In calculating the xWARP projections I’ve been doing, I noticed a worrying problem. The total number of wins projected for each team appeared to be roughly 7% too high. There are only 1230 games, and thus, 1230 wins available for the 30 teams in the NBA to win. In doing my xWARP projections, I had just been posting the win totals as I did each team.
Last night, I finally got around to running the numbers on all of the remaining teams I’d yet to post and summed up team wins for the whole league. Unfortunately, total team wins summed to roughly 1318 wins, which is obviously way too many. There are a number of ways to deal with this problem. The lazy, easy way would be to simply multiply the win totals of each team by 1230/1318 or roughly .933. The top teams then have 55 projected wins and the bottom dwellers would be in the mid to high 20s in wins. This solves the problem, but it doesn’t seem that consistent with reality, so I’ve decided to go another way.
I’ve hinted at wanting to do this before, but my current plan is to go through each and every game on the schedule and utilizing the projected MOV for each team (in the spreadsheets it’s the total Contribution / Game) plus Evan Zamir’s ridge regression work on homecourt advantage, I plan to figure out the win probabilities of each of the 1230 regular season games and then use those probabilities to determine total team win totals. I will then post the updated results. I will be updating the posts that already exist to reflect changes in team win totals after this is all done, and I will also do a master post with the full league projections.
I’m bummed out that the math didn’t work out as well as I’d hoped, but I’m excited to go through the schedule in such detail. The other good thing is that doing things this way acts as a built in strength of schedule adjustment for teams with more or less difficult schedules. Thanks for reading and for your patience while I work out the kinks of all this.
Image from gammaman via Flickr.