A Comfortable Margin

I was watching a rerun of Winning Time on ESPN the other day, which was highlighted by Reggie Miller's 8 points in 9 seconds to beat the Knicks, all after Pacers owner Donnie Walsh had given up and retreated to his smoking room. The Knicks had been up by 6 with about 30 seconds left, was it reasonable for Walsh to leave? At what point in a game can you feel safe your team will win?

Just by observing basketball, a few years ago I had decided 6*(n)^.5 (where n is minutes left and n^.5 is the square root of n) was a good amount. If you were up by 6 or more with 1 minute left, 12 or more with 4 minutes left, 18 or more with 9 minutes left, etc. you'd be pretty safe.

Turns out the square root of n is pretty accurate. That's because the chance a team recovers a given amount of points depends on the standard deviation of the distribution of points. Standard deviation is the square root of variance, and variance can add. So the if the variance of 1 minute is V then for 2 minutes it's 2V, for 3 it's 3V, etc. So the variance is some constant k times n, and standard deviation is (kn)^.5, which is proportional to the safe margin.

Is the 6 constant accurate? Let's take a look at things on a per possession basis. The variance in one possession, based off of last year's numbers, is 1.22. In the final minute there are an average of 6 possessions total (more than the usual 4 or so), so the variance in the final minute is 7.3, and standard deviation is 2.7. Multiplying by 2.33 (so we're 99% sure, it's from z tables) we get 6.3, which is our final constant. 6.3!

Turns out 6*n^.5 was a pretty good guess, and accurate enough to use. It won't always work (it'll probably be a little less than 99% because one possession is not normally distributed), but it's a pretty good guide on when you should head back to your smoking room. Unless of course the other team has Reggie Miller.