Win shares, BPM and VORP: the all-in-one stats
Win shares, box plus-minus and VORP each try to boil a player's season into a single figure. Learn what each measures, how they differ, how to read their scales, and where they fall short.
Basketball fans have always wanted a single number that says how good a player is. The pursuit has produced decades of argument, because the sport is complicated: scoring, passing, rebounding, defending, spacing the floor and a dozen other things all matter, and they do not come in the same units. The attempt to combine them into one figure has produced a family of what analysts call all-in-one statistics. Three of the most widely seen on the public stats pages are win shares, box plus-minus and value over replacement player.
They answer slightly different questions. Win shares asks how many of a team's wins a player is responsible for. Box plus-minus asks how many points per 100 possessions a player adds compared with an average player. Value over replacement player asks how much a player is worth over what a freely available substitute would provide, accumulated across a season. They share a common source of data, the box score, and a common aim, to turn it into an estimate of impact. They also share common weaknesses.
This article explains each measure in plain language, how to read the scales, how the three relate to one another, and what the box score cannot tell any of them. It follows an earlier generation of the idea described in the article on player efficiency rating, and the three measures can be seen as attempts to fix its shortcomings.
Why combine stats at all
The core difficulty is exchange rates. How many assists equal a rebound? How much does a missed shot cost? Is a player who scores 25 points on poor efficiency more or less valuable than one who scores 15 on excellent efficiency? A raw box score does not answer these questions. A well-designed all-in-one statistic does, by tying every event to its effect on winning.
The common method is to ground the weights in what happens on the scoreboard. A made basket adds points. A turnover loses a possession, and the average possession is worth a certain number of points. A rebound gains a possession. A missed shot that the other side rebounds loses a possession. By converting all events into points, or into possessions and then points, you can add them up.
Because of that, these statistics build on the ideas in points per possession and on measures of shooting efficiency such as true shooting percentage. The conversion to wins uses a long-established relationship between points scored and allowed and winning percentage, the same one behind Pythagorean wins.
Win shares
Win shares takes its name and its basic framework from baseball, where the analyst Bill James developed a method of dividing a team's wins among its players. It was adapted to basketball and popularized on a major statistics site, and it is now widely quoted.
The principle is that a team's wins should be fully accounted for by its players. The measure is built so that, for a team, the player win shares add up approximately to the team's actual wins. If a team wins 50 games, its players' win shares should sum to roughly 50.
To get there, win shares is split into two parts.
- Offensive win shares credit a player for the points he produces, relative to the points a team would be expected to produce from the possessions he uses. A player who uses many possessions at high efficiency earns many offensive win shares.
- Defensive win shares credit a player for his contribution to the team's defense. Since the box score does not record most defense, the method starts with the team's defensive rating, which can be measured well, and allocates it among the players based on their minutes and on the defensive events that are recorded, such as steals, blocks and defensive rebounds.
A player's total win shares is the sum of the two. A rate version, win shares per 48 minutes, divides by playing time to produce a per-minute figure. The league average for that rate is about 0.100, so a player at 0.200 is producing at double the average rate, which is a very high level of play.
The scale is intuitive in a way that some other measures are not. A season total of 10 or more win shares is the mark of an All-Star-level season. The very best seasons in history reach well beyond that. A player with 2 or 3 win shares is a rotation player, and a player near zero contributed little.
There are some things to bear in mind about win shares.
It depends on team success. Because the measure divides up a team's wins, a player on a strong team will tend to have more win shares than an equally good player on a weak team. The defensive component, in particular, is built from the team's defensive rating, so a good defense lifts everyone on it. This can be a feature, since winning is the point, but it means the figure is not purely a measure of individual skill.
Its defensive half is a rough estimate. Because the allocation of team defense to individuals relies on box-score events, players who are strong defenders without recording many steals or blocks may be undercredited, and players who record plenty may be overcredited. This is the standard problem described in the article on defensive metrics.
It is a cumulative statistic. Win shares accumulate with playing time. A player who plays more minutes earns more. That is appropriate for a measure of total contribution but means it should not be compared with a rate statistic.
It can be computed for a long span of history. Since it relies on the box score, it can be computed for any season in which the necessary data exist. Steals and blocks were not officially recorded until the 1973-74 season, so earlier seasons use estimates for those components. Three-point shooting data only exist since the NBA adopted the line in 1979-80. This means that comparisons across the full sweep of league history carry extra uncertainty.
Box plus-minus
Box plus-minus, abbreviated BPM, comes from a different line of thinking. It was developed by the analyst Daniel Myers. The goal is to estimate a player's impact in points per 100 possessions relative to an average player, using only box-score data, but with weights that are fitted to the results of a more sophisticated kind of plus-minus.
The reasoning goes as follows. The ideal measure would be the adjusted plus-minus described in the article on plus-minus and on/off, which uses lineup data to isolate each player's effect while accounting for teammates and opponents. But that measure is noisy and needs a lot of data. The box score, on the other hand, is stable but crude. BPM uses statistical regression to find the combination of box-score statistics that best predicts the adjusted plus-minus. Once the weights are found, they can be applied to any player's box-score numbers.
The result is a figure that reads like a plus-minus. A BPM of 0 is an average player. A positive number means the player adds that many points per 100 possessions to his team's margin compared with an average player on the floor; a negative number means he costs that many. The scale is generous at the top.
- Around 0 is league average.
- Around plus 2 or 3 is a good starter.
- Around plus 5 is the territory of an All-Star.
- Around plus 8 and up is the territory of an MVP candidate.
- Around plus 10 and above is reserved for the most dominant seasons ever recorded.
- Around minus 2 is roughly the level of a replacement player, which becomes important in the next measure.
These thresholds are approximate and are best treated as rough guides. BPM is also divided into offensive and defensive components, which can help locate a player's strengths. As with win shares, the defensive side is driven by box-score events and is the weaker half.
What are the advantages of BPM? It is a rate statistic, so a player's playing time does not inflate it. It is adjusted for the strength of the team in a way that tries to avoid giving good teams' players an automatic boost. And it is built to be comparable to a plus-minus, which is intuitive: a player at plus 5 is a player whose team outscores opponents by 5 more points per 100 possessions with him than with an average player. The article on net rating explains the units.
What are the disadvantages? Because it is fitted to data from a particular era, the weights reflect the way the game was played during the period used to fit them. It is also sensitive to role: a big man who rebounds and scores near the rim can post a high BPM that does not translate to a different role. And like every box-score measure, it cannot see what the box score does not record.
Value over replacement player
Value over replacement player, or VORP, is a cumulative version of BPM. It borrows an idea from baseball: instead of comparing a player with an average one, compare him with a replacement-level one, a player of the kind who could be signed for the minimum or called up from the minor leagues. The reasoning is that average players are scarce and valuable, while replacement players are freely available, so the real contribution of a player is what he gives beyond the level a team could obtain at no cost.
For basketball, the replacement level is set at a BPM of about minus 2. That is the convention: a player at that level is treated as contributing zero value. The formula is:
VORP = (BPM − (−2.0)) × percentage of team possessions played × (team games ÷ 82)
In plain terms, take the player's BPM, add 2 to see how far above replacement he is, then scale by how much he played. A player who plays a lot of minutes at a high BPM accumulates a large VORP. A player who plays a few minutes, even at a good BPM, accumulates a small one. The factor involving team games ensures that a shortened season produces a proportionate total.
VORP is a counting statistic expressed in a unit that is hard to state simply, roughly the number of points per 100 team possessions a player adds over replacement over the course of a season. A season total of 2 is a solid starter. A season of 5 or more is a star season. Seasons above 8 are among the rarest in league history.
The strength of VORP is that it values availability and durability. A player who is excellent but plays only a third of the games accumulates little. A player who is very good and healthy all year accumulates a great deal. This makes it useful for evaluating total contributions over a season or career, such as in awards discussions.
The weakness is that it inherits all of BPM's limitations and adds the difficulty of the replacement level itself. The choice of minus 2 is a judgment. A different choice would change the figures, and the proper replacement level may differ with the structure of the league and its roster rules, such as the use of two-way contracts and 10-day deals, which affect what a freely available player looks like.
How the three compare
The three measures are related, but they are not interchangeable.
| Measure | What it estimates | Rate or total | Typical scale |
|---|---|---|---|
| Win shares | Wins a player is responsible for | Total (with a per-48 rate version) | 10 or more is an All-Star-level season |
| Box plus-minus | Points per 100 possessions above average | Rate | 0 is average, about 5 is All-Star level |
| VORP | Value above a replacement-level player | Total | 2 is a solid starter, 5 or more a star season |
Win shares and VORP are cumulative, so they reward playing time. BPM is a rate, so it does not. A part-time player can have an outstanding BPM and a low VORP. A durable good player can have a modest BPM and a respectable VORP.
Win shares is anchored to team wins and to the team's defensive rating, so it is sensitive to team quality. BPM is meant to be context-neutral, though not perfectly so. VORP follows BPM.
When the three agree about a player, the picture is usually reliable. When they disagree, the disagreement is informative. A player with high win shares and a modest BPM may be benefiting from a strong team. A player with a high BPM and low win shares may be a part-time contributor. The comparison can point you to the reason.
What the box score cannot see
All three measures draw on the same raw material, the box score, and so they share the same blind spots. They cannot see screens that free a teammate, defensive rotations that prevent a shot, the effect of a shooter's gravity on a defense, communication, or the difference between a contested and an open look. Defenders whose value comes from positioning rather than events are consistently undercounted.
Analysts have tried to fill the gap with tracking data, which records the location of players and the ball many times a second, and with models of shot quality such as those described in the article on expected points and shot quality. These approaches give a better sense of shot difficulty and of defensive contest, but they produce their own complications, and they are available only for recent seasons.
The measures also say nothing about fit. A player's value depends on the team around him, and the best individual player does not always make the best team. All-in-one statistics describe a player's production in a given context; they do not forecast his production in a different one. For that reason, claims that rest on a single figure deserve skepticism, a point made forcefully in the article on analytics myths.
Reading them well
A few practical rules help.
- Use the rate (BPM or win shares per 48) to judge how good a player is when he plays, and the total (VORP or win shares) to judge how much he contributed over the season.
- Do not read decimal places literally. Differences of a few tenths between two players are within the noise.
- Compare with the league context of that season. The scales are broadly stable, but the league changes.
- Look at several measures together. Agreement is reassuring; disagreement is a prompt to look closer.
- Watch the minutes. A strong figure over a small sample, as discussed in the article on small samples, is likely to regress.
- Remember that defense is the weak point. If a player's reputation rests on defense, the box-score measures will probably understate him.
Seasons can also be placed on a longer arc. A player's year-to-year numbers show his path, which is the subject of the article on aging curves. The leaders page lists many of the counting and rate statistics that feed these measures, so you can see the ingredients behind the totals.
What to take away
Win shares, BPM and VORP are the descendants of the first attempts to summarize a player in a single figure. Win shares divides a team's wins among its players. Box plus-minus estimates a player's per-100-possession impact relative to average. VORP accumulates that impact over replacement level and rewards availability. All three are drawn from the box score, so all three see offense better than defense, and none sees the unrecorded work that makes teams function.
They are best used as a set, alongside efficiency, role and the film. Used that way, they can settle a good many arguments and open a few new ones, which is about as much as any single number should be asked to do.