Win Probability Added

Yesterday’s ESPN Insider article previewed WPA. If I divulge the whole article here, the EIP (ESPN Insider Police) will break down my front door and take custody of my child, so I’ll only give out the definition.

Yesterday’s ESPN Insider article previewed WPA. If I divulge the whole article here, the EIP (ESPN Insider Police) will break down my front door and take custody of my child, so I’ll only give out the definition:

What is WPA? Starting with the beginning of a play, what is the probability of winning the game, given the situation? After the conclusion of that play, recalculate and debit/credit the player for the change in win probability. That’s WPA. This is the essence of sport: each play contributes to a team’s chances of reaching its collective goal of beating the other team.

Tom Tango created this metric for baseball and Brian Burke from Advanced NFL Stats has adapted it to football, but it’s still missing in basketball.

In short, based on the decision of every player on every possession, their WPA will be adjusted. If a player has an open shooter in the corner but decides to take a jumper and misses, chances are his WPA will decrease, thus reflecting his poor decision. It’s a step in the right direction because it’s now been well-established that the box score sucks and doesn’t tell you what and how things transpired in a game. To evaluate a player or a team, you must watch the game and then form an opinion, stats will only tell you so much. Dave Berri and John Hollinger have their critics, but they’ve worked hard to at least quantify the events in a basketball game beyond the obvious. We love to rip on Hollinger because he comments on the Raptors without watching a game, but his PER rankings are pretty close to who the best players in the league are. What the PER fails at is accounting for players who don’t have the ball in their hands, an area which I hope the WPA can address. Unfortunately, from what I can tell, it doesn’t account for defense.

The focus of the popular metrics have so far been offense; defense has been ominously missing from the equation. Aggregated on-off, matchup-stats, and rebounding percentage are some of the defensive stats that carry weight when discussing the unglamorous side of the ball, but these are pieces of the pie, not an overall rating. It’s not surprising that nobody has come up with an authoritative measure of defense, simply because it’s very difficult to do so.

It’s easier to assign blame for a failed offensive possession than for a defensive one. The problem is that usually a person watching the game has no idea what the defensive scheme is, we can take a guess as to whether a defense is encouraging a particular player to shoot, provide early help on dribble penetration, trap someone on the baseline by leaving it open, and so forth. But we don’t know for sure, so when Marco Belinelli seemingly blows a rotation, we don’t necessarily know whether it was even his rotation to blow. Instead of being punished, perhaps he should even be given credit for actually trying to make a rotation that wasn’t even his to make.

A defensive stat would have to at least account for four things:

  • Man-defense
  • Communication (e.g., calling out picks and responsibilities)
  • Degree of the help defense effectiveness
  • Box-out positioning (even if your team gets the defensive rebound despite you not boxing out your man, should you be punished?)

It’s nearly impossible.

Finally, a stat that I’m growing increasingly fond of is percentage based on shot location. Synergy Sports has done some groundbreaking work in this area including counting for the “contestedness” of a shot, but for those who can’t afford the software, HoopData summarizes the data in terms of distance from the net. It does leave out the “sweet spot” since it doesn’t tell you what exact location works for the player, for that you have to see’s HotSpots. For example, the latter reaffirms what we all thought about Jose Calderon’s three-pointer: he shoots it better straight-on. What a stat like this does is tell us what a “good” or “bad” shot for a player might be, and it is based on this that I feel Andrea Bargnani’s mid-range game between 16-23 feet is very poor, but that shouldn’t be interpreted as meaning that he can’t shoot the ball from that range. We all know he’s a decent spot-up shooter, the problem is that he’s not a great pull-up shooter from that range.

I just pointed out a couple holes here and there, but until someone accounts for all these issues, we’re resigned to mix-and-matching and picking out the stats that best suit the argument we’re trying to make.

And that’s about all I want to type this Civic Holiday.

To Top