If you ask a philosopher for his thoughts on economics or any other social science you are likely, amongst some long words and even longer sentences, to hear the time honed phrase that correlation does not imply causation. If you have not heard it before then you can forgive yourself. Very few economics courses will even mention it, let alone go into any detail. Yet this little statement may be one of the most important problems faced by economists, and in many ways is the single largest problem we will face. So what is it?
It has throughout history had many names. Philosophers today call it the problem of induction. At its heart it says that although we observe events happening, we cannot prove that an explanation of the process behind why these events happen is the right explanation. This is because the process behind the events is entirely unknown to us and in many cases will prove exceedingly complicated.
Say you want to look at the effect of alcohol taxes on car accidents. You postulate that higher taxes on alcohol will cut its consumption, and thus cut the amount that people drink and drive. If people drive less while drunk, then less of them will be driving into trees, shopping malls and other objects that get in their way.
You decide, being a classical economist, that the best way to explain this relationship is to form a linear model linking alcohol taxes to car accidents. A linear model is essentially one that says a rise or fall in one variable (taxes) will lead to a proportional change in the other variable (car accidents). You then estimate this model using some form of regression analysis (this is how economists by and large estimate relationships) and find that a rise in taxes of 1 penny on the pound leads to a fall in car accident rates of 0.6%.
So we have found our result! The problem we now have is that we cannot be sure that these estimates are accurate. Economists will now turn to a concept called statistical significance. Essentially, is the relationship between alcohol taxes and car accidents a fluke of the data or did it arise because higher taxes tends to coincide with fewer accidents? Say our data passes all of these tests, what has our study learned?
This is where the problem of induction exists. At heart all we have shown is that higher taxes tend to be seen with lower rates of car accidents. We cannot say how this has come about. It could be because people drink less. But is it because the amount of drunk drivers on the road is lower? Or is because the drivers are now less drunk? Perhaps it is because areas with higher alcohol taxes tend to have safer drivers?
If the last reason is the main cause behind fewer car crashes and the government decides as a result of our study to raise alcohol duty in order to cut car crash numbers then they are likely to be very disappointed. This is why this is such a problem.
Perhaps the worst bit about it is that we won’t know when this problem is important and when it isn’t.
Here are a few links to look at if you are interested in reading more into it: