Being able to separate correlation from causation is the cornerstone of good science. Many errors in reasoning can be distilled to this mistake. Let me preempt this section by saying that making this distinction is by no means a simple matter, and much ink has been spilled over the issue of whether it’s even possible in some cases. However, just because there are some instances where the distinction is indiscernible or difficult to make doesn’t mean we should make a (poor) generalization and conclude that all instances are indiscernible or difficult to make.
We can think of general causal reasoning as a sub-species of generalizations. For instance, we might say that low-carb diets cause weight loss. That is to say, diets that are lower in the proportion of carbohydrate calories than other diets will have the effect of weight loss on any individual on that diet. Of course, we probably can’t test every single possible low-carb diet, but give a reasonable sample size we might make this causal generalization.
A poor causal argument is called the fallacy of confusing causation for correlation or just the causation-correlation fallacy. Basically this is when we observe that two events occur together either statistically or temporally and so attribute to them a causal relationship. But just because to events occur together doesn’t necessarily imply that there is a causal relationship.
To illustrate: the rise and fall of milk prices in Uzbekistan closely mirrors the rise and fall of the NYSE (it’s a fact!). But we wouldn’t say that the rise and fall of Uzbeki milk prices causes NYSE to rise and fall, nor would we make the claim the other way around. We might plausibly argue that there is a weak correlation between the NYSE index and the price of milk in Uzbekistan, but it would take quite a bit of work to demonstrate a causal relationship.
Here are a couple of interesting examples:
Strange but true statistical correlations
A more interesting example can be found in the anti-vaccine movement. This example is an instance of the logical fallacy called “post hoc ergo proptor hoc” (after therefore because of) which is a subspecies of the correlation/causation fallacy. Just because an event regularly occurs after another doesn’t mean that the first event is causing the second. When I eat, I eat my salad first, then my protein, but my salad doesn’t cause me to eat my protein.
Symptoms of autism become apparent about 6 month after the time a child gets their MMR vaccine. Because one event occurs after the other, many naturally reason the the prior event is causing the later event. But as I’ve explained, just because an event occurs prior to another event doesn’t mean it causes it.
And why pick out one prior event out of the 6 months worth of other prior events? And why ignore possible genetic and environmental causes? Or why not say “well, my son got new shoes 6 months ago (prior event) therefore, new shoes cause autism”? Until you can tease out all the variables, it’s a huge stretch to attribute causation just because of temporal order.
Constant Condition, Variable Condition, and Composite Cause
Ok, we’re going to have to introduce a little bit of technical terminology to be able to distinguish between some important concepts. I don’t want to get too caught up in making the distinctions, I’m more concerned about you understanding what they are and (hopefully) the role they play in evaluating causal claims.
A constant condition is a causal factor that must be present if an event is to occur. Consider combustion. In order for there to be combustion there must be oxygen present. But oxygen on its own doesn’t cause combustion. There’s oxygen all around us but people aren’t continuously bursting into flames. However, without oxygen there can be no combustion. In the case of combustion, we would say that oxygen is a constant condition. That is, it is necessary for the causal event to occur, but it isn’t the thing that initiates the causal chain.
When we look at the element or variable that actually initiates a causal chain of events, we call it the variable condition. In the case of combustion it might be a lit match, a spark from electrical wires, or exploding gunpowder from a gun. There can be many variable conditions.
The point is you can’t start a fire without a spark. This gun’s for hire. You can’t start a fire without a spark. Even if we’re just dancing in the dark. Of course, you could also start a fire with several other things. That’s why we call it the variable condition. But despite all the possible variable conditions, there must be oxygen present…even if we’re just dancing in the dark.
As you might expect, when we consider the constant and the variable condition together, we call it the composite cause. Basically, we are recognizing that for causal events there are some conditions that must be in place across all variable conditions and there are some other conditions that have a direct causal effect but that could be “switched out” with other conditions (like different the sources of a spark).
Separating constant conditions for variable conditions can be useful in establishing policy. For example, with nutrition if we know that eating a certain type of diet can cause weight loss (and we want to lose weight) we can vary our diet’s composition or quantity of calories (variable conditions) in order to lose weight. The constant condition (that we will eat) we can’t do much about.
Conversely, we can’t control the variable conditions that cause the rain, but by buying umbrellas we can control the constant condition that rain causes us to get wet. (Water is wet. That’s science!)
The Argument Structure of a General Causal Claim
Someone claims X causes Y. But how do we evaluate it? To begin we can use some of the tools we already acquired when we learned how to evaluate generalizations. To do this we can think of general causal claims as a special case of a generalization (i.e., one about a causal relationship).
I’m sure you all recall that to evaluate a generalization we ask
(1) is the sample representative? That is, (a) is it large enough to be statistically significant (b) is it free of bias (i.e., does it incorporate all the relevant sub-groups included in the group you are generalizing about.;
(2) does X in the sample group really have the property Y (ie., the property of causing event Y to occur).
Once we’ve moved beyond these general evaluations we can look at specific elements in a general causal claim. To evaluate the claim we have to look at the implied (but in good science explicit) argument structure that supports the main claim which are actually an expansion of (2) into further aspects of evaluation.
A general causal claim has 4 implied premises. Each one serves as an element to scrutanize.
Premise 1: X is correlated with Y. This means that there is some sort of relationship between event/object X and event/object Y, but it’s too early to say it’s causal. Maybe it’s temporal, maybe it’s statistical, or maybe it’s some other kind of relationship.
For example, early germ theorist Koch suggested that we can determine if a disease is caused by micro-organisms if those micro-organisms are found on sick bodies and not on healthy bodies. There was a strong correlation but not a necessary causal relation because for some diseases people can be carriers but immune to the disease.
In other words, micro-organisms might be a constant condition in a disease causing sickness, but there may be other important variable causes (like environment or genetics) we must consider before we can say the a particular diseases micro-organisms cause sickness.
Premise 2: The correlation between X and Y is not due to chance. As we saw with the Uzbek milk prices and the NYSE, sometimes events can occur together but not have a causal relation–the world is full of wacky statistical relations. Also we are hard-wired to infer causation when one event happens prior to another. But as you now know, this would be committing the post hoc ergo proptor hoc fallacy.
Premise 3: The correlation between X and Y is not due to some mutual cause Z. Suppose someone thinks that “muscle soreness (X) causes muscle growth (Y).” But this would be mistaken because it’s actually exercising the muscle (Z) that causes both events.
In social psychology there was in interesting reinterpretation of a study that demonstrates this principle. An earlier study showed a strong correlation between overall level of happiness and degree of participation in a religious institution. The conclusion was that participation in a religious institution causes happiness.
However, a subsequent study showed that there was a 3rd element (sense of belonging to a close-knit community) that explained the apparent relationship between happiness and religion. Religious organizations are often close-knit communities so it only appeared as though it was the religious element that cause a higher happiness appraisal. It turns out that there is a more general explanation of which participation in a religious organization is an instance.
Premise 4: Y is not the cause of X. This issue is often very difficult to disentangle This is known as trying to figure out the direction of the arrow of causation–and sometimes it can point both ways. For instance, some people say that drug use causes criminal behaviour. But in a recent discussion I had with a retired parole officer, he insists that it’s the other way around. He says that youths with a predisposition toward criminal behavior end up taking drugs only after they’ve entered a life of crime. I think you could plausibly argue the arrow can point both directions depending on the person or maybe even within the same person (i.e., feedback loop). There’s probably some legitimate research on this matter beyond my musings and the anecdotes of one officer, but this should suffice to illustrate the principle.
Conclusion: X causes Y.
Premise 2, 3, and 4 are all about ruling out alternative explanations. As critical thinkers evaluating or producing a causal argument, we need to seriously consider the plausibility of these alternative explanations. Recall earlier in the semester we looked briefly at Popperian falsificationism. We can extend this idea to causation: i.e., we can never completely confirm a causal relationship, we can only eliminate competing explanations.
With that in mind, the implied premises in a general causal claim provide us a systematic way to evaluate the claim in pieces so we don’t overlook anything important. In other words, when you evaluate a general causal claim, you should do so by laying out the implied structure of the argument for the claim and evaluating them in turn.