!VAR HAGENDOORN

Correlation, Causality and Capacities

Causality is a problem in every part of science, but it is a big problem in cognitive neuroscience. It seems as if it has only recently dawned on researchers that neuroimaging studies only measure correlation (if it is measured correctly that is). Recently a number of papers have been published that apply a Granger causality test to fMRI data in order to examine the direction of information flow. The potential for interdisciplinary cross-fertilization is endless.

Of course neuroimaging is only one technique used in cognitive neuroscience. Neuropsychological case studies demonstrate that damage to different parts of the brain may cause specific impairments. Transcranial magnetic stimulation (TMS) makes it possible to temporarily disrupt the processing in parts of the brain that lie close to the surface, thus creating a virtual lesion.

The fact that damage to a particular part of the brain causes certain cognitive, emotional or behavioral impairments does not tell us what that part of the brain does, what input it receives from other regions and what output it produces. So what do all those studies actually tell us?

In trying to answer that question I found some useful ideas in the work of the philosopher of science Nancy Cartwright. One of her books was on the reading list for a course when I was studying econometrics. When I did a search for her to see what she has published in the intervening years I discovered that she has a namesake who is the voice of Bart Simpson!

I cannot do justice here to the full length (and breadth) of her work and arguments. I can recommend this recent paper.

There's a great chapter by James Woodward, another prominent philosopher of science, in a recent book Nancy Cartwright's Philosophy of Science, in which Woodward elucidates the differences between his and Cartwright's ideas on causality. (Woodward's central claim is that causal relationships are invariant under interventions).

An important concept in the work of Nancy Cartwright is that of capacity, a concept she borrowed from John Stuart Mill. Her book Nature's Capacities and their Measurement (1989) contains a lengthy argument in which she takes some standard econometric practices (aha) as exemplary for deriving causes from probabilities. Incidentally Woodward's work also goes back to the work of some of the pioneers in econometrics. (I much preferred thinking about econometrics than doing regressions).

The concept is best explained by way of an example. An aspirin has the capacity to relieve headaches, even when it doesn't do so at all times and even when it may be necessary to take another tablet. It may seem as if we haven't gained much, why not just say that aspirin causes the headache to go away? Let's take another example. Suppose you have planted several seeds in your garden and that after a few months some have grown big, some have remained small and some have perished. Why? Perhaps you didn't give them enough water, or too much, perhaps you should have given them more manure, perhaps they needed more direct sunlight, perhaps a genetic variation caused some of the plants to grow big. By varying one factor while keeping all others constant horticulturalists try to establish the perfect mix. But the essential point is that each of the factors has the capacity to contribute to the plants' growth.

The reason that cars have safety belts and airbags is that they have the capacity to prevent injury during a car crash. That is not to say that either will prevent injury. Consider a car crash in which the driver gets injured. We might say that the crash caused the driver to get injured. But what if the driver wasn't wearing his safety belt? Then we might say that he got injured because he wasn't wearing his safety belt. And what if due to some malfunction the airbag didn't open? Now we might say that he got injured because the airbag didn't open. If it turns out that in a number of similar cases the airbag didn't open the car manufacturer may be forced to announce a recall of all cars of the same model. The example is simple enough, but the legal and financial stakes can be high.

I am less convinced now than I once was that the concept of capacity solves the problem of causation, it merely rephrases it. It is not clear what the ontological status of capacities is. Are they properties or properties of properties? And what have we gained if all we can say is that, all things equal, capacities manifest themselves?

The reason that I still like the concept is that it offers a way of talking about causality in the presence of multiple causes. It also seems more natural to speak of different parts of the brain having the capacity to produce a certain outcome rather than causing it. Some parts of the brain do one thing when stimulated with one neurotransmitter (e.g. opioid) and another thing when stimulated with another neurotransmitter (e.g. dopamine).

So the scientist and the philosopher in me are as confused as ever, but the artist has learned some valuable lessons. An artwork produces various effects in a spectator, reader, listener etc. But the artwork is not the sole cause, it is just one of many factors. As an artist I am concerned with the aesthetic properties I want to achieve and not with the effect the work has on the spectator. The spectators looking to experience a certain effect search for the artwork that causes it. That is one lesson.

There are various regularities in the way people attribute causality. If a moving object collides with a stationary object, the moving object is seen as the cause. If both objects move, but one diverts its course, the object which exhibits the greatest kinematic variation is seen as cause. If in the car crash example one car is big and the other is small, people tend to see the big car as the cause, until proven otherwise. As an artist I create the illusion of causality. So as an artist I can use the biases in the way people perceive causality to manipulate their feelings. That is the second lesson.

Links

Logothetis, N.K. (2008). What we can do and what we cannot do with fMRI. Nature 453 (7197): 869-878.

Friston, K.J. (2009). Modalities, modes, and models in functional neuroimaging. Science 326 (5951): 399-403.