The Odds – Saybia (from Eyes on the Highway)
I don’t even care about basketball, but I can’t help but point out the breakdown in logic in this recent post by Matt Yglesias. Taking issue with Bill Simmons’ prediction that the Trailblazers will go 41-41 he says:
To merely go 54-28 again would require the team to regress somewhat. To win 41 games would involve a regression as big as the step forward that would be require to win 75 games and become the greatest team of all time.
Yikes.
What is missing in this analysis is a recognition of the importance of ‘regressing to the mean.’ Put simply, on balance any performance is substantially more likely to revert to the mean performance of the pool from which it is drawn than it is to go further toward the extreme.
In this case, the pool is professional NBA teams and the mean is (obviously) .500. All else being equal, we’d assume that any team who wins more than 41 games is going to do worse in the next year, and vice versa. And the further away from the mean, the harder it will be sustain their performance. And it will be particularly difficult to improve.
Like I said, I don’t care about basketball, but the principle is pretty important for a lot of areas. It’s the general idea which helps us remember that exceptional economic performances are…well…exceptional, and thus unlikely to be sustained or improved upon. It’s what reminds us that if we gamble and win, it was still a gamble. And if we try again, the odds are much closer to 50% than they are to 100%.
In short, it’s a principle which we very easily lose track, meaning that we end up making extremely important decisions based on a false sense of security derived from a relatively small sample size. Forgetting to regress to the mean, and failing to gather a significant enough amount of data to minimize the effect this causes, is a deadly combination: whether it’s for sports predictions or the global economy.