With the new NBA season beginning within a week, many NBA fans and analysts have begun to present their preseason predictions, choosing which teams will make the playoffs, finals, and win the championship. I have devised a method of predicting the best NBA teams this coming season using the stats of each team’s players. Using the statistics to predict the best and worst teams for the coming season can allow us to see which teams may be overhyped and which teams are flying under the radar.
I chose to base my preseason predictions off mainly the players on each team as opposed to previous team performance. In order to find this season’s projections, I had to first look back to previous seasons, using the past ten years as the dataset for the project.
When proceeding with all the steps listed below, I not only used the past year’s data to predict the current season, but also the year prior to the past year’s stats. Therefore, a sample of the past two seasons were used. I chose to do this because of the fact that some players miss the majority of a season, such as Kevin Durant, Steph Curry, and John Wall this past year. Additionally, it provides a greater sample to give a more accurate result. However, in order to boost the importance of recency, I weighed the more recent season’s results two times more than the year that was two seasons prior.
In order to summarize the overall impact of a player, I decided to use an average of three metrics: win shares per 48 minutes, BPM, and game score per 36. Each of these metrics have their advantages and their flaws, so averaging them will give the best consensus ranking of a player. Additionally, each has been converted to a rate stat so players with high usage and low efficiency are not unjustly rewarded.
After finding the three main stats for the players in each season, I standardized them based on position in order to account for positional biases that are present in some of the stats (most notably big men in WS/48). For example, the top 15 among qualifying players in win shares per 48 minutes last season consisted of 8 centers and 2 power forwards, placing players like Enes Kanter and Jarett Allen over stars like Luka Doncic and LeBron James. After standardizing, however, the top 15 had a more uniform mix of positions, with only 5 of the original 10 big men. The distributions of all three statistics were very similar as they all were slightly right skewed, displaying the effect of superstars in the NBA as all-NBA caliber players were consistently over 2 standard deviations above the mean.
To include the fact that players will improve if they are very young, I took another step after finding the z-scores of the three main stats. I used a two variable linear regression to predict the subsequent year’s z-score for each of WS/48, BPM, and Game score, using age and current z-score as the two variables. This way, the older players were projected to regress while the young players had improved ratings. To have a quick summary of the results, the top 20 projected players for next year based on their average z-score of WS/48, BPM, and game score is shown below. The results seem to make sense using the eye test, with those who are traditionally regarded as the best placing highly.
Next, after summarizing the effectiveness of each player, I had to estimate how often each would play. I used a multiple linear regression model to find the predicted minutes percentage based on per game statistics. The model used inputs such as field goal attempts, offensive rebounds, steals, turnovers, personal fouls, and points (all significant variables) to estimate the percentage of total minutes that the player played on a per game basis. While the projections below may look low for some notable high usage players like James Harden and Damian Lillard, no projection should be above 20% since there are always five players on the court. Additionally, a minutes percentage around 15%, which is close to most of the players below, equals about 36 minutes a game over the course of a full season.
The final step was combining all the players’ stats into one team stat. First, I found the relative average for each player by multiplying their predicted minutes percentage (which I scaled so they would sum to 1 for each team) with their average z-score. Then, I kept only the top 8 players in minutes percentage (to prevent penalization for teams that happen to have a lot more players, diminishing the effect of their best players) and I summed the relative averages for each team. After that, I used a regression to predict the wins per 82 games based on the team averages, then scaled to 72 games and the average NBA standard deviation in wins.
Lastly, to construct a method of comparing the impact of players against each other, I calculated the estimated wins added by each player. The exact calculation uses a combination of the minutes percentage, average z-score, and league average z-score (which isn’t 0 since I kept only the top 8 from each team). With these values, I can investigate how impactful a player is over a typical player. The most estimated wins added (I’ll use EWA from now on) was Giannis, with 25.88, while the person with the least was Dennis Smith Jr, with -2.88 EWA. The typical player, found by taking the median, had around 3.88 EWA, so in order to see what the estimated record of a team without a certain player would be, you would subtract the player’s EWA from the team’s total projection, then add 3.88. For example, the Bucks estimated record without Giannis would be 55.79 – 25.88 + 3.88 which is 33.79, good for 20th in the NBA.
Results and Analysis
The final predicted number of wins for each team based on player stats are shown below. (The win projections are scaled for a 72 game season for this year as opposed to the normal 82 game season)
The top 5 projected teams for 2021 according to my player based model are the Bucks, Clippers, Nets, Lakers, and Rockets.
The Bucks placed first largely due to Giannis Antetokounmpo, who placed first in the player projections. As already illustrated above, the projection for the Bucks without their star power forward would be 20th in the NBA. The Bucks tried to improve their team by trading guards Eric Bledsoe and George Hill to the Pelicans and Thunder (4 team trade) in exchange for former All-Star Jrue Holiday. However, the model did not see this trade as very beneficial to the Bucks, adding just 2 wins. The Bucks’ huge dependence on Giannis will likely hurt their team as an injury could affect them greatly.
The Nets’ healthy duo of Kyrie Irving and Kevin Durant was the reason for their high projection, adding 22.4 wins combined. Since I could not account for potential rust from not playing for over a year, Kevin Durant’s projection may be off, meaning the Nets will not be as good as their projection if Durant is not as good as he was in 2019.
Meanwhile, the defending champion Lakers are projected to be 4th in the NBA. Their two biggest acquisitions over the offseason were trading for Dennis Schroder and signing Montrezl Harrell. While the Harrell signing added 3.6 wins to the Laker’s projection, the Schroder trade had minimal effect. The model saw both Danny Green and Dennis Schroder as just average players. Schroder’s advanced statistics were average among point guards across the last two seasons, with an average z-score of just 0.26, while Danny Green’s average z-score was 0.30. The advanced statistics like win shares per 48 and BPM likely fail to recognize the difference in the shot creation ability between Schroder and Green, as just 48% of Schroder’s shots last season were unassisted while only 12% of Green’s shots were. Therefore, this is one aspect of my model that I disagree with since Schroder’s contribution should be higher than Green’s.
One team that is different than several other predictions is the standing of the Rockets, as they placed 5th overall in my player based model. However, their standing is almost entirely due to James Harden. Without Harden, the Rockets would be projected 24.73 wins, which would be 24th in the NBA. The model cannot take player morale into account, so if Harden’s production drops or he is traded away, the Rockets would fall substantially. Furthermore, heavy reliance on one player is never good, so the Rockets may have problems heading into the season.
Two other major unexpected results are probably the 76ers placing 6th and the Heat placing 15th, good for 3rd and 7th in the Eastern Conference, respectively. The reason for the 76ers’ high standing was primarily the expected improvement of their young stars in Ben Simmons and Joel Embiid along with the acquisitions of Seth Curry and Danny Green in expense of Josh Richardson and Al Horford. One positive for the 76ers is that they are not so heavily reliant on one player like the Bucks and Rockets are, but a negative is that I did not attempt to predict injuries for the upcoming season. If Embiid and/or Simmons are injured for extended periods of time, as they have for the past few seasons, then the 76ers will have fewer wins than their projection.
The reason for the low projection for the Heat is the fact that the model only took regular season stats into account, meaning the improvement of the Heat players in the playoffs was not included. The Heat also have just 3 players (Butler, Adebayo, Robinson) that are projected to be better than a typical player. Tyler Herro and Goran Dragic, both of whom contributed greatly to the Heat in the playoffs, were both seen as below average players across all three statistics used.
One of the most unexpected results was seeing the Warriors in the bottom 10 after having Steph Curry return. However, after taking a deeper look into the roster, this seems to make sense. With Klay Thompson injured, Curry’s best help consists of Kelly Oubre, Draymond Green, and Andrew Wiggins. None of these players graded well in the statistics I used, with each seen as below average players in all categories. Curry has the most difficult job of any player this season, as his 15.48 EWA is almost half of the Warriors’ entire win projection. Meanwhile, the trio of Oubre, Green, and Wiggins combine for 10.74 of the Warriors 32.75 projected wins.
In this section, I will list some of the notable signings and trades of the offseason, then show how much they helped or hurt each team, in terms of win projections for this season.
Free Agent Signings
- Danilo Gallinari, OKC to ATL: +5.37 wins
- Gordon Hayward, BOS to CHO: +4.38 wins
- Christian Wood, DET to HOU: +4.27 wins
- Montrezl Harrell, LAC to LAL: +3.61 wins
- Hassan Whiteside, POR to SAC: +3.11 wins
- Bogdan Bogdanovic, SAC to ATL: +0.50 wins
The trades below only show which players were acquired, not the draft picks.
Jrue Holiday Trade
- NOP Receives: Steven Adams, Eric Bledsoe (-1.36 wins)
- MIL Receives: Jrue Holiday, Sam Merrill (+1.96 wins)
- DEN Receives: RJ Hampton (+0 wins)
- OKC Receives: George Hill, Zylan Cheatham, Josh Gray, Darius Miller, Kenrich Williams (-0.59 wins)
Chris Paul Trade
- PHX Receives: Chris Paul, Abdel Nader (+3.15 wins)
- OKC Receives: Kelly Oubre Jr (to GSW), Ricky Rubio (to MIN), Ty Jerome, Jalen Lacque (-3.15 wins)
Russell Westbrook Trade
- WAS Receives: Russell Westbrook (+3.54 wins)
- HOU Receives: John Wall (-3.54 wins)