Although APBRmetrics teaches us that all teams' records in close games regress toward .500, at a certain point it becomes statistically unlikely that the Heat's "true" probability of winning those types of games is as good as a coin flip
I was interested in this, so I emailed him about, wanting to know more. He wrote back:
I can't find the original study I was thinking of when I wrote that (apparently the site is now defunct), but this one is probably better anyway:
http://basketballprospectus.com/article ... icleid=573
Basically, a team's wpct in close (margin <= 5) games is going to regress to something like:
true_close% = (1/3) + (1/3) * observed_close%
So an observed .400 team in close games is probably a true .467 team in future close games, etc.
The other component in the Pelton article is that year-to-year correlation is 0 -- there's no persistent ability across seasons for a team to outperform their expected record in close games. Everything eventually regresses toward the mean.
I find the Pelton article pretty interesting, but am looking for more discussion on this. Anyone seen anything else on this phenomenon? Have any thoughts?
To me it makes sense, since even very large Ortg/Drtg differentials lead to very small average margins in a limited possession game.