The final version of my manuscript Philosophy of Epidemiology is off to the publishers, and I hope it counts as self-promotion rather than self-plagiarism if I post the Synopsis here:

The content of the book can be summarised as follows. Causation and causal inference are over-emphasized in epidemiology and in epidemiological methodology, and explanation and prediction deserve greater emphasis. Explanation is a much more useful concept for understanding measures of strength of association (Chapter 3) and the nature of causal inference (Chapters 4 and 5) than causation itself is, because causation is really only part of what we seek to measure and infer respectively. What epidemiologists really seek to do is explain, and their practices are seen much more clearly when described as such. Explanation is also the central notion to an adequate theory of prediction, a topic which has been sadly neglected by both philosophers and epidemiologists (Chapter 6). To predict, one must explain why what one predicts is going to happen, rather than some – but certainly not all – other possible outcomes (Chapter 7). Just like an adequate explanation, a good prediction must be sensitive to the alternative hypotheses it needs to dismiss and those it can safely ignore.

These themes are developed in Chapters 3-7. The remaining chapters tackle more specific problems, and apply the lessons learned where appropriate. Thus Chapter 8 concerns measures of attributability, which are not conceptually straightforward; and the lesson is that an outcome is attributable to an exposure to the extent that it is explained by it. Chapter 9 concerns “risk relativism”, an unfortunate degenerative tendency of thought some epidemiologists have identified in their peers. Risk relativism is diagnosed in Chapter 9 as a case of physics envy, exacerbated by a tendency to seek a univocal measure of causal strength, rather than a context-appropriate explanatory measure. Chapter 10 examines “multifactorialism”, another modern epidemiological ailment – though considered by most epidemiologists to be a profitable mutation. Multifactorialism is criticised for dropping the requirement that diseases be defined in relation to explanatory causes. Chapter 11 discusses the various embarrassments that lawyers have inflicted upon themselves in trying to use epidemiological evidence. Again, the lack of attendance to explanatory questions in favour of causal ones is partly to blame for the mess. Lawyers reasonably resist the explanation of particular trains of events by general statistical facts; but to refuse to admit those facts as evidence for the truth of particular causal explanations is plainly unreasonable. The theme, then, is that explanation deserves more attention epidemiological attention, and causation less. We will revisit the theme in the concluding Chapter 12.

[Philosophy of Epidemiology is being published by Palgrave Macmillan in the series New Directions in the Philosophy of Science and will appear in 2013.]

In 1965, Austin Bradford Hill identified nine “viewpoints” from which an association may be viewed in seeking to decide whether it is causal (Hill 1965). Hill’s nine viewpoints, often wrongly called “criteria”, have been repeated and rehashed very often indeed, with both approval and disapproval. Despite the profusion of developments in technical and non-technical literatures on causal inference since then, Hill’s viewpoints remain a starting point for discussions of causal inference in epidemiology.

There are good reasons for this. Technical advances do not eliminate the need for human judgement, and Hill’s nine viewpoints provide some structure for arriving at these judgements. And it is fair to say that the non-technical literature has not substantially advanced, at least in what it offers for practical purposes. There are other similar lists of guidelines, but it is hard to identify any clear advance, in the non-technical sphere, beyond the basic idea of identifying a few things to bear in mind when trying to decide if an association is causal. For example, Jon Williamson and Federica Russo suggest that, in the health sciences, evidence for causality must come from both population-level studies and experimental studies of underlying mechanisms (Russo and Williamson 2007). This claim may be theoretically interesting (for criticism see Broadbent 2011) but it is clearly intended as a theoretical analysis, and adds little from a practical perspective. Both items in question are covered by Hill’s list; the difference is that Hill does not think any item on his list is necessary, and the claim that evidence concerning underlying mechanisms, in particular, is necessary for a causal inference is highly doubtful in an epidemiological context, and identified as such by Hill. But however that difference is settled, as long as the debate is about what kind of evidence is or is not desired or required for a causal inference, we are not offering anything substantially more useful than what Hill has already offered.

Where should we start, if we wish to move beyond Hill-style lists? Lists of guidelines like Hill’s suffer from notable defects, despite their usefulness. They are open to misinterpretation as criteria, or as primitive algorithms for causal inference. They are a magnet for fruitless debate about exactly what should make the list, what should not, what order they should appear in, what weights to give the various components, and so forth. But most importantly – and this, I think, should be our starting point – they do not provide any clear bar that evidence must clear. The crucial question that making a decision imposes is: is the evidence good enough to act on?

A list of guidelines such as Hill’s has some heuristic value, but it does not tell us, in even the broadest terms, what constitutes enough of each item on the list. The guidelines tell us what the bar is made of, but they do not tell us how high it is. One way we might advance beyond Hill’s viewpoints, then, is to ask how good evidence needs to be before it warrants a causal inference, where that causal inference is important for a practical decision.

Broadbent, A. 2011. ‘Causal Inference in Epidemiology: Mechanisms, Black Boxes, and Contrasts’. In Causality in the Sciences, ed. Phyllis McKay Illari, Federica Russo, and Jon Williamson, 45–69. Oxford: Oxford University Press.

Hill, AB. 1965. ‘The Environment and Disease: Association or Causation?’ Proceedings of the Royal Society of Medicine 58: 259–300.

Russo, F and Williamson, J. 2007. ‘Interpreting Causality in the Health Sciences’. International Journal of the Philosophy of Science 21 (2): 157–170.

If anyone is interested in looking at a book manuscript on philosophy of epidemiology, or on any parts thereof, please get in touch. The manuscript is under contract and has been delivered to the publisher, so this is a good time for comments and criticism. Table of contents can be accessed here:

2012-06-29 contents

…with apologies for the “undefined bookmarks”. The publishers have already indicated that the title should be “Philosophy of Epidemiology” or something similarly descriptive – opinions on this also welcome.

Apologies for the recent inactivity on this blog – the book is a big part of the explanation.

James Woodward’s book “Making Things Happen” does not feature “prediction” in the index. But what is the point of making things happen if you don’t know what?