Analysing data with causal models can prove whether there’s evidence for causation.
Correlation doesn’t imply causation. You’ve probably heard that before. It’s a true statement that’s important in statistical analysis—if more tall people own cats, that doesn’t mean that cats cause people to be tall, or that being tall gives you a deep desire for cat ownership. But there’s a flaw that catchy little slogan; it’s missing a word. Correlation doesn’t necessarily imply causation. Sometimes it does!
As this video from MinutePhysics explains, just because a correlation doesn’t imply causation doesn’t mean that it can’t imply causation. After all, what would be the point of statistics if not to demonstrate that something causes something else?
Causation: drawing conclusions
The key to drawing conclusions? Context is important. In the cat/height scenario, there are other contextual factors that need to be considered to determine causality or the lack thereof. Where do these tall cat owners live? Is their cat ownership dependent on a third factor, like their environment? Who got there first, cats or humans? By answering these questions and then using the results to pare down the number of causation possibilities, we can get a better idea of what is causing what, although there are a few little wrinkles that make it not quite this simple in the real world.
So be careful when you say that green hair causes people to like pickles–but just because it seems unlikely, doesn’t mean it’s impossible. You just need more data.
Source, video credit: MinutePhysics
This article was originally written for and published by Popular Mechanics USA.