Wrong, but Useful – for Whom? The Politics of Pandemic Modeling

This article is part of the series:

On March 17th, after a tepid initial COVID-19 response, Prime Minister Boris Johnson began encouraging citizens to avoid non-essential social contact. President Donald Trump followed suit, introducing travel bans and social distancing policies. These responses marked the beginnings of drastic and unprecedented government measures to prevent the spread of COVID-19 in the United States and the United Kingdom.

A major source of this shift was an epidemiological simulation model developed by the Imperial College COVID-19 response team in London (Ferguson et al. 2020). After assessing the potential outcomes of five different non-pharmaceutical interventions, implemented individually as well as in combination, the modeling team concluded that extreme measures aimed at epidemic suppression would be the only viable strategy to prevent catastrophe. The policy recommendations that followed from the team’s epidemiological modeling soon began to travel, influencing governments’ responses globally. Borders were shut down and citizens around the world were advised or ordered to stay at home as the result of this model’s projections.

Models and modeling are not new phenomena. Throughout history, various forms of models have been used to explain and predict “real-world” phenomena, inform policy, and govern human behavior. Yet, the burgeoning of models and modeling practices in our contemporary world suggests that we live in a modeloscene (Østebø 2021 – in production): an era in which models and modeling have become increasingly hegemonic as forms of knowledge and tools of governance. The increased use and circulation of models within the policy world and as a tool for social and behavioral change more generally can be explained with reference to both technology and ideology. New computing technologies and access to “big data” have, for example, led to an upsurge in mathematical models. The quest for models, particularly those commonly referred to as best practices, can also be linked to the rise of New Public Management and the business-oriented and result-driven ethos that has come to influence contemporary politics and policy making.

The COVID-19 pandemic has, perhaps more than anything else, revealed the effects of this “model reality.” Models are no longer confined to policy experts or scientists, but have become unprecedentedly public and popularized. Graphic models depicting and comparing the spread and consequences of COVID across geographical areas are on the front-page of major news outlets, and numerous newspaper articles and op-eds have been written about predictive models, including the one created by the Imperial College team. Models increasingly also govern our lives, informing us of what the future may look like if we fail to wash our hands or comply with physical distancing rules. 

Recent social science publications have made note of, and discussed the prominent role models and modeling communities play in the COVID-19 pandemic (Hinchliffe 2020, Rhodes, Lancaster, and Rosengarten 2020). In a Somatosphere post, Steve Hinchliffe (2020) expresses concern with “the callous nature of the model-based policy.” While he reminds us that models are “partial technologies” he makes it clear that it is not his intention to criticize modelers or disparage models. Rather, he calls for greater recognition of other voices and forms of knowledge.

While we share Hinchliffe’s concerns, and see the value of knowledge triangulation and interdisciplinary engagement, we caution against what appears to be a reluctance to fully engage in an in-depth, critical analysis of models and modeling communities. In a world where models are everywhere and modeling has gained unprecedented and hegemonic status, it is pivotal that we (and here we speak as anthropologists) “stay with” the models; that we engage in empirical studies of modeling communities and of models themselves. This is, without doubt, a delicate endeavor with potential political ramifications. By critiquing predictive COVID-models and modeling practices we risk being perceived as “anti-science.” We also fear that our critiques will be used or misused by political actors in ways we never intended, feeding into the increasingly strong countervailing social and political narratives which devalue scientific knowledge production. So, when we here engage in a critical analysis of models and modeling practices, we do so for three reasons: First, we believe that a reluctance to engage in a critique of models and modeling communities reinforces their hegemonic status. Secondly, interviews with pandemic modelers and analysis of modeling discourse shows that modelers themselves not only are engaged in and value critique, they also see it as essential for improving their work. Lastly – and this is perhaps the most important point we want to make – the exclusionary and selective practices that are fundamental to all forms of modeling demand a critical analysis of what models are, how they come into being, and what they do. This implies due attention to the ways that models, as simplified, decontextualized representations and traveling technologies, leave out certain groups or elements and hence have intended as well as unintended social and political consequences. To illustrate our argument, we combine findings from interviews we have conducted with pandemic disease modelers over the past two years with an analysis of the Imperial College COVID-19 model. Again, we would like to reiterate that our purpose is not to dismiss the usefulness of modeling for pandemic response, but to encourage critical engagement with models – particularly those that gain hegemonic status – and with modeling as a politically and culturally situated practice. 

“All models are wrong, but some are useful.” This mantra, originally formulated by the statistician George Box (1979), was repeated in almost every interview we conducted with pandemic disease modelers. In a recent article in Nature, Neil Ferguson, the lead scientist behind the Imperial College model, described models as “simplified representations of reality” (Adam 2020). The reduction of complexity is, in fact, a central feature of all model making. When modelers create models of and for real-world phenomenon, they extract a small part of that world, bringing into the model only what they consider important – what they want to “see” in the model – while dropping disruptive elements. This process, in which “the real world is shorn away” (Morgan 2012, 101), is neither neutral nor apolitical. What gets excluded is not something given; the choices and practices that go into model making are influenced, sometimes unconsciously, by the modelers’ education, interests, and worldviews, and the historical and sociopolitical contexts in which they are situated. For example, in interviews, epidemic modelers talked about how their modeling work was not only informed by, but even dictated by the requests or agenda of a particular donor or policy community. As illustrated in the following quote from an epidemiologist who describes their experience working with the WHO on Ebola modeling: 

I have to say that working with the WHO …. there’s a lot of politics going on with the WHO. And they are always very concerned about controlling the narrative. Meaning they always want to make sure that whatever modeling work that is done, they agree with the message that modeling is saying (…). When they give you the data you need some form of agreement with them, that they will review your paper before your paper goes out. And making sure that they agree with your message. Yeah, it is true that we don’t always agree with that approach because sometimes modeling gives you countering results [to currently held beliefs and assumptions]. And it may be that what modeling is saying is true and what you’re saying is wrong. But anyway, every organization has its politics.

The political and exclusionary nature of model making is also evident when we turn our attention to the inner workings of the models themselves, disguised as they often are within technological jargon, intimidating symbology, statistics, and math. The Imperial College COVID-19 model, for example, uses high resolution census population data from the US and the UK, including age, household distribution size, class size in schools, and workplace size, to simulate the spread and effects of COVID-19 under different interventions. In the process of simplifying a complex situation, the modelers extract information they consider important, and eliminate factors they assume are less relevant. For instance, in modeling the ways that COVID-19 could spread, the Imperial College Modelers exclude differences in socioeconomic status: the ways that some people may have to rely on public transportation, may be homeless, or continue to go to work out of financial necessity or as essential employees. The modelers also exclude differential access to healthcare, including insurance status. Their estimates of when hospital capacity will be overwhelmed only take into account the total number of intensive care beds in the country, and not how these may be unequally distributed. Further, in calculating morbidity and mortality, they do not include the ways that preexisting conditions and age may impact the severity of the disease. In short, they have dropped the effects of poverty, race, and inequality from the model, despite their obvious importance for health outcomes. The world that is modeled has been “flattened” in terms of these factors, a move that was made on the basis of what modelers considered important to answering the questions they were asking when they constructed the model. 

These exclusions and assumptions have consequences. If the Imperial College modelers had included the effects of poverty, race, and other forms of inequality in their model, they may have predicted that impoverished communities, and especially communities of color, would be particularly affected by the COVID-19 pandemic. As a result, a response may have targeted at-risk communities. Alternatively, if the model had painted COVID-19 as a disease of poverty, disproportionately affecting people of color, politicians who are less concerned with the wellbeing of these communities may have seen it as license to further downplay danger, and may have ignored recommendations that suggested a strong response. Indeed, if the model had called for drastic economic sacrifice on behalf of impoverished communities of color, it is easy to imagine that the shutdown may not have occurred.

Further, the model had global impact. While the Imperial College modelers made it clear that the findings from their modeling first and foremost were applicable to other high-income countries, their policy recommendations have been uncritically adopted in many countries around the world, with strict lockdowns in countries such as South-Africa and India, for example.  Again, this is reflective of the hegemonic status models have gained.  A widely held assumption about the use of models in the policy field is that they are examples of success or best practices that can be transported, emulated, scaled up, and implemented to bring about a desired change across global spaces. But models are not silver bullets. They are not neutral, universal, or static self-contained entities that exist independently of their historical, political, and economic context. Models are cultural and political constructs that demand contextualization and critical scrutiny. And because models are highly simplified and idealized, they mask complex social relationships and interactions, and only partially mirror what is actually happening on the ground. If the Imperial College model, and its representations of reality that flatten socioeconomic relationships obscure and ignore important social and economic conditions, had deleterious consequences in the USA and UK, what are the implications of this model in low-income countries such as Ethiopia or Haiti, where these dynamics are magnified? The measures that have been taken in the USA and UK attempt to slow the pandemic in order to prevent a breakdown of health systems. To what extent does this model make sense in contexts where the health system is already weak and dysfunctional? What will be the social, political and economic consequences of a quarantine – of any length or strictness – in countries where poverty and political instability, prior to COVID-19, already was a major challenge?

A headline in The Atlantic reads, “Don’t believe COVID models – that’s not what they’re for.” The Imperial College model was not meant as a prediction of the future. As one of our modeler informants put it, “that is not really the point of the model. The point is to test things under a relative scenario and to use it as a decision tool. [The model] is not a crystal ball where you can say I predict that there will be exactly this many cases.”

Models are always wrong. They can never simply depict reality, but in their representations or predictions, they necessarily distort or change it. This also means that models can be useful. A key question is, however, who and what are they useful for? Given that policy models have political implications and goals, it matters who and what is counted within them (Adams 2016). The choice to model COVID-19 without considering differential access to health care and material resources reflects what modelers value, and how they conceptualize health. Such moves may be deliberate political acts, or they may be unintentional, and revealing of the positionality of disease modeling as a profession. As Veena Das writes, the assumptions and choices that go into the model, may also reflect that “most policy makers, bureaucrats, and mathematical modelers (…) simply don’t know how the poor live, which is why they cannot anticipate their actions and consequently take variations in human behavior into account in their modelling” (Das 2020). It is the role of anthropology to critically assess the assumptions, exclusions, and omissions that go into model making, to trace their consequences, and to keep insisting that the kind of knowledge we have and generate, which over and over again has shown that complex problems require complex, place-specific and carefully crafted solutions, is of relevance for policy makers and modeling communities. A critical anthropology of models ensures that political moves are not hidden behind the language of math and science, to become inextricably embedded in our public health responses and policies. This approach only becomes more vital as COVID-19 disease models – as well as other forms of models – become increasingly politicized, disputed, and influential technologies of governance.

Marit Tolo Østebø is an Assistant Professor at the Department of Anthropology at the University of Florida. Her research is situated in the intersection of anthropology of policy, gender and development, science and technology studies and global health. She is the author of Village Gone Viral. Understanding the Spread of Policy Models in a Digital Age (Stanford University Press, in production).  

Rebecca Henderson is an MD/PhD candidate in the Department of Anthropology at the University of Florida. Her research interests include the use of models, case definitions, and algorithms in medical practice and public health, as well as the adaptation of biotechnical processes and systems to low-resource spaces.  Her dissertation focuses on the implementation of oncology care in Haiti.​


Adam, D. 2020. “Special report: The simulations driving the world’s response to COVID-19.”  Nature 580 (7803):316-318. doi: 10.1038/d41586-020-01003-6.

Adams, Vincanne, ed. 2016. Metrics. What Counts in Global Health Durham London Duke University Press.

Box, George E. P. 1979. “Robustness in the Strategy of Scientific Model Building.” In Robustness in Statistics, edited by Robert L. Launer and Graham N. Wilkinson, 201-236. Academic Press.

Das, Veena. 2020. ““Facing Covid-19: My Land of Neither Hope nor Despair.” In “Covid-19 and Student Focused Concerns: Threats and Possibilities,” Veena Das and Naveeda Khan, eds., American Ethnologist website, May 1 2020, [https://americanethnologist.org/features/collections/covid-19-and-student-focused-concerns-threats-and-possibilities/facing-covid-19-my-land-of-neither-hope-nor-despair].

Ferguson, NM, D  Laydon, G  Nedjati-Gilani, N Imai, K Ainslie, M Baguelin, S Bhatia, A Boonyasiri, Z  Cucunubá, G Cuomo-Dannenburg, A Dighe, I Dorigatti, H Fu, K Gaythorpe, W Green, A  Hamlet, W  Hinsley, L Okell, S  van Elsland, H   Thompson, E Verity, E Volz, H Wang, Y Wang, P  Walker, C Walters, P Winskill, C Whittaker, C Donnelly, S  Riley, and A Ghani. 2020. “Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand. .”  Imperial College COVID-19 Response Team, London, March, 16. https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf.

Hinchliffe, Steve. 2020. “Model Evidence.”  Somatosphere (March 31, 2020).

Morgan, Mary S. 2012. The World in the Model: How Economists Work And Think. New York: Cambridge University Press.

Østebø, Marit Tolo. 2021 – in production. Village Gone Viral. Understanding the Spread of Policy Models in a Digital Age: Stanford University Press.

Rhodes, Tim, Kari Lancaster, and Marsha Rosengarten. 2020. “A model society: maths, models and expertise in viral outbreaks.”  Critical Public Health:1-4. doi: 10.1080/09581596.2020.1748310.

Leave a Reply

Your email address will not be published. Required fields are marked *