This video provides some criteria by which a causal theory might be judged
This video was produced in January 2013 for my research methods seminar (SOC 334) at Queens College in the City University of New York. If you are enrolled in this class, you must also complete the assigned readings and exercises. Instructions will be posted on my web site and on Blackboard.
This video is part of an experiment in teaching with technology. In the coming semester, I plan on releasing other videos and an overview of this experiment. If you are interested, please visit my web site (www.josephncohen.org) and share your questions, corrections, thoughts or criticisms. I appreciate any feedback or advice on the video’s content (admittedly poor) production, or the format of moving my lectures to sets of short (5-10 minute) streaming videos.
In statistics class, you might have heard that correlation is not causation. This means that just because you see two things together doesn’t mean that one thing is causing the other. The problem is that we often need to understand what causes what to resolve practical problems. We can’t help students learn better, we can’t cure the sick, or encourage people to drive safely unless we know why students fail to learn, why people get sick, or why people drive dangerously. In this video, I’ll review three criteria for judging a causal relationship: correspondence, precedence, and the elimination of alternative causes. In practice, it’s impossible to prove causality, but these three criteria give us an ideal towards which we can strive. The closer we come to meeting them, the more confident we can be that a causal relationship exists.
Causality is when one thing causes something else. For example, we tend to get sick in the winter. We assume that the cold is somehow playing a role in making us sick, so we tell our kids to bundle up or else they’ll catch a cold. We’re assuming a causal relationship that coldness generates illness, and that assumed causal relationship affects our choices. We nag our kids. The problem is that we often assume causal relationships when none exists. If we base our actions on faulty assumptions about causal relationships, our choices will not achieve what we want them to achieve. In an earlier video, I described a theory as a relationship between concepts. I said that if we had evidence that two concepts were related, the theory was more likely to be true. But I meant that it is more likely to be true than if no relationship existed. I did not mean that the theory is proven true. In fact, it’s nearly impossible to prove a theory true. You can subject them to tests to try to disprove them and accept theories that survived your attempts to disprove them as being reasonably good, but you can’t prove them to be true.
What are these tests? I’ll talk about three: correspondence, precedence, and the elimination of alternative causes. Correspondence means that the cause and effect appear in the same person or country or whatever unit you’re analyzing. If being cold causes you to get sick, then the people who are exposed to cold should be more likely to be sick. If being educated causes you to earn more money, then people with more education should be more likely to earn more money. Correspondence, or what your statistics teacher might have referred to as correlation, is only one basic criterion for ascertaining causality. There are two others. Precedence means that the cause must come before the effect. If we see that smokers are more likely to get cancer, our cancer victims should have been smoking before they got cancer. If people start smoking after they get cancer, well, then we can’t say that smoking is the cause. If we see that children become rich after they accumulate education, then we can say that education causes wealth. However, if only rich kids end up going to school, then the causal relationship might run in the opposite direction. This is precedence: the cause must come before the observed effect. The final criterion is the elimination of alternative explanations. This is a tricky one. It’s ultimately impossible to prove completely, which is why we can’t prove theories to be true. We can only show that they have withstood repeated attempts to prove them untrue. A common mistake that we make when assessing causality is to be tricked by spurious relationships. Spurious relationships occur when we see a relationship between two concepts, but we don’t realize that a third concept is driving them both.
A good example of a spurious relationship comes from Newman’s book on social research methods. Imagine, for example, we see that children who sleep with night lights on tend to need glasses when they’re older. We see a correspondence, and we can assume precedence because children generally don’t get glasses until they’re a few years old. But there might be an outside cause that’s affecting these two things. For example, if parents with bad vision tend to use night lights because it’s hard for them to see in the dark and they genetically pass on vision problems to their children, that would be an example of a spurious relationship. That third factor is causing both what we think to be the cause and its effect. Another example points out that there’s a relationship between the amount of ice cream sold and deaths by drownings. We could look at correspondence and think that the ice cream consumption is causing people to drown, maybe they’re not waiting that one hour you’re supposed to wait before you get in the pool. But the problem is that both of these things are being caused by an outside factor: the weather. When it’s hot, we tend to consume more ice cream and we also tend to swim more. When more people swim, more people drown. Another example comes from Moore and McCabe, who find a relationship between teachers’ salaries and the price of liquor. Maybe we assume that teachers are drinkers and as they get more money, they go out and buy more liquor, allowing liquor vendors to raise prices. But the outside factor here is time: both teachers’ salaries and the price of liquor have been rising over time.
In conclusion, a causal relationship exists when something causes something else, when something affects something else. Studying causes your grades to get better, eating well causes your body to be healthier, and so on. We need to unpack causal relationships to solve problems, but it’s often hard to figure out whether or not causal relationships exist. I described three criteria for figuring out whether or not we have a good basis for believing a relationship is causal. There’s correspondence: the cause and the effect must occur together in the same people or whatever unit of analysis we’re looking at. The second is precedence: the cause must come before the effect. And the third is the elimination of alternative explanations or causes. This is impossible to do in practice, which is why we can’t prove a causal relationship to exist. But if we subject it to a lot of attempts to refute it and we see that a causal relationship withstands many possible alternative explanations and we haven’t found one that refutes it, then we can be pretty confident that it’s a half-decent causal theory.