Don’t Forget the Basics of Correlation vs. Causation

August 29th, 2018
Rich Raquet | Chairman, TRC
Hero Image: Don’t Forget the Basics of Correlation vs. Causation
Last week I was watching the CBS morning news and they had a story about a new study that indicated that your attitude toward gym class as a child shaped your attitude toward exercise your entire life. After watching the story I am convinced that it is another case of causation being confused for correlation.
The basics were that kids who reported loving gym class were far more active decades later than kids who reported finding it stressful (“I was always picked last”).  I don’t doubt this correlation. My problem is they spoke of a need to make gym class more inclusive so that all kids grow up to exercise more. In other words, if we can take the stress out of gym for kids who are not good at sports, we can get them to love exercise more. I’m not sure achieving the first part of this is possible and I’m even more certain that even if we do, it will not alter future behavior.
I don’t see how you can eliminate the anxiety about gym class without eliminating the physical activity. Sure you could eliminate picking teams and save that humiliation, but once the games begin the kids who are poor at sports will continue to feel anxiety. Even if you simply make it an exercise class, the kids who are out of shape will stand out. Short of one-on-one classes I don’t see how you can fix the problem.
Doing so is also not likely to make us more active as adults. The kids who were not good at gym were not good for a variety of reasons but likely they either lacked the natural talent OR more likely the interest in sports that the athletes had. Gym class wasn’t the cause of this! If there had been no gym class I would bet that the kids who didn’t like gym class would still be less active than those that did (those who loved it and/or were good at it). I’d point the “causation arrow” backwards….if you love sports as an adult you probably liked gym class.

Statistical Principles Should Be Explained

It is easy to forget that we have to play a role in explaining statistical principles in our reporting and not just when doing sophisticated work like Discrete Choice Conjoint, Max-Diff, Segmentations and Regressions. Our direct clients likely understand causation and correlation issues, but it is important to know that their internal clients may not. Clear justification for pointing the “causation arrow” must be provided in reports and presentations. Just as important is knocking down attempts to point the arrow based solely on correlation. Otherwise they may walk away with a completely false assumption and not double back with researchers to validate it.

Can You Project Results to the Population?

This study is also useful in highlighting another common mistake made by internal clients. I was telling an old friend about the study and he said “that can’t be right, you hated gym class and you are far more active now than when we were kids”. Imagine me in a focus group telling my story of how much I hated waiting to be picked for a team and how my memory of that humiliation caused me to exercise more and more as an adult. The internal client stands up and says “That’s how we make people healthier…more humiliation in gym class!” In that case, someone will be in the room to point out that one person’s story is not projectable to the population…but that’s another blog.