LARS P. SYLL

What is causality?

Yours truly has been offering a crash course on causality to fellow researchers at Malmö University over the past couple of years.

The course PowerPoint is available here: What is causality? 

Many contemporary research questions in the social sciences are fundamentally concerned with issues of causality. What lies behind rising unemployment? What effects do independent schools have on pupils’ attainment?

Questions of cause and effect are equally central to politics and public policy. What measures are required to curb rising inflation? What impact has the COVID-19 pandemic had on public health? Answering such questions presupposes sound causal reasoning.

Since the mid-twentieth century, causality has increasingly been discussed in probabilistic terms. Using traditional statistical methods, we can identify correlations and describe their properties (statistical significance, robustness, sensitivity, and so on). However, researchers — and policymakers — often wish to go further by identifying the causal mechanisms and relationships that explain these observed correlations. This, in turn, makes it possible, through the active use of interventions or manipulations, to estimate the causal effects of different policy packages and measures.

In contemporary research, it has become critically important — particularly for securing research funding — to employ research designs with high ‘evidential value’. Accordingly, evidence hierarchies are typically structured around the degree of causal knowledge generated by different research strategies, with debates surrounding randomised controlled trials (RCTs) and meta-analyses playing a prominent role.

The course introduces core theories and models of causal inference and explains, illustrates, and demonstrates how to apply the most widely used methods for causal analysis, primarily within the social sciences.

What is the point of doing experiments?

Why bother with the theories? Because science seeks to understand the world and to explain the experimental facts, and you need theories to do that. Theories are not just heuristic aids, “useful instruments that aid the growth of experimental knowledge,” as Chalmers accuses Mayo of thinking … What is the point of doing experiments? It is not just to accumulate reliable experimental knowledge. That is important, to be sure, but we seek reliable experimental knowledge to adjudicate between rival explanatory theories about the way the world is.

Alan Musgrave

Alan Musgrave’s response to Deborah Mayo’s claim that empirical testing does not tell us whether a theory is true strikes at the very heart of his critical rationalism. Mayo maintains that empirical tests can only show whether a hypothesis has passed or failed ‘severe tests’. For her, science is fundamentally about error detection, not truth confirmation.

But although empirical testing never yields certainty, this does not mean testing is epistemically neutral with respect to truth. The very point of ‘severe tests’ is that passing them gives us good reason to believe a theory is true. If empirical success does not count as evidence for truth, then it is unclear why science should be taken seriously at all.

As scientists, we do not merely aim to avoid error — we also seek to discover how the world actually is. That a theory has passed “severe tests” which other theories have not offers no absolute guarantee of its truth. But arguably, it does furnish a warrant for believing it to be so.

The Centrality of Inference to the Best Explanation in Scientific Reasoning

Inference to the Best Explanation (IBE) offers a compelling and robust framework for understanding how scientific knowledge advances. Moving beyond the traditional dichotomy between deduction and induction, IBE is not merely one instrument among many in the scientist’s toolkit, but a fundamental engine of scientific reasoning itself. Its significance lies in its capacity to model realistically the complex, creative, and comparative judgements that underpin theory choice, guide research, and link observable phenomena to deeper ontological understanding. In doing so, IBE captures a central feature of everyday life and scientific practice that purely formal accounts of inference struggle to accommodate: the role of explanatory judgement in navigating uncertainty.

At its core, IBE is a form of ampliative reasoning (cf. Stathis Psillos), in which one infers, from a body of available evidence, the truth — or approximate truth — of the hypothesis that would, if correct, provide the best explanation of that evidence. Unlike deductive inference, IBE is not truth-preserving, yet it is epistemically indispensable precisely because science aims not merely at certainty, but at understanding. Scientific reasoning must often proceed under conditions of incomplete information, and IBE offers a principled way of extending knowledge beyond what is strictly entailed by the data.

The importance of IBE can be articulated across several interconnected dimensions.

(1) IBE provides a superior model for theory choice and evaluation. Science is continually confronted with underdetermination: the logical fact that a given body of evidence can, in principle, be explained by multiple competing hypotheses (cf. the Duhem–Quine thesis). Deductive logic alone cannot resolve this problem, while simple induction is both fragile and superficial. IBE, by contrast, employs a set of explanatory virtues—such as scope, precision, coherence, simplicity, and fruitfulness—as pragmatic criteria for comparison. These virtues function not as strict rules, but as guides to rational preference under epistemic constraint. Scientists infer the superiority of models, theories, and hypotheses not through decisive logical refutations, but through judgements of comparative explanatory power. Darwin’s theory of evolution by natural selection prevailed over special creation not because speciation had been directly observed, but because it uniquely and elegantly explained the nested hierarchy of life, the fossil record, and patterns of biogeographical distribution as the outcome of a single, lawful process.

(2) IBE also structures the very process of scientific investigation. IBE functions as a ‘thinking machine’ that drives discovery. The search for the best explanation is not a passive assessment of existing ideas or hypotheses, but an active generator of new ones. When confronted with anomalous or surprising data, scientists seek causal mechanisms that would render the data intelligible. This ‘abductive’ (cf. Peirce) search frequently precedes hypothesis testing and experimentation, shaping the direction of inquiry itself. In this way, IBE plays a crucial heuristic role, guiding the imagination towards hypotheses that are not only testable, but explanatorily promising.

Peter Lipton famously distinguished between what he termed the ‘likeliest’ and the ‘loveliest’ explanations. The likeliest explanation is the one judged most probable given the current evidence, whereas the loveliest is the one that offers the greatest understanding—the most comprehensive, coherent, and fruitful account. According to Lipton, scientific practice typically aims at loveliness rather than mere likelihood. Even if scientists assess the ‘likeliness’ of competing hypotheses, the ‘likeliness’ is typically based on assessments of the explanatory importance of the hypotheses. There is no perfect match between likelihood and explanatory value. If I take the contraceptive pill, the probability that I will not become pregnant is high, but this fact hardly explains why I do not become pregnant. Even for a committed Bayesian, it is difficult not to concede that the assignment of prior probabilities is standardly guided by explanationist considerations, and that conditionalisation on evidence therefore occurs only after explanatory criteria have been applied to determine the value of the prior.

Scientists often favour bold, unifying explanations that are initially less probable over cautious, ad hoc alternatives that appear more likely on narrow evidential grounds. The history of science is replete with such ‘lovely’ but initially improbable ideas—continental drift and quantum theory being notable examples—pursued because of their explanatory promise. In seeking the loveliest explanation, IBE becomes a logic of scientific progress rather than mere post hoc justification.

(3) IBE plays a crucial role in connecting empirical data with theoretical understanding. Observations — such as statistical correlations within a dataset — are, in isolation, devoid of significance. They acquire evidential status only within an explanatory framework. IBE is the cognitive process through which this framework is constructed, by asking what the world must be like for the observations to make sense. This inferential move allows scientists to posit unobservable entities, causal mechanisms, and structural relations that underpin observable phenomena. In this respect, IBE underwrites critical realist ambitions of science, supporting the view that successful explanations reveal something genuine about the structure of the world.

(4) Some critics argue that IBE is unreliable, claiming that it conflates explanatory appeal with truth. In response, it can be argued that explanatory virtues are not arbitrary aesthetic preferences, but epistemic heuristics refined through the cumulative success of science. The explanatory virtues of IBE have repeatedly proved to be reliable guides to truth because they reflect deep features of how the world is organised. The sustained predictive success of theories selected via IBE provides a pragmatic vindication of its epistemic legitimacy, even in the absence of deductive certainty.

Inference to the Best Explanation is indispensable to science because it captures the dynamic, comparative, and creative character of scientific reasoning. It functions both as a ‘logic’ of discovery and as a rationale for justification. IBE transforms raw data into evidence, guides the scientific imagination towards fruitful hypotheses, and employs explanatory virtues to select theories that offer the deepest understanding of the world. By seeking the ‘loveliest’ explanation — the explanations with the greatest explanatory merit/value — IBE propels science beyond incremental refinement towards genuinely transformative insight, making it the central cognitive process through which we construct an increasingly accurate picture of reality.

People object that the best available explanation might be false. Quite so — and so what? It goes without saying that any explanation might be false, in the sense that it is not necessarily true. It is absurd to suppose that the only things we can reasonably believe are necessary truths …

People object that being the best available explanation of a fact does not prove something to be true or even probable. Quite so — and again, so what? The explanationist principle — “It is reasonable to believe that the best available explanation of any fact is true” — means that it is reasonable to believe or think true things that have not been shown to be true or probable, more likely true than not.

Alan Musgrave

The danger of treating causality as deduction

As I have argued, explanations are not deductive proofs in any particularly interesting sense. Although they can always be presented in the form of deductive proofs, doing so seems not to capturing anything essential, or especially useful, and usually requires completing an incomplete explanation. Thinking of explanations as proofs tends to confuse causation with logical implication. To put it simply: causation is in the world, implication is in the mind. Of course, mental causation exists (e.g., where decisions cause other decisions), which complicates the simple distinction by including mental processes in the causal world, but that complication should not be allowed to obscure the basic point, which is not to confuse an entailment relationship with what may be the objective, causal grounds for that relationship. Deductive models of causation are at their best when modeling deterministic closed-world causation, but this is too narrow for most real-world purposes. Even for modeling situations where determinism and closed world are appropriate assumptions, treating causality as deduction is dangerous, since one must be careful to exclude non-causal and anti-causal (effect-to-cause) conditionals from any knowledge base if one is to distinguish cause-effect from other kinds of inferences.

Per se, there is no reason to seek an implier of some given fact. The set of possible impliers includes all sorts of riff raff, and there is no obvious contrast set at that level to set up reasoning by exclusion. But there is a reason to seek a possible cause: broadly speaking, because knowledge of causes gives us powers of influence and prediction.

John R. Josephson

Science as crossword puzzle solving

The model is not . . . how one determines the soundness or otherwise of a mathematical proof; it is, rather, how one determines the reasonableness or otherwise of entries in a crossword puzzle. . . . The crossword model permits pervasive mutual support, rather than, like the model of a mathematical proof, encouraging an essentially one-directional conception. . . . How reasonable one’s confidence is that a certain entry in a crossword is correct depends on: how much support is given to this entry by the clue and any intersecting entries that have already been filled in; how reasonable, independently of the entry in question, one’s confidence is that those other already filled-in entries are correct; and how many of the intersecting entries have been filled in.

Yours truly — himself an avid crossword puzzle solver — cannot but agree.

In inference to the best explanation, we begin with a body of (purported) data, facts, or evidence and search for explanations capable of accounting for them. To have the best explanation is to have, given context-dependent background assumptions, an explanation that accounts for the evidence better than any competing alternative — such that it is reasonable to regard the hypothesis as true. Even though we inevitably lack deductive certainty, this form of reasoning licences us to regard belief in the hypothesis as reasonable.

Accepting a hypothesis means believing that it explains the available evidence better than any competing hypothesis. Knowing that we have earnestly considered and analysed the available alternatives, and have been able to rule them out, warrants and strengthens our confidence that our preferred explanation is indeed the best explanation — that is, the explanation which, if true, provides us with the greatest understanding.

This does not, of course, imply that we cannot be mistaken. We certainly can. Inferences to the best explanation are fallible, since the premises do not logically entail the conclusion. From a strictly logical point of view, inference to the best explanation is therefore a weak mode of inference. Nevertheless, when sufficiently strong, such arguments can be warranted and may yield justified true belief — and hence knowledge — despite their fallibility. As scientists, we sometimes — much like crossword solvers or detectives — experience disillusionment. We believe we have reached a robust conclusion by eliminating alternative explanations, only to discover that what we took to be true is, in fact, false.

This does not necessarily mean that we lacked good reasons for our belief at the time. If one cannot live with such contingency and uncertainty, then one has chosen the wrong line of work. If it is deductive certainty one seeks — rather than the ampliative and defeasible reasoning characteristic of inference to the best explanation — then one should pursue mathematics or logic, not science.

What I do not believe — and this has been suggested — is that we can usefully lay down some hard-and-fast rules of evidence that must be obeyed before we accept cause and effect. None of my viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non. What they can do, with greater or less strength, is to help us to make up our minds on the fundamental question — is there any other way of explaining the set of facts before us, is there any other answer equally, or more, likely than cause and effect?

Austin Bradford Hill