Bruin Sports Analytics
Are Role Players the Foundation of NBA Team Success?
By: Terrence Liu and Oscar O’Brien
The National Basketball Association (NBA) is filled with superstar players that deservedly receive a significant amount of attention and praise for helping their respective teams. Players such as Lebron James, Stephen Curry, and even younger talents like Luka Doncic are electrifying to watch as they single-handedly dismantle defenses and break records. Their impact on the basketball court is obvious to anyone watching, but no twelve-man roster in the NBA can be filled completely with all-star caliber players. Bench and role players have to fill time during games as starters need to catch their breath or sit out because of injury. Their effect on games is typically more subtle because their role might focus on an aspect of the game besides scores, like rebounds or defense. It is easy to overlook the impact of bench and role players, as some NBA adopt a heliocentric strategy that emphasizes their superstars. The heliocentric basketball model is built upon a star player who dominates the ball, while the rest of the players spread the floor with their shooting ability. One example is the Houston Rockets led by James Harden, who averaged 24.5 shots and 11 free throws attempted per game in the 2018-2019 season.
The purpose of our article is to measure the effect of bench and role players on team success in the NBA by analyzing the 2018-2019 season. We believe that bench players help NBA teams down the stretch because of factors such as injuries, inconsistent starter performance, and load management. In this article, we defined bench/role players as those who averaged between 5 and 25 minutes per game and played at least 10 games with the team. These limitations would mostly avoid garbage-time players and exclude a majority of superstars that usually average just over 30 minutes per game. To measure team success, we use a team’s win/loss ratio because wins are ultimately what decides a team’s playoff fate at the end of the regular season. In order to compare the benches of the 30 NBA teams, we will use box plus/minus, win share percentage, value over replacement, player efficient rating, and age. Daniel Myers, the developer of box plus/minus (BPM), explains that “BPM uses a player’s box score information, position, and the team’s overall performance to estimate the player’s contribution in points above league average per 100 possessions played”. He continues by stating that BPM does not consider playing time, but value over replacement (VORP) is a similar calculation that includes playing time. Win share percentage is the sum of win shares by bench/role players divided by the team’s total win shares. This ratio is important because comparing bench/role players win shares directly would show a clear correlation because teams with more wins have a greater number of win shares that can be attributed to bench/role players. Player efficiency rating (PER) is related to BPM in that it is a rate statistic, but the formula for calculating it is different.
General Overview of Benches in the NBA
The first metric that we used to compare the bench/role players of NBA teams was median BPM because NBA benches are relatively small: the median avoids the influence of extreme outliers (shown graphically below). The red horizontal line in the graph shows the cutoff of playoff versus non-playoff teams for this year. The blue vertical line in the subsequent graphs in this article show the mean value of the statistic on the x-axis for the 30 teams in the NBA. Keeping up with current trends, the bottom three teams from the Eastern Conference barely won 50% of their games and have a much worse record compared to their Western Conference counterparts. The graph below plots the median role player BPM and it appears to show a nice linear trend with a correlation of .679. This result intuitively makes sense because teams that win more would seem like more likely candidates to have a stronger supporting cast. Especially during the grind of the regular season, it makes sense that a basketball team cannot be a one-man show night in and night out because that would just be too much strain over 82 games.
Moving to the next graph, we were curious about restricting every team’s role players to the top four based on average minutes played per game. In reality, most teams play with roughly nine players every game, not 12 or even more, as the statistics show with mid-season traded players that have their stats mixed in with the current players. This graph is less convincing because there is more spread. The Toronto Raptors, who won the NBA Championship that year, had a group of bench players during the regular season that produced a median BPM similar to many mediocre teams. The correlation coefficient drops down to .547, which implies some sort of relationship might exist, but not a very strong one.
Moving away from BPM, win share percentage among bench/role players provides a different picture. In fact, for the graph below, the negative correlation coefficient of -.195 argues that there is actually an inverse relationship between this metric and team success. Compared to the results of box plus/minus, this result is surprising because BPM and win shares are aggregate statistics that try to measure overall player value. Even visually, the playoff teams in the top half of the graph are bunched closer to the left side. One explanation of this outcome is that playoff teams just have much better star players that lead to more wins, which dilutes even above average bench production.
When comparing only the top four bench/role players on each team, the positive correlation (.115) is less to do with the trend but more to do with the abysmally bad Phoenix Suns, New York Knicks, and Cleveland Cavaliers. A possible concern with this analysis is that the median win share percentage of the top four role players could be that some teams are being represented by roughly their seventh-best player, but even the graph of win share percentage by each team’s best role player is so similar it was not worth including. Thus far, it is not overwhelmingly clear that NBA teams derive a noticeable amount of their regular-season success from role players. BPM showed some degree of a linear relationship with a team’s win/loss percentage, but win share percentage shows almost no relationship whatsoever.
The story with bench/role players’ median value over replacement is very similar to BPM. As noted in the introduction, VORP is a similar statistic to BPM, but it factors in playing time to quantify a player’s impact on a team’s success. The graph below has a correlation coefficient of .627, and there is a noticeable trend between team success and median VORP. However, there is still a non-negligible grouping of teams clustered around zero median VORP that vary widely in team success. For example, the vertical line at zero contains the Sacramento Kings that had a sub-500 record and the Toronto Raptors that were the NBA champions. Even teams like the Miami Heat and the Dallas Mavericks had great role players according to this metric but struggled to win games. We believe this graph further emphasizes the significant difference that star players make on team success compared to the impact of role players. It takes a player like Kawhi Leonard to transform an average bench like that of the Raptors to a 70% win percentage.
Considering the median of only the top four role players for each team with the graph below, the linear trend becomes even weaker with a correlation coefficient of .390. With so little horizontal spread across the league, this graph argues that role players are not what drives teams to the top playoff positions. The graph shows three groups of NBA franchises. There are four teams in the bottom left that have statistically bad benches and bad overall team results. In the center near the vertical blue line, there are many teams there that vary widely in win/loss percentage. Finally, there are the teams that have above-average benches and most of those teams made the playoffs. A differentiator for the teams near the center could be the presence and production of a star player. This makes sense too because teams with win/loss ratios above 50% are playoff teams and most of them have big-name star players.
The final nail in the coffin from advanced statistics for the lack of major influence on win percentage from role players comes from player efficiency rating. While comparing the graph below, it shows almost no linear relationship. Interestingly, two playoff teams with a 60% win percentage have the highest and lowest median role player PER. The very successful teams even have the benefit of superstars drawing so much defensive attention to themselves that it opens up opportunities for others to contribute. An example of this could be the Milwaukee Bucks, where Giannis Antetokounmpo requires so much help defense to guard that his teammates might have more scoring or rebounding chances.
Considering only the player efficiency rating of the top 4 role players on each team and the team’s win/loss percentage in the graph below, the trend continues from above where there is really no correlation between the variables. The New York Knicks and the Golden State Warriors have nearly identical median PER values, but the win/loss percentages differ by about 50%. Even though PER is a rate statistic that does not take playing time into consideration, it still argues that role players are not very significant to team wins. It seems like the most common advanced statistics in the NBA that try to condense a player’s total value to one number agree that bench/role players have very little impact on a team’s overall win/loss percentage.
Outside of advanced metrics, could a factor like the age of bench/role players affect a team’s win/loss rate during the grind of the regular season? A veteran bench could be more consistent down the stretch or be more susceptible to injuries, while a younger bench might offer more energy but be more inconsistent with their performances. The graph below compares the median age of role players on each team compared to their win/loss percentage. There is not a very clear trend because teams that made the playoffs had a wide varied to median role player age. The correlation coefficient is only .233. However, there does some to be less variation in team success among teams with median role player age greater than the average NBA value.
Transitioning to just the median age of each team’s top four role players, there is a little bit more of a linear relationship as the correlation coefficient increases to .320. But it is still not a graph that jumps off of the page with an obvious trend to summarize bench age versus team success. The Denver Nuggets and the Phoenix Suns both have median ages just under the NBA average at about 25.5 years old, but the Suns were one of the worst teams in the NBA while the Nuggets had a great year in terms of regular-season wins. Intuitively the lack of a clear trend makes sense because the quality and performance of a player matter much more than their age. Coaches always highlight the importance of a veteran presence in the locker room, but using the median as a measure of spread should remove most of those high side outliers. Like the advanced statistics above, it appears that the age of role players does not have a clear relationship to a team’s win/loss ratio.
Playoff vs. Non-Playoff Teams
Regular season games are important, but what defines the legacy of a franchise, or even a player, is the championships that come with it. Inevitably, teams need to do a lot of work in order to put themselves in the position to contend for a championship by going into the playoffs. In a league where almost half of the teams will be eliminated before going into playoffs, there is not necessarily a clear line drawn between these teams, but for our purposes of the article, it is worth researching the role player performance that may indeed have an effect on how playoff seeding turns out.
I mapped out the win share percentages of role players over their respective teams in box plots to compare the average win shares of the playoff teams and non-playoff teams. In order to correctly reflect the role players on the team, I chose to limit the playing time to between twenty-five minutes and five minutes, excluding potential star players on a certain team and garbage-time players (essentially players who play out games that are either a guaranteed win or loss). In terms of win share percentages for these players, I used the same method as the one used in general statistics, dividing a person’s win share over the team win share -- the summation of every player’s win share -- throughout the regular season. The point of making this adjustment instead of using the win share statistics by itself is that playoff teams have a higher win count by default, thus producing higher win shares by default; the percentages, however, allow us to see the impact a certain player has on the win share of the team, and we can then use this information to compare between the playoff and non-playoff teams.
From this graph, we can see that playoff teams have a higher average percentage win share over the non-playoff teams, at 0.09868% and 0.08633%, respectively. However, we do see outliers in the graphs, namely Nerlens Noel of the Oklahoma City Thunder and Doug McDermott of the Indiana Pacers, to name a few. Therefore, to limit the effect these outliers bring on the average high, I also constructed the medians for the playoff and non-playoff teams. This computation further shows that playoff teams have a higher average win share percentage, at 0.07747%, over non-playoff teams, at 0.05714%.
To show that the differences are significant regarding the dataset, I have used a student’s t-test in order to further assess the correlation between role players and team win percentages during the regular season. The purpose of doing the t-test is to determine if the difference in average percentage win shares is statistically significant enough to show that the teams’ role players make a greater impact on the playoff teams. After conducting the variance analysis, I decided to use the t-test with two different samples. Towards that end, we came up with the null hypothesis that there isn’t a significant difference between playoff and non-playoff teams and if this hypothesis is proven inaccurate, alternatively, we reject the null hypothesis and conclude a clear difference between these teams.
After performing the t-test, we failed to reject the null hypothesis of no difference between playoff and non-playoff teams. In essence, though there is a difference in win percentage in terms of mean win share percentage, it is not significant enough to conclude that the role players’ win shares differ by a significant amount to affect playoff appearances. Because of the discrepancy with my initial expectation, I also decided to do another evaluation with the top four players in terms of minutes played. However, the result indicates an even lower possibility of a correlation between role players and playoff appearances. The underlying meaning of this result is that star players, rather than role players, take most of the credit for bringing teams to the playoffs. That being said, we did not consider intangible factors such as management, health, coaching, and many others into account for this evaluation; however, for the purpose of this article, this is the most plausible result that we can conclude.
Teams with the Best and Worst role players
Although there does not seem to be much of a correlation between role players and team success, basketball is more than pure statistics. As a regular consumer of the NBA who watches games almost every day, role players still maintain very impactful from a visual point of view. There are more than enough examples of how role players eventually stepped up and changed the course of a game: the Last Dance documentary provides two prominent examples — John Paxon and Steve Kerr — who helped Michael Jordan win two of his six championships with their stunning finishing blows. Therefore, I intend to use both statistics and visual analysis to explain how role players affect team success.
The metric I intend to use to decipher which two teams have the best role players is player efficiency rating because it accurately presents the effectiveness of a player individually; in addition, the role attributed to the players are also put into consideration when deciding which team has the best role players.
It is quite interesting to see how the Boston Celtics ended up here as the team that has the best role players. In terms of player efficiency rating, Boston has the highest median out of all the teams that participate in the National Basketball Association. With young talents such as Jayson Tatum and Jaylen Brown and a superstar in Kyrie Irving, they were able to clinch fourth place in the Eastern Conference come playoff time.
Though Robert Williams appears to have the highest PER of all role players at 18.8 on the Boston team, it is worth noting that he only averaged 8.8 minutes per game, and played in 32 out of 82 games throughout the season. The PER could be a bit inflated because of these reasons, so we will not focus on him and anyone who averaged too few minutes for that matter. In today’s game, after Stephen Curry has undoubtedly revolutionized basketball to be oriented around perimeter shooting, many of the role players take up the challenge to score from the outside. Terry Rozier, being the unsung hero of the team, averaged 9 points, 2.9 assists, with 35.3% shooting from the 3-point line, as a great complement and secondary ball distributor to Kyrie Irving, the main star of the team. His versatility was undermined by the halo of Kyrie Irving, but nonetheless allowed him to continually improve: he signed with the Charlotte Hornets in the 2019-2020 season, where he averaged 18 points and 40.7% from three. Aron Baynes, as the defensive rim protector and a stretch big of the Boston Celtics, averaged 4.7 rebounds and 0.7 blocks, and 34.4% from downtown. His PER is 14, which is one of the highest on the team. In the 2019-2020 season, he also exploded and dropped 37 points with 9 3-pointers as a Phoenix Sun. It is quite unfortunate that both players did not have a chance to play for the Celtics in the 2019-2020 season. Other players such as Daniel Theis and Brad Wannamaker all have their moments here and there, thereafter assisting the Celtics to take the fourth seed in the East. However, despite their exceptional role players, the team still lost in the second round of the 2019 playoffs. There were many speculations as to why this team wasn’t successful, but for the purpose of this article, we will not discuss the potential causes of their defeat against the Milwaukee Bucks.
OKLAHOMA CITY THUNDER
Contrary to the Celtics, the Oklahoma City Thunder has always had a strong star lineup. This year, they had both Russell Westbrook, who came off of another amazing season averaging a triple-double throughout the regular season with 22.9 points, 10.7 assists, and 11.1 rebounds, and Paul George, who averaged 28 points, 4.1 assists, 8.2 rebounds, and finished 3rd in MVP rating. However, though the star power is almost incomparable in the whole league, they lack depth in terms of the whole lineup.
The player efficiency ratings look almost incomparable to the Celtics. From the graph above, there is a clear difference between the teams’ individual ratings, with Oklahoma City Thunder being significantly lower. Judging from the bar plot above, there is a cluster of players having around 10 PER throughout the season, while Nerlens Noel has a significantly better performance amongst his peers. Rather than having a group of sharpshooters to accompany the fierce attack that Russell Westbrook executes inside the paint, OKC’s role players’ 3-point shooting percentage simply looks abysmal; none of the players has even over 34% from downtown. To compensate for the lack of depth, the Thunders heavily relied on the rotation of Russell Westbrook, Dennis Shroder, Steven Adams, and Paul George to create offensive threats. One bright spot in this lineup, however, is Nerlens Noel. As the second-best big man on the team, he was able to perform under pressure and become the defensive anchor on the team. Indeed, he racked up a stunning 19.3 player efficiency rating, which is well above the league average, and most definitely much higher than most of his teammates. As a 24-year-old, much of the rim protection duty falls onto his shoulders, and he was able to deliver one time after another. This season, despite playing around 14 minutes per game, he was able to rack up 1.2 blocks per game, 0.9 steals per game, and 5 points per game. To his credit, we can also see an increase in production in all statistical areas in the 2019-2020 season.
The Miami Heat stick out as a team in many of the graphs because their role players are statistically successful, but the team’s win/loss percentage is lower than what would be expected. Especially in the graphs showing only the top four role players on each team, the Heat are either the outright leaders or tied for the lead in BPM, WS percentage, VORP, and PER. However, with a record of just 39 wins and 43 losses, the Heat missed the playoffs even though the bar was set pretty low by the other teams in the Eastern Conference. A closer look at the Heat’s team highlights the lack of a superstar to really lead the team. The most easily recognizable player was Dwyane Wade, but at 37 years old, his 15 points per game was just a highlight of his farewell tour. The team was led in scoring by Josh Richardson, who is a very capable role player, but not someone to build around as the face of the franchise.
Even though the Heat struggled this season, they were ready to reload upon D-Wade's retirement, and their strong supporting cast meant their main goal was signing a star player. Jimmy Butler was the perfect blockbuster signing they needed because of his determination and hard work ethic. Moreover, key additions to the team, such as Jae Crowder, Kendrick Nunn, and Tyler Herro, rounded out the already solid bench to create a threatening team. Offensive production was being distributed fairly evenly as eight players were averaging more than 10 points per game, and no one was averaging more than 20. In the NBA bubble in Orlando, the Heat surprised many by advancing to the NBA Finals as the 5th seed from the Eastern Conference. Huge production by Butler and Dragic paired with clutch performances by Tyler Herro, and Duncan Robinson allowed the Heat to overcome every team they faced except the talent-laden Lakers.
It is rather unexpected to see that role/bench players have less of a role than what we initially thought and believed while watching NBA games. Star players still remain one of the major reasons that teams win championships, as they often have unparalleled talent and prowess. Despite the trivial impact role players have on the game of basketball, teams would still face immense challenges if they refuse to sign any. After all, no human can play 48 minutes per game (and more if games go to overtime) in an 82-game stretch and implement only a few strategies that can be easily cracked using today’s knowledge and technology. The star players would suffer immense health problems that lead to fatigue and injuries, and the audience would think that the NBA is boring to watch. Role players also offer much utility both offensively and defensively across sharp 3-point shooting, playmaking, rebounding, or rim-protecting; leadership and such intangible innate qualities can also be easily overlooked because of its lack of presence on the stats sheet. In addition, many role/bench players gradually transformed into star players due to their hard work and well-thought-out training done with NBA standards. James Harden, for one, made a huge jump from a normal bench player who came out of college as a prospect to sixth-man of the year, then averaging a stunning 36 points per game and becoming the face of the Houston Rockets. Role players may not statistically differ in their impact from team to team, but they nonetheless remain a necessary part of every franchise