparency about the way in which values affect scientific results.[12] Greater transparency promises two benefits. First, transparency encourages novices to believe that scientists are honest about their procedures and findings. This belief in honesty is one of the crucial components in epistemic trust in science. Second, greater transparency could mitigate whatever tendency there might be among the public to think that the collapse of the value-free ideal of science warrants a blanket scepticism about all scientific claims. Values influence science in a variety of ways, some of which merit much less scepticism about the results of science than others. Heather Douglas (2008) has argued that there is a significant difference between direct and indirect influence of values on science. Values play an illegitimate direct role in science when they are used to warrant an empirical claim. Values play a legitimate indirect role in science when they determine what we accept as sufficient warrant for an empirical claim, with the warrant provided solely by epistemic evidence. Thus, for Douglas values should guide our judgements about acceptable levels of inductive risk, but they should not guide our judgement of the probability that an empirical claim is true. If greater transparency reveals that values play a legitimate role in science, then it can help to allay suspicion.

Transparency is by no means an unquestionably effective measure even for building epistemic trust (John 2018; Schroeder 2019). But aside from the problems that arise for using transparency to increase epistemic trust, there are three new problems that arise if we wish to use transparency to build recommendation trust. First, confidence in the sincerity of experts is not enough to meet conditions for rational trust in the recommendations of experts. An individual expert or an expert community could disclose as much information as possible about their methods, procedures, and even value assumptions that lead to their results, without giving the public reason to accept their recommendations. In order to have recommendation trust in experts, we need to know not just that their testimony about their area of expertise is sincere, but also that their recommendations are in our interest. Our confidence in this is not secured by transparency alone.

Perhaps this requires simply that our attempts to increase public transparency of science focus specifically on the values, goals, and ends that inform expert recommendations, thus allowing the public to judge for themselves whether the values of the experts align with their interests. But this fails to address the other two problems with using transparency to build recommendation trust. The second problem for transparency is that the way that values inform scientific results is very often not as simple as we need it to be in order to have reliable ways of communicating this to non-experts. A basic principle of precautionary reasoning used in epidemiological modelling of infection rate may be simple enough to communicate effectively to non-epidemiologists, but transparency about the political assumptions involved in economic modelling of the effects of the pandemic could be much more difficult to achieve. Indeed, the role that values play in setting the parameters for an expert community may not even be transparent to the experts themselves; not all scientists are philosophers of science. Revealing the values that influence scientific recommendations and expert-led policy might sometimes be possible, but often such transparency is highly impracticable.

But even if we can achieve transparency, it is not clear that it will benefit recommendation trust in the same way it can benefit epistemic trust. This leads to the third problem with transparency. One of the purported benefits of transparency is to show non-experts that values are playing the role they should play in science. Provided that everything is running as it should—that the science in question is trustworthy—transparency can allay suspicion that values play an illegitimate direct role. But this relies on a distinction between direct and indirect use of value judgements that does not apply to expert recommendations. Values must play a direct role in a recommendation because values must feature in the reasoning that supports the recommendation. There is thus nothing to be gained simply by showing non-experts what role values play in an expert recommendation, because there is no relevant distinction between more or less legitimate roles for values to play in such recommendations.

In fact, showing the public that their interests are aligned with the values informing science is not just an unhelpful and often unworkable step, but it is an unnecessary step.[13] For my trust in an expert recommendation to be rational, I need to have well-placed confidence in the alignment of the ends of the recommendation and my own ends. But confidence in this can be secured through means other than first learning about the values that inform the expert recommendations and then comparing them with my own. Instead of allowing the public to examine recommendations through transparency of values, we could instead secure confidence in the alignment of values through building public values into the procedures used to arrive at recommendations. If I have good reason to believe that my interests have played a significant role in the procedures used to arrive at expert-led policy, then I need not examine each element of the policy and each expert recommendation for alignment with my interests, because confidence in the procedure can provide confidence in the policy outcome.


How might we build public interests into decision-making procedures in a way that supports recommendation trust in experts? Many have proposed a range of measures for democratizing science to build public trust in experts, but, again, not all of these proposals will help us deal with the extra demands when building trust in expert recommendations. One popular option is James Fishkin’s deliberative polling (2009).[14] The deliberative polling process reaches policy verdicts through face-to-face deliberation between representative samples of the population, informed by relevant scientific information and bound by norms of appropriate discussion. The role of experts in the process is to explain the relevant scientific consensus, where there is one, and as much as possible the relevant concepts, methods, and standards of evidence used in reaching that consensus. Citizen participants are also informed of the most significant robust opposition to the consensus, with an opportunity to learn about the reasons that could be given for dissent, and to hear more from the experts about their defence against objections. Non-expert participants are thus given opportunities to resolve doubts as much as possible before coming to a collective conclusion about the correct policy decision, informed by the expert advice.

As Elizabeth Anderson has noted, one of the virtues of Fishkin’s deliberative polling model is that there is evidence that such a procedure mitigates the effects of cultural cognition, a social psychological phenomenon in which individuals are more likely to accept information that is compatible with their existing values and to deny information that challenges those values (Anderson 2011, 158). Interaction with participants with other political views and a much more extensive interaction with experts than would usually be the case can help to increase the chances that a non-expert will respond to the data regardless of whether they affirm previously held convictions.

But the distinctive challenge of generating recommendation trust in experts is not to overcome cognitive bias. Even if the outgroup prejudice against experts is overcome through deliberative polling, a non-expert participant could still be warranted in refusing to accept what the experts tell them when the experts are issuing recommendations and not just giving factual testimony. Deliberative polling could successfully disabuse a citizen of epistemic prejudice against expert testimony without thereby giving them a reason to accept that the same epistemic-trustworthy expert is issuing recommendations that are in their interest. If this is the case, as I have argued above, the citizen will have good reason to epistemic-trust the expert without having good reason to recommendation-trust them.

Perhaps deliberative polling is used to tackle the problem not by showing citizens they have good reason to recommendation-trust, but instead by eliminating the need for recommendation trust in the first place. It could be that in deliberative polling experts no longer play the role of issuing recommendations for policy, and instead they simply provide information for the non-expert deliberators to use in their collective policy verdicts. But though this might be valuable so far as it goes, it leaves us without a solution for the emergency situations in which expert-led policy, trusted by the public, is desirable and cannot be substituted with policy decisions made solely by novices. As I have argued in Section 2, this is precisely the kind of scenario we find ourselves in during the COVID-19 pandemic.

There is another way to secure confidence in the alignment of public interest and expert recommendation that relies on neither transparency nor novice participation in policy deliberation. We can look instead to the way in which science-led policy is communicated to the public. In addition to Fishkin’s deliberative polling model, Anderson (2011) proposes two other means by which we can overcome cultural cognition and persuade the public to become better disposed to trusting experts where indeed they have good reason to do so.

The first, “expressive overdetermination,” is borrowed from Braman, Kahan, and Grimmelmann (2005, 297): investing policy with multiple meanings such that it can be accepted from diverse political perspectives. Braman et al. cite French abortion law reform as an example of this. After decades of disagreement, France adopted law that made abortion permissible provided the individual has been granted an unreviewable certification of personal emergency. Evidence that such an approach would be effective pre-existed the new law, but such evidence had proved unconvincing to concerned parties. This new policy was sufficiently polyvalent to be acceptable to the most important parties to the debate; religious conservatives understood the certification to be protecting life, while pro-choice advocates saw the unreviewable nature of the certification as protection for the autonomy of women. With this framework in place, acceptable to all, opposing sides to the debate were able to converge on the details of implementation (ibid., 298).

Anderson’s second alternative is to recruit spokespersons who are identifiable to diverse groups as similar to them in political outlook. The strategy is to overcome cultural cognition by convincing citizens that the relevant scientific consensus is likely not to be a threat to their values because that same consensus is accepted by those with similar values. Anderson cites Barack Obama establishing links with Evangelical Christians such as Rick Warren, one of the 86 evangelical leaders who had signed the Evangelical Climate Initiative two years before the beginning of Obama’s presidency (Frazier-Crawford Boerl 2010, 153). She suggests that Obama’s public association with Warren was an attempt to win over conservative Christians to his climate-change policy. Braman et al. call this “identity vouching” (2005, 297).

Can these strategies for communicating science to sceptical publics also work for building recommendation trust? I believe they can, with the right caveats. First, given that we are looking for measures to build well-placed recommendation trust, it must be the case that the evidence-based policy that is communicated through expressive overdetermination and identity vouching is in fact in the interests of the public whose trust we are soliciting. These strategies could also be used to mislead the public into accepting recommendations from experts that are not in their interest, if, for instance, the spokespersons are dishonest about whether this is the case. But this would not be a strategy to cultivate well-placed recommendation trust, because the public would not have good reason to trust such recommendations.

Second, if we are to use these strategies specifically for recommendation trust, the goal of the strategies is different, and this will likely affect the details of how we implement them. Expressive overdetermination and identity vouching support epistemic trust by encouraging citizens to believe expert-testimony that would otherwise be resisted as a threat to those citizens’ values. The same strategies support recommendation trust only if they encourage citizens to see that the relevant expert recommendations are in their interests. But it is not difficult to imagine how these strategies could do this. Braman et al. (2005, 297) suggest that French religious conservatives were more willing to accept the evidence in favour of abortion law reform once they saw that the law affirmed the sanctity of human life. However, it is also likely that this adjustment gave those same religious conservatives good reason to believe that this policy was in their interests (“If you want a policy that respects the sanctity of life, this is the policy for you.”). Similarly, evidence-led policy could be presented to the public by spokespersons whose interests are already thought to align with the relevant members of the public. Thus, an American Evangelical Christian might think that if recommendations to cut carbon emission are good for Rick Warren, then they are good for them too.

With these caveats, using these communication strategies to cultivate recommendation trust is, I submit, preferable to other proposals. Expressive overdetermination and identity vouching avoid the problems of transparency measures: they demonstrate specifically that public interests align with expert recommendations, and  they do so without relying on novices (or even experts themselves) understanding the complex ways in which expert values shape scientific recommendations. And these communication strategies also provide an approach that is preferable to deliberative polling in times of crisis. When an emergency situation precludes a lengthy process of expert-informed citizen deliberation, expert-led policymaking might be necessary. Where this is the case, trust must be won by demonstrating to the public that expert recommendations are in their interest.


The account I have given offers reasons for a cautious optimism about the possibility of building well-placed trust in science-led policy. I am afraid I will conclude on a more pessimistic note. I have argued for the view that there are additional challenges to rational trust in experts when those experts are leading public policy and not just providing policy-relevant information. The additional demands of this kind of trust boil down to three claims that I have argued for: recommendation trust is different from both epistemic and practical trust; the conditions for rational recommendation trust are more demanding than the conditions for rational epistemic trust; and many measures that have been proposed to cultivate trust in experts do not help to cultivate recommendation trust in science-led policy. The measures that can help involve communicating science-led policy to the public in a way that allows the public to see, where it is indeed the case, that it is in their interest to follow the policy.

But the success of such a strategy is likely to be limited by factors that I have not had space to address. One such factor is the fragility of public confidence in the very idea of science-led policy. Where governments present their policy as technocratic, only for their claim to be “following the science” to be exposed as false, citizens will have good reason to be suspicious of future policy that is sold to them on a similar basis. Another significant factor is the broader climate of public trust, most notably the levels of public trust in politicians and government. Perhaps science-led policy is the product of experts and politicians working together. Perhaps science-led policy is the product of a full-blown technocracy, where experts become our political leaders. Either way, once science is corrupted by politics, its authority is vulnerable to public trust in the political system. Rebuilding this kind of trust is a much more difficult task.


Anderson, Elizabeth. 2011. “Democracy, Public Policy, and Lay Assessments of Scientific Testimony.” Episteme 8 (2): 144–64.

Atchison, Christina, Leigh Bowman, Charlotte Vrinten, Rozlyn Redd, Philippa Pristera, Jeffrey W Eaton, and Helen Ward. 2020. “Perceptions and Behavioural Responses of the General Public During the COVID-19 Pandemic: A Cross-sectional Survey of UK Adults.” medRxiv doi: 10.1101/2020.04.01.20050039

Braman, Donald, Dan Kahan, and James Grimmelmann. 2005. “Modelling Facts, Culture, and Cognition in the Gun Debate.” Social Justice Research 18 (3): 283–304.

Brennan, Johnny. 2020. “Can Novice Trust Themselves to Choose Trustworthy Experts? Reasons for (Reserved) Optimism.” Social Epistemology 34 (3): 227–40.

Burge, Tyler. 1995. “Content Preservation.” Philosophical Issues 6: 271–300.

Croce, Michel. 2019. “On What it Takes to be an Expert.” Philosophical Quarterly 69 (274): 1–21.

Douglas, Heather. 2000. “Inductive Risk and Values in Science.” Philosophy of Science 67 (4): 559–79.

–––––– 2008. “The Role of Values in Expert Reasoning.” Public Affairs Quarterly 22 (1): 1–18.

Fishkin, James. 2009. When the People Speak: Deliberative Democracy and Public Consultation. Oxford: Oxford University Press.

Frazier-Crawford Boerl, Christopher Wayne. 2010. “American Evangelicals and the Politics of Climate Change.” St Antony’s International Review 5 (2): 147–63.

Fricker, Miranda. 1998. “Rational Authority and Social Power: Towards a Truly Social Epistemology.” Proceedings of the Aristotelian Society 98 (2): 159–77.

Hardwig, John. 1985. “Epistemic Dependence.” Journal of Philosophy 82 (7): 335–49.

–––––– 1991. “The Role of Trust in Knowledge.” Journal of Philosophy 88 (12): 693–708.

Holton, Richard, and Jacopo Domenicucci. 2017. “Trust as a Two-place Relation.” In The Philosophy of Trust, edited by Paul Faulkner and Thomas Simpson, 149–260. Oxford: Oxford University Press.

Ipsos MORI. 2019. “Trust in Politicians Falls Sending Them Spiralling Back to the Bottom of the Ipsos MORI Veracity Index.” https://www.ipsos.com/ipsos-mori/en-uk/trust-politicians-falls-sending-them-spiralling-back-bottom-ipsos-mori-veracity-index (Accessed 15 May 2020)

John, Stephen. 2016. “From Social Values to P-Values: The Social Epistemology of the Intergovernmental Panel on Climate Change.” Journal of Applied Philosophy 34 (2): 157–71.

–––––– 2018. “Epistemic Trust and the Ethics of Science Communication: Against Transparency, Openness, Sincerity and Honesty.” Social Epistemology 32 (2): 72–87.

Kappel, Klemens. 2014. “Believing on Trust.” Synthese 191 (9): 2009–28.

Kitcher, Philip. 2001. Science, Truth, and Democracy. Oxford: Oxford University Press

Kukla, Rebecca. 2007. “How do Patients Know?” The Hastings Center Report 37 (5): 27–35.

O’Neill, Onora. 2020. “Trust and Accountability in a Digital Age.” Philosophy 95 (1): 3–17.

Pew Research Center. 2019a. “Public Trust in Government: 1958–2019.” Accessed May 15, 2020. https://www.people-press.org/2019/04/11/public-trust-in-government-1958-2019/.

–––––– 2019b. “Trust and Distrust in America.” July. Accessed May 15, 2020. https://www.people-press.org/2019/07/22/trust-and-distrust-in-america/.

Rolin, Kristina. 2002. “Gender and Trust in Science.” Hypatia: A Journal of Feminist Philosophy 17 (4): 95–120.

Schroeder, S. Andrew. 2019. “Democratic Values: A Better Foundation for Public Trust in Science.” The British Journal for the Philosophy of Science axz023. https://doi.org/10.1093/bjps/axz023.

Watson, Jamie Carlin. 2020. “Hunting the Expert: The Precarious Epistemic Position of a Novice.” Social Epistemology Review and Reply Collective 9 (4): 51–58.

Wilsdon, James, and Rebecca Willis. 2004. See-through Science: Why Public Engagement Needs to Move Upstream. London: Demos.

Wylie, Alison. 2003. “Why Standpoint Matters.” In Science and Other Cultures, edited by Robert Figueroa and Sandra Harding, 26–48. New York: Routledge.

[1] Ipsos MORI data on public trust in the UK in November 2019 showed that 14% of the public trusted politicians to tell the truth, matching previous recorded lows in 2016, 2011, and 2009 (Ipsos MORI 2019). Data from the Pew Research Center in April 2019 showed that 17% of respondents in the US trusted the government to do what is right at least “most of the time” (Pew Research Center 2019a), but this same figure has passed 50% only three times since 1972 (ibid.).

[2] Ipsos MORI’s data show that 86% of respondents said they generally trusted professors to tell the truth, 84% scientists (Ipsos MORI 2019). Pew Research Center data from 2018 showed that 83% of respondents said they have at least a fair amount of confidence in scientists, with very little (1%) variation between age groups (Pew Research Center 2019b, 19).

[3] What is an expert? The account of trust in experts I give here will remain neutral on debates about how best to define experts. However, I am inclined to what Michel Croce (2019) calls novice-oriented accounts because the experts I am concerned with are those that play the role of providing information and recommendations to non-experts.

[4] Where decisions really are exclusively within the remit of politicians, science-led policy can only be such if politicians trust experts. The norms governing when it is reasonable for politicians to trust experts, and the measures that we can take to build this trust, might well differ from those relating to public trust in science.

[5] It is not universally accepted that trust is tripartite. For more see Holton and Domenicucci (2017).

[6] The domain restriction can also reflect a principle that underpins some standpoint epistemology: some facts about, for instance, social and political injustice are better understood by those who suffer those injustices (Wylie 2003). Thus marginalized groups might be more epistemic-trustworthy in relevant domains.

[7] Hardwig is also not without his critics (see e.g., Kappel 2014; Rolin 2002). Nonetheless, the criticisms of Hardwig that I am aware of do not give us reason to adjust his principles in ways that would threaten the distinction in norms of rational trust that I am arguing for. Demonstrating this would, I suggest, take me beyond the scope of this paper.

[8] Note my argument in this section is not an argument against Hardwig, because Hardwig does not claim his principle would apply to anything other than epistemic trust.

[9] Does this mean that Hardwig’s principle (T) is not true? (R) appears to be a more specific version of (T) insofar as it specifies the nature of the belief professed by B and accepted by A. But if (R) is entailed by (T), and (R) is false, then (T) must also be false. Given that the truth of (T) is not at stake in this paper, I will remain agnostic on this issue, though I provisionally suggest that (T) might be saved were we to specify that B’s belief is about non-moral facts—this adapted (T) would not entail (R). I thank an anonymous reviewer for raising this.

[10] Other concerns include the difficulty for novices of identifying experts. See e.g., Brennan 2020 and Watson 2020.

[11] I am also setting aside matters of practical trust in experts, not to mention practical trust in government.

[12] See e.g., Kitcher (2001, chapter 6) and Wilsdon and Willis (2004).

[13] It could also be a dangerous step, as Schroeder (2019) has argued, insofar as it could encourage non-experts to be selective about the experts they prefer to follow, and as a result it might politicize science such that scientists and the public become divided along ideological lines.

[14] This has been endorsed as a means to build trust byAnderson (2011) and Kitcher (2011, chapter 8).

Special Issue, Uncategorized

How Government Leaders Violated Their Epistemic Duties during the SARS-CoV-2 Crisis

Eric Winsberg, Jason Brennan, & Chris W. Surprenant[1]

[This is an advance copy of an article that will appear in print in September 2020 as part of the KIEJ’s special double issue on Ethics, Pandemics, and COVID-19.]

ABSTRACT. In spring 2020, in response to the COVID-19 crisis, many world leaders imposed universal lockdowns. We argue that these restrictions have not been accompanied by the epistemic practices morally required for their adoption or continuation. While in theory lockdowns can be justified, governments did not meet and have not yet met their justificatory burdens. We will not argue that less stringent policies were optimal or better justified. Rather, we explain how government leaders failed and have continued to fail to meet their epistemic duties by relying upon data, models, and evidence of insufficiently good quality to justify their actions.

Sovereign is he who provides the exception.…The exception is more interesting than the rule. The rule proves nothing; the exception proves everything. In the exception the power of real life breaks through the crust of a mechanism that has become torpid by repetition. (Schmitt 2010, 1, 15)


In spring 2020, in response to the COVID-19 crisis, world leaders imposed severe restrictions on citizens’ civil, political, and economic liberties. These restrictions went beyond less controversial and less demanding social distancing measures seen in past epidemics. Many states and countries imposed universal lockdowns. Lockdowns, as we define them here, require people to stay home; in some countries and places, citizens must have ad hoc licenses to leave their homes for any reason. Citizens are often forbidden from playing outside, e.g., by jogging alone in the park. Citizens are forbidden from gathering in groups larger than ten, and in some cases they are forbidden from visiting friends and family even in small groups. Lockdowns do not merely prohibit large gatherings, such as conferences or concerts, but also prohibit small backyard parties. Most places of work are ordered to close, resulting in mass unemployment.

In this paper, we argue that these restrictions have not been accompanied by the epistemic practices morally required for their adoption or continuation. While in theory, lockdowns can be justified, governments did not meet and have not yet met their justificatory burdens.

This paper will not attempt to assess or determine which suppression mechanisms governments ought to have imposed, either in light of the information they had or have now. We will not argue that less stringent policies were optimal or better justified. Rather, our goal is to explain how government leaders failed and have continued to fail to meet their epistemic duties. We will argue that states relied upon bad data and flawed models, and they lacked the other kinds of evidence they would need to justify lockdowns. Again, we do not thereby claim that lockdowns were bad policy, nor are we assessing how dangerous COVID-19 is. Instead, we argue that most governments have failed to meet their epistemic duties.

As a partial analogy, imagine the state strongly suspects a person is a dangerous serial killer. Suppose there is indeed some evidence he is. To ensure he does not further endanger the public, they arrest and detain him. Months later, however, he remains in prison, yet the state has not convicted him; in fact, it has barely begun to collect the evidence it needs to demonstrate his guilt. Moreover, suppose we learn that the state has made demonstrable errors in its reasoning in accusing the person of the killings. Here, civil rights lawyers might well complain that the state has not met the epistemic obligations needed to hold the prisoner. This does not mean the lawyers necessarily deny the suspect is a killer. They may not even want him set free. But to justify infringing the suspect’s rights, the state needs to be more than factually correct: it needs to have strong epistemic grounds for its claims. For state agents to imprison someone without proper evidence is a severe ethical failing. Note that we are not, in this analogy, claiming that lockdowns are equivalent to imprisonment; our point is simply to provide an example where governments are required to possess a certain level of justification before they may restrict citizens’ liberties.


Liberal political philosophies regard liberty as the fundamental political value. All citizens possess an extensive sphere of individual liberty. Governments may restrict such liberties only in exceptional cases and upon meeting high justificatory burdens.

Consider John Rawls’s theory as an exemplar. Rawls’s theory claims that each person is “entitled to a fully adequate scheme of equal basic liberties…compatible with like liberties for all” (2001, 42). “Basic liberty” here is a technical concept, referring to a liberty which may not easily be overridden by concerns for social stability, economic efficiency, economic fairness, or general welfare. While non-basic liberties (such as the right to invest) may be restricted in order to promote other values (such as equity or welfare), basic liberties may not. Any reduction of basic liberties must meet standards of strict scrutiny. While trade-offs among the basic liberties are permitted, trade-offs between the basic liberties and various other social goals generally are not, except perhaps in extreme cases.

Rawls claims that not all liberties are basic. He defends an enumerated list of particular liberal freedoms, including liberty of conscience, freedom of thought, freedom of association, rights of due process and equal protection under the rule of law, freedom of occupation, and a right to own personal property (Rawls 1996; Freeman 2006, 46).

Of course, a major debate within liberalism is which liberties are “basic” in Rawls’s sense. We will not resolve this debate here. We simply remark that lockdowns restrict and reduce citizens’ basic liberties according to any major liberal theory.

The equivalent of the distinction between basic and non-basic liberties appears in most liberal democratic constitutions. For instance, in the United States, political speech is more strongly protected than commercial speech, while the right of free association for religious, political, or social purposes is more strongly protected than commercial freedom of association. For Congress to restrict citizens’ religious expression or to forbid their gathering for friendship or private events, the state must meet a high burden of justification, both in terms of the values it purports to promote through such restrictions, and in terms of the evidence it must give in support of any causal claims (Killion 2019).

Liberals have a variety of grounds for such views. Some appeal to the long run utility of rights (Mill 1859; Schmidtz 2008), others to autonomy, equality, and personhood (Rawls 1971; 1996; Gaus 2011). Others claim rights prevent state overreach (Spaulding 2009).[2]

Note, however, that the liberal position is not that basic liberties can be impeded or reduced only under conditions of strict scrutiny, while all other liberties can be reduced at will. Instead, all liberals believe in a “presumption of liberty” (Feinberg 1984, 9; Benn 1988, 87; Gaus 1996, 162–66; Rawls 2001, 44, 112; Gaus, Courtland, and Schmidtz 2018): Liberty is presumed to be normatively basic. By default, citizens are presumed free to do as they please, and by default, liberty does not need to be justified. However, any restrictions on liberty must be justified by appeal to various public values. Basic liberties can be restricted only if justifications survive strict scrutiny, while restrictions on non-basic liberties still require significant justification. The stronger the imposition and the greater the potential harm it imposes, the stronger the needed justification.

Our discussion here abstracts from these theoretical details in order to appeal to the generic principles that liberals share, and which supposedly undergird modern democratic nation-states. Liberals and constitutional democrats generally believe that (a) all restrictions on freedom must be justified, and (b) freedom cannot easily be overridden or silenced in the name of the common good, though some freedoms are more easily restricted than others. Further, liberals believe that (c) the justifications governments offer for overriding basic rights must be grounded in and appeal to public reasons and information that is appropriately available to all citizens.


Nevertheless, many liberals do believe that restrictions on basic liberties, including forced quarantine and social isolation (Parmet and Sinha 2020), are in principle justifiable. Even many libertarians, whose rejection of state interference is especially strong, share this view. For instance, although Robert Nozick argues that the state may not violate rights just because doing so produces better consequences, he suggests that rights may be violated to “avoid catastrophic moral horror” (Nozick 1974, 31). Along similar lines, Jessica Flanigan (2014; 2017) and Jason Brennan (2018) argue that mandatory vaccination can be justified. Although both reject paternalistic grounds for mandatory vaccination, they agree that states may mandate vaccines to prevent citizens from imposing undue risk onto others. Flanigan argues that firing a gun in the air over a crowded place imposes an undue risk of harm upon innocent bystanders (Flanigan 2014, 6). She claims that infected people who venture into crowds behave analogously. Brennan (2018) revises Flanigan’s argument by accounting for problems of uncertainty, collective action, and overdetermination, but reaches a similar conclusion.

Thus, even libertarian liberals, despite their anti-statism, often defend restrictions on basic liberties, particularly in the name of preventing harm. In the case of mandatory vaccination, this argument is made on the grounds that the people have no right to expose others to undue risk of infection. Considering these arguments, then, one might think it trivial to justify COVID-19 lockdowns on the same grounds. Such restrictions on liberty prevent citizens from exposing their neighbors to undue risk—potentially resulting in catastrophic moral horror as infections spread rapidly through the population—and thus they are justified from a liberal point of view.


Liberals and constitutional democrats agree that under the right conditions, states may restrict or remove people’s liberty, force them to accept medicines, deprive them of their jobs, imprison them, or even kill them. But in order to do these things justifiably, the state must meet certain conditions, including certain epistemic conditions. For example, it may not mandate an untested vaccine. It may not imprison a suspected killer without proving his guilt. It may not start a war concerning possible weapons of mass destruction on poor intelligence. It may not place citizens in internment camps on mere suspicion of disloyalty.

Epistemic norms are sometimes also moral duties (Chignell 2018). In some cases, individuals or groups have moral duties to collect good evidence, reason carefully about that evidence, engage in proper self-skepticism, and overcome their cognitive biases. This often occurs when one person is the fiduciary of another, when one person exerts significant power and authority over another, or when two people have certain contracts with each other. For instance, parents owe their children duties of care; these duties of care require parents to reason properly about issues related to their children’s welfare. Similarly, a financial advisor owes it to her clients to assess possible investment plans with high levels of competence and rational evaluation.

In recent work, Jason Brennan (2011; 2016) argues that governments have strong epistemic duties when making high stakes decisions. He motivates this idea with the example of a murder trial.

Imagine a defendant is charged with first degree murder. During the trial, both sides present evidence, question witnesses, and make arguments. The defendant will likely be executed or imprisoned for life if found guilty. Suppose the members of jury find him guilty. However, they are ignorant of the facts of the case, decide on the basis of false or pseudoscientific information, lack the cognitive capacity to understand the case, or process the information presented in the trial in irrational and biased ways. Alternatively, suppose they have improper motivations, such as malice toward the defendant, a conflict of interests, or simply want to please the judge and the press with a guilty verdict regardless of the defendant’s actual guilt.

If we knew a jury found the defendant guilty for any of these reasons, we would conclude they have acted unjustly. The jurors owe it to the defendant—and to society, as our representatives—to conduct a fair, impartial, and unbiased trial, and to reason in truth-conductive, reliable ways. In this case, we would conclude the jury’s decision should be thrown out and the trial conducted again. This judgment is reflected in the laws of many US states, which entitle a defendant to a new trial if he shows his jury had these problems.

We would not excuse the jury’s behavior if they claimed they acted on the best information available, but the available information was very bad. For instance, suppose the trial is held three days after capturing the defendant. Because of the lack of time, neither the prosecution nor the defense have much evidence for their side, and the evidence they have is of poor quality. In this case, if the jurors find the defendant guilty, they act wrongly. Saying they acted on the best available information is not sufficient justification. Instead, the evidence must meet an objective rather than relative standard; in this case, there must be no reasonable doubts that the defendant is guilty.

It would not be acceptable for a government to convict a person on the basis of poor evidence, and then collect good evidence later, after the fact. If critics complained about this behavior, it would make little sense for apologists to say, “Sure, everyone admits the government needs better evidence, which thankfully they are now, two months after the conviction, starting to collect.” The evidentiary bill comes due before conviction. Even if we discover later that the defendant was guilty, any liberal or constitutional democrat must nevertheless condemn the state’s behavior and demand the state follow pre-established rules of evidence in the future.

Liberals or democrats in the public reason tradition (e.g., Benhabib 2002; Christiano 2010; Eberle 2002; Estlund 2008; Freeman 2009; Gaus 1996; 2003; 2011; Habermas 1995; 1996; Larmore 2008; Rawls 1996; 2001; Tomasi 2001; 2012; Vallier 2018)—now the dominant paradigm in English-language political philosophy—hold that governments are subject to additional constraints. When they impose policies upon citizens, these policies must be justifiable to those citizens in light of certain publicly shared values and publicly available evidence which all reasonable citizens can accept. Governments are generally forbidden from acting on private, inaccessible, or non-public sources of information. They must instead appeal to widely shared values implicit in a democratic conception of personhood, which views everyone as free and equal (Rawls 1996; Gaus 2011). Public reason liberals in particular have reason to avoid claiming that citizens should blindly follow government leaders without demanding a public justification for their decisions.

Note that in using these analogies, we are not claiming that COVID-19 lockdowns are like imprisonment or punishment, though governments’ use of the word ‘lockdown’ does tend to push public rhetoric in that direction. Nor are we arguing that the appropriate remedy here is the same as in the case of a trial. In the case of a trial, if the state fails to meet its epistemic duties, the defendant goes free. We are not arguing that when the state fails to meet its epistemic duties, a quarantine must be immediately ended. Instead, Brennan argues that the point about the capital murder trial generalizes. When governments make high-stakes political decisions, decisions which can greatly harm people, or deprive them of livelihood, property, liberty, or even life, they are morally obligated to make such decisions competently and in good faith. What should happen when the state fails to meet its duties is a separate question that we do not address here. We simply argue that meeting the epistemic duty means relying on good information—not the best information available, but good information, period. The jury example motivates this intuition, but it generalizes to a wide range of political decisions. Below, we will explain why the COVID-19 lockdowns are “high stakes” in the relevant way, though we suspect this point is obvious.

Liberal democrats have good reason to endorse something like this in light of their own principles. They hold that certain liberties are basic and that liberty in general is normatively fundamental. Overriding, silencing, or forfeiting freedom requires that governments meet a strong justificatory burden. Governments must make such decisions using proper epistemic reasoning procedures, on the basis of good information, and while acting in good faith.

The foregoing comments provide the basic normative background of our argument. Appealing to ideas and principles shared within democratic or liberal traditions, we will show why governments have failed to meet the justificatory burdens required to legitimate the COVID-19 lockdowns. First, we will argue that the quality of the data and models used by officials was poor. We will argue that work on the philosophy of science and the reliability of experts gives us further reason to be cautious in deferring to such models. Second, we will argue (though this is far more obvious) that the decisions were extremely high stakes, imposing significant harms and costs upon people everywhere, especially those in extreme poverty. Together, this provides strong evidence that governments violated the Competence Principle and have failed to meet their justificatory burdens. We will not try to draw a precise line at which governments would meet their epistemic obligations to justify the lockdowns. Any precise line would be controversial. Instead, we will argue the information, models, etc., that governments used were sufficiently poor that they fall below any plausible line we might draw.

One might object to this entire line of argument by saying that while imprisoning a defendant is “high stakes,” so is letting him go. In the same way, lockdowns are high stakes—involving mass suppression of freedom of movement and association, serious psychological trauma, and severe economic loss—but refusing to impose lockdowns is also high stakes—as it could lead to serious death. First, as we have emphasized, we do not argue for the analog of “letting the suspect go.” We argue for no general form of remedy to the situation of states failing to meet their epistemic duties when they deprive their citizens of rights. Second, even though it is true that there is a parity of risks, it also misses the point. If one simply rejects the ideals of constitutional democracy or simply rejects liberalism, then the question of whether to impose lockdowns or not becomes a utilitarian issue. At the time lockdowns were imposed, the quality of information in support of any choice was quite poor (as we will explain below), and so from a utilitarian standpoint, it is just as difficult to justify staying open as it is to justify closing things down. But our point here is that constitutional democrats and liberals do not take all options to start on equal footing. They regard freedom as the default from which departures must be justified; the greater the imposition, the stronger the justification needed. While not all readers are liberals or constitutional democrats, these are nevertheless the dominant paradigms in political philosophy and actual political practice in the West.


           Cooper/Smith epidemiologist Dylan Green reports the following:

I’ve been asked to generate modeling results in a matter of weeks (in a disease which I/we know very little about) which I previously would have done over the course of several months, with structured input and validation from collaborators on a disease I have studied for a decade. This ultimately leads to simpler rather than more complicated efforts, as well as difficult decisions in assumptions and parameterization. We do not have the luxury of waiting for better information or improvements in design, even if it takes a matter of days. (Cowen 2020)

When epidemiologists model an emerging epidemic, data are sparse. In constructing their models to make forecasts, they have myriad methodological decisions to make, many of which are unconstrained by data or existing background knowledge.

Consider the Imperial College London (ICL) model, which had significant impact on policy decisions in the UK and US.[3] The model was used to estimate what public interventions would be needed to prevent hospital systems from becoming overwhelmed. The model’s primary job was to predict the impact of various policy choices on demand for hospital beds, intensive care unit (ICU) beds, and the like. Thus, the model needed inputs for the expected death rate, hospitalization rate, and ICU admittance rate for each 100,000 people infected. In all, the model employed almost 700 different parameters.

At the beginning of the COVID-19 epidemic, and even now as we write this sentence, these magnitudes were not well estimated. The WHO’s early estimates used case rates from China and other early areas of infection. But case rates are directly a product of surveillance /selection bias. When medical professionals predominantly test people who demand care, the resulting data are biased toward more severe results. Not all infected people become sick, and not all sick people need treatment, but the most severely ill people are most likely to seek treatment. Early WHO estimates were extremely high, with fatality rates as high as 3.4% and hospitalization rates well into the double-digit percentages. The correct numbers are still unknown, but early estimates were clearly too high.[4]

A bar graph comparing predicted weekly ICU cases. It is labeled "Figure 4: Illustration of adaptive triggering of suppression strategies in GB, for R-0=2.2, a policy of all four interventions considered, an "on" trigger of 100 ICU cases in a week and an "off" trigger of 50 ICU cases. The policy is in force approximate 2/3 of the time. Only social distancing and school/university closure are triggered; other policies remain in force throughout. Weekly ICU incident is shown in orange, policy triggering in blue."
Figure 1. (Ferguson et al. 2020)

The ICL model is a massive extension of so-called “SIR” models. SIR models divide an epidemiological population into three groups: Susceptible, Infected, and Recovered. SIR models excel at explaining, in retrospect, why epidemics tend to fit a familiar curve pattern. But to be useful for making policy recommendations, such as closing schools, ordering people in general to stay home, ordering the elderly in particular to stay home, closing restaurants, etc., SIR models must be considerably more complex. Each of the three main groups must be now divided—for example, into age categories, into those that stay at home, go to school, go to work, etc.

Let’s examine the ICL model in further depth. The ICL code creates a hypothetical random population for each country it models. Each individual is assigned to a household and, depending on age, to a school/university or workplace. (The sizes of these are chosen in proportion to their real values in the world.) The model is then simulated in six-hour steps; it determines the probability that each individual gets infected based on where s/he is in the model, and then randomly decides (against a background probability estimate) whether each individual is infected, and what happens to infected individuals (hospitalization, death, etc.).

There are a huge number of decisions behind such models. One must choose and code in a death rate, hospitalization rate, and rate of admittance to intensive care; one must choose a time step (which can have a large impact, since people move around in the world on a diurnal basis, and the time step is a significant fraction of the day), as well as the probability of infection at work, school, or at home.

For example, the ICL model assumes that when people socially distance, their probability of getting infected at home increases by 25%. But why 25%? Why not 35%? In fact, there is no data or research to support any particular choice in the model, since we have few well-established rates for any past virus, let alone rates for the novel SARS-CoV-2 virus. In retrospect, it appears now that SARS-CoV-2 is particularly virulent at home, relative to other places and relative to other viruses (Qian et al. 2020). There was enormous uncertainty concerning nearly every parameter built into the model’s coding. Most modeling choices were relatively unconstrained by data or background knowledge; when there was data, it was of poor quality. Our complaint here is not that the ICL model relied upon hundreds of parameters, but that the inputs into these parameters were largely arbitrary and unsupported by evidence.

A single run of the ICL model requires about 20,000 processor hours. It was impossible, on short notice, to explore how varying the (largely arbitrary) parameter values would impact the model’s predictions.[5] It was impossible to determine to what extent the model’s predictions were robust under varying parameter values. Indeed, now that a cleaned-up version of the model’s code is available, it is clear that the ICL model can generate significantly different estimates even with the same parameters inputted. In the end, therefore, the outcome of the simulation was highly dependent on the largely unconstrained choices that the modelers had to make, as well as on chance.

In a recent working paper, economists Christopher Avery et al. (2020) identify many other shortcomings of the major models, including the ICL model. These include failing to account for heterogeneity in degree of viral exposure, failing to account for endogenous behavioral changes (such as that people will self-isolate or reduce their contact with others as the disease spreads), a lack of parameters for hospital capacity, and a lack of parameters for underlying comorbidities. The authors complain, as we do, that many of the assumptions in the SIR and related models are ad hoc and unsupported by evidence, that the arbitrary choice of parameter values greatly changes the models’ predictions, and further that the data fed into these models suffer from heavy selection bias. They conclude that the “each type of model can be reasonably well-calibrated to an initial period of spread of disease, but further assumptions, often necessarily ad hoc in nature, are needed to extend either type of model to later phases of an epidemic” (Avery et al. 2020, 13).

It is no wonder, then, that the model performed poorly at anticipating ICU demand, which was at the heart of the policy recommendations that emerged from the model. Recall that the ICL scientists recommended a policy of “maximum suppression”(Ferguson et al. 2020). This was the most draconian set of policies the group imagined. They anticipated that even maximum suppression would at first barely avoid overwhelming the UK’s existing ICU and ventilator capacity, and it would then require cycling the economy on and off until a vaccine was available. Despite less than maximum suppression, this did not occur. It projected, for the US, that unless maximum suppression measures were used to “reverse epidemic growth, reducing case numbers to low levels and maintaining that situation indefinitely,” the US would experience over a million deaths (Ferguson et al. 2020). On March 20, 2020, Ferguson told reporter Nicholas Kristof that the US’s “best-case” scenario with moderate social distancing would be 1.1 million deaths (Kristof 2020).

A strong indictment of the ICL model comes from examining what it would have predicted for Sweden, which has not implemented any lockdowns. Of course, ICL never ran their model on Sweden, but the model has few country-specific inputs. A group of epidemiologists based in Sweden, Belgium, and the United States (Gardner et al. 2020, 31) ran a model very closely based on the ICL model[6] using parameters adjusted for Sweden’s population density, demographics, etc. They reported, “Our model for Sweden shows that, under conservative epidemiological parameter estimates, the current Swedish public-health strategy will result in a peak intensive-care load in May that exceeds pre-pandemic capacity by over 40-fold, with a median mortality of 96,000 (95% CI 52,000 to 183,000).” Their best-case estimate, if Sweden used maximal suppression and lockdown techniques, was that Sweden would have over 15,000 deaths by the end of April. Of course, Sweden is not actually suffering from overload of its healthcare system. According to their article, Sweden under its current policies should be crossing 70,000 deaths sometime in the next week. As of this writing on May 19, 2020, Sweden has experienced 3,743 deaths from COVID-19.

It’s unclear what the ultimate death toll will be in the US or the UK. But it is clear that this and other models’ projections of ICU and ventilator demand were overly pessimistic. (See Figure 1 above.) Note, also, that the ICL and most similar models did not make projections for deaths of individuals in nursing homes or other critical care facilities.

Another egregious (though not particularly exceptional) example of modeling failure can be seen in the projection stated by New York State Governor Andrew Cuomo on March 25, 2020, that, unless the state went into severe lockdown, it would need 40,000 ventilators by April 7, but that the best case scenario was a need for 40,000 ventilators by April 14. Here are the actual data according to covidtracking.com/data/

A chart, showing the two predictions of levels of ICU beds needed, compared to actual ICU beds used in New York. The predictions called for 40,000 beds needed by 4/7 or 4/14, and in fact, at most only 5,039 beds were needed.

Figure 2. (Adapted from https://twitter.com/ElonBachman.)

Here, Cuomo relied on a regression model of the kind used by the now famous Institute for Health Metrics and Evaluation (IHME). The IHME is a “mixed effects non-linear regression framework” (IHME 2020). It basically takes death, hospitalization, ICU, and ventilator data, as well as the date that particular location has gone on lockdown as inputs, and then fits it to a modified Gaussian curve that looks like the red and yellow lines in Figure 2. The projections of the IHME models of hospitals’ needs, ICU needs, and deaths, for each state, are now legendary for their poor performance and frequent massive updating. These failures have not been limited to New York. For example, consider the IHME’s central projections for how many ICUs the state of Florida would need on April 20, 2020. On the April 7 version of the projection, it was 2,409 units. By April 13, that number had fallen to 763. By the seventeenth, it had fallen again to 354.[7] These were not minor changes, given that Florida was said, at the time when the first projections were released, to have only 1,695 ICU beds available. So, the model predicted on the seventh a large and rapidly approaching shortage, which turned on the thirteenth into a long-delayed shortage, and then finally into a lasting surplus on the seventeenth. As one recent survey put it, “the true number of next day deaths fell outside the IHME prediction [95% confidence] intervals as much as 70% of the time, in comparison to the expected value of 5%” (Marchant et al. 2020). The IHME model has consistently performed far worse than chance, even as the modelers revise it in light of new data.

Given the internal deficiencies of the models being used to justify the policy responses to COVID-19 (such as lockdowns), we might hope that the models themselves (and the policy recommendations stemming from them) would be bolstered by empirical evidence from past pandemics. However, a literature search reveals there are no published, peer-reviewed papers demonstrating the effectiveness of universal lockdown procedures to combat any epidemic. To be clear, there are papers showing that closing schools reduces flu transmission in children (e.g., Chowell et al. 2014). There many papers demonstrating the effectiveness of centralized quarantines, in which infected individuals are confined in designated state facilities. But we lack empirical evidence that extensive lockdown policies or maximal suppression work at all, never mind that they are superior to other, less draconian practices. In Paediatric Respiratory Review, Rashid et al. (2015) survey and review eighty major studies examining various kinds of mild to moderate social distancing (though not lockdown) measures imposed in response to the 2009 influenza pandemic. They note that most papers conclude that social distancing measures are “moderately effective,” but at the same time, they find that “overall, the quality of the evidence was quite weak, drawing primarily on observational or simulated data.” Only one of the eighty papers used “more established methods” such as quasi-randomized control trials (Rashid et al. 2015).

The best paper we can find defending lockdowns is a working paper by Friedson et al. (2020), but this paper has significant limitations. In particular, it counts drops in deaths five days after California’s closing as evidence that lockdowns work. Since the virus takes longer than that to incubate, this drop could not have been caused by the lockdowns.

Issues like these are not unique to the field of epidemiology. On the contrary, we have strong grounds in general to be skeptical about experts’ predictions on hard problems. For instance, in Expert Political Judgment, Philip Tetlock (2005) examined nearly 83,000 predictions made by experts in a variety of fields. He focuses on what the experts themselves consider hard problems rather than easy problems. In general, he finds that on such questions, experts performed poorly, barely better than Berkeley undergraduates. Tetlock’s work warns us against simply “deferring to the science” on hard predictions, since the science in fact shows the scientists are bad at such predictions.

Basic liberties are not to be suspended lightly. Governments must meet high standards of evidence before doing so. We might debate just what the standards need to be to justify lockdowns. However, as the forgoing discussion shows, the actual quality of evidence was quite poor. No plausible theory claims governments may engage in the mass suppression of civil and economic liberty on the basis of poor evidence.


Why did so many expert epidemiologists fail so badly, or rely on speculative parameters within their models? Why did so many liberal democracies massively restrict their citizens’ civil and economic liberties on the basis of poor levels of information? Here, we turn from critiquing the models to reminding readers of what the literature on the philosophy of science tells us about the modelers themselves (Douglas 2000; 2009; Winsberg 2012; 2018; Parker and Winsberg 2018; Rudner 1953; Hempel 1965). Philosophers of science have long recognized that when scientists face unconstrained modeling decisions, their choices are often strongly influenced by their social and ethical values—as well as the various pressures the scientists are under. Insofar as the way we design our models has a strong effect on which policies the models will tend to make look attractive or unattractive, these underlying choices can play an important role in determining how useful a model is for guiding complex public policy decisions.

By way of illustration, suppose you would like to use a scientific model to help decide which of two policy choices you ought to implement. But in making the model, there are two ways you can proceed. On the first way of designing the model, the first policy option ends up looking attractive. On the second approach, the second policy option looks more attractive. Which model do you go with? What if both approaches seem reasonable on the basis of the limited evidence you have? This kind of dilemma is very common when model builders face methodological choices.

One way of resolving this dilemma is to ask which version of the model aligns the balance of inductive risks in the way that accords with your values. For instance, consider the ICL model and the choice of value for the parameter representing the probability of infection transmission at home while socially isolating. If you assume that the probability of getting infected at home goes up by 25% while socially isolating, this makes social isolation look far more attractive than if you assume that the probability of getting infected at home goes up by 35% while socially isolating. If you think that social isolation is the more prudent policy, because you think that risking losing lives to disease is a more serious risk than risking losses to the economy or to political freedoms, this may be reason enough to choose the former specification. The reader might think this is a small change. And indeed, maybe it is. But a model with almost 700 such unconstrained choices, each of which produces non-linear effects on the model output, creates a highly flexible model.

Alternatively, you might consider what will happen if you choose the wrong approach. Imagine that you are an epidemiologist who faces the kind of pressures that Dylan Green describes—asked to instantaneously deliver policy-defining predictions about a disease you know little about, with potentially hundreds of thousands of lives on the line. What are you to do, particularly when you have no independent evidence upon which to determine the correct value of that parameter? Any choice you make will reflect a value decision about the danger of overpredicting deaths vs. the danger of underpredicting deaths. We hope it will not be terribly controversial to say that epidemiologists, faced with Green’s pressures, are inclined to avoid underpredicting rather than overpredicting deaths (see, e.g., Green and Farahany 2014). They will be inclined to recommend policy choices that minimize the risk of death as opposed to, say, minimizing the chances of overreaction. Moreover, they will focus primarily on reducing the risk of death by disease—given that this is the subject of their expertise—and not on potential collateral damage resulting, for example, from hunger and dislocation that might result from an overly aggressive policy choice. The nature of the work they do directs their attention more to the damage caused by viruses like COVID-19 then to damage done by economic loss or reduction of political freedom. The consequences to themselves, their careers, their discipline, their own sense of moral culpability will be much larger if they underpredict rather than overpredict death by disease. It is the primary social role and responsibility of epidemiologists to focus on avoiding disease. It is their role to make their best guess when information is lacking. In contrast, it is the social role and responsibility of policymakers, and our political representatives, to make policy decisions that reflect the whole spectrum of our moral values. It is their role and responsibility to make the hard decisions and to take uncertainty and scientific ignorance into account.

Given these influences, it is unsurprising to find a great deal of evidence from past experiences that epidemiologists favor a balance of inductive risks that leads to over-forecasting the severity of diseases. The infection fatality rate of Mad Cow Disease, H1N1, H5N1, H7N9, and MERS all were considerably lower than what epidemiologists predicted. And while SARS 2002 actually ended up being twice as fatal as originally predicted, its infectious spread was tiny compared to what they predicted (Yu et al. 2013; Wang, Parides, and Palese 2012; Lipsitch et al. 2015; Cauchemez et al. 2014). Repeated cases of overprediction can even be diagnosed in single individuals. For example, Neil Ferguson, the famous epidemiologist behind the ICL model, has often overestimated disease dangers. To cite one example, he claimed in a 2001 New York Times article that it would be “unjustifiably optimistic” to think Mad Cow Disease would kill only a few thousand people; his group claimed it would kill around 136,000 (Blakeslee 2001). So far, the actual number of deaths, after 20 years, is under 200. In 2005, he told the BBC that the deaths from bird flu could be between 5,000,000 and 150,000,000; the actual number was around 300 (Sturcke 2005).

These reflections should give us pause in endorsing restrictions on citizens’ basic liberties that are rooted solely in expert policy recommendations during crises. Government leaders may claim that their actions are justified purely in deference to expert recommendations regarding SARS-CoV-2. But we have strong empirical evidence that experts in most fields are systematically awful at making predictions in difficult situations that require them to predict the effects of untried policy measures on a brand new, poorly studied, and poorly understood problem. As a general matter, the demand that we simply defer to what scientists tell us is based on a largely falsified theory of scientific expertise.

Even so, it might be appropriate, at the beginning of a potential catastrophe, for policymakers to adopt a very cautious stance. In doing so, it might be excusable to accept, provisionally, the extremely cautious predictions of epidemiologists, despite the problems in their data and models. It might be fine to act first and ask questions later. It should be stressed that even this concession is questionable—after all, governments must have strong and solid evidence, rather than poor evidence, that a potential disaster of a certain size is occurring in order to justify their behavior. Historically, “we must avert disaster” has been the main excuse for government overreach. But even so, as Nozick (1974) rightly observes, the potential to avert “catastrophic moral horror” through speedy action can license many responses that would normally go beyond the pale.

Regardless, this kind of justification will not do beyond the very short term. Even in the direst emergencies when immediate action is required, we expect policymakers to supply the needed justification shortly thereafter, to rely upon established standards of evidence, to rely on high quality evidence, and to show their work in which they balance various social and ethical values against each other.

For all the reasons outlined above, it will not do, in more than the very short run, for policy makers to declare, as Governor Newsom of California has done, that they are simply “following the science” in responding to a crisis like the COVID-19 pandemic. In the interest of transparency, they should make it clear that they are adopting precautionary reasoning and inform their constituents what the plan is to quickly move to a more substantive cost-benefit analysis—and explain what values are to undergird that analysis. But states are under more substantive obligations as well. They should begin collecting the data needed to properly assess their strategies and determine whether continued restrictions of citizens’ basic liberties are justified. The longer they neglect to take measures like these, the more their impositions look incompatible with the foundational commitments of liberalism.

Making decisions under uncertainty is hard. It is likely impossible to avoid over- or undervaluing various considerations depending on social mood and other similar factors. But one thing that can help mitigate the influence of individual scientists’ values on the advice they offer to policy makers is to follow established methodological standards.

Policy makers have a moral obligation, as soon as they are even considering restricting the political and economic rights of citizens, to immediately begin gathering the best and most systematic data available. We do not try suspected criminals in the absence of standards of how to evaluate DNA or fingerprint evidence. Likewise, we should not be reacting to fears of pandemics by limiting people’s rights in the absence of clear standards regarding how to collect and evaluate evidence of the severity of the threat we face from such a pandemic. In responding to the 2020 SARS-CoV-2 pandemic, however, Western governments have largely failed to put such standards in place, or even to collect evidence in a minimally adequate way.

Consider that it quickly became clear that SARS-CoV-2 case counts undercount actual infections. To some extent, this undercounting was inevitable: we could have easily predicted that some amount of infection would be asymptomatic, and the widespread lack of adequate testing capacity meant that tests could not be administered to all suspected victims. But it is indisputable that undercounting went beyond these factors. In every case where there have been “natural experiments” with SARS-CoV-2 infections—on cruise ships, navy ships, among women giving birth in hospitals, in testing people experiencing homelessness, prisons, etc.—it has been made clear that infection is much more widespread than case counts suggest, although we do not know quite by how much.

One very promising avenue for filling this hole is antibody serological testing. If you can test a representative sample of a population of people with such tests, and you know their rate of false positives and false negatives, you can very easily, with a reasonably large sample, get a very good picture of how much infection there is. But very little of this testing has taken place, and what testing has been done has failed to appease skeptics who have legitimate worries about how representative the sampling is. As a general matter, we know that small n studies will be biased toward false positives. Worse, we still have no clear sense of what the rate of false positives and false negatives of these tests are.

Yet, these problems would be relatively trivial for well-organized policy makers to fix. Around the world, governments have imposed unprecedented and dramatic restrictions on citizens’ civil and economic freedom. For instance, at least 30 million Americans have so far lost their jobs. Governments could easily have opened a few blood banks storing SARS-CoV-2-free blood and run 5000 tests on these antibody kits to determine their rate of false positives. They could have done extensive representative sampling of citizens in various locations around the country and sample the rate of infection using both blood testing and PCR testing. But governments have not done this, even now. They should have done much of this testing beforehand. The balance of civic considerations here makes little sense. It is as if generals decided to invade a foreign shore but chose not to acquire aerial photographs of the enemy’s defenses.

We have been criticizing the major public models and data which various world leaders reference as justifying their actions. We admit it is possible government leaders have private, classified, and otherwise non-public data and models of higher quality which would justify their actions. Nevertheless, we remind readers that in liberal, democratic, constitutional governments, acting on such private information and refusal to disclose such information is prohibited except in truly exceptional circumstances. We can understand not disclosing military secrets, but the SARS-CoV-2 is not a strategic actor which would take advantage of classified information. Governments must disclose their best information to the public.

Before moving on, it should be emphasized that while although we have criticized the quality of the COVID-19 data and the models which policymakers have used—and while we criticize policymakers’ deference to such models—our core contention is not that the danger of SARS-CoV-2 has been overstated, or that lockdowns were the wrong policy to adopt. Nor is our aim to establish what the optimal suppression strategy would have been in light of what information governments had. (Doing so would require an extensive cost-benefit analysis, which would take another paper’s worth of work at least.) Our concern is more procedural in nature. Whether or not governments have encountered correct information or adopted the right policies, the process by which they have made their determinations cannot be reconciled with basic liberal commitments. States must meet strong epistemic standards if they are to justifiably restrict their citizens’ basic liberties, and they have failed to do this. This failure cannot be dismissed by saying that the governments got it right in the end. By analogy, if we criticize a colleague’s data and evidence, her model, and her reasoning process, we are not thereby claiming to know the paper’s conclusion is false. A poorly researched paper could still have a correct conclusion. But without performing the appropriate epistemic work, our colleague would still be unjustified in drawing that conclusion, and we would be justified in criticizing her right to assert it.

Academics frequently make bold claims in journals or public opinion pieces, and the bolder an academic’s claims the more likely he will receive attention for his work. For academics, there are often no negative consequences for being wrong, even horribly wrong. But for policymakers, especially chief executives, the story is far different. State governors, mayors, and other chief executives can order their citizens to stay in their homes, to close their businesses, and otherwise make themselves dependent on the state for their survival because all of these orders can be backed up by overwhelming force. Even when policymakers do not issue formal orders but provide strong suggestions for how citizens should behave, these suggestions are taken seriously by most people and impact how they choose to live their lives.


As of April 20, 2020, governors of 42 US states have issued stay-at-home orders to slow down the spread of COVID-19 (Mervosh, Lu, and Swales 2020). Political leaders around the world implemented similar measures. Almost without exception, political leaders claim such drastic measures are necessary because people will otherwise die. But the lockdown has caused and will cause deaths as well—along with a range of other maladies. Deaths connected to layoffs that are the result of COVID-19 might already be in the same ballpark as the number of deaths caused by the virus itself (Cordle 2020), and, over the long-term, we are likely to see more deaths and a decreased life expectancy connected to rising unemployment (Forster 2018). Reports of child abuse and domestic violence have both increased significantly since the stay-at-home orders have taken effect (Taub 2020; Da Silva 2020). Hospitals are laying off staff and closing from a lack of revenue as most procedures are postponed. Deaths from untreated cancer will increase in the long-run. Many businesses deemed “nonessential” will also die because they have been forced to close, even if there was no good reason to close them—hobby shops, specialty food stores, cobblers and tailors, art studios, various factories. We do not equate the death of a business with a death of a person, of course. But for many business-owners, their businesses are not merely the means to support their families but also life projects from which they derive meaning and fulfillment. Further, mass job losses and workplace closures will have serious negative effects on citizens’ welfare. These decisions should not be taken lightly, especially as we do not know how to model the long-term economic effects of shutdowns.

The stakes are higher in poorer places. UN officials complain that the COVID-19 shutdowns may lead to “famines of biblical proportions” (McNamara 2020). Of course, such dramatic claims partnered with requests for money should be taken with a grain of salt. (After all, the same dynamics that led epidemiologists to overpredict the impacts of SARS-CoV-2 hold for the UN’s forecasts.) Nevertheless, the point remains that putting, say, 30 million relatively rich Americans out of work is one thing; putting those in extreme poverty out of work (while also possibly shutting down food supply chains) is another.

As we have emphasized throughout this paper, these mandates also impact civil liberties. Evacuation and shelter-in-place orders normally are issued when there’s an immediate threat that is visible or otherwise easily recognized by everyone in the community—natural disasters, active shooters, etc. For the COVID-19 pandemic, there were far more unknowns than knowns about the level of the danger when shutdown orders were given. Citizens will tolerate government restrictions to basic civil liberties from immediate, known dangers. But when we allow these restrictions even under circumstances where there are so many unknowns, we create conditions susceptible to abuse and oppression, especially for members of historically disadvantaged groups. And we are seeing this situation play out now.
Expanding the conditions under which the state is willing to impinge on civil liberties requires us to broaden the conditions under which agents of the state are directed to use force against citizens who are not complying with these mandates. In the US, this will often create situations in which citizens are subjected to police interventions. Invariably, some of these end in the death of citizens who are unarmed or otherwise doing nothing wrong. These encounters are especially dangerous for members of historically disadvantaged groups. Initial data surrounding the enforcement of COVID-19 orders have shown that these orders have been disproportionately enforced against minority citizens. In New York City, nine out of ten people arrested for COVID-19-related issues have been Black or Hispanic (Associated Press 2020). In Ohio, Black Americans were four times more likely to be charged with violating stay-at-home orders than White Americans (Kaplan and Hardy 2020), even though in Ohio White Americans make up 79% of the population while Black Americans make up only 12% (State of Ohio 2019). As more arrest data starts to trickle out in the coming weeks and months, similar data is likely to come out from cities across the US.

It’s clear that lockdown orders are high-stakes decisions which significantly harm certain people, impede their liberties, and deprive them of their livelihoods. They reduce people’s freedom to work, freedom to associate, and freedom of movement. To what degree they impede basic vs. non-basic liberties will vary from liberal theory to liberal theory. Our point here is simply that these are high-stakes decisions, and thus subject to the epistemic considerations we defended above.


Government officials must meet certain evidentiary standards before they detain someone. They must meet stricter standards before they arrest them. They must meet stricter standards to hold that person in prison before trial. They must meet even stricter standards to convict that person and imprison them over the long term. Even if one thinks a particular suspect is in fact guilty, it is nevertheless crucial in the name of preserving the rule of law and protecting constitutional rights to hold the government accountable if it fails to meet its epistemic duties. Likewise, in cases of such failure, it is important to demand better behavior in the future. Emergencies and dangers are often pretexts for government overreach and abuses of power, and it is precisely when the stakes are highest that government officials must use the best possible epistemic practices.

This paper offers a general indictment of government leaders across the world, though the specifics vary from leader to leader. The models and data used in support of lockdowns were poor. There was not sufficient evidence to justify lockdowns over other less restrictive policies. Governments did not and have not yet collected the data needed to continue their practices. Even if aggressive actions were initially excusable in the name of precaution, such pretexts cannot be sustained now that governments have broadly failed to remedy the deficiencies in their epistemic positions. Even if governments had acted on the best available evidence­ at the time—a highly controversial claim—nevertheless, the information and evidence available was objectively poor, as we argued above. To suppress liberty, they must act on sufficiently good information, not merely the best available information.

Again, we are not thereby making any claims about which suppression policies governments should have implemented in the short or longer term. We claim only that governments have systematically failed to meet their epistemic obligations in this crisis and that, for this reason, their actions cannot be reconciled with the values of a free society.


Associated Press. 2020. “9 Out of 10 People Arrested for Coronavirus-Related Offenses in NYC Have Been Black or Hispanic.” Time, May 12. Accessed May 14, 2020. https://time.com/5835815/nypd-coronavirus-related-arrests/.

Avery, Christopher, William Bossert, Adam Clark, Glenn Ellison, and Sara Fisher Ellison. 2020. “Policy Implications of the Spread of Coronavirus.” National Bureau of Economic Research, Working Paper 27007. http://www.nber.org/papers/w27007.

Benn, Stanley. 1988. A Theory of Freedom. New York: Cambridge University Press.

Benhabib, Seyla. 2002. The Claims of Culture: Equality and Diversity in the Global Era. Princeton: Princeton University Press.

Blakeslee, Sandra. 2001. “Estimates of Future Human Death Toll from Mad Cow Disease Vary Widely.” The New York Times, October 30. https://www.nytimes.com/2001/10/30/health/estimates-of-future-human-death-toll-from-mad-cow-disease-vary-widely.html.

Brennan, Jason. 2011. “The Right to a Competent Government.” Philosophical Quarterly 61: 700–24.

––––––. 2016. Against Democracy. Princeton: Princeton University Press.

––––––. 2018. “A Libertarian Case for Mandatory Vaccination.” Journal of Medical Ethics 44: 37–43.

Cauchemez, Simon, Christophe Fraser, Maria D. Van Kerkhove, et al. 2014. “Middle East Respiratory Syndrome Coronavirus: Quantification of the Extent of the Epidemic, Surveillance Biases, and Transmissibility.” The Lancet – Infectious Diseases 14 (1): 50–56. https://doi.org/10.1016/S1473-3099(13)70304-9.

Chignell, Andrew. 2018. “The Ethics of Belief.” Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/ethics-belief/.

Chowell, Gerardo (editor), Charlotte Hackson, Punam Mangtani, Jeremy Hawker, Babatunde Olowokure, and Emilia Vynnycky. 2014. “The Effects of School Closures on Influenza Outbreaks and Pandemics: Systematic Review of Simulation Studies.” PLoS One 9 (5): e97297.

Christiano, Thomas. 2010. The Constitution of Equality: Democratic Authority and Its Limits. Oxford: Oxford University Press.

Cordle, Vaughn. 2020. “Pandemic 2020: Layoff-related Deaths.” April 23. https://www.linkedin.com/pulse/pandemic-2020-layoff-related-deaths-increase-covid-19-cordle-cfa/?fbclid=IwAR18TeBQA3Lna61kdMPzPcWyrBnj70n6EP9qF7sFIL10JZlPdVeX9Mnt6Ok.

Cowen, Tyler. 2020. “More on Economists and Epidemiologists.” Marginal Revolution, April 19. https://marginalrevolution.com/marginalrevolution/2020/04/more-on-economists-and-epidemiologists.html.

Da Silva, Chantal. 2020. “Texas Hospital Child Abuse Cases Rise in COVID-19 Outbreak: ‘It’s Hard to Think That It’s Just Coincidental.’” Newsweek, March 22. https://www.newsweek.com/texas-hospital-child-abuse-cases-rise-covid-19-outbreak-1493642.

Douglas, Heather. 2000. “Inductive Risk and Values in Science.” Philosophy of Science 67 (4): 559–79.

———. 2009. Science, Policy, and the Value-Free Ideal. University of Pittsburgh Press.

Eberle, Christopher J. 2002. Religious Conviction in Liberal Politics. New York: Cambridge University Press.

Estlund, David. 2008. Democratic Authority. Princeton: Princeton University Press.

Ferguson, Neil M., Daniel Laydon, Gemma Nedjati Gilani, et al. 2020. “Report 9: Impact of Non-Pharmaceutical Interventions (NPIs) to Reduce COVID19 Mortality and Healthcare Demand.” Imperial College London. https://doi.org/10.25561/77482.

Feinberg, Joel. 1984. Harm to Others. New York: Oxford University Press.

Flanigan, Jessica. 2014. “A Defense of Compulsory Vaccination.” HEC Forum 26: 5–25.

–––––––. 2017. Pharmaceutical Freedom: Why Patients Have a Right to Self-Medicate. Oxford: Oxford University Press.

Forster, Nicky. 2018. “AP Analysis: Unemployment, Income Affect Life Expectancy.” AP News, December 18. https://apnews.com/66ac44186b6249709501f07a7eab36da

Freeman, Samuel. 2006. Justice and the Social Contract. Oxford: Oxford University Press.

––––––. 2009. “Constructivism, Facts, and Moral Justification.” In Contemporary Debates in Political Philosophy, edited by Thomas Christiano and John Philip Christman, 41–60. Oxford: Wiley-Blackwell.

Friedson, Andrew, Drew McNichols, Joseph J. Sabia, and Dhaval Dave. 2020. “Did California’s Shelter-in-Place Order Work? Early Coronavirus-Related Public Health Effects.” NBER Working Paper No. 26992.

Gardner, Jasmine, Lander Willem, Wooter Van der Wigngaart, Shina C. L. Kamerlin, Nele Brusselaers, and Peter Kasson. 2020. “Intervention Strategies Against COVID-19 and Their Estimated Impact on Swedish Healthcare Capacity.” https://www.medrxiv.org/content/10.1101/2020.04.11.20062133v1.full.pdf.

Gaus, Gerald. 1996. Justificatory Liberalism: An Essay on Epistemology and Political Theory. New York: Oxford University Press.

––––––. 2003. Contemporary Theories of Liberalism: Public Reason as a Post-Enlightenment Project. London: Sage.

–––––– 2011. The Order of Public Reason: A Theory of Freedom and Morality in a Diverse and Bounded World. Cambridge: Cambridge University Press.

Gaus, Gerald, Shane Courtland, and David Schmidtz. 2018. “Liberalism.” Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/liberalism/.

Green, Robert C., and Nita Farahany. 2014. “Regulation: The FDA is Overcautious on Consumer Genomics.” Nature 505: 286–7.

Habermas, Jurgen. 1995. “On the Internal Relation between the Rule of Law and Democracy.” European Journal of Philosophy 3 (1): 12–20.

–––––– 1996. Between Facts and Norms. Cambridge, MA: MIT Press.

Hempel, Carl G. 1965. “Science and Human Values.” In Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, 81–96. The Free Press.

IHME. 2020. “Forecasting the Impact of the First Wave of the COVID-19 Pandemic on Hospital Demand and Deaths for the USA and European Economic Area Countries.” Institute for Health Metrics and Evaluation. www.healthdata.org/research-article/forecasting-impact-first-wave-covid-19-pandemic-hospital-demand-and-deaths-usa-and

Kaplan, Joshua, and Benjamin Hardy. 2020. “Early Data Shows Black People are being Disproportionally Arrested for Social Distancing Violations.” ProPublica, May 8. https://www.propublica.org/article/in-some-of-ohios-most-populous-areas-black-people-were-at-least-4-times-as-likely-to-be-charged-with-stay-at-home-violations-as-whites.

Killion, Victoria. 2019. “The First Amendment: Categories of Speech.” In Focus, Congressional Research Service, www.crs.gov. URL= https://fas.org/sgp/crs/misc/IF11072.pdf

Kristof, Nicholas. 2020. “The Best-Case Outcome for the Coronavirus, and the Worst.” The New York Times, March 20. https://www.nytimes.com/2020/03/20/opinion/sunday/coronavirus-outcomes.html.

Larmore, Charles. 2008. The Autonomy of Morality. Cambridge: Cambridge University Press.

Lemoine, Philippe. 2020. “Are We Headed toward an Unprecedented Public Health Disaster?” Nec Pluribus Impar. March 21. https://necpluribusimpar.net/are-we-headed-toward-an-unprecedented-public-health-disaster/.

Lipsitch, Marc, Christl A. Donnelly, Christophe Fraser, et al. 2015. “Potential Biases in Estimating Absolute and Relative Case-Fatality Risks during Outbreaks.” PLoS Neglected Tropical Diseases 9 (7). https://doi.org/10.1371/journal.pntd.0003846.

Marchant, Roman, Noelle I. Samia, Ori Rosen, Martin A. Tanner, and Sally Cripps. 2020. “Learning as We Go: An Examination of the Statistical Accuracy of COVID-19 Daily Death Count Predictions.” ArXiv:2004.04734 [q-Bio, Stat], April. http://arxiv.org/abs/2004.04734.

McNamara, Audrey. 2020. “UN Food Agency Chief: World Could See Famines of ‘Biblical Proportions’ Within Months.” CBS News, April 22. https://www.cbsnews.com/news/coronavirus-famines-united-nations-warning/

Mervosh, Sarah, Denise Lu, and Vanessa Swales. 2020. “See Which States and Cities Have Told Residents to Stay at Home.” The New York Times, April 20. https://www.nytimes.com/interactive/2020/us/coronavirus-stay-at-home-order.html

Mill, John Stuart. 1859. On Liberty. London: John W. Parker and Son. https://archive.org/details/onlibertyxero00milluoft/page/n9/mode/2up

Nozick, Robert. 1974. Anarchy, State, and Utopia. New York: Basic Books.

Parker, Wendy S., and Eric Winsberg. 2018. “Values and Evidence: How Models Make a Difference.” European Journal for Philosophy of Science 8 (1): 125–142.

Parmet, Wendy E., and Michael S. Sinha. 2020. “Covid-19––The Law and Limits of Quarantine.” The New England Journal of Medicine 382: e28.

Qian, Hua, Te Miao, Li Liu, Xiaohong Zheng, Danting Luo, and Yuguo Li. 2020. “Indoor Transmission of SARS-CoV-2.” MedRxiv, April, 2020.04.04.20053058. https://doi.org/10.1101/2020.04.04.20053058.

Rashid, Harunor, Iman Ridda, Catherine King, et al. 2015. “Evidence Compendium and Advice on Social Distancing and Other Related Measures for Response to an Influenza Pandemic.” Paediatric Respiratory Reviews 16 (2): 119–26.

Rawls, John. 1971. A Theory of Justice. Cambridge: Harvard University Press.

––––––1996. Political Liberalism. New York: Columbia University Press.

–––––– 2001. Justice as Fairness: A Restatement. Edited by Erin Kelly. New York: Columbia University Press.

Rudner, Richard. 1953. “The Scientist Qua Scientist Makes Value Judgments.” Philosophy of Science 20 (1): 1–6.

Schmidtz, David. 2008. Person, Polis, Planet: Essays in Applied Philosophy. Oxford: Oxford University Press.

Schmitt, Carl. 2010. Political Theology: Four Chapters on the Concept of Sovereignty. Chicago: University of Chicago Press.

Spaulding, Norman. 2008. “The Rule of Law in Action: A Defense of Adversary System Values.” Cornell Law Review 93: 1377-1412.

State of Ohio. 2019. “Ohio Population Overview.” Report of the Development Services Agency. https://www.development.ohio.gov/files/research/P7001.pdf

Sturcke, James. 2005. “Bird Flu Pandemic ‘Could Kill 150m.’” The Guardian, September 30. https://www.theguardian.com/world/2005/sep/30/birdflu.jamessturcke

Taub, Amanda. 2020. “A New Covid-19 Crisis: Domestic Abuse Rises Worldwide.” The New York Times. April 6. https://www.nytimes.com/2020/04/06/world/coronavirus-domestic-violence.html.

Tetlock, Philip. 2005. Expert Political Judgment: How Good Is It? How Can We Know? Princeton: Princeton University Press.

Tomasi, John. 2001. Liberalism Beyond Justice: Citizens, Society, and the Boundaries of Political Theory. Princeton: Princeton University Press.

––––––. 2012. Free Market Fairness. Princeton: Princeton University Press.

Vallier, Kevin. 2018. “Public Justification,” in The Stanford Encyclopedia of Philosophy, ed. Edward N. Zalta. https://plato.stanford.edu/entries/justification-public/

Wang, Taia T., Michael K. Parides, and Peter Palese. 2012. “Seroevidence for H5N1 Influenza Infections in Humans: Meta-Analysis.” Science 335 (6075): 1463. https://doi.org/10.1126/science.1218888.

Winsberg, Eric. 2012. “Values and Uncertainties in the Predictions of Global Climate Models.” Kennedy Institute of Ethics Journal 22 (2): 111–37.

———. 2018. Philosophy and Climate Science. Cambridge: Cambridge University Press.

Yu, Hongjie, Benjamin J. Cowling, Luzhao Feng, et al. 2013. “Human Infection with Avian Influenza A H7N9 Virus: An Assessment of Clinical Severity.” Lancet 382 (9887): 138–45. https://doi.org/10.1016/S0140-6736(13)61207-6.

[1] Authors’ names appear in random order.

[2] See also In re Winship, 397 U.S. 358 (1970).

[3] See (Ferguson et al. 2020). See also (Lemoine 2020) on which we draw heavily for this section.

[4] In places where testing is extensive, even case fatality rates are lower than this. There is much debate about various serological testing, but little disagreement that it indicates an upper bound of at most 1%.

[5] In climate science this is called running a “perturbed physics ensemble” and it plays a central role in estimating model forecast uncertainty (Winsberg 2018).

[6] “We employed an individual agent-based model based on work by Ferguson et al. Individual-based models are increasingly used to model epidemic spread with explicit representation of demographic and spatial factors such as population distribution, workplace data, school data, and mobility” (Gardner et al. 2020, 8).

[7] Source http://www.healthdata.org/covid/data-downloads. This is not even a tiny bit cherry picked. We picked Florida because the author who researched this lives in Florida and the date for its humorous overtones. Any other state and date could have revealed a similar pattern.