A Deep Dive on Win Shares
When did they start being used? How do we calculate them? What purpose do they serve?
An Introduction
Win Shares. If you’re involved in NBA discourse you’ve almost certainly come across them before. They’re one of the most commonly-cited stats in NBA Twitter arguments, but where did they originate from? What are their limitations? And what do they even show? If you’re interested in finding out, read on - as I take a deep dive into the origin, meaning, flaws, and uses of Win Shares.
The Origin of Win Shares
Initially invented by Bill James, Win Shares started as a metric used in baseball, to try and quantify the number of wins an individual player contributed towards their team’s win total throughout a season. The term was first coined in 2002, in James’ book of the same name. This was the same year that saw the Oakland Athletics employ a rigorous stats-based (or sabermetric) approach to scouting players, with great success (watch Moneyball if you haven’t already). Though the A’s focussed on On-Base Percentage and Slugging Percentage, this year marked the rise of ‘advanced statistics’ in American sports, a revolution that would soon find its way to the NBA.
The concept of Win Shares in baseball is relatively simple to understand.
Players on a team are assigned three different ‘Win Share’ statistics - Hitting, Fielding and Pitching, corresponding to the three different ways a player can contribute to their team’s success. One Win Share is given to each of these statistics per team win, meaning that 3 win shares would be divided amongst the players for each win. For each player, this meant that a ‘Win Share’ actually accounted for a third of a team win.
As it happened, win shares would not be divided equally among batting, pitching and defensive contributions. Batting contributions typically account for up to half of the total win shares, with pitching receiving around 35%, and the rest going to fielding. This creates an imbalance between the Win Shares of Batters and Pitchers, with pitchers receiving much fewer Win Shares than hitters. This wasn’t much of an issue, however, as pitchers don’t tend to hit as well (unless they are called Shohei Ohtani), meaning that comparison was not needed.
One of the key advantages of the statistic was that it factored in the era, making it possible to compare players who played in different decades, which wouldn’t otherwise be possible.
I’m not going to go into detail about how win shares are calculated in baseball, because that took up 85 pages of James’ book, and I don’t have the space, time, or knowledge to of so here. (I also have not read the book, so I couldn’t tell you how they do it, even if I wanted to)
Despite the promise of Win Shares as a catch-all statistic for quantifying player contributions, people still had their criticisms. Firstly, as Win Shares were awarded based on team wins, there was no way of quantifying the negative effects a player might have had on their team (i.e. as a ‘loss share’). Furthermore, if a team won more games than they ‘should’ based on the number of runs they scored and allowed, those players would be awarded more Win Shares than players on teams who lost more than they ‘should’. This meant that average players on great teams could suddenly be comparable to great players on average teams. If used as intended (a stat to measure player contributions to their team), this would not be an issue. However, it could become misappropriated in discussions of player ability, which it was never developed to quantify.
Calculating Win Shares in Basketball
Win Shares were brought to the NBA through Basketball-Reference.com; a website dedicated to providing basketball statistics to the masses, for free.
The creator of the website created an adaptation of James’ Win Shares metric, to fit with the game of basketball. Rather than splitting Win Shares by role, they were instead divided into Offensive and Defensive Win Shares. The calculation of Win Shares for basketball, as in baseball, is incredibly complex. However, for this deep dive, I will (attempt to) break it down into digestible chunks, so that we can understand what exactly it shows.
Offensive Win Shares
Offensive Win Shares, as defined by Basketball-Reference.com, are calculated as follows:
OWS = Marginal Offense / Marginal Points Per Win
Marginal Points Per Win
Marginal Points Per win is the easier statistic to calculate of the two, and accounts for the overall scoring ability of the league, as well as the pace of the player’s team. calculated using the formula:
Marginal Points Per Win = 0.32 * (League Points Per Game) * (Team Pace / League Pace)
Marginal Offense
is a measure of the offensive output of a player relative to the league average, and is calculated as:
Marginal Offense = Points Produced - 0.92*(League Points Per Possession)*(Offensive Possessions)
Points Produced and Offensive Possessions are statistics that were created by basketball statistician Dean Oliver. Points Produced is a measure of the number of points an individual creates (through scoring, assisting and offensive rebounds), per 100 possessions, and is calculated according to the following formula:
Points Produced = (Field Goal Attempts + (0.44 * Free Throw Attempts) + Turn Overs) * (Offensive Rating)/100
This formula uses yet another advanced statistic, Offensive Rating. This was also developed by Dean Oliver, and accounts for a player’s “Scoring Possessions, Missed FG Possessions, Missed FT Possessions, and Turnovers”, according to Basketball-Reference.com. The calculation for Offensive Rating is more complicated than can be explained here, but a great in-depth explanation exists on the Basketball-Reference website, which you can find here https://www.basketball-reference.com/about/ratings.html.
Offensive Possessions, the other statistic included in the Marginal Offense calculation (and also the complex Offensive Rating calculation), is simply an estimate of the total number of possessions an individual player has on the offensive side of the ball.
Overall, Offensive Win Shares are effectively a measure of the offensive output of an individual player, adjusted to factor in the pace of the league/team, the number of possessions the player gets, and the scoring rate of players in the league.
Defensive Win Shares
Defensive Win Shares are calculated in much the same way as offensive win shares, but for the other side of the ball:
DWS = Marginal Defence / Marginal Points Per Win
The same denominator is used in this equation as in the calculation of Offensive Win Shares, which accounts for the overall scoring ability of the league, along with the pace of the player’s team.
Marginal Defence
Marginal Defence is calculated according to the following formula:
Marginal Defence = (Player Minutes Played / Team Minutes Played) * (Team Defensive Possessions) * (1.08 * (League Points Per Possession) - ((Defensive Rating) / 100)
This is similar to Marginal Offense, in that it accounts for the number of possessions, and scoring rate of the league. However, unlike offensive possessions, defensive possessions are much harder to quantify at the level of the individual. This means that the total number of possessions an individual faces while on the floor is approximated instead, and used alongside Defensive Rating. In contrast to Offensive Rating, Defensive Rating is designed to represent the number of points an individual concedes for every 100 possessions. It is comprised of numerous calculations (as with Offensive Rating) and accounts for a player’s steals, blocks and defensive rebounds, as well as estimating the number of opposition turnovers and misses that can be attributed to a player.
Overall, Marginal Defence is a statistic to represent the defensive impact of a player - through their box score defensive stops, and estimated indirect defensive stops, based on the amount of time they spend on the court, giving a value as the number of points they concede for every 100 possessions. Converting this to Defensive Win Shares is as simple as dividing by the Marginal Points Per Win, which factors in the pace of the team relative to the rest of the league.
The Failings of Win Shares
Despite their many proposed advantages, there are still some flaws with win shares, which are intrinsic to the way they are calculated.
Flaws with Defensive Win Shares
Firstly, since some of the box-score metrics used in calculating Offensive and Defensive Rating are not available for the NBA before 1973, proxies and approximations have to be used instead.
For example, in the calculation of Defensive Rating, the most important individual defensive metrics are steals, blocks and defensive rebounds. None of these statistics were recorded pre-1973. This means a completely different calculation has to be used for crediting Defensive Win Shares in those years, which may or may not align with the modern calculation. This means that, while Win Shares are available for players of all eras, they may not be directly comparable.
Another major flaw also comes from the overestimation of Defensive Win Shares for big men, and poor defenders on good defensive teams, stemming from the calculation of Defensive Rating.
As mentioned earlier, Defensive Rating takes into account an individual’s box score defensive metrics (Steals, Blocks, Defensive Rebounds) and an estimate of the number of opposition turnovers and misses which can be attributed to a player. The latter is a major issue with Defensive Rating, as it works on the assumption that players on the court are equally responsible for forcing offensive errors (not captured by steals or blocks). This is a large assumption to make and may lead to overestimates of the defensive impact of players who play lots of minutes in good defensive systems or alongside strong defensive players, even if they are poor defenders themselves. On top of this, there is also the assumption that each player on the court faces the same number of defensive possessions per minute. This disregards the fact that poor defenders are likely to be targeted by offensive systems, and thus face more defensive possessions. This is especially the case in the modern NBA, where teams will hunt out poor defensive players and take advantage in isolation. Also, big men are likely to face much fewer defensive possessions than guards, due to the nature of their position, but are credited with the same amount of defensive stops.
These flaws are best illustrated through examples.
Luka Doncic, regarded by some as one of the worst ‘superstars’ from a defensive standpoint, has the same number of Defensive Win Shares (2.9) this season as Jrue Holiday, arguably one of the best point-of-attack defenders in the league.
This may be influenced by Holiday’s team forcing the fewest turnovers in the league (11.7 per game to the Mavericks' 13.5), along with Doncic grabbing a higher percentage of available defensive rebounds (13.9% to 8.1%), steals (1.9% to 1.7%), and blocks (1.2% to 0.9%) than Holiday, whilst also playing more total minutes (2378 to 2183). This is despite Holiday defending more PnR Ball Handler possessions than any other player (7.4 per game), and holding attackers to just 0.85 points per possession on those plays. Luka isn’t a horrible defender by any stretch of the imagination, but he’s certainly not on Jrue Holiday’s level, despite what Defensive Win Shares might lead you to believe.
Another example comes from the case of big men being over-estimated in their Defensive Win Shares. At the time of writing, the top 11 players in Defensive Win Shares this season all play either the PF (3) or C (8) position, according to Basketball-Reference.com. Among these are some incredible defensive players - Evan Mobley, Brook Lopez, Jarrett Allen, Nic Claxton, Giannis Antetoukounmpo and Jayson Tatum, to the credit of Defensive Win Shares. However, the lack of wings and guards, as well as the inclusion of Nikola Jokic in 11th place, highlight some of the statistic’s flaws.
Furthermore, looking at the top guards in Defensive Win Shares, the names which top the list are not necessarily those that are considered premier defenders in the league - with Donovan Mitchell, Anthony Edwards and Caris LeVert making up the top 3. Having 2 guards from the Cleveland Cavaliers is fitting, given that they have the best Defensive Rating in the league (110.6); although this should be attributed more to their incredible defensive bigs (Jarrett Allen and Evan Mobley), who are very much the anchor of their defence.
Overall, Defensive Win Shares is a big-man-biased metric, which overestimates the impact of poor defenders in good defensive systems, especially in the case of guards. I must add, although Defensive Win Shares has its flaws, it still does a pretty good job of identifying defensive talents and attributing wins to defensive efforts by individual players, which is what it was designed to do in the first place.
Offensive Win Shares
Offensive Win Shares avoid some of the flaws of Defensive Win Shares, as offensive contributions are much more tangible in the box score, and thus easier to formulate into a catch-all metric. We can also account for the exact number of possessions a player is involved in on the offensive end, meaning that individual contributions are easier to quantify. Furthermore, there isn’t as much disparity between guards and bigs, as both can contribute in similar capacities on offence, with the offensive rebounds of bigs being balanced out by the assists of guards.
Despite this, there is still a couple of flaws with Offensive Win Shares. Offensive rebounds, whilst important components of a team’s Offence, can heavily bias the Offensive Win Share calculation. Whilst there isn’t as much of a disparity between bigs and guards as there is in Defensive Win Shares, it can lead to some interesting results.
Take, for example, the Golden State Warriors, who currently sit at 6th in the West, with a record of 42-38. Who would you expect to lead them in Offensive Win Shares? Would it be their 29.6ppg, 49/43/92, 6.3 assist superstar Stephen Curry? Or perhaps his Splash Brother Klay Thompson? Maybe even Draymond Green, their leader in assists and chief playmaker. Whatever you guessed, you would be wrong. The leader of the Golden State Warriors in Offensive Win Shares this season is Kevon Looney.
This isn’t to say that Looney is a bad offensive player - he is a fantastic role player and an integral part of what GSW do, but to attribute more of their wins this season to his offensive performance than Stephen Curry is wild, at least to me. The reason for Looney’s incredible 5.6 Offensive Win Shares this season (13th overall in the league) is his exemplary performance on the offensive glass. He ranks 3rd in the league in Offensive Rebound Percentage (15.5%), trailing only Mitchell Robinson and Clint Capella, who are less involved with their teams’ offence than Looney, highlighted by his superior Assist Percentage (13.4% compared to 4.4% for Robinson and 4.7% for Capella).
So Why Use Win Shares?
Despite their flaws, Win Shares remain a great metric for determining how much an individual player contributes to their team’s success. They are great for comparing players within the same system or the same player across multiple seasons. This is evidenced by the graph below, which shows the division of win shares among players with more than 2000 minutes played, across different teams.
In each case, you can clearly see who the most impactful players are for each team. Jokic is by far and away the most impactful player for the Nuggets. The Kings are split between Domantas Sabonis and De’Aaron Fox, and OKC are lead by Shai Gilgeous-Alexander. For these purposes, which Win Shares were originally designed for, it is a very useful metric.
Additionally, if you’re prepared to accept the flaws of Win Shares, they can also be used to compare players from around the league and across different eras, thanks to the inclusion of pace and other statistics to normalise Win Shares across teams and generations. Win Shares can also be converted to a ‘per 48 minutes’ metric, which allows for comparison of players who played different numbers of games or minutes per game.
The extent of Win Shares’ ability to show overall ‘player impact’ is best demonstrated through their correlation with the awarding of MVP. Of the last 46 MVPs (dating back to the NBA/ABA merger), 27 have been awarded to the player with the highest Win Shares Per 48 Minutes (WS/48). This is visualised in the graph below, which shows the MVP winners and leaders in WS/48 for each season since 1977.
Interestingly, 11 of the last 12 MVPs were also leaders in WS/48. This may be due to the increased weight that statistics are playing on MVP conversations and voting, or simply a product of the NBA’s evolution.
Furthermore, many of the MVP winners who were not WS/48 leaders are regarded as some of the more ‘contentious’ MVPs.
For example, many believe that LeBron James deserved MVP in 2011 over Derrick Rose, but that Rose won due to his meteoric rise at such a young age. LeBron had already won two MVPs, and people had already become numb to his greatness, and voter fatigue had kicked in. The same could be said of Iverson’s 2001 MVP, which saw him become the least-efficient MVP of all time, shooting 42% from the field. Steve Nash was also regarded as a contentious two-time MVP, with many believing that Shaquille O’Neal, Kobe Bryant or LeBron James deserved those titles instead (even though WS/48 would suggest that Dirk Nowitzki deserved them).
Conclusions
Overall, Win Shares are a great metric for measuring the impact a player has on their team’s success on both ends of the floor. When used for what they were originally intended, they can prove highly useful in identifying the most impactful players in a team. They may have their flaws (overestimating the impact of big men and poor defensive players in solid defensive systems), but as a catch-all metric for player impact, they still do a pretty good job. As the great statistician George Box one said “All models are wrong, but some are useful”.