By Harrison Chase, Matt Goldberg, and Kurt Bullard
Earlier this year, we debuted NFL POWER rankings based on win probability curves. The general idea was that we would used a modified version of the Colley method, where instead of treating all wins equally, we would weigh the win by the ‘dominance’ of the win, as measured by the team’s average win probability throughout the game. To us, a dominant win should be valued more than a game that is closely-contested throughout, controlling for team strength. To us, this was an improvement on the Colley method—which weights wins and losses equally. There is more information that you can gain from a game than just whether a team just won or lost, and our modified Colley method does just that.
With the NBA playoffs having just started, we figured it would make sense to calculate these same POWER (Probabilistic Outcomes by Win Expectancy Regularization) Rankings for the 2015-2016 season. We used Basketball Reference’s amazing database of play-by-play to estimate the expected win probability of a team throughout the game, training our model on data from 2001 – 2014.
The model takes as input the time remaining in the game, the point differential, who has possession, and the Vegas line in the game, and attempts to predict the likelihood that the home team will go on to win the game. This model is far from complete – ideally we would like eventually include fouls, timeouts, and other variables that might have a large effect at the end of the game – but we wanted to used this model to create these POWER Rankings before the NBA playoffs were over, so we were under a bit of a time constraint. The Vegas line is included as a measure of team strength entering the game. For example, the win probability of the Warriors when they play the Sixers is not 50% at the start of the game, nor is it 50% if the game were tied at halftime. These Vegas lines essentially act as a prior, and because we weren’t entirely sure we wanted to include a prior for each game we calculated the rankings both using these lines and without them.
The POWER Rankings for each team, prior to the start of postseason, are below.
Two top tiers immediately float to the top: the Warriors, Spurs are in a league of their own, with the Cavs and Thunders in Tier 2, with a large gap between them and the rest of the league. The Pacers also appear to be better than a No. 7 seed in the East, as they are the fifth-best team in the East according to this chart. Meanwhile, the Grizzlies and Mavericks are clearly the worst teams in the NBA playoffs, as there are three non-playoff teams ahead of them. The Jazz were a better team than either of them this year, but a few late season losses cost them a chance to play for the Larry O’Brien trophy.
Another interesting finding is that, besides the Top 3 teams, the East and West are are relatively equal footing. Even including the three clear frontrunners from the West, the East has six of the ten best teams in the Association this year.
It is also possible from this to plot each team’s win average win probability over the course of the game. For this, we used the win probability without the Vegas prior, as it gave a better visual. Below is the interactive Tableau visualization.
We also thought it would be interesting to see how our model ranked teams last year, now that we know what happened. The rankings as of the end of the 2014-2015 regular season are below.
As you can see, our rankings had the Warriors, the eventual champions, in a class of their own. After that came the Spurs, Clippers and Rockets —although the Clippers beat the Spurs (ranked above them) and then the Rockets beat the Clippers (ranked above them), they are close enough by these rankings that those ‘upsets’ may be explained by home court advantage. However, our rankings seemed to either underestimate the Cavs or overrate the Hawks, as despite being the number one seed and having home court advantage the Hawks got swept by the Cavs.
Who will end up winning it all? Only time will tell. But hopefully our rankings can shed some light on the question of who will be hoisting the Larry O’Brien trophy come June.