Last Sunday’s newspaper headlines read, “Complaints down after liquor curbs,” (The Sunday Times, July 12 2015) which made me pause for a bit. Was it true that limiting the time takeaway-liquor sales could be sold, and banning public consumption of liquor after 1030pm, had caused complaints of ill-behaviour to drop? Reading the story, it would seem so, yet a critical thinker cannot accept such a claim without diving deeper; after all, it could well be a correlation than a causation. In this article, we discuss why we should not be so quick to accept explanations that seem related, but may well not be.
Causation or correlation?
Consider these two statements:
The construction worker dropped a hammer on his foot. His foot is now swollen.
Nancy ate wild mushroom salad for lunch. She now has an upset stomach.
For these two situations, we can quite likely say that the initial event caused the second, especially so for the first statement. In the second statement, if Nancy hadn’t eaten any other thing out of the ordinary, then the likelihood of the wild mushrooms causing her upset stomach is high. But if she had mixed other new exotic ingredients in with her wild mushroom salad, then we cannot be sure that it was the mushrooms, although we can narrow it down to something in the salad. If nothing else has happened that links to the swollen foot or the upset stomach, they can be seen to be causative, meaning one thing caused the other.
Let’s take two other statements:
Every time my dog barks in the morning, it rains in the afternoon.
Mr Jones hired a new sales employee on 1st December. Sales increased 200% for the month!
Does this mean that my dog’s bark had caused it to rain? That’s absurd! My dog may be able to sense the change in atmospheric pressure and barked for us to take the clothes in, but it certainly didn’t cause the rain to fall. And was it the new staff who caused sales to increase 200%? It could be, but it might also not be, because it was December, after all.
These two statements are correlations; the dog barks, it rains. The new staff is hired, sales increase. There is a correlation when two or more events are seen happening together. But are they causative? This requires deeper understanding.
We can all agree that my dog didn’t cause it to rain because we understand the physics. But the sales person? How much of the sales increase was caused by the new employee, and how much of it was caused by the season? For this, we will need to de-seasonalize the information, and look at how sales have been over all Decembers before the new sales person arrived. We also need to look at sales figures after December, and understand both the impact of time, and of the sales person’s contribution, before we conclude that it was indeed because of the new staff that sales had increased.
What this means is that we cannot simply take two correlated events and consider them as causative. You should develop your systems diagrams, and uncover the centre of gravity of the situation before you actually decide if indeed there was indeed causation. Only after you have drawn your system diagram will you understand the complex inter-relationships of drivers that can cause the outcomes that you see. It will make you realize that what you thought to be causative, was merely correlated.
So we come back to the headlines; did the liquor curbs cause the complaints to come down, or was it merely correlative? More work needs to be done, but unfortunately, there is neither time, nor interest, in the powers that be, to pull in the causative studies. It is probably best left to the media to spin the causative allusions and move on.