How not to construct an NBA lineup

By Stephen Shea (@SteveShea33) and Chris Baker (@ChrisBakerAM)

February 21, 2016

Building a successful NBA lineup is more than blindly assembling 5 talented NBA players. It’s much more than that.

Every player is unique, and sometimes the difference between a good fit and a poor fit for a particular player is subtle.

There isn’t one correct formula, one blueprint every team should follow. The NBA’s biggest stars don’t fit any molds. The optimal lineup around Steph Curry doesn’t look like the optimal lineup for Russell Westbrook. A team with LeBron James tends to have a great deal more flexibility in their lineup constructions than a team without LeBron.

It’s not just the superstars that create and remove lineup flexibility. Boston’s Evan Turner provides tremendous playmaking out of the small forward position. He allows Boston to play two guards that work well off the ball. At the same time, his lack of perimeter shooting forces the team to play shooters around him in order to create proper spacing on the offensive end. In contrast, a wing like Kyle Korver is a great floor spacer, but he needs a teammate that can create off the dribble. Playing Rondo with a player like Korver could work. Rondo with Turner, Tyreke Evans or Monta Ellis is likely to implode.

Lineup construction is a challenging puzzle where the teams can’t always choose the pieces. To build on the previous example, Evan Turner has worked with lineups that feature stretch fours such as Olynyk or Jerebko. If Blake Griffin were available at the trade deadline for the right price, the Celtics weren’t going to pass because he didn’t fit well with Turner. They would have taken Griffin and then shuffled the pieces as best they could to match.

The following is an attempt to navigate the lineup labyrinth by identifying positive and negative trends, all the while understanding that sometimes the best things happen when all the rules are broken.

Rim Protection and Floor Spacing

The modern NBA is won at the 3-point line and at the hoop. The mid-range shot is not efficient. Even the best teams can’t generate 0.90 points per shot from mid-range. In addition, spacing to the 3-point line creates better driving lanes than spacing to 15 feet.

When evaluating players and lineups, I tend to think in terms of 4 major categories. Offensively, I look at perimeter shooting and ability to get to the hoop (including drawing fouls and FT%). Defensively, I think in terms of rim protection and perimeter defense. Perimeter defense includes positional versatility to switch screens.

My optimal (but still realistic) lineup would include the following. The big man would be an elite rim protector. Ideally, this big would be agile enough to hedge on screens (or even switch if need be). Perhaps this is what Willie Cauley-Stein will be after a year or two of development. Perimeter defense and rim protection from a big is unlikely, and so I’d settle for a good interior defender with Rudy Gobert the prototype.

Offensively, the big needn’t be more than a capable roll man that can leap for the lob and finish at the hoop. Post skills, a jump shot, or passing would be a bonus.

The lineup would have three wings with positional versatility. They’d be able to switch nearly every screen on the perimeter and guard most teams’ 2 through 4s.  Players like Jae Crowder, Trevor Ariza, Demarre Carroll, and Shane Battier come to mind.

In addition, all three wings would need to shoot at least moderately well from 3-point range (33% or better).

One of those 3 wings would be an offensive “star.” It would be the type of player that earns max money, is ball dominant, and who can draw doubles on drives to open his teammates for open catch-and-shoots. He would be the go-to option in crunch time.

Point guard is a deep position in the NBA, and so this is a place teams don’t need to overinvest. Steph Curry would be amazing, but I’d be happy with a number of other options. With a very few exceptions (such as Russell Westbrook, whose talent breaks the rules), I’d want the PG to be an adequate perimeter shooter as well.

It would be absurd to ask if this optimal lineup, a lineup that would perform on both the interior and perimeter on both offense and defense would be successful. Of course it would.

For a better analysis, we’ll lower the standards and focus on two areas, rim protection and perimeter shooting (floor spacing).

There are a number of reasons to focus on these two areas. Here are just two of them.

NBA teams almost always play lineups with at least one threat going to the hoop (and usually more). In contrast, there can be great variety in the amount of perimeter shooting a lineup features.

Defensively, it is very hard to quantify an individual’s perimeter defense and positional versatility.

At the team level, rim protection and perimeter shooting have been a tremendous predictor of team success. We’ll use a team’s 3-point FG% (3P%) as a measure of their perimeter shooting and their opponents’ FG% in the restricted area (Opp.RA%) as a measure of rim protection. We’ll combine them into one number by taking 3P% and dividing by Opp.RA%. The following chart plots this simple formula to win % for each team this season and in the last 3 seasons.

Image may be NSFW.
Clik here to view.

Our simple formula has a high correlation to win %. Even more remarkable, the formula has correctly predicted the eventual champion in each of the last 3 seasons. This season, Golden State is in the lead. San Antonio is not far behind, and they are actually posting a score better than any team in the previous 3 seasons.

The team-level correlation shows that teams should strive for floor spacing and perimeter shooting, but what does this mean at the lineup level? For example, how much perimeter shooting does a lineup need?

A lineup study

Basketball is a team sport. Very few activities are solely the product of one player’s efforts. Proper rim protection and floor spacing are the result of all players working well in unison. Still, there are certain individuals that can have a much bigger impact in rim protection (such as Rudy Gobert) and others that can have a much bigger impact on stretching defenses (such as JJ Redick). It is exceptionally rare to find a player that can do both.

To understand how many rim protectors and floor spacers a lineup needs, we must first identify players as 3-point threats (3P) and/or rim protectors (RP). We will have two levels for each (a light and a strong). The following table contains the definitions. For example, a 3P Light will be all players that shot at least 33.3% from 3 and shot at least 0.05 3PA per minute the player played.

Image may be NSFW.
Clik here to view.

The next table shows how many players qualified in each category in each of the last two seasons.

Image may be NSFW.
Clik here to view.

The Light qualifications are loose enough so that every team had ample opportunity to have qualified players. There were enough 3P Lights in every season so that every team could have about six and enough RP Lights so that every team could have 2. That is enough for two distinct lineups each with three 3P Lights and one RP Light.

We first ran (in-sample) studies for 2014 and 2015. The next table shows the net points per 48 minutes for lineups grouped into 3 categories. It also shows the percentage of time in the NBA that each type of lineup played.

Image may be NSFW.
Clik here to view.

The in-sample studies use player qualifications from the season of interest. So, for 2014, we looked at all lineups with at least three 2014 3P Lights and 1 2014 RP Light.

To show that there is some predictive power to this strategy, we ran an out-of-sample study. We started with every player that played in 2015. If that player played at least 25 games in 2014, we used the player’s 2014 classification (as a 3P Light or not, etc.). If the player played less than 25 games in 2014, we used his 2015 classification (so that we wouldn’t have to toss out a significant portion of our sample). The results are in Table 4.

Image may be NSFW.
Clik here to view.

A league-wide predictive study of this nature is going to lose some power because it cannot determine on a case-by-case basis (without more information) the true rim protectors and shooters. It wouldn’t be too much trouble for teams that are considering adding depth rim protection or 3-point shooting to assess whether or not a particular individual is likely to continue his production (or even improve if given a better context).

Next, we looked at lineups that had at least three 3P Strongs and at least one RP Strong. This is (in some sense) a modern ideal for how a lineup should be rounded out (around its stars). Table 5 displays the results for the 2014, the 2015 and the out-of-sample 2015 study (2015 OS).

Image may be NSFW.
Clik here to view.

Perhaps the most important information in Table 5 is how rare these “Strong” lineups are. In both 2014 and 2015, just a few teams drove the numbers.

In 2015, the Clippers had the highest % of time with a Strong lineup at 43%. Cleveland was second with 24%. In both cases, the Strong lineups were dominant (at least 12.4 net points per 48 minutes).

In 2015, Atlanta (82%) and Golden State (74%) were the top two teams in % of time with a Light lineup on the court. For GSW, Steph Curry, Klay Thompson, Harrison Barnes and Draymond Green all qualified as 3P Lights from the starting lineup. Andrew Bogut was the rim protector. Atlanta also featured four 3P Lights in their starting lineup.

The main conclusions of this study are that rim protection and floor spacing (through 3-point shooting) are especially important components for a modern NBA team. In particular, lineups with at least one rim protector and three 3-point shooting threats have far outperformed other lineups.

But why? We expected these results from the lineup study after seeing the team-level analysis, but neither of the two analyses tells us precisely why rim protection and floor spacing are so valuable. We have our intuition, but can analytics actually explain how spacing the floor helps an offense? Is 3 the magic number for floor spacers? Why is one (and perhaps just one) rim protector so crucial?

An explanation with spatial-tracking data

On May 10, 2015, Phil Jackson famously gave the world his “how’s it goink?” tweet. In that tweet and surrounding messages, Phil suggested that the NBA is more about penetration than 3-point shooting.

Why am getting cranky stuff? Seriously, bball, it's about penetration.
Those cranky the Heat/Spurs got it..Heat via s/roll, Spurs all ways.

— Phil Jackson (@PhilJackson11) May 10, 2015

Phil has a point. Penetration is important. The issue is that Phil suggested that penetration and 3-point shooting are at odds with each other.  In his defense, he later tried to clarify his position.

Some corrected thoughts:1)like 3pt shooters, check it out, but to play for 3pt shot is an error. Penetration, first principal of offense.

— Phil Jackson (@PhilJackson11) May 24, 2015

We’re afraid he’s missing a key link between the two activities.

In a basketball offense, 3-point shooting is not in competition with penetration. Instead, 3-point shooting and penetration are perfect complements. When an offense is a serious threat to attack at the rim (through drives, post ups or cuts), the opposing defense must consider playing more compact and bringing help defenders near the lane. When a team has good 3-point shooters stationed around the perimeter, the help defenders off the ball must play tight to their men so that they do not give up an open catch-and-shoot 3 on a kickout.

The off-the-ball defenders cannot be two places at once. They cannot be both near the lane to help on penetration and out on the 3-point line taking away the perimeter shot.

NBA offenses that have the formation, action and 3-point shooting to keep help defenders at bay during penetration will typically be more efficient.

Let us demonstrate this effect with data from the 2014-15 NBA season. We first filtered all plays to isolate the halfcourt possessions. (We used a conservative formula that is more likely to err on the side of excluding too many possessions than including transition.) Within the halfcourt possessions, we identified the instances where the offense “penetrated.” These were instances when the ball went from being in possession outside of 15ft from the hoop to being in possession inside of 10ft. This could be a drive or post up, for example. At the moment the ball crossed to within 10 ft., we looked at how many defenders were inside of 12 ft. from the hoop.

The offense should want less help defenders around the hoop. Here are the 15 lineups that averaged the fewest defenders while penetrating (and that met a minimum of 50 such possessions in our data). We loosely label the number of defenders within 12 ft. of the hoop as “help defenders.”  This group will typically include the ball handler’s defender as well.  The average number of help defenders ranged from 2.23 to 3.24.

Notice that 11 of the top 15 lineups (73%) had an ORtg (per NBA.com) above 111. The best team offenses in the NBA last season, the Clippers and Warriors, had ORtgs of 109.8 and 109.7, respectively. So, an ORtg better than 111 is more than a hair better than the most efficient team offenses last season.

There were 148 lineups that met our minimum possession requirement. Of the bottom 133 lineups (those that weren’t in the top 15), only 16 (12%) had an ORtg better than 111. None of the bottom 15 (and only 1 of the bottom 30) had an ORtg above 111.

But wait, we have yet to directly link having less help defenders on penetration with having more perimeter shooting.

Let’s call a player a 3-point threat if that player shot at least 33.3% from 3 last season and attempted least 0.05 3-pointers per minute played (which is the same as our 3-point Light qualification above). (If a player switched teams midseason, let him qualify if he reached those thresholds for at least one team. The definition is meant to be inclusive.) There were 196 such players.

Of the 15 lineups with the fewest help defenders on penetration, 14 had at least three 3-point threats on the court. Nine of those top 15 had four 3-point threats on the court. The one exception was Portland, which had two 3-point threats (Matthews and Lillard). Aldridge just missed the cutoff for rate of 3-point attempts. Batum just missed on 3P%. Batum’s career 36.3 3P% suggests that he’s a threat from the perimeter. Portland’s label as having only two 3-point threats is misrepresenting their true perimeter-shooting ability.

The table below provides the average ORtg and number of help defenders on penetration split by the number of 3-point threats on the court.

Each additional 3-point threat on the court leads to less help defenders on penetration and a better ORtg.

We agree with Phil Jackson that penetration is important for NBA offenses. However, we do not see this as a reason to not focus on 3-point shooting. The importance of penetration is a great reason to emphasize 3-point shooting in the offense.

3 is not a magic number for the number of floor spacers. More is better, and less can be ok. 3 was chosen because it is a reasonable expectation for a modern NBA lineup. We expect that the lineup’s rim protector will not be a 3-point threat (although some are). That leaves most lineups with the option of having up to 4 3-point threats. We understand that there is more to the game than shooting. Teams have to field good perimeter defenders and players that can attack the rim. We leave open one slot for a defensive stopper like Tony Allen, or a penetrator like Tyreke Evans. In summary, 3 is good and feasible.

The spacing analysis has a direct implication on the number of rim protectors desired for an optimal lineup. Again, most rim protectors are not 3-point threats. Having 2 rim protectors usually implies the lineup can have at most 3 3-point threats, and in those cases precludes the lineup from including a defensive stopper like Tony Allen. Having 2 rim protectors constrains the rest of the lineup construction.

Furthermore, many of the best teams can space with both forwards. Cleveland has Kevin Love, Golden State has Draymond Green, Atlanta has Paul Millsap, Toronto has Patrick Patterson, and Oklahoma City has Serge Ibaka. A lineup with 2 rim protectors would have to send one of those rim protectors to the perimeter, negating the player’s rim protecting ability. Even worse, many rim protectors are not great perimeter defenders. They are a liability on the perimeter on these types of players.

Ideally, a team will have one player that can rim protect. This puts a huge responsibility on that one player. He will often be the only true help defense under the rim, and he’ll be responsible for grabbing a lot of rebounds.

Putting it all together, one rim protector and three floor spacers is a solid blueprint to follow.  (Although we still acknowledge it’s not the only way to win.)

How not to construct a lineup

Stars break the rules. The analysis thus far suggests that it’s not wise to play a guard that can’t shoot or a big that can’t space or defend the rim. If that big is Blake Griffin or that guard is Russell Westbrook, toss out the general trends. These players are so talented in other areas that they more than make up for not fitting the mold.

On the other hand, playing a big that can’t defend or space and is not quite star caliber is questionable. It can work in the right situation, with appropriate teammates, but it constrains the lineup construction.

Similarly, playing a guard that can’t shoot and is not quite star caliber can also be difficult. Alone, it’s not necessarily a problem. For example, lineups with Tony Allen can be excellent. However, playing this type of player limits options to round out the lineup. It would be hard to play Tony Allen with another guard that can’t shoot.

Try to construct a lineup with both a big that can’t defend and a guard that can’t shoot (where neither player is a star), and you’ve got a recipe for a disaster.

At the 2015 trade deadline, the Milwaukee Bucks were 30-23, heading for the playoffs and developing an intriguing core of young players. At that deadline, the team moved Brandon Knight and acquired Michael Carter-Williams.

Knight could shoot. MCW cannot.

There were concerns beyond the on-the-court performance that led to this move. Knight was heading for free agency, and Milwaukee had decided they weren’t going to pay him the salary he would command.

For our purposes, it’s important to note that the team opted for a guard that can’t space the floor, and MCW is not a star (at least not yet).

Milwaukee followed up the MCW addition with a big free agency contract for Greg Monroe. Monroe is a gifted offensive big. He has great touch around the hoop and can dominate smaller lineups. However, Monroe is not a great defender (nor a floor spacer).

It is incredibly difficult to construct a solid lineup with MCW and Monroe on the floor. Remember that we want three 3-point threats and one rim protector. We’ve got three spots in order to complete that task.

First, we can’t include promising young forward Giannis Antetokounmpo. He’s not a floor spacer.

It’s difficult to even play Jabari Parker, the team’s first round pick (2^nd overall) from the 2014 draft. Parker profiles as a player that can develop a consistent 3-point shot, but he hasn’t demonstrated this yet. (He’s 0 for 8 from 3 this season.)

The troubles continue. Milwaukee needs one of the players in the lineup with MCW and Monroe to be both a rim protector and a floor spacer. If they had Serge Ibaka or Draymond Green, that would be fine. But they don’t have that type of player.

Milwaukee needs a PG and C that properly fit with their young core of Middleton, Parker and Giannis. That’s not MCW and Monroe.

This year, the Bucks have taken a step back. They are 23-33, and Coach Jason Kidd has taken both Monroe and MCW out of the starting lineup. Those are the ramifications of misunderstanding lineup construction.

Milwaukee isn’t the only team that appears confused. Orlando has invested in the pairing of Elfrid Payton and Nikola Vucevic. Like Monroe, Vucevic is a gifted offensive player (inside the arc), but limited defensively. Payton’s jump shot is improving, but he isn’t good enough yet to command the attention required to stretch the defense.

It’s near impossible to find an Orlando lineup with Payton and Vucevic that the analytics project will be successful. The combination of the two puts an amazing strain on the forwards to be capable on the perimeter offensively and on the interior defensively.

Remarkably, the pair of Tobias Harris and Channing Frye pulled it off. The combination of Vucevic, Payton, Harris and Frye outscored opponents by 9 points per 48 minutes before Harris and Frye were traded. At that same time, Vucevic and Payton without the versatile forward pair was outscored by 11 points per 48 minutes.

This will sound strange, but I’m not totally against playing Payton with Vucevic. The Magic have a young player in Aaron Gordon that might be the key to making such a lineup work. Gordon has the athleticism to be a great rim protector. If he can develop a consistent perimeter shot, he can be an Ibaka-like contributor. Other young players on the team (like Oladipo, Fournier and Hezonja) are floor spacers.

Orlando has the pieces that could eventually work. Gordon could develop into rim protector and floor spacer they’d need. Payton could also improve his jump shot. Heck, Vucevic can still improve his interior defense. It’s young group with lots of potential.

I’m fine with playing a Vucevic and Payton lineup if the objective is player development. The lineup might eventually work, but it’s not there yet. What’s surprising is that Orlando’s lack of production with this group this year seems to have disappointed the organization.

If the primary objective was to compete this season, then Orlando needs some lessons in lineup construction.

In 2015, Orlando fielded a Strong lineup (as defined in the lineup study above) only 3% of the time, but when they did, they were very effective. They outscored opponents by 6.9 points per 48 minutes.

An example of a Strong lineup would be Fournier, Frye, Harris, Oladipo and O’Quinn. Our analysis of the proper construction of lineups suggests that this should be one of Orlando’s best lineups. It was. It had the best net rating (offensive rating minus defensive rating) of any lineup that played at least 80 minutes. The group’s offensive rating of 116.5 and defensive rating of 100.7 were both outstanding. A quick calculation shows that this group outscored opponents by 15.8 points per 100 possessions. The opponents outscored every other Orlando lineup with at least 80 minutes.

If Orlando wanted to compete for a playoff spot this year, the pieces were there.

Final Thoughts

Dennis Rodman was excellent at certain elements of the game, but lousy in others. A team of all Rodmans wasn’t going to win a championship, but in the right lineup, Rodman was a tremendous asset.

Basketball is a fast-paced game where all five players must work in unison for the team to succeed. Every player impacts the performance of the others on the court.

Piecing together a lineup is a challenging puzzle where the difference between something that works and a total disaster can be subtle. We hope that our work sheds some light on the problem.

Lineup construction provides a natural metaphor for the use of analytics in general. An organization needs collaboration from all of its members to build and support a competitive roster. Decision-makers must take information from a variety of sources. Ideally, that information would be from sources that complement each other through taking distinct perspectives.

Analytics is a unique perspective on the game. It’s the Rodman for your lineup. A front office shouldn’t be composed of all “analytics Ph.D.’s.” That won’t work. However, in an appropriate collaboration with “basketball Ph.D.’s,” analytics is a tremendous asset.