Some thoughts on causation


What does it mean for one event to cause another? One place to start with this question would be to distinguish between causally necessity and causal sufficiency. We can say A is a causally sufficient condition of B if A always causes B, and a causally necessary condition of B if B cannot occur without A. Fulfilling one or more of these conditional relationships might be a property of a cause, but it cannot be a necessary or sufficient condition of causality. This can be illustrated on a case-by-case basis.

First off, we would have to add the additional condition that for condition A to cause be it also has to occur prior to B. This is not true of A merely being a condition of B. Sufficiency and necessity are about the logic of occurrence, not the mechanics of processes. Getting prescription lenses is a (quasi) sufficient condition of having poor eyesight, but not a cause (all people with prescription lenses have bad eyesight). The causality is the other way round. This restriction also has to be made of necessary causes. Someone might always wear a coat when it is cold outside. This would make wearing a coat a necessary condition of coldness (it cannot be cold without this person wearing a coat). But the cause here is also backwards.

Even if we add in the restriction of a condition being prior to its effect to be a cause, I don’t think this conditionality can be a necessary or sufficient condition for causation. Shooting someone at close range is neither a necessary nor sufficient condition for someone’s death, but it can be a cause. Causation therefore does not necessarily fulfil either conditionality relationship. But is the fulfilment of this conditionality, if the events are in the right sequence, a sufficient condition of causation? Also no, I think. Imagine someone who only orders a particular dish when they go to a certain restaurant. They also always drink a glass of water with any meal. Ordering that dish is a sufficient condition for drinking a glass of water, and drinking a glass of water is a necessary condition for having that dish, but neither delivers any kind of causal explanation of the other. We can shift round the order in which the events happen in any way we like, by imagining some additional conditions (perhaps drinks are always served first, perhaps this person prefers to have their meal before their drink.)

What about fulfilling both relations, i.e A being a necessary and sufficient condition of B? This is the suggestion of the philosopher David Lewis.  The problem with this is it implies an event can be caused by another if and only if it is impossible for that event to be caused by something else instead of that particular cause. This cannot be a necessary condition of causality. What’s more, if we modify the restaurant example, it cannot be a sufficient one either. Perhaps someone always and only orders a particular drink at a restaurant and only and always orders a particular meal at that same restaurant. Both can be necessary and sufficient conditions for one another without the implication of causation.

We might then say, OK, so this necessity and sufficiency are not necessary or sufficient conditions of causal relationships, but might they still be common properties? My problem here is that when dealing with phenomena on the macro level, this is almost never actually true of causal relationships. A politician may win an election by a landslide, but perhaps if they were found to have committed a murder the week beforehand they would not have done so. As for necessity, when it comes to the interesting conditions, it is equally always possible to imagine different conditions, which might have produced the same result. Perhaps Trump would not have won in 2016 without winning Pennsylvania and Florida, but these conditions aren’t strictly necessary. Perhaps he might have won California.

One way of trying to resolve this problem is to think about these conditions in probabilistic terms. It is, after all, logically possible for Trump to have won the Presidency without winning Pennsylvania or Florida, but it is however highly unlikely that that ever would have occurred. Smoking may not always cause cancer, but it sure does increase the probability of it occurring. This too has some problems with it. Imagine I can only just afford my rent, and I only pay the rent on time every other month, when I manage to restrain myself at the weekend. If I buy a lottery ticket, in all probability I am decreasing my chances of having enough money to pay the rent. But in the event that a surprised landlord asks me how I managed to pay the rent for the last 6 months on time, the fact that I bought a lottery ticket could indeed be cause, if it wins, even though buying it actually decreased the probability of the event occurring.

One way one might try and get out of these difficulties is through a kind of modified ceteris paribus. Causal explanations do not require absolute necessity, it simply has to be the case that all other things being equal an event would not have occurred without a particular condition being the case. A similar counterpart can be thought of for sufficiency. The problem with this, is we can never really specify what the other things are which have to be equal. It might then be said that we don’t have to, we just have to assume no additional change. This is fine for a model, where we can think of a closed system with a finite number of elements. But the world is complex, and the counterfactual has to be presumed to occur in this complex world, one where the result of the other conditions being ‘the same’ is unclear in the counterfactual (or whether, indeed, it is coherent to think of them as being such). 


Perhaps the best way forward is to ditch this kind of thinking about causation in large scale historical developments, and to leave the formal logic of conditions to models which seek to provide abstract generalisations. This is possible if we take explanations of historical developments as a fundamentally different activity from modelling. Models seek generalities as to what will happen (or assign a probability to an outcome) under certain precisely specified conditions. Historical explanations accept the complexity of the world and the fact that the conditions are not fully known or describable. They are therefore more about resolving mysteries: they make an event comprehensible and seek to render it unsurprising. It is not the same as finding a way of making the event likely, let alone necessary, merely getting us to the point where nothing seems to violate our fundamental expectations of how the world works. We might replace causal sufficiency and necessary statements more like the following: given what we know about the situation, does it make sense that under these circumstances, something like this would occur? Or would we, but for this particular fact, find the development puzzling, in a way that violates the way we assume the world works? 

No comments:

Post a Comment