Our Blog


Daily Fantasy NBA Player Correlations

Daily Fantasy Correlations

With daily fantasy football coming to an end in January, I want to look at a topic that is popular in the NFL but hasn’t been explored as thoroughly in NBA: Teammate correlations and stacking.

In football, it’s easy to understand why teammates are correlated: Fantasy points are largely scored by high-leverage events — namely, touchdowns — and you want access to as many of those events as possible in your lineups. If Aaron Rodgers is going to throw four TDs, you likely want to roster him with Jordy Nelson, because 1) Nelson combines with Rodgers for a large chunk of the Packers’ TDs and 2) both players can separately benefit from that one event.

In NBA, it’s different. You could potentially stack a guy like Chris Paul and DeAndre Jordan: Paul is fourth in the league this year with 9.5 assists per game, and, like a QB, he can rack up points with his teammates when they score on his passes. However, since his 441 DraftKings points (on 294 assists) make up only 32.4 percent of his DFS value, stacking in the NBA doesn’t function as it does in the NFL.

To study NBA stacking, I created a test group with every player who has played at least five games this season and is averaging at least 15 minutes per game. Then I looked at how they correlate with their teammates: 1 is perfect correlation, 0 is no correlation.

Here’s the data, first in interactive graphs (hover to see data points) and then in an excel sheet.

NBA Player Correlations

The Clippers Revisited

In the intro, I touched on CP3 and Jordan. Using the information above, we can see that Blake Griffin actually has a higher correlation with CP3 than Jordan has, even though Jordan is much more dependent on Paul for offensive production.

clippers1

So why is Jordan less correlated with CP3? Jordan relies more on rebounding than scoring for DFS value. Jordan’s 449 rebounds account for 49.0 percent of his DFS value this season. What Paul does for them means more to Griffin than to Jordan.

Game Stacking

There’s a lot to take away from this correlation data.

Carmelo Correlations

Take a player like Carmelo Anthony. He’s averaging 34.59 FD points per game and a +2.97 Plus/Minus this season. He hasn’t been consistent — he’s hit salary-based expectations in 53.3 percent of his games (for reference, Kevin Durant has 72.7 percent Consistency) — but he’s been an asset on many nights. Per our Trends tool:

anthony1

For the most part, the players correlated positively with Melo are his fellow starters:

melo1

That’s what we might expect: A starter to correlate with other starters. After all, if the bench has a good game, that likely means that the starters have performed poorly and the game is a blowout. Right?

That’s partly true. The Knicks have lost four games this year by more than 13 points — twice against the Cavs and once versus the Rockets and Celtics. In those games the Knicks’ ‘Big 3’ performed poorly:

knicks1

However, only three Knicks posted positive Plus/Minus values in those four games, and they did so only because they were cheap.

knicks2

The Knicks all had strongly correlated outcomes in those games. They just weren’t positive outcomes.

Carmelo + Porzingis = Carzingis

Whereas Melo and the Knicks bench players don’t compete for production, Melo and Porzingis shouldn’t be highly correlated, as they cut into each other’s usage. Per our new NBA On/Off tool, Melo actually increases his usage rate by 6.2 percent in his 9.9 minutes per game with Kristaps off the floor:

melo2

However, look at those other numbers. Although Melo does slightly better (+0.2 DK points per minute) without Porzingis, the Knicks offense overall is worse: It drops from 1.112 points per possession with both on the court to 1.098 with Melo on and Porz off. The team as a whole is better with both of them.

Projected Game Flow

Just last night (12/28/16), C.J. McCollum was a chalky play without Damian Lillard, as he averages a 31.4 percent usage rate and a +7.9 DK Plus/Minus differential in games without Lillard. Situations like that are great because of the extra value created, but what about situations in which McCollum and Lillard or Melo and Porzingis play together?

There will be more studies, but initially this correlation data suggests that teammates don’t cannabilize each other to the point that they can’t be rostered together.

My theory is that projected game flow is much more important than balanced usage rates. This might be an extreme example, but Lillard and McCollum put up 51.3 and 44.4 FD points (+11.73 and +14.49 Plus/Minus values) in the same game versus the Kings last week. That game went for 247 points, which is high, but the point is that positive game script outweighs usage cannibalization.

The Cavs: LeBron and Love

Take Kevin Love: Although he increases his usage rate by 8.7 percent with LeBron James off the floor, he also sees a 2.7 percent drop in true-shooting percentage, and the team allows 4.0 points per 100 possessions more than when both are on.

The concept of weighted usage rates can help us put the LeBron-Love pairing in context. Although Love uses 35.8 percent of the Cavs’ possessions when LeBron is off the floor, he does so for only about seven minutes per game. Considering that the Cavs play at a slower pace and Love commits more turnovers in that time, Love’s usage bump in those seven minutes doesn’t mean much.

Love is valuable in games that LeBron doesn’t play . . .

love1

. . . but we already knew that. We’re looking at using Love and LeBron together when they both play — which has been the case for 25 of the Cavs’ 30 games this season.

Love’s usage is relatively consistent from game to game. What is not always consistent is the Cavs’ performance.

Performance and Correlation

Offensive and defensive efficiency, projected shooting, and other efficiency metrics are much more important in predicting game flow than usage distribution among players. Even though usage is important to daily fantasy NBA — especially when considering injuries — the NBA is not so different than MLB and NFL when it comes to player correlations and teammate stacking.

Finding games that have the potential to over- or underperform their Vegas totals could be much more important than analyzing players who could over- or underperform their projected usage rates. Starters do cannabalize each other’s usage rates, but when they’re on and scoring a lot of fantasy points generally the whole team performs better.

Enjoy the data above and look for more pieces studying this topic as the NFL season winds down and we get heavy into NBA in 2017.

Daily Fantasy Correlations

With daily fantasy football coming to an end in January, I want to look at a topic that is popular in the NFL but hasn’t been explored as thoroughly in NBA: Teammate correlations and stacking.

In football, it’s easy to understand why teammates are correlated: Fantasy points are largely scored by high-leverage events — namely, touchdowns — and you want access to as many of those events as possible in your lineups. If Aaron Rodgers is going to throw four TDs, you likely want to roster him with Jordy Nelson, because 1) Nelson combines with Rodgers for a large chunk of the Packers’ TDs and 2) both players can separately benefit from that one event.

In NBA, it’s different. You could potentially stack a guy like Chris Paul and DeAndre Jordan: Paul is fourth in the league this year with 9.5 assists per game, and, like a QB, he can rack up points with his teammates when they score on his passes. However, since his 441 DraftKings points (on 294 assists) make up only 32.4 percent of his DFS value, stacking in the NBA doesn’t function as it does in the NFL.

To study NBA stacking, I created a test group with every player who has played at least five games this season and is averaging at least 15 minutes per game. Then I looked at how they correlate with their teammates: 1 is perfect correlation, 0 is no correlation.

Here’s the data, first in interactive graphs (hover to see data points) and then in an excel sheet.

NBA Player Correlations

The Clippers Revisited

In the intro, I touched on CP3 and Jordan. Using the information above, we can see that Blake Griffin actually has a higher correlation with CP3 than Jordan has, even though Jordan is much more dependent on Paul for offensive production.

clippers1

So why is Jordan less correlated with CP3? Jordan relies more on rebounding than scoring for DFS value. Jordan’s 449 rebounds account for 49.0 percent of his DFS value this season. What Paul does for them means more to Griffin than to Jordan.

Game Stacking

There’s a lot to take away from this correlation data.

Carmelo Correlations

Take a player like Carmelo Anthony. He’s averaging 34.59 FD points per game and a +2.97 Plus/Minus this season. He hasn’t been consistent — he’s hit salary-based expectations in 53.3 percent of his games (for reference, Kevin Durant has 72.7 percent Consistency) — but he’s been an asset on many nights. Per our Trends tool:

anthony1

For the most part, the players correlated positively with Melo are his fellow starters:

melo1

That’s what we might expect: A starter to correlate with other starters. After all, if the bench has a good game, that likely means that the starters have performed poorly and the game is a blowout. Right?

That’s partly true. The Knicks have lost four games this year by more than 13 points — twice against the Cavs and once versus the Rockets and Celtics. In those games the Knicks’ ‘Big 3’ performed poorly:

knicks1

However, only three Knicks posted positive Plus/Minus values in those four games, and they did so only because they were cheap.

knicks2

The Knicks all had strongly correlated outcomes in those games. They just weren’t positive outcomes.

Carmelo + Porzingis = Carzingis

Whereas Melo and the Knicks bench players don’t compete for production, Melo and Porzingis shouldn’t be highly correlated, as they cut into each other’s usage. Per our new NBA On/Off tool, Melo actually increases his usage rate by 6.2 percent in his 9.9 minutes per game with Kristaps off the floor:

melo2

However, look at those other numbers. Although Melo does slightly better (+0.2 DK points per minute) without Porzingis, the Knicks offense overall is worse: It drops from 1.112 points per possession with both on the court to 1.098 with Melo on and Porz off. The team as a whole is better with both of them.

Projected Game Flow

Just last night (12/28/16), C.J. McCollum was a chalky play without Damian Lillard, as he averages a 31.4 percent usage rate and a +7.9 DK Plus/Minus differential in games without Lillard. Situations like that are great because of the extra value created, but what about situations in which McCollum and Lillard or Melo and Porzingis play together?

There will be more studies, but initially this correlation data suggests that teammates don’t cannabilize each other to the point that they can’t be rostered together.

My theory is that projected game flow is much more important than balanced usage rates. This might be an extreme example, but Lillard and McCollum put up 51.3 and 44.4 FD points (+11.73 and +14.49 Plus/Minus values) in the same game versus the Kings last week. That game went for 247 points, which is high, but the point is that positive game script outweighs usage cannibalization.

The Cavs: LeBron and Love

Take Kevin Love: Although he increases his usage rate by 8.7 percent with LeBron James off the floor, he also sees a 2.7 percent drop in true-shooting percentage, and the team allows 4.0 points per 100 possessions more than when both are on.

The concept of weighted usage rates can help us put the LeBron-Love pairing in context. Although Love uses 35.8 percent of the Cavs’ possessions when LeBron is off the floor, he does so for only about seven minutes per game. Considering that the Cavs play at a slower pace and Love commits more turnovers in that time, Love’s usage bump in those seven minutes doesn’t mean much.

Love is valuable in games that LeBron doesn’t play . . .

love1

. . . but we already knew that. We’re looking at using Love and LeBron together when they both play — which has been the case for 25 of the Cavs’ 30 games this season.

Love’s usage is relatively consistent from game to game. What is not always consistent is the Cavs’ performance.

Performance and Correlation

Offensive and defensive efficiency, projected shooting, and other efficiency metrics are much more important in predicting game flow than usage distribution among players. Even though usage is important to daily fantasy NBA — especially when considering injuries — the NBA is not so different than MLB and NFL when it comes to player correlations and teammate stacking.

Finding games that have the potential to over- or underperform their Vegas totals could be much more important than analyzing players who could over- or underperform their projected usage rates. Starters do cannabalize each other’s usage rates, but when they’re on and scoring a lot of fantasy points generally the whole team performs better.

Enjoy the data above and look for more pieces studying this topic as the NFL season winds down and we get heavy into NBA in 2017.