When Virtues are Vices: 'Anti-Science' Epistemic Values in Environmental Politics

2.1 Air pollution and control of error rates

My first example comes from the work of S. Stanley Young, a statistician with ties to, among other organizations, two conservative think tanks, the Heartland Institute and the Hoover Institution; the pharmaceutical industry; several North American universities; the American Statistical Association; and the American Association for the Advancement of Science (“Who We Are: S. Stanley Young,” n.d.). In a series of recent papers funded in part by the American Petroleum Institute, Young and collaborators have criticized work in environmental epidemiology on the hazards of air pollution, especially particulate matter (PM) (Young 2017a; Young and Kindzierski 2019; Young, Acharjee, and Das 2019).

Young levels three related but distinguishable arguments. Here I consider just one, which appeals to the idea, from frequentist hypothesis testing, that corrections must be made for multiple testing to avoid inflating the false positive rate.⁵

Recall that frequentist hypothesis tests are designed around the control of error rates (Mayo 1996).⁶ The conventional 0.05 threshold for statistical significance corresponds to a false positive rate of 5%, that is, when a test with this threshold is used correctly, the probability of rejecting a true null hypothesis is 5%. However, when multiple tests are conducted in the context of a single study, the probability that at least one result is a false positive (assuming that all test results are statistically significant) is greater than 5%. For example, if two tests are conducted and both have p=.05 (observed), the probability that at least one result is a false positive is 1− .95² =9.75%. This is called a multiple comparisons scenario, and statisticians have developed various ways of correcting for multiple comparisons to control the error rate.

Young does not argue that the air pollution research he targets has incorrectly conducted multiple comparisons without correction. Rather, he argues that the studies could have conducted alternative analyses, and that this requires the same kind of correction. For example, for one study (Koken et al. 2003), Young observes that the study includes 5 different health outcomes, 6 different predictors (variables of interest, i.e., pollutants), 5 different lags (values of predictors from previous days in the time series), and 5 different covariates. Young calculates that there are 150 = 5×6×5 different “questions” that could be asked in this study (is predictor x with lag t correlated with outcome y?); 32=2⁵ different “models” that could be used to answer each question (one for each subset of covariates included as controls); and all together that the “search space” for this study is at least 4800=150×32 “question-model” pairs (Young 2017a, table 1, 178). Young argues that the p-values reported in environmental epidemiology studies should be corrected for these (potential) multiple comparisons. For the studies scrutinized in Young (2017a) and Young, Acharjee, and Das (2019), the “search space” is generally on the order of 10⁴ to 10⁶.

Based on this line of argument, Young and his collaborators have concluded that “causality of PM10/PM2.5 [particles smaller than 10 or 2.5 microns, respectively] on heart attacks is not supported” (50), “There is no convincing evidence of an effect of PM2.5 on all-cause mortality” (Young and Kindzierski 2020), and indeed that “regulation of PM2.5 should be abandoned altogether” (Young 2017b). Young’s potential multiple comparisons argument exemplifies the epistemic value of skepticism, in the sense of requiring strong evidence before accepting a substantive claim.

However, in this case, skepticism imposes a demanding burden: some rough calculations assuming Bonferroni corrections for multiple comparisons suggest that, for one paper targeted by Young (Koken et al. 2003), collecting sufficient data to satisfy this demand would take well over a century. And this study has a relatively small “search space.” Thus, satisfying the epistemic value of skepticism—and, in the meantime, accepting the claim that PM does not cause heart attacks—would effectively block air pollution regulation. Skepticism contrasts with a precautionary attitude, a willingness to accept a claim and take action accordingly, based on relatively provisional, weaker evidence (Steel 2015b, 9–10).⁷ A precautionary attitude is a pragmatic value: it leads us to take action earlier to address possible threats to human health (ch. 4).

Thus, this case gives us an example of a tension or conflict between the epistemic value of skepticism and the pragmatic value of precaution.

2.2 Open science and regulatory delay

My second example is a strategy of appealing to transparency, open data, and open science to delay environmental regulation. The strategy appears to have originated in the tobacco industry in the 1990s (Lerner 2017) and has been adopted by several industries whose products pose environmental threats to public health. I first discuss a case in which this strategy was used to attack an individual study. I then discuss some more general efforts, in the US Congress and US Environmental Protection Agency (EPA), which would have imposed open science requirements on all science used in US EPA’s regulatory decision-making.

2.3 Anti-science epistemic values?

Like the critics of Strengthening Transparency, I am strongly inclined to say that something has gone wrong when open science is used to argue against environmental regulation. Young’s concerns about control of error rates in environmental epidemiology have not been widely discussed, but I expect the environmentalist critics of Strengthening Transparency would say that his argument goes wrong in the same general way. Both of these arguments are “anti-science,” even as they appeal to widely shared epistemic values. How can we make sense of this thought? In this subsection, I consider two approaches, one from Karen Kovaka and the other from the literature on “normatively inappropriate dissent.” I then sketch my approach.

In an analysis of climate denial, Kovaka (2021) observes a common argument pattern:

1. In order for us to accept its results, science must meet the standards for good science.

2. Climate science does not meet these standards.

3. So we should not accept the results of climate science. (2364)

Kovaka considers the possibility that the standards in question “are widely accepted …, shared by accepters, deniers, experts, and non-experts.” But she rejects this as “implausible” (2364), and argues that “pro-science science-denial” is more likely to be based on common misconceptions about science, including the value-free ideal (2365), the idea of “a universal, step-by-step scientific method” (2366), and “the mythic treatment that scientists such as Galileo, Mendel, and Darwin receive in science textbooks” (2368).

Swapping out climate science for environmental epidemiology, the argument pattern that Kovaka identifies also seems to fit Young’s critique and the various calls for publicly-accessible data. But, while Kovaka’s diagnosis of common misconceptions might be apt for climate denial, I don’t think it will work for the cases that I am interested in here. Control of error rates and open science are widely-accepted by both philosophers of science and scientists themselves.

Another recent line of work has concerned the concept of “normatively inappropriate dissent” (Biddle and Leuschner 2015; Melo-Martín and Intemann 2018; Miller 2021). It might be thought that our two examples of “anti-science epistemic values” can be covered by this concept. Both Young and proponents of Strengthening Transparency do appear to be dissenting from the scientific mainstream, in ways that their critics think is inappropriate. I consider two accounts of normatively inappropriate dissent, both based on the concept of inductive or epistemic risk. First, Biddle and Leuschner (2015) give a set of four jointly sufficient conditions for dissent to be normatively inappropriate. However, one of their conditions—which Miller (2021) labels Standards—is that “The dissenting research that constitutes the objection violates established conventional standards” (Biddle and Leuschner 2015, 273). Again, in the two case studies of interest in this paper, the dissenters are appealing to established conventional standards. So Standards is not satisfied. Because Biddle and Leuschner’s criteria are jointly sufficient, not individually necessary, their account does not logically entail that this dissent is not normatively inappropriate. But this just means that their account does not have much to say about these two cases.

Second, Miller (2021) offers a revised version of the criteria from Biddle and Leuschner, to address criticisms from Melo-Martín and Intemann (2018). Miller’s account involves three individually necessary criteria for normatively appropriate dissent; so, any case of dissent that violates at least one criterion is inappropriate. Based on concerns about excessively lax conventional standards, Miller replaces Standards with a criterion he calls Workable Evidential Thresholds: “The dissent adopts evidential thresholds within a range that allows researchers to make meaningful knowledge claims and that is not knowingly likely to rule out a priori the attainment of certain empirical results” (Miller 2021, 925). Both Young and proponents of Strengthening Transparency present themselves as supporting evidential standards that allow researchers to make meaningful knowledge claims, so this clause seems satisfied. Do they “rule out a priori the attainment of certain empirical results,” such as conclusions about the hazards of air pollution? Miller’s criterion is ambiguous here. On the one hand, the proposals from these dissenters would significantly delay drawing such conclusions. But, on the other hand, they would not make drawing such conclusions impossible. So whether or not this dissent is appropriate depends on the importance of timeliness. From a point of view that understands the aim of science in narrowly epistemic terms, timeliness is likely to seem less important than skepticism and transparency, and so this dissent might seem appropriate. I discuss this, and a rival point of view of the aims of science, below.

One of Miller’s (2021) other criteria, Fairness, might seem more promising: “The inductive risks that would stem from accepting or adopting the dissent, and the respective harms and benefits that they entail, would be fairly and justly distributed among relevant members of or groups in society” (921). The mainstream view is that air pollution, and other environmental hazards, are highly unjustly distributed (Buzzelli 2017). But the evidence for this unjust distribution uses much the same methods as the evidence for the hazards of air pollution (333ff). So assessments of the inductive risk of, say, Young’s potential multiple comparisons argument depend on whether we find his argument plausible. From Young’s point of view, it’s reasonable to think that distributive environmental justice research is highly unreliable, we should remain skeptical about how air pollution (if it turns out to be hazardous) is distributed, and so the inductive risks of adopting more stringent research standards are not necessarily unjustly distributed.

I suggest taking another approach to analyzing our cases of “anti-science epistemic values.” The underlying issue in these cases can be understood as a tension between epistemic and pragmatic aims of science. On the one hand, control of error rates and open science are—both generally and in the particular cases of environmental epidemiology presented above—put forward as epistemic values, that promote the aim of accepting true claims about the world. On the other hand, environmental epidemiology also has pragmatic aims. Environmental epidemiology is valuable because it is useful for doing things in the world. Specifically, it is useful for designing regulation to protect human health. If, on the account that I am proposing, it makes sense to say that environmental regulation is (the? a? among the?) aims of science (in general? certain fields?), then it seems to make sense to say that things (anything? including epistemic values?) that work against environmental regulation are “anti-science.” Such things would be scientific vices, rather than virtues—they would frustrate, rather than promote, the aims of science. So (perhaps) epistemic values can be vices.

As the parentheses suggest, developing this account will acquire some work. I begin that work in the next section.

3.1 The narrow view

The narrow view seems to provide us with a simple account of the relationship between epistemic aims and pragmatic goods: the sole aim of science is knowledge, and so, any pragmatic goods that science produces are mere accidents. Thus, when there is a conflict between them, pragmatic values should be sacrificed for the sake of epistemic values. To return to the example of Strengthening Transparency, from the narrow view, open science requirements promote transparency, an epistemic value, and so promote the epistemic aims of environmental epidemiology. While it may be regrettable that requiring open science would delay protective regulation and result in preventable harms to human health and the environment, this is an unhappy accident. Similarly, for the conflict between the epistemic value of skepticism and the pragmatic value of precaution in Young’s concerns about potential multiple comparisons in air pollution epidemiology, the narrow view seems to imply directly that precaution should be sacrificed for the sake of maintaining skepticism. The epistemic aims of environmental epidemiology take priority over any merely accidental practical application. Doing otherwise would be just like engaging in research misconduct for the sake of prestige.

This simple analysis becomes more complicated when we recognize that science has a plurality of epistemic aims.¹⁰ Magnus (2013) draws on William James to distinguish the aim of believing the truth from the aim of not believing falsehoods. In these terms, skepticism can be understood as an epistemic value because it promotes the epistemic aim of not believing falsehoods. But precaution can likewise be understood as an epistemic value because it promotes the epistemic aim of believing truth. As Fraser (2020) puts it, “Sure, if you only believe things for which you have good evidence, you are less likely to have false beliefs. But you will also miss out on a lot of true ones.” On this analysis, Young’s potential multiple comparisons argument would protect us from believing falsehoods, but also prevents us from believing truth claims. So it raises a conflict between two epistemic values. Steel (2016) makes a related point, arguing that the problem with climate skeptics is that they delay knowledge. As Steel sees it, the tension is between modestly-secure knowledge now and highly-secure knowledge in the future. We can apply this to the conflict between transparency—securing our knowledge claims in the future, after other scientists have scrutinized a study’s data and analysis code—and timely but less secure knowledge claims that can be used by policymakers now. In the debate surrounding Strengthening Transparency, both timeliness and transparency can be understood as epistemic values.

This analysis is plausible, and so far fits neatly within the narrow view of the aims of science. But it doesn’t seem to provide any resources for resolving the tensions between these values. Should we choose believing truth or avoiding error, knowledge now or knowledge later? Notably, Magnus, Fraser, and Steel authors make their points to introduce arguments that what are usually called “non-epistemic values” can legitimately inform the judgments we use to resolve these kinds of tensions. Magnus presents William James’ observation as an early version of an inductive risk argument (Magnus 2013). Fraser argues that “epistemic fear of missing out” among conspiracy theorists leads to a lack of shared belief that is necessary for democratic public deliberation. And Steel argues for a generalized version of inductive risk, “from risk of error to risk of ignorance” (Steel 2016, 705) where the problem with ignorance is that “prolonged suspension of judgment … leads to serious difficulties if policy decisions are expected to be ‘science based’ in the sense that the course of action chosen is justified by scientific knowledge” (706–7). That is, Magnus, Fraser, and Steel identify conflicting epistemic values, and go on to argue that the conflict should be resolved by appeal to non-epistemic considerations.

At this point, it seems to me that the narrow view has followed Magnus, Fraser, and Steel into a box canyon. The narrow view can recognize that science has a variety of epistemic aims, but this recognition does not help us resolve tensions between different aims. Magnus, Fraser, and Steel provide a way forward. But their routes appeal to what the narrow view regards as accidents or by-products of scientific research. It may be appropriate to bring in these kinds of considerations, if there really is no way to move forward on epistemic grounds. But this kind of route should be taken only with great reluctance and care. Specifically, on the narrow view, pragmatic considerations should not influence our conception of the epistemic aims of science—to allow this would be to risk “corrupting” or “infecting” the practice.¹¹

3.2 The broad view

Like the narrow view, the broad view takes science to have epistemic aims, plausibly including some combination of accepting true claims, rejecting false claims, and achieving understanding, perhaps among others. The broad view adds that science also has pragmatic aims.¹² For example, epidemiologists and public health researchers regularly characterize their fields in terms of pragmatic aims, usually some version of protecting or promoting human health:

Stated another way, the objective of epidemiology, as long-argued by many leading epidemiologists, and as underscored in the “Ethics Guidelines” issued by the American College of Epidemiology in 2000, is to create knowledge relevant to improving population health and preventing unnecessary suffering, including eliminating health inequities. (Krieger 2011, 31, citations to 10 works removed; see also Krieger 2015, 5–6)

As epidemiology is one of the essential disciplines of public health, its major aim is to contribute to fulfilment of the definition of public health as “a science and art to promote health and prevent disease by organized effort of society” (Gulis and Fujino 2015, 179; see also US EPA 2017; Nature Communications 2018; Fernández Pinto and Hicks 2019)

In light of these self-characterizations, I propose that the primary pragmatic aim of environmental epidemiology is protecting human health. This means that the benefits that environmental epidemiological research produces for protecting human health are not accidents or by-products of that research. Rather, they are among the primary reasons for pursuing that research. If environmental epidemiology is producing research that helps protect human health, then it is being done well (flourishing). At the same time, if environmental epidemiology is not (or prevented from) producing research that helps protect human health, then it is not being done well (degenerating).

On the broad view, a scientific field typically promotes its pragmatic aims by achieving its epistemic aims. Environmental epidemiology protects human health by producing knowledge of the distribution and causes of environmental-related injury and disease. In this relationship, epistemic aims are proximate or direct, while pragmatic aims are distal or indirect. In their everyday work, scientists will often attend exclusively to epistemic aims, but pragmatic aims provide a broader context and purpose for these everyday activities.

Borrowing a pattern from action theory (Boesch 2019, 2311–12; MacIntyre 1973; and compare the “significance graphs” of Kitcher 2001, ch. 6), consider the following exchange of questions and answers about the research activity of an environmental epidemiologist:

Q: Why are you writing code in R?

A1: To fit a linear regression model.

Q: Why are you fitting a linear regression model?

A2: To understand how air pollution exposure is correlated with race.

Q: Why are you trying to understand how air pollution is correlated with race?

A3: To reduce racial disparities in asthma.

Using an analogous example, Boesch (2019) points out that the answers (A1-A3) “are not separate actions, but rather separate descriptions of one and the same action” (2312). In this example, it’s not that the scientist is currently (in some “core” or “epistemic” phase, stage, or context) engaged in fitting a model, then later will pursue the epistemic aim of understanding air pollution and race, and only subsequently (in a distinct “applied” or “practical” phase, stage, or context) will pursue the pragmatic aim of reducing racial health disparities. Rather, the activity of fitting the model is (part of) the activity of understanding air pollution and race, which is (part of) the activity of reducing racial health disparities (Hicks 2014, 3289). Fitting the model is pursuing both epistemic and pragmatic aims. And so it’s appropriate to evaluate this activity (and the resulting model) in terms of whether and to what extent it promotes both epistemic and pragmatic aims.

This action-theoretic pattern is focused on individual actors. As noted in section 1, I am not focused on individual actors here, because I think in most cases the available evidence is inadequate to make claims about individuals’ intentions.¹³ For example, alternatives to A3 might include:

• to publish a high-impact paper;

• to get tenure;

• to get my next grant;

• to win the Nobel prize in medicine;

• to delay regulation (because it would harm my industry funder’s profits);

• to delay regulation (because I’m opposed to regulation on philosophical grounds);

• to delay regulation (because environmentalism is a socialist conspiracy); and/or

• to accelerate regulation (because it’s one of the goals of my nonprofit funder).

A research community is composed of individual researchers, with diverse material interests. Even members of a close-knit community—such as the lab of a single faculty member and their graduate students—might take the series of why questions in discordant directions. So why think that something like the entire field of environmental epidemiology could have a remotely coherent set of aims?¹⁴

Consider the quotations from epidemiologists above. I suggest that these statements of the pragmatic aims of epidemiology and public health are best read normatively, rather than descriptively. That is, the authors are not claiming that all or even most individual epidemiologists would agree that promoting human health is the (primary) aim of epidemiology. Rather, I suggest that these authors are claiming that promoting human health should be regarded as the aim of epidemiology.

This does not mean that epidemiologists will all agree on a specific list of aims for their field, especially a fine level of detail. Krieger (2011), a textbook in “epidemiological theory,” is organized historically as a series of debates about conceptualizations of public health—from miasma to germ theory to the social determinants of health. These debates all presumed that epidemiology should promote “health”; the disagreement was over how this key concept should be conceptualized. Similarly, one can be a pluralist about the conceptualization of human health in epidemiology while still maintaining that promoting human health (understood generically or ambiguously) is the overarching pragmatic aim of the field.

This also does not mean that every individual epidemiologist must have the exclusive intention to promote human health. There is no necessary tension between working to promote human health and working to get tenure. Tenure could even be seen as a means to promote human health, by giving researchers secure positions from which to do their work. Indeed, an epidemiologist might even work for an industry sponsor to design changes to production processes that preserve profits while also protecting human health.

But job security and industry profits are not—that is, should not be regarded as—aims of epidemiology. They are mere accidents. So if an epidemiologist sacrifices the protection of human health for the sake of getting tenure or their funder’s profits, then they have sacrificed the pragmatic aim of epidemiology for the sake of mere accidents. And so they have acted viciously.

The analysis of the last several paragraphs does not apply to the arguments presented in section 2. If we take those arguments at face value, they are not cases in which there’s a tension between the aims of environmental epidemiology and mere accidents, such as industry profit. Rather—again, taken at face value—they are cases of conflicts between epistemic and pragmatic aims. Skepticism and transparency promote the epistemic aims of environmental epidemiology, while precaution and timeliness promote the pragmatic aims. On the broad view, the conflict between epistemic and pragmatic values in these cases are conflicts between epistemic and pragmatic aims.

3.3 Conflicting aims and “anti-science” epistemic values

The arguments in section 2 appeal to epistemic values, yet are “anti-science.” This seems paradoxical: how can appeals to epistemic values—skepticism and transparency, things that promote the aims of science—be “anti-science”? My development of the narrow and broad views of the aims of science suggests a more precise diagnosis. Both arguments draw on values that would (I assume) promote certain aims of science, but in ways that require sacrificing other aims. On the narrow view these other aims are epistemic; on the broad view they are pragmatic. In either case, the paradox comes from the underlying conflict between aims.

Thus, in order to maintain—or deny—that these arguments are “anti-science,” we need to resolve the conflict between the relevant aims. We need to be able to say that skepticism and transparency are scientific vices, not virtues. This requires some additional conceptual tools.

4.1 Instrumental and Intrinsic Goods

I have supposed that the aims of a practice are, at least for practitioners,¹⁵ intrinsic goods, valuable for their own sake and without need for further justification. Axiology (theory of value) traditionally distinguishes intrinsic goods from instrumental goods, which are valuable because they are useful for promoting some other goods. Havstad (2021) develops something like this distinction in her review of Brown (2020). On Brown’s Deweyan account, inquiry “begins in doubt or perplexity” arising from a material breakdown in established habits or routines (25); Havstad proposes that science can also be “spurred by positive sensations like curiosity, wonder, adventure, and delight” (Havstad 2021). Brown’s emphasis on doubt and perplexity reflects the pragmatist emphasis on the instrumental value of scientific research, while Havstad’s contrasting aesthetic responses correspond to science as intrinsically valuable.

Often these categories are treated as mutually exclusive: an instrumental good is not intrinsically valuable, and vice versa. But this does not follow from the definitions. As Richardson puts it, “it is also coherent to say that we pursue something both for its own sake and for the sake something else” (Richardson 2002, 101).

Consider the activity of coding, writing computer software. While not everyone finds coding an intrinsically valuable activity, many do (including myself). We enjoy the intellectual challenge of explicating a complicated task as a precise sequence of computational steps. Coding is done in many different contexts (practices), with different aims, which can have different standards for good code. One context with unusual standards is the game of code golf. The aim of code golf is typically to produce a given output using source code with the fewest number of characters. This is a purely intrinsic good, with no instrumental value. Good code golf code is often extremely difficult for humans to understand. For example, the following line is a complete program in the Befunge language, which does the calculation 45²–11 and then prints the target result, “2014,” using 4 characters (Justin 2015): @.ߞ’ .

By contrast, for most practicing software engineers coding is both intrinsically valuable and useful for some other purpose. Good code is an instrumentally valuable intrinsic good. Importantly, producing useful software is often a complex, highly collaborative activity. Some popular open source projects might have hundreds or even thousands of contributors. As an extreme example, there are approximately 12,000 official contributors to the GitHub repository for the open source Linux kernel as of 2021–12-15 (https://github.com/torvalds/linux), and anyone with a GitHub account can submit a “pull request” to propose changes to the code. In this radically collaborative context (Winsberg, Huebner, and Kukla 2014), it’s essential that software engineers write code that potential collaborators are able to understand relatively quickly and easily. For this reason, software engineers are taught to follow style guides and use techniques such as “self-documenting” to produce highly readable code, while code that is difficult to read is derided as “smelly” or “messy.” In this context good code golf code would be bad code. While good code is an intrinsic good, the standards for good code depend on the context, in ways that correspond to differences in the instrumental value of the code.

I suggest that scientific knowledge often has a similar status, valuable both for its own sake as well as because it is useful. That is, I suggest that the knowledge produced by environmental epidemiology is both intrinsically and instrumentally valuable. If a line of research has good prospects for producing new knowledge, it does not require further justification to claim that the research is valuable. But it can be given further justification by showing that the knowledge it will produce has good prospects for being useful for protecting human health.

Further, I claim that there is an asymmetry between environmental epidemiology’s epistemic and pragmatic aims. Knowledge of the distribution and health effects of pollutants is useful for protecting human health, but human health is not useful for producing knowledge of the distribution and health effects of pollutants. While human health is instrumentally valuable for many other goods, from the perspective of environmental epidemiology it is a non-instrumental or purely intrinsic good.¹⁶

Richardson notes that “When we pursue x for the sake of y, … we judge it appropriate or acceptable to regulate the manner and extent of pursuit of x by reference to y” (Richardson 2002, 101). That is, if we do x in order to do y, then we should do x in a way that tends to promote y, and we do x well only if we have, by so doing, also (contributed to) doing y well. When we write code that is purely intrinsically valuable (as in code golf), we don’t need to take into account whether it promotes any further goods. But most practicing software engineering is producing code for some further purpose. Because code golf ‘s extremely succinct style generally frustrates these further purposes, this style is unacceptable in everyday software engineering. In these contexts, good golf code is bad code; or, in terms of virtues and vices, an extremely succinct style can be a virtue in code golf but a vice in everyday software engineering. In terms of science, when we pursue the epistemic aims of science for the sake of some pragmatic aims, then it is appropriate to regulate the manner of the epistemic pursuits by reference to the pragmatic aims. In other words, when the epistemic aims of science are instrumentally valuable for promoting some pragmatic aims, then epistemic virtues must also be pragmatic virtues. And so epistemic values that are pragmatic vices—things that help us avoid false belief but frustrate the protection of human health, for example—are actually epistemic vices—they frustrate, not promote, the epistemic aims of science.

To be clear, on an alternative version of the broad view, fields such as environmental epidemiology have two independent sets of aims, in the sense that we can achieve one set (say, the epistemic aims) without achieving the other. Janet Kourany seems to have this kind of view (2010, 62). By contrast, here I am proposing that the aims are interdependent. When and insofar as the epistemic aims of science are valuable because (along with any intrinsic value) they are useful for achieving the pragmatic aims of science, the epistemic aims are not achieved if they are not useful in this way. Apparent epistemic achievement is merely apparent unless it is also pragmatic achievement. This means that things that appear to promote epistemic achievements while frustrating pragmatic achievements do not actually promote epistemic achievements. In this way, vices can be mistaken for virtues.¹⁷

Now consider an objection.¹⁸ Suppose it is December 2021 (the beginning of the Omicron wave of Covid-19 in the US), I am attending a social gathering tomorrow, and I want to take a Covid-19 test. I might use a faster (15-minute) but less accurate (80%, say) at-home antigen test, or a more accurate (95%, say) but slower (48-hour) PCR test. The slower test is pragmatically useless, and so my argument in the last few paragraphs implies that it is also epistemically useless. But, so the objection goes, it seems like the results of the PCR test are still epistemically valuable.

Some sense of value for the PCR test might come from pragmatic value on a longer time scale. Suppose I take both tests, the home test is negative, and the PCR test is positive. Because of its higher accuracy, the PCR test is regarded as authoritative, and I can conclude that the home test gave me a false negative. I can then let the other attendees know that I exposed them to Covid-19, which will hopefully reduce further downstream cases. Here the epistemic features of the PCR test contribute to its pragmatic value, just on a longer time frame than the initial decision to attend the social gathering. PCR tests in general are also pragmatically valuable for making these kinds of decisions, even if the pragmatic value of this particular test is limited. If I had gone to get my Covid-19 test a few days earlier, all else being equal a PCR test would have been more pragmatically valuable than a home test, because of the greater accuracy.

Some readers might insist that the PCR test has some purely intrinsic epistemic value, completely independent from any pragmatic value on any time scale. I can imagine some readers feeling some sense of wonder or awe at the scientific accomplishment of making the PCR process incredibly inexpensive and reliable over just a few decades, or perhaps at the high degree of accuracy of the PCR test. (In personal communication, Havstad says that she would not have these feelings herself if she were in this situation.) I think these feelings seem more appropriately directed at PCR tests in general, not at this particular PCR test, even if they are prompted by contemplating this particular test.

We might also say that the epistemic aims of the particular test have been achieved but in a compromised way. Insofar as the epistemic aims of the test are intrinsically valuable, and so have standards independent of any pragmatic aims, they have been achieved; but because they are also instrumentally valuable, and they have failed to contribute to those further uses, they are flawed or incomplete.

In general, the 48-hour turnaround time and high accuracy of PCR tests are impressive epistemic achievements. But, given my needs in this particular situation, the 48-hour turnaround time makes this particular PCR test seriously epistemically deficient, even with its high accuracy. Even if we want to say its epistemic aims have been achieved, they have only been achieved in a compromised, flawed, or incomplete way. So this kind of case shows how epistemic virtues in one context can be epistemic vices in another.

4.2 Disunity and reconciliation

In virtue ethics, the unity of the virtues is the view that—at least for the “cardinal” or primary moral virtues—having one virtue, such as benevolence, requires having all of the others, such as courage and temperance. The view has been controversial, in part because it seems highly plausible that someone could be both kind and cowardly, for example. I take no position here on whether the unity of the virtues is true. Instead, I suggest that Annas’s (2011) defense of the unity of the virtues can provide some guidance when we are faced with a practical dilemma that appears to require choosing between conflicting virtues.

The crux of Annas’s argument is a recognition that a virtue must be both stable and contextsensitive. On the one hand, it must be stable, in the sense that it “involves more than the activity performed in the situations in which it is first learned: it involves something on the person’s part which can also be shown in entirely different activity in a different context” (84). Comforting a child with a scraped knee and granting a paper extension to an undergraduate student are both acts of benevolence. If we say someone has a benevolent character, we are saying that we can rely on them to do these kinds of actions in a variety of contexts. On the other hand, the specific actions used to “do benevolence” are quite different across contexts. The specific things that you do to comfort a child (giving a hug, gently applying a bandage, perhaps giving a cookie) have little or no superficial resemblance to the specific things you do to grant a paper extension (writing an email, adding a paragraph to a syllabus, perhaps changing a due date in a learning management system). So if we say someone has a benevolent character, we are also saying that we can rely on them to act differently in different situations, according to differences in what benevolence requires.

Further, the stability of a virtue typically requires its context-sensitivity. Someone who responds to an extension request by offering a hug, bandage, and cookie isn’t exemplifying a benevolent character. As an epistemological point, insofar as our conceptualization of a virtue is tied too closely to the specific ways that virtue is exemplified in certain contexts, we will be unable to recognize that this same virtue is exemplified in different ways in other contexts.

The next key move in Annas’s argument is to claim that the context-sensitivity of a virtue— and so also its stability—often (or perhaps typically) requires the exercise of other virtues, even ones that are conceptually distinct.

The compassionate person might well need courage to insist that a victim be treated properly, or to stand up to a bully on someone else’s behalf. If he [sic] lacks courage, his compassion will be flawed too; victims can’t rely on it, and others generally can’t rely on him to be compassionate in appropriate circumstances. In general a virtue which is unreliable in its exercise because of facts about the person (rather than external circumstances) is a compromised virtue. (Annas 2011, 88)

Consider the paper-extension request. At least in North America, the culture of higher education is often highly suspicious of students. We instructors are encouraged to think of students as lazy and mendacious. In this cultural context, granting paper extensions—especially explicitly adopting a policy that invites extension requests and grants them freely—requires being prepared to defend the policy if challenged by employers or colleagues. That is, exemplifying benevolence in this way requires also exemplifying courage. Generalizing, as we move between different situations, we may need to clarify or reconceptualize virtues, to ensure that they are context-sensitive in a way that maintains their stability.

Again, Annas is primarily interested in the theoretical question of whether exemplifying one of the primary moral virtues requires also exemplifying the others. Here I am interested in the practical question of how to reconcile a conflict between two apparent virtues of a specific practice, such as the apparent scientific virtues of skepticism and precaution, or transparency and timeliness.

One response to such a conflict would be to maintain the independence of the virtues. Transparency and timeliness are conceptually independent, and so we would expect both happy cases in which they agree and unhappy cases in which they disagree. When they disagree, we need to pick one, or bring in some other kind of consideration to decide how to move forward. (Compare Neurath’s 1983, concept of the auxiliary motive—which might resolve such dilemmas with a coin flip—and the way the narrow view seemed required to bring in “non-epistemic” values above.)

Another response to apparent conflict, inspired by the doctrine of the unity of the virtues, would see reconciliation between these virtues as a practical challenge. It may not be obvious how we might reconcile these seemingly-conflicting virtues. But perhaps we can do it if we try.

Consider an apparent conflict in paper-extension cases between benevolence and justice. We might have a conception of justice as procedural fairness and formal equality, according to which justice is exemplified by announcing paper deadlines well in advance and holding students responsible for managing their various responsibilities. On this conception of justice, giving some students additional time on the paper because they failed to manage their responsibilities well would be ad hoc and unjust to the rest of the class. Thus there appears to be a conflict between the virtues of justice (having the same expectations for all students) and benevolence (granting exemptions).

Deliberating on this apparent conflict, we might learn that today traditional-aged higher education students (18–25) may have little or no control over their work schedules (Lambert 2008; Henly and Lambert 2014) and are significantly more likely than other age groups to experience debilitating depressive episodes (Lipari 2018, fig. 46).¹⁹ These sorts of factors may mean that many higher education students are much more likely to face unexpected challenges than their instructors did as students. At the same time, we also might realize that many students have not had explicit instruction or guidance in time management, and so it is unreasonable to expect them to have mastered this challenging skill; and that students might reasonably choose to misrepresent the reasons for their time management challenges in order to preserve their privacy. All together, we might conclude that we need to modify our conception of justice in our assignment schedules, to be more flexible and more accommodating of unexpected challenges in our students’ lives. According to this modified conception of justice, different students face different challenges and so it would actually be unjust to hold all students to the same rigid assignment schedule. Instead, the assignment schedule should be flexible, to accommodate unexpected complications. Now benevolence and justice agree rather than conflict: both virtues direct us to adopt a generous extension policy. Put more generally, we respond to a seeming disunity between benevolence and justice by reconceptualizing justice in a way that allows benevolence and justice to be unified. I will call this activity of reconceptualizing one virtue, in order to resolve a seeming conflict with another virtue, reconciliation. Note that reconciliation will often involve the exercise of creativity (Brown 2020) and judgment (phronesis). There is no deterministic procedure for reconciling an apparent conflict between two virtues.

After this reconciliation, we can recognize that the previous conception of justice was not just mistaken, but actually vicious. This is in part because of the way the first conception of justice was context-insensitive, namely, the way it neglected differences between traditionalaged higher education students today and otherwise similar students decades ago. Because of this context-insensitivity, the procedural and formal conception of justice actually frustrated the pursuit of both benevolence and justice.

4.3 Resolving the paradox

In the last two subsections, I’ve developed two conceptual tools: the concept of intrinsicallyvaluable instrumental goods, with a corresponding asymmetry between the epistemic and pragmatic aims of some scientific fields; and the reconciliation of one virtue to another as a way to ensure that virtues are appropriately sensitive to context. Here I apply these tools to the apparent conflicts between epistemic and pragmatic values: between transparency and timeliness, and between skepticism and precaution. Reconciliation suggests that, when confronted by this kind of apparent conflict between a pair of epistemic and pragmatic values, we can remove the conflict by reconceptualizing one. The asymmetry between epistemic and pragmatic aims suggests that the epistemic value is the one that should be reconceptualized.

Let us begin with transparency and timeliness. The epistemic goods promoted by open science are valuable on their own, but also must be valuable for the way they promote the pragmatic aims of protecting human health and the environment. Specifically, they should not delay research just because a field like environmental epidemiology doesn’t (yet) have the infrastructure and know-how that is required for open science to be relatively time-efficient. This means that open science initiatives, to be virtuous, must include infrastructure development, training for both students and professional researchers, expanded research funding (to cover additional researcher time and maintenance expenses), and ready access to data management and informatics experts. These efforts must continue over a substantial period of time—perhaps a decade or two—in order for the vast majority of research groups in the field to be able to do open science efficiently. Until that time, a sharp open science requirement such as Strengthening Transparency is almost certain to delay research, and hence be vicious.

This proposal expands the conception of transparency in open science, from a single action performed by an individual researcher—“release your data”—to a complex collaborative activity, involving significant infrastructure and the ongoing work of different kinds of specialists (Levin et al. 2016; Bezuidenhout et al. 2017; Elliott and Resnik 2019). That is, the previous paragraph reconceptualizes transparency from an individual virtue to a social one (compare Longino’s (1990) social reconceptualization of objectivity), in a way that reconciles it with the pragmatic virtue of timeliness. After this reconciliation, from the perspective of the social conception of transparency, the completely individualistic conception of transparency—the way transparency is implicitly understood by Strengthening Transparency—is recognized as a vice, not a virtue, in two ways. First, the completely individualistic conception conflicts with the pragmatic aims of environmental public health; and second, it fails to recognize that individual researchers can only exemplify transparency when the appropriate infrastructure is available.

Next, skepticism and precaution. To reconcile skepticism and precaution, we might apply skepticism to skepticism itself. Recall that Young’s argument was not that error rates in air pollution studies are actually inflated by multiple comparisons across different model specifications, but rather that they might be. A skeptical attitude towards this skeptical argument would point out that the mere possibility of a problem is not yet very compelling evidence of an actual problem. A special line of research funding might be developed to examine the way alternative model specifications might make a difference to the conclusions of air pollution studies (perhaps using the method of “multiverse analysis,” which attempts to fit every alternative model specification; Steegen et al. 2016). Until this kind of research has produced evidence of an actual problem, the pragmatic aim of protecting human health and the environment recommends a precautionary attitude—namely, accepting the conclusion that air pollution is hazardous.²⁰ This skeptical attitude—directed towards hasty claims of problems with air pollution research—supports, rather than undermines, a precautionary attitude, and thus promotes both the epistemic and pragmatic aims of environmental epidemiology.²¹