Need scientists worry about philosophy? Or should philosophers get off their backs and let them do their work in peace? Unsurprisingly, many scientists want to stay clear of philosophical discussions. What is more disturbing is when I hear philosophers themselves announce that our discipline has nothing useful to offer science. In my view, they could not be more wrong.
I study causation. Scientists search for causation. It is clear that we have some interests in common, but our approach is different. I want to know what causation is and what it means for something to cause something else. Scientists use methods to discover causes, choosing the ones they think are best suited for the task. But how can one judge which method is better or worse for picking out causation unless one already has some idea of what causation is?
There are many philosophical theories of causation available and no general agreement about which one is correct. Is causation the same as regularity, difference-making, probability-raising, manipulability, energy transference, tendencies, or perhaps all of these? And what difference does it make to the scientist, one way or the other?
As it happens, scientific methods are not philosophically neutral. They bring with them all sorts of philosophical commitments, especially but not exclusively about causation. These days I am writing a book with Stephen Mumford on causation in science. During this work I’ve come to realise just how much positivism and Humean philosophy has influenced the normative and methodological foundation of science in general. In the book we encourage the scientist to critically reflect upon this philosophical foundation, while also offering an alternative. In our research project, Causation, Complexity and Evidence in Health Sciences (CauseHealth), we offer an alternative to positivism for medicine.
So how do scientific methods involve philosophical assumptions about the nature of causation? A quick glance at some basic methodological approaches might help illustrate this point.
Epidemiological and other statistical methods use correlation data to search for causation, emphasizing large amounts of data and proportion of outcomes. The idea is that more data will lead to more accurate causal conclusions. Ideally, one might think, if we had a complete set of correlation data—past, present, and future—one would also have complete causal knowledge. Philosophically speaking, this fits well with Hume’s regularity theory of causation and empiricist agenda.
Comparative methods allow us to search for causes by looking at the difference between two set-ups: one in which the cause is present (test) and a second in which it is absent (control). In randomized controlled trials (RCTs), if the outcome is more frequent in the test group than in the control group, one concludes that the increase is due to the intervention rather than the background conditions, which should be evenly distributed between the groups. The cause is then understood as a difference-maker, as suggested by Hume and Lewis.
Most scientific methodologies will include both regularity and difference-making in some way. In a lab experiment, one compares what happens in the case of intervention with what happens without it, and usually with some repetition. Instead of randomization of background conditions, these are carefully controlled for. By isolating the cause from interfering factors, one expects to better see its causal role. Experimental methods also involve an assumption of manipulability: that by changing the cause one can also change the effect. This is the basic idea of the interventionist theory of causation, with Woodward as its main proponent. Unlike statistical methods, however, one might here think of causation as a singularist and intrinsic matter, rather than as an instantiation of a general causal claim or law, à la Hume.
What we see, then, is that the scientific methods carry with them assumptions about the nature of causation. But there is more. As a strict empiricist, Hume thought that we can only know what can be observed. One might observe two events, A and B, but we cannot see whether A caused B. The causal link itself is not directly observable. Instead, we must find ways to derive causation from occurrences, for instance, of the relevant types of events. Causal relations are then studied via observables, or data. Unlike causal theories about underlying mechanisms—which might force us into metaphysical speculations—regularities, difference-makers, probability-raisers, and robustness can all be derived from data alone. If causation is accessible via observation data in this way, we are living the inductive dream of Bacon, generating causal hypotheses and theories directly from data.
This empiricist ideal is reflected in evidence-based medicine, which emphasises the need for data, and preferably lots of it, in order to draw causal conclusions. Population studies are ranked higher than mechanistic knowledge and clinical experience, and randomized controlled trials and their meta-studies are thought to give the highest evidence of causation.
Evidence based medicine was introduced in the 90s, and today its methodology has spread: public policy, education, management, and many others. Within medicine, however, we can detect a move away from the ideals of evidence-based medicine and practice, and towards a more person-centred healthcare. Criticisms of current medical thinking come from medical researchers, practitioners, and philosophers of science, and include a wide range of issues, including
- methods (quantitative vs. qualitative methods)
- models (biomedical vs. bio-psychosocial model)
- ontology (reductionism and dualism vs. holism)
- causation (multi-factorial vs. mono-causal; statistics vs. mechanisms)
- probability (frequentism vs. propensities)
- practice (evidence-based vs. person-centred)
Kuhn saw it as a sign of crisis in a paradigm when its members start participating in philosophical discussions, which many of these debates seem to involve. Is medicine going through a scientific revolution? I would say no. This debate is not a disagreement over scientific theories or interpretation of results. In a recent paper, I argued that if there is a paradigmatic revolution going on within medicine, it is one of ontology. While evidence-based medicine has a solid foundation in positivism and Humeanism, person-centred healthcare is more consistent with dispositionalism.
One major difference between Humeanism and dispositionalism is the role of context in our thinking about causation. Hume thought that same cause should give same effect. For him, causation is linked to robustness, or perfect regularities, which is the opposite of context sensitivity. If something really is a cause, it should be able to bring about its effect over a variety of contexts. The way to test causation would then be to put a causal factor in different contexts and see whether the effect still follows. Such an idea can be found in the notion of invariance, for instance, which some think is helpful for understanding causes.
This methodology is used in all the sciences, and also in medicine. While the default assumption is that all patients are different, causation itself is thought of as something robust and context insensitive. Causes of symptoms, diagnosis, treatment, and recovery are typically approached by systematically searching for same cause and same effect:
- same medical history, same symptoms (aetiology)
- same symptoms, same diagnosis (diagnostics)
- same diagnosis, same intervention (standardised treatment)
- same intervention, same effect (tested through RCTs)
Individual variations are included in the studies, of course, but they are not what is in focus when trying to establish causation. Within RCTs, one studies the link between an intervention and an outcome. In retrospective case-control studies, one starts with a condition and searches for common causes. Perhaps something was the same for all the patients? If so, this might be the cause of the condition.
Such a common cause has been found for many medical conditions, but far from all. 30–50% of all health complaints in industrialised countries today are so-called medically unexplained symptoms. These conditions include low back pain, tension type headache, fibromyalgia, chronic fatigue syndrome, multiple chemical sensitivity, general anxiety disorder, irritable bowel syndrome, and many other chronic conditions. Medically unexplained symptoms remain a problem for medicine, since no one has been able to find common causes, common sets of symptoms, a clear psyche–soma division, clear-cut classifications, or common experiences for these conditions. Instead, each patient seems to have a unique combination of symptoms and a unique expression of the condition, what is called ‘medical uniqueness’.
If we assume a Humean notion of causation—universal, repeatable, and robust—then uniqueness, heterogeneity, and causal complexity represent a methodological challenge. From a dispositionalist perspective, however, these are not problems for causation, but typical of it.
The dispositionalist, unlike Hume, emphasises complexity, singularism, and context sensitivity as essential features of causation. One should expect that the same causal factor will tend to give different effects when placed in different contexts. Adding a lit match to a candle will give a very different outcome than it would in combination with dynamite or gasoline. Understanding how a causal factor will interact with a certain context is thus crucial for understanding causation. But for this we need to also look at causal mechanisms and the unique context to which the cause is added. In the case of medicine, that means the patient. Dispositionalism reveals the importance of tailoring a treatment to the patient by looking at their total situation. Each patient will meet the treatment with a whole set of causal factors—lifestyle, diet, biology, and medical and personal history—and many of these will be unique to them.
By replacing the positivist and Humean scientific framework with a dispositionalist one, we get a shift in focus: away from same cause, same effect, homogeneity, regularity, universal treatment, statistical correlations, and frequencies; towards causal singularism, heterogeneity, context sensitivity, tendencies, medical uniqueness, and individual propensities.
Does this affect how we search for causes in medicine? Absolutely. Evidence-based medicine was motivated by the idea that everyone should get the same treatment for the same medical condition, rather than treatment being decided by individual clinicians. Today we have clinical guidelines that recommend interventions based on what has been proven to work best for most patients in RCTs. But what works best for most might not work so well for some individual patients.
Dispositionalism explains why it is possible, and even to be expected, that patients respond differently to an intervention. Since no causal factor does anything in isolation (no pill will cure a headache while lying in the box), the same intervention is not even the same treatment within two different patients. Learning more about how context influences causal processes will then create more causal knowledge than repeating the same causal set up to confirm the theory. If so, the dispositionalist should give epistemic priority to qualitative studies over quantitative ones, to patient context over population studies, and to causal mechanisms over robust regularities.
To search for causation via epidemiological invariance might be the perfect method for a regularity theorist, assuming that diagnosis, treatment, and patient are all ‘normal’ or statistical averages. If we instead assume that all patients are different, and that causation is essentially sensitive to context, then we should not expect average patients, standard expressions of illness, or one treatment that can benefit all.