Friday, April 26, 2013

How Westbrook Changes the 2013 Playoffs

The problem with predictions is that the future will not be like the past. Random events occur, throwing off everything to come. Westbrook allegedly hasn't missed a game since junior high, but after tearing his meniscus he's out for weeks, where he may not even make it back in time for the start of the NBA finals. As many NBA analysts have discussed, it was looking like a rematch of Thunder-Heat, especially when one considers the historically strong point differential of the Thunder. With Westbrook out, however, suddenly the west is wide open, and we may see an intriguing, veteran Spurs team in the finals or, as strange as it sounds, the Clippers.

Since Westbrook hasn't missed an NBA game before (439 including the playoffs consecutively), this is a fantastic challenge to the NBA statistics community on predicting the Thunder's performance without him. There are extremely few minutes with backup point guard Reggie Jackson with the other starters, much less ancient veteran Fisher. Lineup analysis is difficult, as the Thunder have never had to deal with extended stretches without Westbrook. It will take careful, precise predictions of how the team performs.

Additionally, here are also two schools of thought on Westbrook -- he shoots way too much, considering Durant is on the same team, damaging his team in the process; or his aggressive style is overall a tremendous positive, even overriding the bad shots. The debate will be armed with new information, but I think the latter has a stronger case. The Thunder's offense was arguably the best during the 2013 season, especially adjusting for strength of schedule, after coming in second in 2012 and Westbrook was the point guard orchestrating the affairs and the player who shot the most. Although Durant is an amazing shooter, Westbrook often receives the short end of the stick at the end of shot clocks, responsible for creating offense out of thin air. But now low usage players well be replacing his minutes, and if he is indeed a monkey wrench in a well-oiled machine the Thunder should hardly miss a beat.

Methods: Win Shares, PER, and IPV (TalkingPractice)

As a limited dataset of lineups featuring non-Westbrook point guards inhibits 5-man analysis, and the fact that there are no missed games to analyze how the Thunder fare without him, the effect of losing Westbrook has to be estimated through individual player stats. I'll present three different methods and a handful of estimates through guessing the players whose minutes will increase without him. I'll also judge the Thunder's strength with and without him through SRS (point differential adjusted for strength of schedule.)

(The next part explains how I calculated the new team strength, so you may skip that if you want.)

The first metric I used is the popular all-in-one player metric: Win Shares. This is easy to work with because Win Shares are in a system where it's easy to convert player changes into team level win percentage changes. For the win percentage of the Thunder before the injury, I used the projected regular season win percentage based on adjusted point differential because it's generally more predictive than plain ol' win percentage. From there, it's a quick back calculation to the new Westbrook-less point differential using the Pythagorean win formula (Points^14/(Points^14 + Opposing points^14)).

The next all-in-one metric is PER, the flagship of the ESPN empire. Frankly, I'm using this out of curiosity, as I don't think it will perform well and due to its heavy bias to high usage players like Westbrook, and the low PER's from the bench guys, it'll be interesting to see the pessimistic outlook. For estimating the loss in SRS, I converted his EWA (estimated wins added through PER) to EWA per game and then subtracted that from the Thunder's win percentage. PER assumes Westbrook was displacing a replacement level player, but this isn't entirely accurate so I added in the estimated boost in win percentage from a set of replacement minutes heavy in crappy backups (i.e. Fisher) and another set with more minutes devoted to better players (i.e. increasing the workload for Durant and Ibaka.) I did the same boost with the other metrics. ESPN gives the full details on Estimated Wins Added on the advanced stats page, if one is curious.

VA: Value Added - the estimated number of points a player adds to a team’s season total above what a 'replacement player' (for instance, the 12th man on the roster) would produce. Value Added = ([Minutes * (PER - PRL)] / 67). PRL (Position Replacement Level) = 11.5 for power forwards, 11.0 for point guards, 10.6 for centers, 10.5 for shooting guards and small forwards
EWA: Estimated Wins Added - Value Added divided by 30, giving the estimated number of wins a player adds to a team’s season total above what a 'replacement player' would produce.

The last metric used is a form of regularized adjusted +/- from TalkingPractice. Plus/minus is just looking at the point differential when a player is on the court, adjusted means regression is used to find the best fit, and regularized is basically a fancy mathematical way to reduce wildly high/low estimates for players with low minutes. The site calls their metric IPV (individual player value), and due to the NBA statistical analyst exodus (they're being hired by teams) it's the best, and one of the few, publicly available +/- stats. A few box score stats are also used to help shape IPV, so it's also not a complete departure from conventional stats.

For player minutes, I have two different situations. One is called the "pessimistic" model: no other player minutes will increase except point guard backups Reggie Jackson by 18 minutes, Fisher by 12 minutes, and shooting guard Sefolosha by 5 minutes. Note that the total minutes displaced here is 35, which is roughly Westbrook's average for the regular season. I'm calculating an adjusted point differential based on the regular season stats instead of a minutes distribution you'd see in the playoffs (Westbrook played 38 a game last year in the playoffs.) Since I've used regular season point differential in playoff predictions, this is fine, as I'll be comparing alike-things. The other minutes distribution, the "optimal model," gives more time to their better players, and suggests the Thunder play less smallball because they lost one of their best small players: Reggie Jackson 13 minutes, Fisher 8 minutes, Sefolosha 6 minutes, Durant 2 minutes, and Ibaka 6 minutes.

Oklahoma City's new team strength

Based on a few different player metrics, there's a range of estimated team strengths in point differential for the Thunder without Westbrook. Win Shares doesn't love Westbrook because it's built on efficiency, and as I've discussed before it gives a lot of unearned credit just for simply being on a great defensive team. In replacing him, they'd still be an elite team, even with the pessimistic minutes distribution. PER, however, views his high usage play as extremely valuable, and finds the backup options as the flotsam of the league -- even Sefolosha doesn't have a good PER. IPV, however, finds an estimate in between the two metrics, and sees the dropoff as significant but not debilitating.

OKC pre-injury SRS
Westbrook per game
OKC post-injury SRS, pess. mins
OKC post-injury SRS, opt. mins
Win Shares

I don't believe Win Shares or PER are up to this task, and IPV nicely finds a middle ground between the total collapse of not having a creator and the underrated contributions of their other players. Given our limited information on Reggie Jackson and Fisher, as well the future minutes distribution in the playoffs, any estimate will be a rough guess at best, but the magnitude is important to note. A point differential from 5.1 to 6.5 (IPV's range with a little leeway) is in the same range as the Nuggets (+5.4) and the Clippers (+6.4.) PER and Win Shares also tend to clump all the player value at the very top, evident from Westbrook's estimated +/- per game.

However, Fisher's IPV value is strangely decent, near 0 instead of the typical negative value you'd see from an end of the bench type. This is where knowledge of the metrics you use can come in handy -- IPV uses a prior value that heavily influences the final number, and Fisher has had consistently strong +/- values in the past. Since he's played few minutes recently, there is not enough information to downgrade Fisher's IPV any further (for example, Granger's +/- is very similar to the one he had last year because he had very few possessions to convince the model his value was different.) That type of adjustment may bring down their point differential incrementally, and you can make another one due to hitherto unused lineups being rusty as players try to adapt to each other.

I'd estimate this Thunder team is now somewhere around the +5.5 SRS level, depending on who they play and how Fisher looks. This should be enough to finish off the Rockets, but don't be surprised if they drop a couple games. It looks like the Clippers will meet them in the next round, and they're more likely the stronger team. With homecourt advantage, the series based on my estimates is nearly a coin flip. I'd take the Clippers in seven, maybe even less -- they have a slight edge in adjusted point differential even adjusting for homecourt, and the Thunder have to go against Chris Paul without their all-star point guard. The Clippers often use a lot of players on offense you can hide weaker defenders like Fisher, but it's not enough to overcome the loss of Westbrook. Against a healthy Spurs team? They'd have little chance.

It's time to see what Westbrook's value really is.

No comments:

Post a Comment