What methods should scientists use to study the world? Robert Northcott’s Science for a Fragile World addresses this general question about the methodology of science by carefully examining some instances of the methods employed in political science, economics, ecology, and epidemiology.

Northcott’s approach to the methodology of science centres around one crucial distinction—the distinction between stable relations and fragile relations. He draws this distinction in terms of predictability: a stable relation is one where scientists can easily predict how the relation will be instantiated differently from one case to another. Take, for example, the causal effect of an object’s electrical charge on its acceleration (in the presence of another charged object nearby). On the one hand, an object’s acceleration may be difficult to predict, because it may be difficult to measure all the forces acting on the object in question. In this respect, the outcome of interest (acceleration) is subject to a lot of ‘noise’ generated by factors other than the cause of interest (electrical charge). But, on the other hand, the causal difference that electrical charge makes to an object’s acceleration is easy to predict. That’s because we know Coloumb’s law, which tells us how this causal effect depends upon, and only upon the degree of electric charge, and the distance between the object and any other electrically charged objects in the vicinity. So even though the outcome of interest (acceleration) differs from one case to another, and even though the causal effect of interest (electrostatic force) differs from one case to another as well, this causal effect differs in ways that scientists can predict, given that they know Coloumb’s law. In virtue of this, Northcott counts this causal effect (electrostatic force) as a stable relation.

In contrast, a fragile relation is one where scientists find it difficult to predict how the relation will be instantiated from one case to another. Take lockdown effectiveness, for instance—namely, the causal effect of a national lockdown on the spread of an infectious disease. Lockdown effectiveness probably depends upon a large set of other variables (contextual factors), such as vaccination uptake, prior infections, testing capacity, the timing and duration of the lockdown, trust in government, cultural norms, policing capacity, household size, schooling arrangements, urbanization, the job market, the possibility of remote working, and financial support for social isolation. On the one hand, God knows exactly how the effectiveness of a lockdown will depend upon this long list of contextual factors. In other words, God possesses a theory of lockdown effectiveness that correctly describes a wide variety of cases. (Admittedly, this ‘wide-scope’ theory is probably rather complex.) On the other hand, scientists do not have warrant for any such wide-scope theory of lockdown effectiveness. And, given our human limitations and given the complexity of disease dynamics, Northcott contends that it’s wishful thinking to suppose that scientists could ever get such warrant. As a result, it is much more difficult for scientists to identify whether a given lockdown was (or will be) effective. In virtue of this, Northcott counts lockdown effectiveness as a fragile relation. Indeed, this illustrates how he defines fragility epistemically rather than metaphysically: fragility denotes scientists’ lack of warrant for any wide-scope theory, and it denotes the corresponding difficulty in predicting how the relation in question will be instantiated in a given case.

Since the fragility of lockdown effectiveness rules out appealing to wide-scope theory, scientists will instead need to become ‘case workers’, Northcott argues; that is to say, whenever scientists want to identify whether a particular lockdown was (or will be) effective. Being a case worker involves examining the particular lockdown in question in painstaking detail.

Northcott criticizes scientists and philosophers for spending too much time addressing the question of what methods scientists should use to study stable relations, and not enough time addressing the question of what case-worker methods scientists should use to study fragile relations. His book considers all the major research programmes in philosophy of science over the last half century—such as the realism versus instrumentalism debate, the debate on the epistemic import of idealized models, the new mechanist literature, the debate on reflexivity in the social sciences—and it shows how each research programme has ignored or obscured the need for case-worker methods. Northcott then suggests how work in these literatures might be re-orientated to focus on fragile relations and case-worker methods. This, I think, is the major contribution of the book. It’s a manifesto for researchers to invest more time thinking about the proper methodology for studying fragile relations.

If you are not already sold on the importance of fragile relations and case-worker methods, then I wholeheartedly recommend this book. In particular, I’d recommend that you peruse chapter 2 to get the gist of the distinction between stable relations and fragile relations. You might then study chapters 3 and 4 in detail, which is where Northcott illustrates the methodological problem posed by fragile relations (by means of a fascinating case study of informal truces in the trenches of the First World War). These chapters suffice to make these important ideas crystal clear. You can then easily dip into and out of chapters 6 and 7 (which speed through a long list of philosophical research programmes that have obscured the issue of fragile relations and case-worker methods). The same goes for chapter 8–10 (which look at three further cases studies drawn from economics, epidemiology, and data science).

But what if you are already sold on the importance of methods for studying fragile relations? Perhaps you are a researcher studying external validity and extrapolation: given the fragility of causal effects, what extra knowledge is needed to extrapolate a causal effect from one context (a randomized controlled trial, for example) to another (Khosrowi 2022)? Or perhaps you are a researcher studying idealized models as tools for inductive inference: given the differences between an idealized model and your target system, what is needed to warrant an induction from what happens in an idealized model to what happens in the real-world target system (Sugden 2022)? For researchers working in the above literatures, the lessons of the book are subtler. Indeed, if you fall into this category, then I’d recommend you focus your attention mostly on chapters 2 and 4.

First, chapter 2 draws a rigorous and clear distinction between stable relations and fragile relations. This gives us philosophers of science a vocabulary with which to describe, as incisively as possible, the exact methodological problem posed by fragile relations. This is a vocabulary of fragile versus stable relations, and of wide-scope versus narrow-scope theories, and of theoretical methods versus case-worker methods.

Second, researchers in the above literatures have inherited a way of thinking about these issues pioneered by Cartwright. But Cartwright’s (1983) early work implicitly assumed that the relations of interest are stable. Indeed, she develops the concept of a causal capacity, which by definition only applies when causal relations are stable, not fragile. In her more recent work on ‘support factors’, however, it becomes less and less clear whether Cartwright takes herself to be studying fragile relations or stable relations (Cartwright and Hardie 2012). To what extent does Cartwright now believe that social scientists, for example, can, with a lot of effort, get warrant for a wide-scope theory that identifies most of the relevant support factors (contextual factors) upon which a given causal effect depends? It’s not fully clear. Northcott, in contrast, nails his colours to the mast: in most social scientific contexts, fragility abounds, and wide-scope theories are simply not available.

Third, Northcott doesn’t put science up on a pedestal. Some scientific research programmes are fruitful; others are not. And we philosophers of science should be honest about this. For example, social scientists have written thousands of theoretical articles in which they use game theory to offer ‘how-possibly’ explanations of a huge variety of social phenomena. But this how-possibly theoretical work is often little more than speculation, Northcott points out. This work usually provides no warrant to favour one how-possibly explanation over the countless other explanations that we could have dreamt up for the phenomenon to be explained. If you disagree, then I thoroughly recommend chapters 3 and 4 to you. They constitute a lively, lucid, and incisive challenge to the (in my view exaggerated) epistemic value of much theoretical modelling in the sciences.

Having recommended this book to two different audiences, I want nevertheless to manage your expectations—as well as point towards some directions for future research. This book does not look at the logic behind case-worker methods. Yes, it mentions some examples of case-worker methods, such as interviews or ‘process tracing’ in political science. And it gives a clear and engaging description of some of the conclusions that process tracers and interviewers have drawn—for example, della Porta’s (1995) ground-breaking work on the dynamics of violent left-wing groups in Germany and Italy. But Northcott’s book does not make explicit the rationale (formal or informal) whereby the evidence della Porta gathers warrants the causal conclusions that she draws. The book shows the need for studying the logic behind case-worker methods, but it does not engage in that study itself.

For readers wanting to engage in that study, I recommend Crasnow’s (2017) work on narratives, Fairfield’s (2022) Bayesian explication of case-worker methods, and work by myself and Rosa Runhardt on the logic behind process tracing (Runhardt 2020; Clarke 2023) and mixed methods research (Clarke 2024; Runhardt unpublished). There’s also the seminal work in Beach and Pedersen (2016) and Brady and Collier (2010) on qualitative case-worker methods in political science. However, if you are not yet convinced for the need for case-worker methods, then I can think of no better place to start than Science for a Fragile World.

Christopher Clarke
Erasmus University Rotterdam
clarke@esphil.eur.nl

References

Beach, D. and Pedersen, R. (2016). Causal Case Study Methods: Foundations and Guidelines for Comparing, Matching, and Tracing, University of Michigan Press.

Brady, H. E., and Collier, D. (2010). Rethinking Social Inquiry: Diverse Tools, Shared Standards, Rowman and Littlefield.

Cartwright, N. (1983). How the Laws of Physics Lie, Oxford University Press.

Cartwright, N. and Hardie, J. (2012). Evidence-Based Policy: A Practical Guide to Doing It Better, Oxford University Press.

Clarke, C. (2023). ‘Process Tracing: Defining the Undefinable?’, in J. van Bouwel and H. Kincaid (eds), The Oxford Handbook of Philosophy of Political Science, Oxford University Press.

Clarke, C. (2024). ‘Mixed Methods and Causal Ontology’, in Y. Shan (ed.), Philosophical Foundations of Mixed Methods Research, Routledge, pp. 210–39.

Crasnow, S. (2017). ‘Process Tracing in Political Science: What’s the Story?’, Studies in History and Philosophy of Science A, 62, pp. 6–13.

della Porta, D. (1995). Social Movements, Political Violence, and the State, Cambridge University Press.

Fairfield, T. (2022). Social Inquiry and Bayesian Inference: Rethinking Qualitative Research, Cambridge University Press.

Khosrowi, D. (2022). ‘What’s (Successful) Extrapolation?’, Journal of Economic Methodology, 29, pp. 140–52.

Runhardt, R W. (2020). ‘Concrete Counterfactual Tests for Process-Tracing’, available at .

Runhardt, R W. (unpublished). ‘Limits to Evidential Pluralism: Multi-method Large-N Qualitative Analysis and the Primacy of Mechanistic Studies’, Synthese, 200, available at .

Sugden, R. (2022). ‘Credible Worlds: The Status of Theoretical Models in Economics’, Journal of Economic Methodology, 7, pp. 107–36.