Here’s a tricky thing with statistics that constantly throws us off: correlation does not imply causation. I forget this all the time. Here are the basics: Birds have feathers. Birds fly. Therefore, I must have feathers in order to fly. As a statement of logic, this works marvelously. It is perfectly consistent and Gary Klein taught us that we really want consistency.
And hey, it’s actually sensible. Feathers are a marvel of natural design—a lightweight material that creates fabulous draft with minimal effort. So when you see early experimentation with human flight, you kinda get where these brave souls were coming from. If not feathers, then something very close:
These were proper experiments and, misguided though they were, the pioneers should be lauded because they took an obvious correlation and tested it for cause. We should all do this in some capacity. Testing correlations for cause is the hallmark of a great experiment and the start of great discovery.
It’s only when we treat correlation as causation from the very start that things go bad. Imagine if someone built one of the contraptions in the photos above and then jumped off a cliff in pure, total confidence, resolute in their claim that feathers = flight. Scary, huh?
Big bets, failed efforts, and horrible medical travesties are all tied to such acts of baseless conviction.
So what is correlation good for? It’s powerful for identifying relationships. As a planner, one of my favorite correlations is the relationship between building intensity and net fiscal impact of the built environment for a City’s general fund revenue. In that instance, we found a sensible correlation coefficient above 0.80. Such a finding, which precisely measures the direction of relationship (positive or negative) and the degree of effec, feels like finding new knowledge. Just fantastic. But it doesn’t mean you then change city governance.
As a manager, I’ve enjoyed workplace surveys that uncover the relationship between staff engagement and performance. Engagement increases performance; it’s a positive relationship. But anecdotally, I don’t think performance improves harmony. I’ll explore that in future posts after I run a few tests.
So that’s the key to correlation. It points you in a better direction. That’s all. And that’s enough. The next step is inference, prediction, and experimentation. So let’s admire correlation for what it is: as one of the most powerful statistical elements affecting our world today. It is the device that powers our Netflix recommendations and the growing filter bubble.
Who’s The Bozo?
Most important of all, it is the best test for uncovering a bozo. If a person claims to know about a thing but can’t articulate the correlations and devise a way to test them for cause, they are a bozo.1 Thankfully, bozo isn’t a terminal condition. I’m a bozo on anything where I forget correlation ≠ causation. This week’s featured book, “Naked Statistics”, helps me not do that so much.