Just ask the 2012–13 Lakers (Kobe, Nash, Dwight, Gasol), the 2021–22 Nets (Durant, Irving, Harden), or more recently, the Suns (Durant, Booker, Beal). Despite headline-making rosters, none of those had success. They had the star power, but not necessarily the structure.

The truth is, winning consistently in the NBA isn’t just about who’s on the roster, but how a team performs in the margins: possessions, efficiency, discipline, and second chances. The below radar charts compare some star-heavy teams to league averages, and by the end of this article, those charts should make perfect sense.

We explore what actually drives winning through a data-driven lens and specifically the Four Factors of Basketball Success. We introduce the framework, test its validity in today’s game, and then use it to analyze a real-world challenge: the Los Angeles Lakers’ current center position problem.

Despite a late-season surge and a playoff appearance, the Lakers were exposed in the first round, revealing fundamental weaknesses that current star trio could not cover up. We show how the Four Factors can help identify these gaps, and how they can inform smarter personnel decisions moving forward.

Why Use the Four Factors?

Originally introduced by Dean Oliver in Basketball on Paper, the Four Factors provide a statistical framework that explains how teams win games. These aren’t exotic metrics but rather basketball fundamentals, quantified:

• eFG% (Effective Field Goal Percentage): Measures shooting efficiency, giving extra weight to more valuable three-pointers.

• TOV% (Turnover Percentage): How often a team loses possession.

• OREB% (Offensive Rebounding Percentage): Ability to generate second-chance opportunities.

• FTR (Free Throw Rate): How often a team gets to the line relative to shot attempts.

We use this model because it's simple, transparent, and most importantly, powerful. But does it still hold up in today’s faster, spacing-focused NBA?

Do the Four Factors Still Predict Winning?

While the four factors where introduced in the early 2000s, the game has evolved dramatically since then. We now live in the 3-point and spacing era, where offensive systems emphasize perimeter shooting and faster pace.

To test the continued relevance of the Four Factors, we ran an Ordinary Least Squares (OLS) regression using 15 seasons of team-level data (2010–11 to 2024–25), modeling winning percentage (W%) as a function of the Four Factors.

Despite being developed over two decades ago, the framework still holds strong in today’s NBA. Our regression shows that a simple linear combination of these four metrics explains 37.4% of the variance in team winning percentage, a strong result given the complexity of the sport.

Updated Factor Importance: A Modern Weighting System

Dean Oliver’s original model assigned fixed weights:

• 40% to eFG%

• 25% to TOV%

• 20% to OREB%

• 15% to FTR

To reflect the modern game, we recalculated these weights by normalizing the coefficients from our regression model over the past 15 seasons.

The comparison reveals both continuity and change in what drives winning. eFG% remains the most important factor, with nearly identical weight in Oliver’s model (40%) and ours (c.37%). The most significant shift is in TOV%, which now accounts for about 32%, up from 25%.

While points off turnovers have only modestly increased since the early 2000s, the sharp rise in fast break points (usually stemming from live-ball turnovers) suggests that giveaways are now punished more efficiently. With the game being faster, as reflected in the steady rise in fast break scoring, often resulting in high expected value shots, taking care of the ball has become more important than ever.

Meanwhile, FTA Rate has remained relatively stable, and OREB% has declined significantly, from 20% to just under 15%. This drop implies a strategic shift: teams now prioritize getting back on defense over crashing the offensive glass, trading off transition defense and efficient shot selection over second-chance opportunities.

The Four Factors framework doesn’t just explain offense, it also applies seamlessly to the defensive side of the ball, using opponent-based metrics to capture how well a team limits efficiency, controls possessions, and avoids costly mistakes. And just like on offense, these defensive factors show strong predictive power for winning.

Our regression analysis using defensive Four Factors yields an R² of 36.7%, nearly identical to the offensive model (37.4%). This reinforces the idea that winning is a balanced equation, both scoring efficiently and disrupting the opponent’s ability to do so matter nearly equally. But the real power of the Four Factors emerges when they’re combined into a single model. The combined model achieves an R² of 92.4%, a remarkable jump in explanatory power. This shows that offense and defense are not just separate levers of success, but complementary pieces of a unified system that, together, offer a near-complete explanation of team performance.

To help bring this analysis to life, we’ve built an interactive dashboard that includes all the data used so far. You can explore the Four Factors and underlying stats for every NBA team/player across seasons, compare performance, and even dive into your favorite team’s trends. Click on the image below to access the dashboard and run your own analysis.

The Four Factors in Action: Solving the Lakers’ Center Problem

After the blockbuster Doncic–Davis trade reshaped the Lakers' roster midseason (and a failed medical voided their deadline deal for Hornet's Mark Williams) questions around the center position surfaced. Yet a late-season surge and a third-place finish in the West temporarily masked those concerns.

The playoffs, however, delivered a harsh reality check. Los Angeles was eliminated in just five games by the Timberwolves, exposing the roster’s underlying structural issues, particularly at the center spot.

Our Four Factors analysis confirms the story. Offensively and defensively, the Lakers remain highly efficient shooters (56% eFG% vs. 54% allowed) and excel at getting to the line (FTR of 27% vs. 23% for opponents). But they struggled throughout the season with Turnover Percentage (TOV%) and mainly Offensive Rebounding Percentage (OREB%), where they rank 27th in the league (and 21st with regards to OPP OREB%).

These weaknesses, while apparent in the regular season, became especially costly in the playoffs. The Lakers allowed the most opponent offensive rebounds (13.0 per game) and consequently gave up a league-high 18.8 second-chance points per game. They also ranked top three in points conceded off turnovers and fast breaks.

If the Lakers are serious about contending with a core of Doncic, LeBron, and Reaves, they’ll need to address this glaring hole.

Filling the Gap: A Four Factors-Based Assessment of UFA Centers

After reading Evan Massey’s 4 Realistic Center Options for the Lakers in 2025, we decided to apply our data-driven framework to evaluate the potential impact of each candidate. Using our Four Factors model, we assessed how each unrestricted free agent (UFA) big man would influence the Lakers’ performance, focusing on improving key weaknesses in offensive rebounding (OREB%) and turnover rate (TOV%), without sacrificing established strengths in shooting efficiency (eFG%) and free throw rate (FTR).

The methodology is simple, yet analytically powerful. We computed team-level Four Factors using the combined 2024–25 stats of the 11 Lakers currently under contract for next season according to Spotrac.

We then simulated the effect of adding each UFA center (Brook Lopez, Al Horford, Clint Capela, and Myles Turner) using their season totals, focusing on offensive contributions due to limited availability of opponent-specific player data, and holding opponent defensive rebounding constant based on the Lakers’ actual 2024–25 metric. While this approach doesn’t account for contracts, playing time, defensive schemes, or system fit, it offers a theoretical lens to understand marginal gains and trade-offs.

The results are telling, and based on the chart below, all four players provide marginal improvements across some metrics, but with varying strengths. Brook Lopez offers the highest shooting efficiency (eFG%) while modestly enhancing rebounding and turnover rates. Al Horford contributes slightly across the board but decreases the team’s free throw rate (FTR), one of their existing strengths. Clint Capela delivers the most significant boost in offensive rebounding (30.1%), the Lakers’ primary weakness, while maintaining balance in other areas. Myles Turner shows the most well-rounded profile with strong gains in eFG% and FTR, but offers less impact on offensive rebounding and a slight increase in turnovers.

Ultimately, while this is not a forecast or cap-informed model, it helps prioritize fit over name value, giving front offices a simple decision-support layer: plug the right holes without creating new ones.

Recommendation: Capela Checks the Boxes

Among the four UFA center options, Clint Capela emerges as the most logical fit for the Lakers in our opinion. He delivers the most significant boost to the team’s biggest weakness, offensive rebounding, pushing OREB% to 30.1%. He also slightly improves turnover percentage and maintains stability in shooting efficiency and free throw rate.

While others like Myles Turner offer more shooting range and all-around talent, his market value is likely to be high coming off a strong season with the Pacers, and his next contract could create cap pressure. Lopez and Horford bring valuable traits like floor spacing (both shooting around 37% from three this season) and playoffs experience, but at ages 37 and 39 respectively, they represent very short-term fixes. While their age may make them more affordable than in previous free agency periods, the impact of the aging curve can’t be ignored when assessing their long-term value to a contending roster.

In contrast, Capela is younger (will turn 31 in a few days), on a similar past contract to Lopez, and could be more affordable given his current team context. For a Lakers squad already rich in perimeter shooting (with Doncic, Reaves, James, Hachimura, Finney-Smith, Kleber, etc..), a traditional center like Capela who thrives on interior play and second-chance creation, while being a solid rim protector, could be the missing piece. He fills the need without creating new ones, and in a model grounded in the Four Factors, that’s the kind of efficient upgrade championship teams want to make.

In summary, winning in today’s NBA is about more than assembling stars, it’s about mastering the margins. Through the lens of the Four Factors, we’ve shown that shooting efficiency (eFG%), ball security (TOV%), offensive rebounding (OREB%), and free throw rate (FTR) still explain most of what drives success, but their relative weights have shifted in the modern era. Our model confirms their predictive power both offensively and defensively, and when combined, they provide a near-complete picture of team performance (R² = 92.4%).

Applied to the Lakers’ roster, the analysis highlights clear weaknesses in offensive rebounding and turnovers, both exploited in the playoffs. Among four free agent center options, Clint Capela stands out as the best fit according to us: he significantly boosts offensive rebounding while preserving the team’s existing strengths in eFG% and FTR.

The takeaway: identify what matters, measure it, and fill gaps without creating new ones. That’s how good teams become great.

Sources: NBA Stats, Basketball Reference, Spotrac.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

To answer these questions, we analyzed the last 10 seasons of EuroLeague data, classifying players based on their actual contributions rather than predefined positions. Then, we examined how different team constructions influence winning percentage (W%), looking for trends that could redefine how rosters should be built.

Identifying New Positions Based on Clustered Player Tendencies

Before diving into player clustering, we first need to ensure a level playing field. Basketball statistics exist on vastly different scales; a player's height (often over 190cm) and a shooting percentage (typically between 20-50%) don’t carry the same magnitude. Without proper adjustment, larger values would overshadow smaller but equally crucial stats. To prevent this, we normalized the dataset, ensuring that each metric contributed meaningfully to the clustering process.

With the data prepared, the next challenge was defining how many clusters would best separate players into distinct yet meaningful groups. We turned to the Elbow Method, a technique that identifies the point where adding more clusters no longer significantly improves differentiation. To further validate the groupings, we used the Silhouette Score, which measures how well-separated and distinct the clusters are, essentially confirming whether players within the same group truly belong together.

After testing different configurations, seven clusters emerged as the sweet spot, providing enough differentiation to highlight key playing styles while avoiding unnecessary complexity.

From there, we applied K-Means clustering, a machine learning algorithm that assigns each player to one of these groups based on their statistical profile. This step allowed us to group players not by their traditional positions, but by how they actually contribute on the court.

To make sense of these clusters visually, we used Principal Component Analysis (PCA) to reduce our dataset into two dimensions. With over 20 statistical features in play, reducing the data while preserving its most important differences helps reveal clear player groupings. On a 2D scatter plot, these clusters appear as distinct formations, offering an intuitive look at how different players align with one another based on their skill sets.

Once the clusters were established, the real insight began. By analyzing the defining characteristics of each group, we identified the top five and bottom five statistical features that set them apart (see example below). This step helps understand what makes each cluster unique, whether it's elite shooting, rebounding dominance, defensive versatility, or playmaking prowess. What emerges is a new way of classifying players, one that reflects the modern game far better than traditional position labels ever could.

And with that foundation in place, let's break down the newly formed groups…

As we all know, basketball has long been defined by rigid positional labels (point guard, shooting guard, small forward, power forward, and center). But in today’s game, those designations often fail to capture the essence of what a player actually brings to the court. The above skill-based approach to classifying players opens up a far more interesting way to analyze the game.

A playmaking, high-assist guard might belong to an entirely different group than a high volume-scoring guard, even though both historically play the same position. And while centers tend to show less differentiation (most falling into clusters 2 or 7) this approach still allows for distinctions between big men based on their offensive and defensive production. Instead of asking whether a player is a “good shooting guard,” the question shifts to whether they are an elite off-ball shooter, a defensive disruptor, or a secondary ball-handler. By rethinking how we categorize players, we move closer to a data-driven model that reflects the evolution of the sport, where versatility is king, and traditional labels are more misleading than they are helpful.

Analyzing Roster Construction In The EuroLeague

EuroLeague rosters in 2024 show a fascinating contrast in team-building approaches. Some teams have embraced a diversified player profile, blending different skill sets across multiple clusters, while others rely heavily on one or two archetypes. Take EA7, for instance, they have a noticeable emphasis on playmakers and traditional bigs but lack representation from Cluster 7 (scoring big). It is no surprise the team ranks 14 out of 18 in two-pointers attempted per game. Zalgiris (ZAL) follows a similar trend, leaning into versatile wings and bigs while steering away from playmakers and scorers (clusters 4 & 5). The team is last in points scored per game. On the other hand, Maccabi (MTA) has a much more balanced spread across clusters, with an emphasis on Cluster 7 players. Logically, only Barcelona (BAR) makes more two-pointers that MTA on average per game.

Meanwhile, Real Madrid (RMB) and AS Monaco (ASM) have built their rosters without a true floor general, instead relying on high-usage shot creators and high-volume three-point shooters to drive their offenses. Their approach prioritizes offensive firepower and spacing over traditional playmaking and interior dominance. Players like Facundo Campazzo and Mario Hezonja for Madrid, and Mike James and Elie Okobo for Monaco, exemplify this philosophy, thriving as primary scorers rather than traditional facilitators.

Partizan (PAR) and Fenerbahçe (FBB) exhibit an unusually high concentration of two-way forwards, lacking diversification across other player profiles. While this could suggest over-reliance on a specific playstyle, Fenerbahçe’s strong position in the standings challenges the idea that balance is always necessary for success.

This contrast raises a key question: does roster diversification actually correlate with better team performance? To answer this, we’ll analyze the last 10 EuroLeague seasons, using regression models to uncover whether certain roster structures consistently lead to success, or if adaptability and execution matter more than positional variety.

Does Roster Construction Shape Success?

To determine whether roster composition impacts team success, we built a dataset spanning the last 10 EuroLeague seasons. Using statistical clustering from the previous section, we classified players into positions for each season, then grouped teams based on their roster makeup and linked it to their winning percentage (W%) for that year.

The heatmap highlights a clear shift in roster composition over the past decade, reflecting the evolution of modern basketball. Two-way forwards have become the most dominant position, emphasizing versatility on both ends of the floor. Meanwhile, traditional floor generals are declining, replaced by scoring guards who create their own shots rather than feeding the post. The fall of scoring bigs further reinforces this trend, as teams prioritize spacing and perimeter play over low-post scoring. Instead, bigs are now valued more for their rebounding and rim protection than offensive output.

Obviously, not all teams construct their rosters the same way. To quantify these differences, we developed two key metrics. The Unique Clusters metric captures how many distinct player positions a team has in a given season, revealing its level of roster diversity. The Cluster Balance Score takes it a step further, measuring whether a team’s roster is evenly spread across multiple roles or built around a few dominant clusters. A high balance score suggests the team is structured around key roles with complementary pieces, while a lower score indicates a more evenly distributed lineup. Together, these metrics help us understand how teams shape their rosters and whether diversity and/or specialization leads to greater success.

We ran a linear regression testing how the number of unique player positions and the balance of those clusters within a team correlate with winning percentage.

While the R-squared suggests that roster diversity alone doesn’t explain all of a team's success (which is expected given the complexity of basketball), both variables show statistically significant relationships with winning:

Individually, Unique Clusters is the stronger predictor of success, with a statistically significant relationship (p = 0.001) when analyzed on its own. However, when both metrics are included in the model together, Cluster Balance Score also becomes significant, suggesting that roster construction is a balancing act. Teams that maximize diversity and structure their lineups around a few key positions tend to perform better.

Interestingly, when only Unique Clusters is included in the model, its impact on W% drops from 8.1% to 5% (increased in win percentage per additional cluster), implying that part of its effect is shared with roster balance.

This reinforces the idea that while diversity in player positions is crucial, a completely even distribution across clusters is not necessarily optimal. Teams still need to lean into specific strengths rather than spreading talent too thinly.

What Does the Ideal Roster Look Like?

We’ve established that diversity in player positions matters, but the next question is just as important: What is the best way to build a winning roster?

Rather than assuming an ideal team structure, we let the data do the talking. We began by comparing the top and bottom 25% teams in terms of winning percentage to see if a clear trend emerged in roster construction.

Surprisingly, successful teams don’t necessarily lean more on dominant ball scorers or scoring bigs. Instead, they show a clear preference for perimeter-oriented players, with higher proportions of Cluster 1 (Perimeter Threats) and Cluster 6 (High-Volume Three-Point Shooters). This aligns with the modern era of basketball, where spacing and outside shooting are paramount. Notably, there’s also a significantly lower reliance on classic floor generals, reinforcing the shift away from traditional playmaking toward a more fluid, three-point-driven offensive approach.

With 176 unique roster compositions out of 185 team-seasons in our dataset, it became clear that nearly every team constructs its lineup in a distinct way. But are there patterns among the most successful teams? To push our analysis further, we clustered these roster compositions into 6 groups, to identify which ones correlated with the highest winning percentages. By grouping similar roster structures, we aimed to uncover whether certain lineup formulas have consistently driven success over the past decade.

The results confirms once again that not all roster compositions are created equal, with some structures consistently lead to higher winning percentages. The most successful roster cluster stands out with a clear edge, reinforcing the idea that certain team constructions are better suited for success. However, to truly understand why these clusters perform better, we need to break down their composition. Next, we’ll dive into what defines each cluster and which player positions are driving success.

The highest-performing roster cluster stands out as a unique case, featuring an unusually high concentration of High-Usage Scorers (Cluster 5), High-Volume Three-Point Shooters (Cluster 6), and Scoring & Rebounding Bigs (Cluster 7). This mix suggests that while high-scoring players play a critical role, they are most effective when complemented by strong outside shooting.

However, the heatmap makes it clear that Player Cluster 5 alone does not guarantee success. After the top-performing roster cluster, Roster Cluster 4 has the next highest concentration of High-Usage Scorers (1.73 Cluster 5 players on average). Yet, it is among the least successful roster compositions, largely due to its lack of perimeter shooting threats, both Perimeter Threats (Cluster 1) and High-Volume Three-Point Shooters (Cluster 6) are underrepresented. A similar pattern appears in Roster Cluster 2, which also struggles despite a high number of scoring options.

Another key takeaway is Roster Cluster 3, which is structurally very different from the others. Despite featuring a solid presence of Two-Way Forwards (Cluster 3)—offering defensive versatility, and a decent balance of Scoring Bigs (Cluster 7) and High-Volume Three-Point Shooters (Cluster 6), the key weakness is in the backcourt. Nearly all guards in this cluster are Floor Generals (Cluster 4), with an average of less than 0.35 High-Usage Scorers (Cluster 5). This once again suggests that this traditional playmaking-heavy structure might no longer be effective in modern basketball, as teams are shifting away from pass-first guards and towards dynamic scorers who can create for themselves.

In summary, our analysis confirms that roster construction is a significant driver of success in the EuroLeague, but not in the way traditional basketball wisdom suggests. By applying clustering techniques to classify players based on skills rather than positions, we uncovered seven distinct positions that better reflect the modern game. Regression analysis further revealed that while roster diversity is crucial, balance is not the key. Successful teams concentrate talent in key roles rather than evenly distributing player types.

The most effective rosters prioritize perimeter threats and high-volume three-point shooters over traditional playmakers and post-heavy offenses. Simply stacking elite scorers doesn’t guarantee success; instead, winning teams blend offensive firepower with floor spacing and defensive versatility. Our historical clustering of rosters over the past decade identified clear formulas for success while exposing outdated team structures that struggle in today’s pace-and-space era.

The takeaway for EuroLeague front offices? The future of roster construction isn’t just about acquiring talent, it’s about data-driven precision, structuring teams around the right mix of skills rather than outdated positional labels.

Sources: EuroLeague Stats, Eurobasket

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

Alongside the above, defensive shifts were restricted to ensure a more traditional setup, requiring four infielders to be on the dirt with two on each side of second base. This rule aimed to counteract the decline in the offensive metrics like batting average.

In this article, we will delve into the effects of these rule changes by analyzing key trends before and after their implementation. More than 4,300 matches have been played since the rule changes, and we will explore whether the decline in offensive metrics has been reversed, whether the anticipated rise in stolen base attempts and success rates has materialized, and whether shorter games, combined with the above, have indeed led to increased fan engagement.

Defensive Shifts Ban

“The 2010s will be known as the decade when these defensive shifts exploded in baseball” Matt Kelly for MLB.com.

Defensive shifts in baseball involve repositioning fielders, often stacking them on one side of the field, to better defend against a batter's hitting tendencies. From 2016 to 2022, it became a dominant strategy in MLB, dramatically altering how the game is played.

Analysis of Defensive Shifts Trends (2016-2024)

In 2016, shifts were used in just 17.1% of plate appearances, but by 2022, this figure had more than doubled to 37.5%. Left-handed hitters (LHH) were particularly affected, with shift usage against them soaring from 30.3% in 2016 to 61.8% in 2022, while right-handed hitters (RHH) saw a more modest increase.

In the following sections, we will analyze the effects of shifts on individual players’ performance by comparing wOBA against shifts and non-shifts. Additionally, we will explore the relationship between the league-wide increase in shift percentages and key offensive metrics, aiming to understand why the league ultimately viewed the continued proliferation of shifts as a potential threat to the game's appeal.

Effects of shifts on individual players’ performance

The below line plot illustrates the average wOBA (a version of on-base percentage that accounts for how a player reached base) for both LHH and RHH under shifted and non-shifted defenses over the period from 2016 to 2022.

The analysis of wOBA trends reveals that defensive shifts consistently lowered the offensive performance. For LHH, wOBA under shifted conditions was generally lower than in non-shifted situations, especially from 2016 to 2019, indicating the effectiveness of shifts during this period. Over time, particularly from 2020 onwards, the gap between wOBA with and without shifts narrowed for both LHH and RHH, suggesting that hitters may have adapted to the frequent use of shifts. Despite this, the more significant difference for LHH highlights why (alongside their tendency to pull the ball to the right side) they were more frequently targeted by shifts.

Effects on League-Wide Offensive Metrics before and after 2023

Here, we're examining the trend of major league-wide offensive metrics alongside the rise in defensive shifts and evaluating the impact of the 2023 shift ban on these metrics.

The trends in wOBA, BA, OBP, and SLG from 2016 to 2024 show a clear decline, aligning with the rise in defensive shifts and the analysis in prior sections. Although there was a spike in all metrics in 2023 following the shift ban, by 2024, these metrics had dropped back to almost pre-ban levels. This suggests that while the shift ban initially boosted offense, its long-term impact may be limited, as other factors or adjustments by teams likely offset the intended effects.

Pickoff Limits and Wider Bases

In addition to the defensive shifts ban, the MLB introduced in 2023 limits on pickoff attempts and larger bases to boost stolen base attempts. These changes aimed to make it harder for pitchers to control runners and increase the likelihood of successful steals, creating a faster, more dynamic game that enhances both strategy and fan excitement.

Analysis of Stolen Base Trends (2016-2024)

This time, trends in stolen bases and stolen base attempts per game from 2016 to 2024 clearly reflect the impact of these rule changes. Prior to 2023, both stolen bases and attempts were on a downward trend, indicating a decline in aggressive baserunning strategies. However, with the introduction of the new rules in 2023, there was a sharp increase in both metrics. Stolen bases per game rose significantly, and stolen base attempts per game followed a similar trajectory, highlighting a renewed focus on baserunning as a key offensive strategy.

Additionally, the stolen base success rate saw a noticeable spike following the rule changes. This suggests that the larger bases not only encouraged more attempts but also made stealing more successful. Although there is a slight dip in 2024, the overall success rate remains much higher than in the pre-rule change years, indicating that the effects of the rule changes might have a lasting impact on the game.

Pitch Timer

This rule aimed to reduce the length of games, which had steadily increased over the years, reaching over three hours per game in 2021 and 2022. The goal was to speed up the pace of play, making games more engaging and accessible for fans, both at the stadium and watching from home.

Analysis of Game Duration and Attendance Trends

To evaluate the effectiveness of the pitch timer, we first look at the trend in average game time per nine innings from 1980 to 2024. We can observe this trend using the below line chart to see how game durations have evolved, especially before and after the implementation of the pitch timer in 2023.

The introduction of the pitch timer in 2023 had a significant impact on reducing game duration, with the average time per 9 innings dropping from over 3 hours in 2021 and 2022 to around 2 hours and 36 minutes so far in 2024. This reduction brought the game closer to levels we had not seen since the early 1980s.

We thought it would be also interesting to analyze the trend in attendance, particularly looking at how attendance per game has changed over the years, in conjunction with game times.

The plot shows steady growth in MLB attendance from the 1980s to the early 2000s, followed by fluctuations that stabilized through the early 2010s. A declining trend was observed from 2015 to 2019. The sharp drop in 2020 reflects the COVID-19 pandemic, with a recovery in the following years, though attendance hasn't fully returned to previous highs. While recent rule changes aimed at enhancing the game experience coincide with this period, their direct impact on attendance is difficult to pinpoint, suggesting that broader factors continue to influence fan turnout.

The long-term trend in attendance since 1980 is influenced by population growth and baseball's expansion, making it less reflective of current dynamics. Focusing on data from 2000 onward, however, reveals a different story. The below chart shows a moderate negative correlation (-0.44) between game duration and attendance during recent years, suggesting that longer games are less appealing to modern audiences. This supports MLB's recent efforts to shorten games to boost fan engagement.

In summary, the 2023 MLB rule changes have had a mixed impact on the game. The pitch timer has successfully reduced game durations, aligning them with levels not seen since the 1980s, and this appears to be a positive step in addressing the modern fan’s preference for shorter, more engaging games. The ban on defensive shifts initially boosted offensive metrics, but these gains seem to be leveling off, indicating that teams may be finding new ways to adapt. Meanwhile, the changes to pickoff limits and base sizes have revitalized baserunning strategies, leading to an increase in both stolen base attempts and success rates. While these rule changes have certainly influenced the game, the broader impact on fan engagement, as reflected in attendance figures, remains complex and multifaceted. The continued evolution of the game will reveal whether these changes can sustain long-term interest and excitement in baseball.

Source: MLB, statscast, teamrankings, Baseball-Reference.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

To truly understand how Boston managed to triumph over such a powerhouse, we turn to a meticulous analysis of play-by-play data. The Mavericks were no ordinary adversary; they eliminated the number one seed Oklahoma City Thunder in six games and dominated one of the main favorites, the Minnesota Timberwolves, conceding only one game in the conference finals.

Each action on the court, from three-point shots to defensive rebounds, tells a story of execution and resilience. By dissecting these plays, we can quantify the impact of key players, highlight successful strategies, and pinpoint the turning points that led to Boston’s victory.

Despite a heavy loss in Game 4 (84-122), the Boston Celtics consistently kept the Dallas Mavericks under the 100-point mark in the other games, securing some tight wins like in Game 2 and Game 3. Overall, Boston scored 508 points while allowing only 496 and this defensive effort was crucial in their victory.

Strategic Shot Selection and Execution: Analyzing Action and Shot Types

Despite Boston's overall victory, the data reveals some surprising trends. Dallas made more shots and missed fewer, implying a higher shooting percentage. Additionally, Dallas secured more rebounds, likely due to Boston's higher number of missed shots.

Before delving into the types of shots taken by each team, we will compare the stats (per game) over the five games with a correlation analysis to win percentage, as performed in a previous article.

This comparison will help us understand which metrics were most strongly associated with winning in this series.

Boston averaged 101.6 points per game, slightly higher than Dallas's 99.2, with points scored having a significant positive correlation (0.38) with winning percentage. Dallas outperformed Boston in rebounds (43.2 vs. 41.2), with the latter showing a quasi-similar positive correlation (0.31). Boston led in assists (24.2 vs. 16.8), indicating that team play and ball movement were crucial.

Despite Dallas having a higher field goal percentage (45.8% vs. 43.6%) with a strong positive correlation (0.57), Boston outshot Dallas from beyond the arc (33.8% vs. 31.6%) with three-point shooting showing a strong positive correlation (0.59). Both teams averaged 11.4 turnovers per game, with turnovers negatively correlated (-0.3) with winning. Dallas’ significantly lower free throw percentage (68.1% vs. 86%) costed them crucial points (and victories) in tight games. Additionally, Boston led in steals (6.2 vs. 5.8) and blocks (5.4 vs. 2.2), contributing to their defensive efforts.

While Boston led in points and other metrics, Dallas had more rebounds and a better field goal percentage, both strongly correlated with winning, prompting a deeper look into the types of shots taken by each team to uncover further insights.

When analyzing the different types of shots using NBA play-by-play data, we can observe some drastic differences between Boston and Dallas. Boston attempted significantly more jump shots, likely catch-and-shoot shots, which ties back to their higher number of assists, suggesting more ball movement to create open shots. Boston also had more cutting layup shots, emphasizing their off-ball movement and team play. On the other hand, Dallas had a higher number of shots from drives and fadeaways, implying a greater reliance on one-on-one type of offense, isolation basketball, and less ball movement due to Boston's defense and Dallas's intrinsic style of play.

Additionally, Dallas attempted fewer alley-oop dunks than usual, a big part of their pick-and-roll offense, especially after the addition of Daniel Gafford to the squad and increased minutes from rookie Dereck Lively. This reduction suggests that Boston's defensive strategy effectively disrupted one of Dallas's key offensive plays.

As the previous shot type analysis suggested, Boston took more uncontested shots (275) compared to Dallas (255), and consequently fewer contested shots (142 vs. 164). This aligns with Boston's higher number of assists and superior ball movement, which helped create open shots. Despite this strategic advantage, Boston's field goal percentage for uncontested shots was relatively low (37.8%) compared to Dallas (44.3%). However, Boston partially offset this with a better percentage for contested shots (c.7% higher). This discrepancy suggests that, in addition to the type of shot, the location of the shot played a crucial role in why Boston managed to score more points overall despite having a lower field goal percentage.

Precision and Placement: Analyzing Shot Location

The concept of expected value in basketball refers to the average points a team can expect to score from a particular type of shot, based on historical performance data. By analyzing the shot distribution and calculating the expected value, we can assess the efficiency of shot selection.

We discovered that despite taking two fewer shots overall, Boston's shot selection generated an expected value of c.7 points more than Dallas. This equates to 1.345 points per game over the finals (five games) and partially explains why Boston averaged 2.4 more points per game.

The main difference in shot selection between the teams was Boston's greater emphasis on high-value shots, such as corner threes (c.37 more expected points) and above-the-break threes, which have higher points per shot averages (1.10-1.20 and 1.05-1.10 respectively). For reference, Boston was the team in the regular season that attempted the most 3 pointers (47% of FGA). Conversely, Dallas relied more heavily on shots in the paint and while the restricted area is highly effective, it did not yield as high an expected value per shot as Boston's strategy.

From Shot Selection to Player Efficient Offensive Production: Analyzing EOP

Efficient Offensive Production (EOP) quantifies a player's ability to maximize scoring opportunities while minimizing wasted possessions. We covered this metric extensively in part of a previous post. It highlights players who score with high precision and consistency. Comparing a team's overall EOP helps better understand the efficiency of the points produced, providing a deeper insight beyond just points or FG%.

For the Boston Celtics, Jayson Tatum and Jaylen Brown exhibited solid performances with high EOP values of 14.03 and 12.82, respectively. Jrue Holiday and Al Horford also contributed significantly to the team's overall offensive efficiency, reflected by their impressive OE values of 0.71 and 0.72 respectively. Derrick White and Kristaps Porzingis (the latter played only 3 games) also played an important role.

On the other hand, the Dallas Mavericks were logically led by Luka Doncic, who demonstrated outstanding offensive production with an EOP of 16.54, despite a lower OE of 0.49. Kyrie Irving also showed contributed offensively, with an EOP of 11.89. Players like Dereck Lively and Daniel Gafford had high OE values of 1.75 and 1.12, respectively, indicating their efficiency (high FG% shot, many offensive rebounds and few turnovers).

The heatmaps reveal that while both teams had players with high offensive efficiency, the Celtics had a broader distribution of players contributing effectively to their offense. This balanced approach, reflected in the high OE and EOP values across multiple players, underscores why Boston managed to outscore Dallas despite a lower overall field goal percentage. It is further confirmed when looking at the team total EOP. Over the series, Boston produced 7 more efficient points per game than the Mavs.

Defense Wins Championships? Analyzing Hustle Plays

In the adage "defense wins championships," the importance of hustle plays cannot be overstated. To win a championship, players must be willing to "die" on the court, exemplified by Derrick White chipping his teeth while corralling a loose ball in the Celtics' Game 5 victory over the Mavericks. Hustle plays like contested shots, deflections, loose ball recoveries, and box outs disrupt the opposition's offense and secure crucial possessions. Thanks to play-by-play data provided by the NBA, we looked at the key hustle plays during this series.

Celtics head coach Joe Mazzulla emphasized that when you study all NBA Finals games, the team that plays the hardest, is the toughest, most physical, and executes at the highest level consistently, while withstanding mistakes, comes out on top. It seems the Celtics players received this message loud and clear, as they showcased superior hustle plays. Leading in deflections (67 to 58), loose balls recovered (26 to 21), and box outs (49 to 39) compared to the Mavericks, the Celtics' relentless defensive efforts were pivotal in their success.

In the NBA Finals, Jaylen Brown stood out, leading his team in deflections and loose ball recoveries. His all-round performance, paired with a high Efficient Offensive Production (EOP), earned him the Finals MVP title. Brown's efforts were bolstered by key contributions from Jrue Holiday, Derrick White, and Al Horford, who made significant defensive plays and impacted both ends of the floor.

For the Dallas Mavericks, Kyrie Irving posted notable defensive stats, while Luka Doncic, often criticized for his defense, led his team in deflections. Despite these individual efforts, the Mavericks fell short, underscoring the importance of consistent and comprehensive defensive play in winning championships.

Jayson Tatum demonstrated his versatility by being a co-leader in box outs for the Boston Celtics, an impressive feat for a forward rather than a center. This relentless effort on the boards complemented his performance, leading his team in points, assists, and rebounds in the Finals.

In summary, the Boston Celtics clinched their 18th championship by strategically outplaying the Dallas Mavericks through superior teamwork, defensive hustle, and efficient shot selection. Higher assists and ball movement created more open shots, resulting in a higher number of uncontested attempts. Boston's emphasis on high-value shots like corner threes generated a higher expected value. Jaylen Brown's all-round performance earned him Finals MVP, complemented by significant contributions from Jayson Tatum, Jrue Holiday, and Al Horford.

Source: NBA.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

In the context of the NBA, we can think of each player as an 'asset' whose performance has both 'returns' and 'risks'. Here, the 'return' is the player’s ability to contribute to wins, quantified by a metric known as Win Shares. This statistic measures the total victories a player produces for their team through their on-court actions. The 'risk' is defined by the variability (standard deviation) of these Win Shares over multiple seasons, representing how consistent a player is.

Just as in the investment world, where the balance of assets is tailored not only to maximize returns but also to align with the investor’s risk tolerance, our application of MPT to NBA team composition includes crucial practical constraints to ensure feasibility such as the salary cap and position requirements.

Understanding Win Shares: The ‘Return’

Win Shares (WS) is a comprehensive statistic designed to quantify the total contribution of a player to their team’s success within a single number. Developed to bridge the gap between individual performance and team victory, WS attributes a player's involvement in team wins, integrating both offensive and defensive metrics. This formula is based on principles outlined in Dean Oliver’s book, “Basketball on Paper”, and was officially created by John Zubatko. The idea and name was based on the win share statistic established in baseball’s sabermetrics community.

Traditionally, the calculation of WS involves several key steps, central to which is the estimation of 'points produced' by a player. This metric is critical as it quantifies a player's offensive contribution in terms of points, which is directly linked to winning games. However, this conventional method uses basic scoring and assisting figures, potentially overlooking the nuances of modern basketball playmaking and efficiency.

Introducing Offensive Efficiency in Win Shares Calculation

In an attempt to refine the estimation of a player's offensive value, we have integrated the concept of Offensive Efficiency (OE) into the WS formula. OE enhances the traditional 'points produced' by considering the efficiency with which these points are produced. For more details, we covered OE in a previous article.

If we take a look at the above boxplots, average OEs logically suggest a pattern where positions traditionally closer to the basket (C and PF) tend to have higher OE, due to higher FG%, more offensive rebounds, and being less prone to turn the ball over.

It's become clear that a one-size-fits-all approach does not accurately reflect the varied contributions of each role. To address this, we propose an adjustment to the points produced by players, factoring in their offensive efficiency relative to the average OE for their respective positions. Specifically, we will enhance the points produced for each player by multiplying them by the below factor:

This adjustment not only accounts for the positional differences but also rewards players who exhibit higher efficiency than the norm for their position.

Win Shares

To calculate WS, we used our efficient points produced (EFF PP) and subtracted the product of 0.92, league points per possession, and a player's offensive possessions. Then, we divided this marginal offense by marginal points per win, which factors in team and league pace. We used defensive WS directly provided by the NBA, and both components are summed to derive total WS for each player.

While surprising at first glance, the presence of role players like Josh Hart, Derrick White, Josh Giddey, or Michael Porter Jr. among the top 20 in WS highlights their significant impact and underscores why their teams are among top contenders for the championship, finishing the regular season in the top three of their respective conferences.

To ensure a fair comparison among players, particularly in contexts where injuries or rest days lead to differing numbers of games played across the season, we converted WS on a per-game basis.

Upon recalculating WS (per game) using EFF PP instead of traditional points produced, an improvement in the correlation with team success was observed. As we can see in the correlation matrix above, while the classic WS method showed a correlation of 0.62 with team wins, this correlation increases to 0.7 when using EFF PP. This indicates a stronger alignment with actual team outcomes and underscores the importance of efficiency in points production.

Furthermore, a regression analysis comparing the total WS per team per season (using both methods) with a team win percentage revealed that the R-squared value increased from 0.39 to 0.50 when using EFF PP. This substantial increase validates the enhanced predictive power of our modified WS.

Risk-Return

In this next part of our analysis, we wanted to look at the relationship between risk and returns for NBA players. Imagine players as investments, where a higher return usually comes with higher risk. Ideally, you want high returns with as little risk as possible.

To assess 'risk,' we looked at the variability in performance, calculated as the standard deviation of a player's win shares. A three-year span is utilized for these calculations to smooth the impact of one season's peaks or troughs and provide a more reliable assessment of a player's stable performance.

In the context of applying MPT to NBA roster building, players with above-average returns and below-average risk are highly valuable. They contribute consistently to wins while maintaining stable performance, similar to "blue-chip" stocks in finance. Below is a list of 20 players with below average risk and above average return, sorted by the last three seasons’ average modified WS per game.

Similar to how higher-returning financial assets tend to be more expensive, especially if they also exhibit low volatility, players who provide higher returns in terms of win shares generally command higher salaries.

This correlation underscores the economic principle that high performance comes at a premium. However, consistency in performance doesn't necessarily lead to a proportional increase in salary. When performing a linear regression, we found that adding risk as a factor alongside return didn't significantly improve the model's predictive power. Risk's impact on salary wasn't statistically significant, and the R-squared value remained unchanged.

Moving into constructing our player portfolio, the insights from our risk-return analysis are crucial. They reveal opportunities to assemble a team that mirrors the principles of MPT (maximizing performance while managing risk). By identifying high ‘return’ players whose salaries don't fully reflect their ‘risk’, we can build a competitive, cost-effective team.

Building The Team

To create optimal NBA team portfolios, we used Mixed-Integer Linear Programming (MILP), treating each player as a binary decision (selected or not). The optimizer operates by examining all possible combinations of players that respect our constraints, which include staying under a $141 million salary cap, selecting exactly 15 players, and not exceeding three players per position, to ensure a realistic and balanced team composition. The optimizer then calculates the ‘total return’ or the ‘total risk’ for each combination, seeking the one that best achieves our specified objective.

We began by exploring two extreme strategies: maximizing ‘total return’ and minimizing ‘total risk’.

The above roster targeted the highest possible return, focusing on assembling a team capable of achieving the maximum modified WS per game but with an elevated risk. Conversely, the below strategy prioritized minimizing the risk associated with player performance fluctuations, aiming for a stable yet less aggressive team performance.

The risk/return value for each team, being calculated as follows:

helps us evaluate the balance between the potential rewards and the risks associated with each portfolio. A higher value indicates a more favorable balance, where the returns sufficiently compensate for the risks.

In financial theory, the efficient frontier represents the set of portfolios that provide the highest expected return for a given level of risk, or the lowest risk for a given level of return. Applying this concept to NBA team building, we explored 10 potential team compositions, each designed to be within a defined range of expected returns, from the average total return of teams that have made the playoffs over the last three seasons (benchmark for competitive success) to the maximum return team we've just analyzed.

By treating each team as a portfolio of players, we optimized each lineup to achieve the best mix of total return and risk, similar to maximizing the Sharpe ratio in investment portfolios. This method allowed us to pinpoint the team with the highest risk/return value, effectively identifying the optimal team.

The above showcases the optimal NBA team derived from our MPT model. This lineup features a well-balanced mix of players across all positions:

• Centers: Mitchell Robinson, Jarrett Allen, and Clint Capela. These players bring significant rebounding and shot-blocking capabilities, contributing heavily to the defensive backbone of their respective teams. They exhibit high efficiency around the basket, which is evident from their high average returns.

• Forwards: The team includes versatile forwards like Scottie Barnes (averaging 21.72 efficient points produced over his first 3 seasons in the NBA) and Brandon Clarke (top 10 defensive rating this season amongst players with more than 15 minutes per game), who contribute effectively on both ends of the court. At just 22 years old, Mobley brings a unique blend of size, skill, and agility that allows him to be effective on both ends of the court as well. Kevin Love adds depth, provide veteran experience, and three-point shooting.

• Guards: The backcourt is composed of players like Tyrese Maxey, who is one of the most entertaining players to watch this season, perfectly filling Embiid’s role of offensive leader at Philly this year, averaging 22.70 efficient points produced and shooting 41.13% beyond the arc over the last 3 seasons. Russell Westbrook brings veteran leadership and dynamic playmaking, while Alex Caruso and Jose Alvarado add defensive ability.

The model crafted a team that not only respects the salary cap and maintains a balanced mix of positions but also showcases a lot of synergy among the players. It goes beyond assembling consistent scorers or great defenders and provides a perfect blend of both qualities across all positions with a team built with a few stars but important role players as well.

Limitations

Analyzing past performance to build an NBA team faces similar limitations as those encountered in financial markets, where past returns are not always indicative of future ones. This approach can undervalue young players whose current statistics may not fully reflect their potential, as they have not yet reached their peak performance years yet. Conversely, it might overvalue veterans like Russell Westbrook, whose future performance may decline as they approach the end of their careers or consider retirement.

Furthermore, the model's reliance on the last three years of data excludes rookies or sophomores who might be central to a team's future success (i.e. players like Victor Wembanyama, who, despite their immense potential, have limited NBA track records). Additionally, the rookie contract structure of the NBA introduces another layer of complexity. Players such as Tyrese Maxey, who is nearing the end of his rookie contract (only $4.34 million this season), will surely be a maximum contract player this summer, significantly impacting the team's salary cap dynamics. The same goes for other youngsters like Scottie Barnes and Evan Mobley, whose current contract values do not accurately reflect their market value and impact on the court.

Future Directions

To improve the model, future directions could include adjusting for player age and potential growth, incorporating more sophisticated metrics for player valuation, and dynamically modeling salary cap implications over multiple seasons. This would allow for a more nuanced understanding of player development or decline, and contract cycles, ultimately leading to more strategic and forward-looking team compositions.

In summary, we applied Modern Portfolio Theory (MPT) to NBA team management by assessing players as assets with quantifiable returns and risks, using a modified version of Win Shares to measure contributions to team wins. By optimizing team compositions while respecting the NBA salary cap and roster requirements, exploring strategies to maximize returns and minimize risks. We employed financial concepts like the efficient frontier to identify the optimal team composition, achieving a balance between player performance, risk, and salary efficiency. This approach demonstrates the practical application of MPT in sports management, offering a strategic framework for building balanced and competitive NBA teams. This methodology is particularly useful for crafting strategies during expansion team creation or evaluating potential trades to see how replacing one player with another affects the team's risk/reward profile.

Source: NBA, Basketball Reference, Spotrac

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

The Offensive Contribution

Stephen Shea and Christopher Baker developed (in their book “Basketball Analytics”) the Efficient Offensive Production (EOP). Before we can delve into the intricacies of EOP, we must first understand Offensive Efficiency (OE). The authors stated that they “believe efficient players make for efficient teams, and efficient teams win” and we could not agree more. Their statement is later demonstrated in their book where the regression from OE to WIN% for seasons between 2003 and 2013 has an R-squared of 0.44 when it is only 0.15 for the regression from points per game to WIN%.

OE serves as a crucial intermediary metric, assessing a player's effectiveness in generating points. Calculated as the sum of a player's field goals and assists, divided by their total field goal attempts minus offensive rebounds plus assists plus turnovers, OE encapsulates the essence of offensive contribution in a single figure. We decided to enhance the original formula by including screen assists.

From the above bar chart, we can observe the average OE per position for NBA players playing more than 15 minutes per game this season. These averages logically suggest a pattern where positions traditionally closer to the basket (centers and power forwards) tend to have higher Offensive Efficiency, due to higher FG%, more offensive rebounds and screen assists, and less prone to turn the ball over. In contrast, positions that operate more on the perimeter (espacially shooting guards and small forwards) show lower OE, due to more ball handling potentially leading to a turnover, less offensive rebounds and lower FG% (perimeter shots). Hence, it is important (in our opinion) to compare a player’s OE with the average for his position and not only looking at the absolute value.

Above are the top 10 outperformers (difference between OE and position average OE) per position. And apart from some well-known stars like Giannis Antetokounmpo, Rudy Gobert, or Tyrese Haliburton, most of the names in the above scatter plot are relatively unknown to the general public. This observation is understandable, as top-tier players often shoulder a heavier offensive load, necessitating more shot attempts, including those that are heavily contested or executed under high-pressure situations like the last seconds of a possession or critical moments of a game.

This is for this reason that OE is only an intermediary metric. To get a more complete measure of the offensive production, the authors decided to incorporate points scored and the value of an assist relative to a point scored (0.76). EOP is then calculated as OE multiplied by the result of 0.76 times a player’s assists plus his amount of points scored.

When looking at EOP, names become more familiar. With Nikola Jokic leading the way (26.1 pts, 12.3 reb, 9 ast per game, shooting at a 58.1% this season). Since not all players played the same amount of game this season (injuries, suspensions…), we thought it would be interesting to look at the EOP per game and below is the 20 players with the highest EOP per game this season.

We all became used to Nikola Jokic almost averaging a triple double for the season, Luka Doncic and Shai Gilgeous-Alexander dropping constantly more than 30 points each night (SGA had 45 30-point games in the first 60 of his team, or 75%), or Lebron James having the highest EOP per game among small forwards in the NBA at 39 years old and after more than 20 seasons in this league. But we decided to plot the EOP per game distribution to really acknowledge how unique those players are.

The Defensive Contribution

The NBA's Defensive Rating (DEFRTG) is a metric designed to quantify a player's defensive effectiveness by estimating the number of points the player allowed per 100 possessions they defensively faced. In simpler terms, it measures how good a player is at stopping the opposing team from scoring. The lower the DEFRTG, the better a defender is considered, as it implies they allow fewer points.

Calculating DEFRTG involves a complex formula that accounts for various factors so to understand what drives this rating, we decided to run a correlation matrix between some defensive stats (opponent FG%, steals, blocks) and defensive hustle stats (deflections, charges drawn, constested shots).

Keep in mind, a lower DEFRTG signifies a more effective defender. Therefore, it's expected to see a significant positive correlation of the latter with the weighted opponent field goal percentage (wOFG%). What's unexpected, however, is the lack of significance of almost all other metrics, apart from contested shots, in relation to DEFRTG. Key statistics such as steals (STL) and blocks (BLK) display almost no correlation with it.

In the broader context of a player's defensive value, we believe DEFRTG must be used alongside other statistics, such as steals, blocks, and defensive rebounds, to provide a comprehensive view of a player's defensive contributions leading to his team’s success. To confirm our belief, we looked at the relationship between DEFRTG and other defensive stats (at team level) with WIN%

Although DEFRTG shows a notable inverse relationship with WIN% (indicating that lower DEFRTG typically corresponds to higher WIN%), other defensive metrics such as steals (STL), blocks (BLK), and deflections also play a role in determining a team's success.

The first linear regression, with an R-squared value of 0.29, demonstrates that DEFRTG alone offers a partial glimpse into a team's winning potential.

In contrast, when running a multiple linear regression, we were able to improve the R-squared to 0.48 however, most coefficients (except DFRTG, blocks, and contested shots) were not statistically significant (p-value > 0.05). This is due to the presence of strong multicollinearity, and the estimated regression coefficients become unstable and difficult to interpret. To help dealing with multicollinearity we ran a Lasso regression since it penalizes the size of the coefficients and potentially reduce some coefficients to zero. The model reduced most coefficients to zero hence we will assume that DEFRTG significantly captures a player's defensive impact on their team’s win percentage.

Surprinsingly, the favorite player to win Defensive Player Of the Year (DPOY), Rudy Gobert, is only 10th when looking at the DEFRTG. Victor Wembanyama is not even in the top 20.

Overall Player Value

To translate a player’s defensive prowess into a more intuitively understandable metric, we’ve reengineered the DEFRTG into a positively oriented scale, where higher numbers reflect stronger defensive performance. This rating is calculated by reversing the DEFRTG relative to its mean, so that a player’s score above the mean signifies superior defensive contribution. We’ve further refined this metric by normalizing it to have a mean of one, creating the Defensive Impact Score (DIS). Players with a DIS greater than one are considered to have above-average defensive impact, while those below one are below-average.

It was a crucial step to compute an Overall Player Value (OPV). We can multiply EOP and DIS to synthesize a singular value that encapsulates a player's total contribution. Indeed, it gives us a nuanced perspective of a player’s value, rewarding those who excel offensively and defensively. It ensures that a player's overall impact isn’t totally skewed by proficiency in just one aspect of the game but is instead a balanced reflection of their abilities on both ends of the court.

From the plot, we observe that while the top players remain relatively consistent in rank when comparing solely by EOP, incorporating the DIS adjusts the overall player valuation to reflect both offensive and defensive contributions. For instance, players like Nikola Jokic and Joel Embiid lead among centers, with their OPV reflecting their dominant two-way play.

The NBA provides us with a Net Rating (NETRTG) being simply the difference between OFFRTG and DEFRTG. On player level, this statistic is the team's point differential per 100 possessions while they are on court. So why not simply use it as the overall value of a player?

The moderate correlation between EOP per game and OFFRTG suggests that while the metrics are related, OFFRTG is only one part of a larger offensive picture. Conversely, the lower correlation between NETRTG and OPV, implies that NETRTG might not fully capture an individual player's contribution to a team's success. Specifically, it may overlook the nuanced ways that a player's efficient scoring and defensive actions contribute to winning games.

A deeper dive into the discrepancy between players' NETRTG and their OPV rankings reveals intriguing patterns. We noted that players marked in green in the below chart, who exhibit higher OPV rankings compared to their NETRTG, predominantly belong to teams with losing records. This suggests that NETRTG can be significantly swayed by the team's overall performance, which is less reflective of individual contribution as evidenced by their OPV. For instance, players like Desmond Bane and LaMelo Ball find themselves well within the top 40 for OPV, despite being ranked 245th and 267th, respectively, by NETRTG among qualified players.

On the flip side, those with negative differences are typically associated with winning franchises (yes even Jae'Sean Tate from the Houston Rockets). Payton Pritchard of the Celtics serves as a case in point: his high NETRTG benefits from the team's success. We believe Pritchard has a spot in every NBA team’s roster, he is an offensively efficient player with average defensive impact but his engagement in the game is relatively low, with an average of just under three shot attempts per game, raising questions about whether his efficiency and NETRTG would hold up with a larger role. He ranks 6th in NETRTG but only 205th in OPV. Similarly, Jose Alvarado's defensive reputation is well-known, yet his offensive production falls (very) short.

While the value of a solid role player cannot be overlooked, our analysis suggests that OPV offers a more nuanced view, distinguishing between up-and-coming talents in challenging situations, such as LaMelo Ball and Cade Cunningham, and crucial role players like Pritchard and Sam Hauser, who contribute to a title-contending team.

In summary, our analysis highlights the nuanced complexities of measuring a player’s true value. While NETRTG offers a broad view, influenced significantly by the team’s overall success, our exploration into the OPV provides a more individualized perspective. This metric affords a multifaceted look at player performance, acknowledging the essential roles of players in various contexts.

Future Direction: our next step in a future article would be to juxtapose players' roles and OPV against the backdrop of their contracts with the aim to discover potential under- and over-valued players, contributing to a more informed and strategic approach to team building and player acquisition.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

Why examine pace? Because it offers a window into the tactical evolution of the sport, showing how teams adapt to leverage speed for competitive advantage. From deliberate, slow-paced offenses to rapid, transition-focused attacks, the tempo of the game reveals much about the changing nature of how the game of basketball is played.

This article embarks on a journey to unpack the evolution of pace, its implications on the game, and its relationship with team success. Through the analysis of historical data from InPredictable, Basketball Reference, and the NBA, we aim to uncover the reasons behind the strategic shift towards quicker possessions.

Evolution of Pace and Possession Dynamics

Since the 1996-97 season, there's been a clear shift towards a speedier style of play, as evidenced by the gradual decrease in the average time teams hold onto the ball before attempting to score. The line chart below detailing the trend of seconds per possession from 1996 to 2024, paints a picture of transformation. The early years exhibited a more deliberate and measured approach, with possessions often eating deep into the shot clock. However, as the millennium turned, a new philosophy began to take hold. The data shows a steady decline in possession length, suggesting an eagerness to push the ball up the court and capitalize on scoring opportunities with increased urgency.

Complementing this narrative is a scatter plot revealing a robust inverse relationship between the pace of play and the average number of possessions per game. As teams slashed seconds off their possession times, the number of possessions naturally increased. The correlation is unmistakable, quicker possessions fuel a higher volume of scoring attempts, a strategy modern teams seem to be embracing wholeheartedly.

Not all possessions are created equal, and the distribution of possession types over time has also undergone significant shifts. Our stacked bar chart captures the changing proportions of possessions following made shots, defensive rebounds, and turnovers. The portion of possessions after a turnover reduced significantly from c.11.50% in 1996 to less than 8.40% in the current season. Indeed, the estimate of turnovers per 100 plays decreased from 14.8 to 12.2 over the period. While defensive rebounds has remained a steady source of possession initiation (c.34%), there's a discernible uptick in the fraction of possessions following made shots (from 52.44% to 55.43%).

Diving deeper, the trend line chart for seconds per possession type crystallizes the narrative. While all possession types have seen a reduction in duration, it's the possessions following turnovers that have seen the most dramatic acceleration. The data points to a strategic impetus on capitalizing quickly on opponents' errors, turning defense into offense far more quickly.

Shooting Efficiency and Scoring

Next, we decided to look at the impact of pace on the number of shots attempted, the efficiency of those shots and ultimatly the number of points scored per game.

The Paradox of Pace and Precision

The below scatter plot with a regression line shows an expected negative correlation between pace (seconds per possossion) and shots attempted (FG + 3P) per game. A faster pace, which lead to more shots due to shorter possession times, shows an increase in the number of shots attempted.

The below visual is a striking scatter plot that charts the relationship between the game's pace and field goal percentage (FG%). Contrary to initial expectations, as the pace increases, the FG% shows resilience. Instead of declining, the shooting accuracy remains robust or even improves slightly, suggesting that players are adapting and their shooting skills are sharpening amidst the hastened hustle.

We can conclude from the above that, as teams accelerate their gameplay, opting for shorter possessions and higher volume of shots attempted, whithout impacting negatively their FG%, it is no surprise that shorter seconds per possession yields more points per game.

However, while the pace acts as the maestro, it would be ignorant to park the impact of 3-point revolution in recent years on the side, which might conduct a more significant sway than pace itself. As the three-pointer has become a staple in modern playbooks, its potency might overshadow the impact of pace on scoring trends.

As we prepare to delve deeper in our analysis, we will try to isolate the pace factor. Ranking basketball teams per season based on seconds per possession allows for a consistent comparison across different eras, sidestepping the distortions caused by the 3-point shooting trend. This approach offers a season-contained view that highlights a team's pace strategy relative to its contemporaries, providing insight into the correlation between a faster pace of play and team success metrics such as win-loss records, scoring efficiency, and plus-minus statistics.

Pace and Team Performance

Before delving into the intricacies of pace analysis, it is crucial, using data spanning from the 1996-1997 season, to identify the elements that enhance or detract from a team's winning percentage (WIN%). Our correlation matrix highlights that a team's WIN% is strongly influenced by their plus-minus (naturally) and offensive output, notably points scored (PTS) and field goal percentage (FG% & 3P%). The data also emphasizes the significance of defensive prowess, with defensive rebounds (DREB) showing a notable positive correlation (less second chance opportunites given, more opponent shots missed). Conversely, turnovers (TOV) are negatively linked to winning, reinforcing the importance of possession control. These metrics collectively paint a comprehensive picture of the dynamics that dictate a team's success on the court.

In this step of the analysis, we are ranking NBA teams within each season according to their performance in various key statistical categories that have been shown to correlate with winning percentage. By assigning ranks, we establish a clear hierarchy within the data, where a rank of 1 represents the best performance in a given metric. This methodology allows us to compare teams' pace (how fast they play) with their rank in these key metrics. The goal is to investigate whether teams that play at a faster pace tend to have higher (better) ranks in these critical areas that contribute to winning games.

The heatmap analysis reveals that in the NBA, a team's pace of play has a nuanced relationship with other performance metrics. There's a positive correlation between pace and scoring suggesting, as explored earlier, that faster play can lead to more points. Shooting efficiency shows only a weak link with pace, implying that quick play does not necessarily impact shooting percentages. There's a slight positive association with defensive rebounds, yet a negative one with turnovers, hinting at the trade-off between a swift offense and ball security. Overall, pace has a negligible effect on a team's point differential.

To confirm, the scatterplot above illustrates the relationship between Pace Rank and WIN%. The regression line, almost horizontal with a formula WIN% = 0.51 - 0.00*Pace Rank, indicates a negligible slope, suggesting that changes in Pace Rank have no effect on WIN%. The dense clustering of data points across all ranks of pace further supports the conclusion that there is no clear trend or correlation. Thus, we can conclude that the pace of play, as per this analysis, does not have a direct impact on a team's likelihood of winning. For example, the reigning NBA champions (Denver Nuggets) currently rank 29th in terms of seconds per possession (15.0 sec), but exhibit a WIN% above 0.670 after 43 games so far. Being the fastest playing team in a given season and posting the best WIN% only occured 5 times over the period and does not guarantee success.

In summary, the evolution of pace in professional basketball, particularly in the NBA, appears less a direct catalyst for success and more a reflection of a broader stylistic evolution within the game. According to a stacker.com article, the hand-check penalty implementation in 2004 dramatically sped up the game, diminished the size advantage of the biggest players, and ushered in the modern era, where smaller, faster guards rule the roost. Consequently, the game has become faster, not necessarily because teams believe speed alone leads to victory, but because the modern player's skill set and the strategic emphasis on 3-point shooting have reshaped the game's tempo. The rapid increase in 3-point attempts contributes to more rebounds, more possessions, and thus a quicker pace of play. While the raw pace of a team's play is not a standalone indicator of success, it is an outcome of the strategic shift towards a more open, dynamic, and perimeter-oriented game. The data shows that teams have adapted to these changes, and while pace itself is not the cause of victory, it is a byproduct of the game's continuous evolution.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

Our journey through the data reveals the proportion of games that feature these critical moments, offering a glimpse into how common—yet special—such situations are in the broader context of the NBA regular and post-season. By dissecting the dynamics of home and away team performances, we delve into the age-old question of home-court advantage and its impact on the likelihood of a comeback. Moreover, our analysis extends beyond win-loss outcomes to scrutinize the strategic choices made on the court, examining shot distribution, efficiency, and the role of free throws (if any) in these high-stakes moments.

The Essence of Clutch: Defining the Crucial Moments

Upon analyzing approximately 6,600 games, our initial findings unveil that a significant portion of these contests enter such clutch moments. Specifically, 4,469 games featured situations fitting our clutch criteria, translating to about two-thirds of the matches analyzed. This high incidence underscores not only the competitive balance within the league but also the frequent occurrence of scenarios where outcomes hang delicately in balance until the final buzzer.

Home vs. Away: The Comeback Battle

In clutch moments, the distinction between playing at home versus on the road becomes starkly apparent. Out of 4,469 clutch scenarios in our dataset, 2,172 instances where the home team trailed by 5 or fewer points within the last 3 minutes of the 4th quarter, compared to 2,297 instances for the away team. The question at the heart of this investigation: does the home-court advantage play a significant role in a team's ability to orchestrate a comeback?

Upon examining the outcomes, we discovered that home teams managed a comeback in approximately 26.4% of their clutch scenarios, while away teams succeeded only about 19.6% of the time. This differential not only underscores the psychological and strategic edge of playing on familiar territory but also highlights the tangible impact of home support in tilting the odds during game-defining moments.

Shot Distribution: Tactics Unfolded

Our comprehensive analysis of the shot types across the entire dataset, containing more than 1.14 million shots, has revealed a remarkably balanced trichotomy. 2-point jump shots account for roughly 34.3% of all attempts, 2-point layups (including dunk and tip-in) for 31.8%, and 3-point jump shots closely follow at 33.9%. This distribution not only reflects the strategic diversity in the league but also underscores the evolving nature of the game where the three-point shot is just as prevalent as the time-honored layup.

These numbers set the stage for a deeper exploration into clutch moments, inviting us to examine whether teams cling to their shooting habits or pivot drastically.

As expected, we can observe a striking shift in shot selection during clutch moments by trailing teams, as depicted in the below chart. It can be attributed to the urgency of chasing points with limited time remaining on the clock. Teams prioritize three-point jump shots, which account for a substantial 45.3% of their attempts, a clear indication that they are strategically opting for plays that can yield higher points per shot.

Next, we decided to look at the shooting percentages during these critical clutch moments, juxtaposing them with the regular season averages. In clutch situations, trailing teams make 33.62% of 2-point jump shots and 59.70% of layups, but 3-point accuracy significantly drops to 31.38%, below the 5-season average of 36.14%. This decline likely results from a higher volume of attempts that may not always be the most open or high-quality opportunities, and more defensive pressure towards the end of the game. The decrease in shooting efficiency has a tangible impact on the expected value of each 3-point shot (from 1.083 using 5-season average to 0.941 points in clutch situation when trailing).

Next, we'll explore whether this approach pays off for teams attempting to stage a comeback.

The Anatomy of a Comeback

In dissecting the strategies of NBA comebacks, it becomes evident that the commonly perceived approach of ramping up three-point attempts during clutch moments (espacially with lower shooting efficiency) does not emerge as the winning formula. This insight aligns with our earlier observation that teams trailing in clutch moments increased their three-point attempts from 33.9% to a staggering 45.3%, only to suffer a decline in shooting efficiency.

The successful teams, however, took a different path. Their shot distribution maintained a closer resemblance to regular gameplay, with a balanced mix of three-pointers (34.8%) and a pronounced focus on two-point layups (39.8%). It wasn't the sheer increase in volume of three-point shots that paved the way to victory, but rather a calculated shot selection that leveraged the higher expected value of layups. With a 51.64% success rate on layups, these teams optimized their scoring efficiency, highlighting the importance of shot quality over quantity in high-pressure scenarios.

This strategic insight challenges the conventional clutch-time mentality, suggesting that a disciplined approach to shot selection, favoring higher-percentage attempts, may be a more reliable route to overturning deficits and clinching victories.

The Surprising Role of Free Throws in Comeback Scenarios

Contrary to conventional wisdom that free throws significantly contribute to the dynamics of a successful comeback by providing high-percentage points and stopping the game clock, our analysis suggests a more nuanced reality. Across the entire trailing clutch scenarios, teams averaged 2.54 free throw attempts per game at a success rate of 77.25%. Surprisingly, in games where comebacks were successfully mounted, these numbers slightly dipped to an average of 2.30 attempts per game at a 75.29% success rate. This data indicates that, despite the inherent advantages of free throws, their impact on the outcome of comebacks might not be as substantial as previously thought. While every point counts in the clutch, strategies beyond the free-throw line are equally crucial in flipping the scoreboard in a team's favor.

In summary: Our analysis of 6,600 games unveils that successful comebacks hinge less on the expected increase in 3-point shots and more on strategic shot selection, particularly favoring layups. Contrary to the conventional belief in the pivotal role of free throws for their high-percentage points and clock-stopping capability, our findings suggest their impact on comeback victories might not be as significant as previously thought. The essence of clutch moments, therefore, does not solely revolve around ramping up three-pointers or free throw attempts but in a balanced approach that emphasizes the higher expected value of layups. This insight challenges traditional strategies, highlighting the nuanced tactics teams employ in the adrenaline-filled final minutes to alter the game's outcome.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

This trend has been alredy well documented and numerous articles can be found on the internet. In the spirit of adding to this body of knowledge, we delve deeper into the theory and practice behind the rise of the long-range game.

Our exploration will dissect the elements that contribute to a winning formula in the modern era of basketball. By understanding the key factors influencing winning percentage, analyzing the shot mix—particularly the distribution between 2-pointers and 3-pointers—and trying to find the optimal mix, we aim to distill the essence of success in today’s game. It's clear that the sheer number of 3-point attempts (3PA) has a significant impact on team strategy and game outcomes however, as we will explore, the type of 3-point attempts may be even more crucial. Teams like the Houston Rockets and Golden State Warriors have not only championed the 3-point shot but have also revolutionized its strategic implementation, highlighting the critical role of shot selection and placement in securing victories.

As we continue to contribute to the ongoing conversation, we'll unpack these components, demonstrating how the 3-point shot is not just a trend but a vital piece of the puzzle in the quest for basketball excellence.

Understanding the Key Factors Influencing Winning Percentage

In our quest to unravel the mysteries of what drives success in the NBA, a thorough statistical analysis was conducted, focusing on the correlation between various team statistics and their winning percentage (WIN%) over the last 10 seasons. The findings are intriguing and offer some expected as well as some surprising insights.

As anticipated, there are strong positive correlations between WIN% and certain key statistics. Scoring more points (PTS) has a moderate positive correlation (0.38) with winning, highlighting the straightforward but crucial role of offense in securing victories. Defensive rebounds (DREB) show a stronger correlation (0.42), underscoring the importance of controlling the defensive glass to limit opponents' scoring opportunities. However, it's worth noting that offensive rebounds (OREB) show a negative correlation (-0.10) with WIN%, which might seem counterintuitive initially but could be indicative of teams missing more shots and thus having more opportunities for offensive rebounds.

The Paradox of Field Goal Attempts: Quality Over Quantity

A fascinating aspect of our analysis is the relationship between WIN% and the number of field goal attempts (FGA). Contrary to what one might expect, there is a slight negative correlation (-0.05) between the number of FGAs and WIN%. This suggests that merely increasing the number of 2pts shots taken does not necessarily lead to more wins. It underscores the importance of shot quality over quantity.

The Rising Significance of the Three-Point Shot

In contrast to the general field goal attempts, the number of three-point attempts (3PA) has a positive correlation (0.14) with WIN%. This finding aligns with the contemporary trend in basketball where the three-point shot is increasingly recognized as a crucial element of modern offenses.

The Paramount Importance of Shooting Efficiency

Most notably, the strongest correlation with WIN% comes from shooting efficiency, both from the field (FG%) and from beyond the arc (3P%). FG% has a correlation of 0.57, and 3P% is even higher at 0.59. These statistics reinforce the critical nature of not just taking shots, but making them. Efficient shooting, therefore, is paramount to a team's success.

Given the above findings, understanding the most effective shooting areas and the ideal balance between two-pointers and three-pointers will offer deeper insights into crafting winning basketball strategies. In the following sections, we’ll continue to dissect the intricacies of basketball analytics to uncover the formula for success in the NBA.

Analyzing the Shot Mix (2-Point vs. 3-Point Distribution)

The heatmap visualization below offers us a nuanced understanding of the relationship between shot distribution and winning percentage. While our initial focus was on the raw number of shot attempts (FGA and 3PA), our further analysis reveals that the volume of shots (TOT_SHOTS) has an almost negligible impact on winning, with a correlation close to zero. This finding challenges the conventional wisdom that more shooting opportunities inherently lead to more victories.

The data decisively points to the composition of shots being more influential on a team's success than the volume of shots taken. Teams with a higher proportion of three-point attempts out of their total shots (3P%_TOT_SHOTS) have a positive correlation with winning percentage, suggesting that a strategic focus on three-point shooting is more beneficial for winning games. In contrast, a greater percentage of two-point attempts (FG%_TOT_SHOTS) correlates negatively with success, indicating a strategic pivot in basketball towards the higher value of three-pointers over the traditional two-point shots.

Finding the Optimal Mix

The below heatmap provides a compelling visual representation that underscores the significance of expected values (EV) in our quest to determine the optimal shot mix for NBA teams.

EV = (FG% × FG Shot Proportion × 2) + (3P% × 3PT Shot Proportion ×3)

The concept of EV is pivotal because it encapsulates not just the likelihood of a shot being made, but also its point value. This dual consideration is crucial for understanding which shots are most conducive to winning games.

Analyzing the heatmap, it's evident that the expected value of 3-point shots (3P_EV) has a stronger correlation (0.32) with winning percentage (WIN%) compared to the expected value of 2-point shots (FG_EV), which shows a lower correlation (0.24). The total expected value (TOT_EV), a synthesis of the two, shows an even more robust correlation (0.62) with WIN%, suggesting that a balanced approach that optimizes both 2-point and 3-point attempts can be particularly effective for teams looking to enhance their winning prospects.

The stronger correlation for 3P_EV is rooted in the formula itself, which factors in the higher point value of 3-point shots (worth 3 points versus 2 for a FG), the shooting percentage from the 3-point line (3P%), and the distribution of 3-point shots within the team's overall shot selection. Given that 3P% is not dramatically lower than FG% for 2-point shots, the additional point gained from 3-pointers makes them statistically more rewarding. This insight suggests that teams should consider increasing their proportion of 3-point attempts within their offensive strategy to leverage the higher point value.

It's important to note that if we were to solve this optimization problem (optimal mix between FG and 3P) under the assumption of constant field goal percentage (FG%) and three-point percentage (3P%), the logical outcome would be a 100% allocation to three-point shots. This is because, as previously demonstrated, three-pointers inherently possess a higher expected value (EV) than two-pointers.

However, basketball is not a game of absolutes, and such an extreme allocation is impossible in reality. The distribution should be increased only to the extent that it doesn't substantially diminish the 3-point shooting percentage. The trade-off is beneficial as long as the increase in 3-point attempts doesn't lead to a decrease in 3P% that would outweigh the additional point gained per shot.

To address this challenge and provide actionable insights, we turned to historical data. By analyzing the shot distributions and performance metrics of successful teams over the years (WIN% above .500), we uncovered a compelling trend. Teams that strategically balanced their shot selection, allocating a proportion of their attempts to three-pointers falling within the 25% to 30% range, were more likely to achieve higher average total expected values (TOT_EV). This 'sweet spot' in the shot distribution reflects the real-world complexity of basketball strategy, where optimizing the mix of two-pointers and three-pointers can lead to a trend of enhanced performance.

As we've established that an optimal shot distribution falls within the 25% to 30% range for three-pointers, and recognized the practical limitation of not being able to allocate 100% of shots to three-pointers therefore, to further enhance a team's performance, the focus shifts to improving field goal percentage (FG%) and three-point percentage (3P%). The fundamental question becomes: What types of field goals (FG) and three-pointers (3P) are more likely to result in higher percentages, and how can coaches design systems that lead to these types of shots?

The Number of 3PA Matters, but the Type Might Be Even More Important

The excerpt from Kirk Goldsberry's book "SprawlBall" discusses the transformation of NBA shot selection analysis due to advanced cartography and spatial analytics. His meticulous analysis dovetails with our own above findings, the success rate of jump shots diminishes only modestly with increasing distance from the hoop. Despite only a slight variance in shooting percentages between extended 2-pointers and 3-pointers, the latter garners a significantly higher point yield (50% reward).

Source: How Mapping Shots In The NBA Changed It Forever by Kirk Goldsberry

Building upon this framework, it's advisable for teams to prioritize corner 3-pointers and layups due to their superior expected value (points per shot). Strategically, teams should implement systems that enhance the frequency of these high-value shots.

Plays like well-timed cuts to the basket can increase chances for successful layups and shots with points per shot (PPS) of theoretically more than 1.20 according on the above shots map. Hence, it is not surprising to see that cuts yield the highest points per possession (1.303 on average) among all play types so far. Since the beginning of the current NBA season, cuts are used as low as 4.6% (Portland) and as high as 9.9% (Golden State) of the time (total possessions). In addition to not rely on cuts, Portland exhibits the 23rd 3P% in the league (out of 30) despite being among the first 14 teams attempting the most 3-pointers per game.

As discussed before, even if Portand attempt a relatively higher number of 3-pointers, the poor 3P% combined with a very little reliance on cuts to the basket potentially leading to shots with high EV, can be the reason why only Memphis score less points per game on average than the Trailblazers this season (108.1, 29th).

Tactics

Basketball Immersion has conducted a concise yet insightful video breakdown showcasing how the Shanghai Sharks execute plays to create open three-point opportunities for Jimmer Fredette, a shot with a high expected value (EV). The offensive system is designed to set Fredette up for a swift three-pointer, targeting either a wing position following an initial screen or a corner shot. Of particular note in this analysis is the system's built-in response to defensive overplay. If a defender anticipates the move and obstructs Fredette’s route to the screen, the play adapts seamlessly. Fredette counters by using the screen to execute a sharp cut to the basket instead, converting the defender’s denial into an opportunity for an easy layup, which carries an even higher EV. By integrating such tactics, teams can exploit the most economically sound areas of the court, as underscored by our analysis and Goldsberry's work.

In summary, in examining the strategic evolution of basketball, our analysis confirms the increasing importance of the 3-point shot, evidenced by the rising trend in attempts over the years. Our study delves into the winning percentage's relationship with shot types, revealing that while volume matters, the efficiency and expected value (EV) of shots, especially 3-pointers, are crucial. We highlight the ‘sweet spot’ for the balance between 2- and 3-point shots, noting the higher EV of corner 3s and 2s in the restricted area. This insight into shot selection and efficiency offers teams actionable strategies to enhance their performance in the evolving game of basketball.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

This article delves into a data-driven analysis, focusing on the top 5 European leagues - Premier League, La Liga, Bundesliga, Serie A, and Ligue 1. By harnessing a comprehensive dataset encompassing player statistics and market values, we try to uncover players who not only exhibit strong on-field performance but also present significant value for money in the current market - a critical consideration during the winter mercato window.

Our methodology is rooted in evaluating players based on a range of position-specific metrics. For forwards, the spotlight is on goal-scoring prowess and efficiency. Midfielders are assessed through a blend of offensive and defensive contributions, while defenders are scrutinized based on their effectiveness in disrupting opposition plays. This exploration isn't only about identifying top talent; it's about finding players whose market values may not align with their performances.

As we navigate through this analytical expedition, we aim to offer clubs and enthusiasts insights into players who could be the key to success in the season's second half and beyond.

Data Insights and Position-Specific Metrics

Our analysis begins with a comprehensive dataset encompassing detailed player statistics and current market values from the top 5 European leagues from FBREF and Transfermarkt respectively.

To ensure a nuanced assessment, we have selected specific metrics for each position - forwards, midfielders, and defenders - based on their relevance and impact on the game. These metrics include traditional stats like goals and assists, as well as advanced metrics such as Expected Goals (xG), Key Passes, and the difference between Assists and Expected Assisted Goals (A-xAG).

The dataset contains data on more than 1,500 players distributed as below.

For a tailored evaluation, we've pinpointed metrics (per 90 minutes) that resonate with the specialized functions of each position on the field:

Forwards (FW)

• Goal Metrics (Gls, G-xG): Measure scoring efficiency and the ability to exceed expected outcomes.

• Playmaking Metrics (Cmp%, Ast, A-xAG): Assess quality in link-up play and creating opportunities beyond scoring goals.

• Impact Metrics (GCA90): Reflect the rate of goal-creating actions, underscoring influence on the team's offensive play.

Midfielders (MF)

• Distribution Metrics (Cmp%, PrgDist%): Indicate precision and ability to drive the ball forward effectively.

• Creativity Metrics (Ast, A-xAG, KP): Capture the ability to craft goal-scoring opportunities.

• Advancement Metrics (PFT, PrgP): Quantify impact on moving the ball into key attacking zones.

• Opportunity Metrics (SCA90, GCA90): Highlight contributions to both shot-creating and goal-creating actions.

• Defensive Metrics (Tkl%, Tkl+Int): Reflect prowess in ball recovery and defensive interventions.

Defenders (DF)

• Ball Management Metrics (Cmp%, PrgP, PrgDist%): Evaluate the defender's contribution to building up play from the back.

• Challenge Metrics (TklW, Tkl%, Tkl+Int): Show effectiveness in winning tackles and asserting defensive dominance.

• Protective Metrics (Blocks, Clr): Indicate the ability to shield the goal and clear threats, essential for maintaining defensive solidity.

Calculating Player Performance Scores and Financial Efficiency

We then calculated percentiles for the aforementioned metrics based on the player’s position on the field. These percentiles are a way to standardize player performance across the board, making it possible to compare players in different leagues and positions. By ranking each player against their peers and calculating the percentage of players they outperform, we obtain a percentile score for each metric.

To synthesize these metrics into an overall performance score, we took the average of the percentile ranks across all relevant metrics. This aggregate score reflects a player's all-around capabilities and efficiency on the field, distilled into a single comprehensible figure.

Moving beyond performance on the pitch, we sought to integrate a financial perspective into player valuation. To achieve this, we divided the overall performance score by the player's market value, creating what we refer to as the 'Financial Efficiency Ratio'. This ratio serves as an indicator of value for money, highlighting players who deliver high performance relative to their potential cost in the transfer market.

This financial efficiency ratio allows clubs, scouts, and analysts to identify players who may be undervalued in the market — those whose performance merits a higher valuation. It's a tool that can be particularly valuable for clubs operating under budget constraints, looking to make strategic signings that promise both high on-field impact and sound financial investment.

Highlighting Top Performers with Low Market Value

In assessing the financial efficiency of soccer players, it's crucial to recognize that a player's market value is not solely a reflection of on-field performance. Factors such as the remaining length of a player's contract and their age play significant roles in determining this figure.

It is a common occurrence in the transfer market to observe experienced, prolific players, especially those approaching the twilight of their careers or nearing the end of their contracts, being positioned favorably in terms of financial efficiency. This phenomenon can be attributed to the reduced transfer fees they command due to the shorter term left on their contracts or the expectation of a diminishing future contribution as they age. Olivier Giroud and Andrea Belotti are perfect examples in the scatter plot above.

To refine our analysis, we have decided to highlight the financially efficient players who are under 30 years old. This age criterion ensures that we are focusing on players who are likely to sustain their performance levels for a longer period, thus providing a balance between immediate impact and potential long-term contribution.

Moreover, it's important to acknowledge that limited playing time can skew a player's performance score, particularly when metrics are normalized on a per-game basis. (e.g. Luka Jovic above, with minutes played equivalent to 5 games). To mitigate this, our analysis will concentrate on players who have amassed playing time equivalent to 10 games or more during the current season. This threshold helps to ensure that the performance scores we consider are based on a substantial amount of playing time, thus providing a more reliable and accurate reflection of a player's capabilities on the field.

While there is an understandable positive correlation between performance scores and market value — with the latter often accounting for a player's potential and future performance — our model is specifically tailored to gauge immediate on-pitch effectiveness. Consequently, we are interested in uncovering players, valued below €15 million, who can offer immediate benefits to a team. These players represent a sweet spot of financial prudence and immediate performance readiness, making them highly sought-after targets for clubs looking to strengthen their squads smartly without breaking the bank.

Defenders

Our pick: Jorge Cuenca, a 24-year-old centre-back playing for Villarreal in La Liga - market value: €5 million.

The radar chart below visualizes the relevant statistics, showing that Cuenca's standout attributes are his tackle success rate, pass completion, clearances, and blocks with his performance in these areas being significantly above players with a similar market value. His progressive passes and distance metrics are around average, suggesting that while he is capable in these areas, they are not his standout attributes.

Overall Performance Score: Jorge Cuenca's overall performance score is around the 72nd percentile, indicating that he performs better than approximately 72% of all defenders among the 5 major european leagues. This high overall score reflects a well-rounded and solid defender.

Financial Efficiency Ratio: With a value ratio of 14.43 (average of 8.75 for players playing at a similar position), it suggests that compared to other players with a market value of €5 million, Cuenca offers a high performance per monetary unit. This ratio can be particularly appealing for clubs seeking cost-effective transfers, as it suggests that Cuenca's on-field contributions are of high value relative to his market price.

In conclusion, Jorge Cuenca's stats profile him as a strong defensive asset, particularly in clearances, pass completion, and tackling, with an overall performance that makes him a financially efficient player to consider for teams looking to reinforce their defense. Next summer, Jorge Cuenca’s will only have one year remaining on its contract.

Midfielders

Our pick: Angelo Stiller, a 22-year-old defensive-midfielder playing for VfB Stuttgart in Bundesliga - market value: €14 million.

The radar chart for Angelo Stiller accentuates his proficiency in key playmaking metrics such as key passes, passes to the final third, and a notably high pass completion rate, illustrating his effectiveness in ball distribution and chance creation. His ability to exceed his expected assists indicates a knack for setting up scoring opportunities that surpass the average forecast for players within his market value range. Stiller delivered 3 assists in 15 appearances in Bundesliga this season so far. His progressive passing and distance metrics are commendable, denoting a reliable capability in advancing play, although they do not stand out as his primary attributes.

Overall Performance Score: With an overall performance score placing him in the 82nd percentile, Angelo Stiller outperforms roughly 82% of midfielders in the major European leagues (only Joshua Kimmich ranks higher if we consider only midfielders of 30 years old or less). This impressive percentile ranking demonstrates Stiller's robust and impactful presence in the midfield.

Financial Efficiency Ratio: Stiller's value ratio is attractive, indicating that for a market value of €14 million, his performance is substantially more efficient monetarily compared to peers. This metric positions him as an exceptionally attractive proposition for clubs aiming for high-value acquisitions without the premium expense. His performance relative to cost suggests a savvy investment for teams prioritizing strategic financial planning and on-field competence.

Accounting for Stiller’s young age, his value as a strategic acquisition is further amplified, offering clubs the opportunity to capitalize on his advanced playmaking abilities and defensive contributions at a cost-effective rate.

With a comprehensive skill set that enhances team dynamics both in possession and during transitions, Angelo Stiller stands out as a financially prudent and tactically advantageous target for clubs looking to bolster their midfield with a player who has both immediate impact and potential for growth.

Forwards

Our pick: Jan-Niklas Beste, a 25-year-old left-winger playing for Heidenheim in Bundesliga - market value: €12 million.

The radar chart for Jan-Niklas Beste reveals a player with a particular prowess in goal contribution, highlighted by his goals exceeding the expected goals (G-xG) metric, which indicates a clinical finishing ability beyond what is statistically anticipated. Furthermore, his assist ratio surpasses the expected assist (xA) value, showcasing his capability to create goal-scoring opportunities. Beste was involved in 13 of its team goals this season (52% goal participation rate), including 5 goals and 8 assists in only 14 league games.

Overall Performance Score: Positioned in the higher echelons with an overall percentile score of 76.71, Jan-Niklas Beste ranks above the vast majority of his peers in the forward position across the major European leagues, signifying a dynamic and effective attacker.

Financial Efficiency Ratio: With an exceptional value ratio of 6.39—substantially greater than the average for his position—Beste's contributions on the field offer tremendous financial value for his market value of €12 million. This ratio is particularly enticing for clubs looking for a high-impact player at a reasonable cost.

His performance is further underscored by his advanced metrics in goal-creating actions per 90 minutes (GCA90) and a high percentile in assists, which are reflective of his direct impact on the game. Beste's adeptness at finding the back of the net, as well as setting up his teammates, combined with his overall performance score, solidifies him as a valuable asset for any team in need of an attacking boost.

Heading into a the final year of its contract, Jan-Niklas Beste's proven ability to outperform in crucial offensive parameters makes him an attractive target for clubs seeking a financially efficient yet highly effective forward. His age, coupled with his demonstrated performance, suggests a player in his prime who can contribute immediately while also offering potential for a good return on investment.

In summary, this analysis provides a snapshot of soccer player performance relative to market value. By meticulously examining position-specific metrics, we have highlighted some players who demonstrate high performance on the pitch, which does not seem to be adequately reflected in their market valuations.

It's important to reiterate that our model, while robust in its current form, is not without limitations. We have focused on a mathematical interpretation of present performance juxtaposed with current market value, intentionally setting aside predictions of future potential or performance. The model is grounded in statistical data available up to this point and does not factor in the unpredictable nature of a player's future trajectory or growth.

Furthermore, we must acknowledge that there is often a substantial disparity between a player's market value and the actual transfer fee involved in their movement between clubs. Market value is but a starting point for negotiations, with the final transfer fee frequently influenced by a myriad of factors including contract length, club needs, player desires, and the competition among bidding clubs.

Our findings, highlighting players like Jorge Cuenca and Angelo Stiller, suggest opportunities for clubs to make financially efficient signings. However, decision-makers should consider this data as one part of a broader recruitment strategy that includes a player's fit for the team, potential for growth, and adaptability to new environments.

The purpose of our model is not to provide definitive answers but to guide clubs towards more informed decisions by identifying undervalued talent. In the dynamic and unpredictable world of football transfers, marrying data analysis with traditional scouting can pave the way for successful squad building and strategic acquisitions.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

The Pythagorean expectation posits that a team's winning percentage can be accurately estimated by comparing the number of points it scores (points for) against the points it concedes (points against). This formula has historically been a reliable indicator of a team's performance, offering insights into their future success based on scoring statistics. Our project seeks to test and expand upon this concept in the context of the NFL, exploring its efficacy and potential enhancements for predicting outcomes.

Data Collection

To construct a comprehensive analysis, we've amassed a dataset encompassing key statistical elements:

• Points For and Against: The total number of points scored by and against each NFL team up until Week 11.

• Win Percentages: The percentage of games won by each team (as of season end)

• Game Schedules: The schedules of the games including the list of opponents for the remaining matches in the season.

This data spans from the start of the current NFL season up to Week 11 and also includes historical data from the last three seasons.

In the following sections, we will delve into the methodology of our analysis, compare our predictions with actual outcomes, and explore additional factors that could further refine our predictive model.

Understanding the Pythagorean Expectation Formula in NFL Analytics

This formula, a brainchild originally for baseball and later tailored for football, operates on a straightforward but insightful premise: a team's likelihood of winning is closely tied to their scoring efficiency – how many points they score compared to how many they concede.

The Formula, Simplified

For baseball, it looks at the square of the points (runs) a team scores and divides it by the sum of the square of the points (runs) they score and concede. However, the distinct scoring nature of football necessitated an adjustment, leading to the adoption of a different exponent (2.37), a figure derived from empirical analysis of football games.

Observations from the last three NFL seasons

When applying this formula to the NFL, we made an intriguing discovery:

• Teams with higher winning percentages (PCT) tended to have an even higher calculated winning percentage when we used the 2.37 exponent, as opposed to the original 2.

• Conversely, teams with lower winning percentages showed a decrease in their calculated percentages with the 2.37 exponent compared to the 2.

In essence, the range of winning percentages predicted using the 2.37 exponent is broader than when using 2. This suggests that the 2.37 exponent amplifies the distinction between stronger and weaker teams more than the original exponent.

However, the true revelation came when we examined the standard deviation of these predictions compared to the actual end-of-season winning percentages. Despite the broader range of predicted percentages with the 2.37 exponent, the standard deviation – a measure of how scattered or close these predictions are to the actual outcomes – was generally lower. This means that even though the 2.37 exponent creates a larger spread in the predicted winning percentages, these predictions are more closely aligned with how teams actually performed by the season’s end.

Correlation Between Actual Performance and Pythagorean Expectation

The scatter plots for each NFL season reveal a compelling story about the relationship between the actual winning percentages (PCT) and the Pythagorean Winning Percentages (P_PCT). Notably, it is evident from the charts that teams with higher actual winning percentages tend to surpass their Pythagorean expectations as well as the opposite.

What is particularly striking is the strength of the correlation between PCT and P_PCT after Week 11, recorded at a robust 0.78. This high correlation coefficient underscores a significant predictive power within the Pythagorean formula. It suggests that, despite some outperformance by high-performing teams, the Pythagorean expectation after Week 11 offers a reliable indicator of a team's likely success as the season progresses.

What About This Season?

As we delve into the 2023 NFL season data, the following observation can be made when comparing each team's actual winning percentage (PCT) after 11 weeks with their Pythagorean Winning Percentage (P_PCT).

• Philadelphia and Detroit's Remarkable Outperformance: Philadelphia leads the pack with a significant outperformance, registering a performance difference of 0.254, which is notably higher than their adjusted Pythagorean expectation. Detroit follows suit, outperforming its statistical prediction by nearly 0.200. These figures suggest that both teams are winning more games than what their scoring trends would typically forecast.

• Contrasting Stories of Baltimore and Dallas: Baltimore and Dallas, while maintaining strong actual winning percentages, exhibit underperformance against their Pythagorean expectations. Dallas, with a -0.084 difference, and Baltimore, at -0.055, show that despite their successes, their win rates are slightly lower than expected based on their scoring.

• Buffalo and LA Chargers Falling Short of Expectations: On the other end, Buffalo and LA Chargers are trailing behind their Pythagorean predictions, with Buffalo showing a substantial negative difference of -0.192. This discrepancy might point to factors such as key game moments (LAC lost 6 games by an average deficit of only 4.6 points including 5 of them by only 3 points or less) or player performances (Josh Allen is the QB with most interceptions since the beginning of the season) that are not fully encapsulated by the Pythagorean formula.

Interestingly, the standard deviation in team performances (difference between current winning percentage and expected) is significantly higher than in previous full seasons. This increase, from an average of around 0.07/0.06 and 0.08 in the last three seasons to above 0.10 in the current season, suggests that teams are currently performing in a manner less predictable by the Pythagorean Winning Percentage. However, statistical principles such as mean reversion imply that extremities tend to normalize over time. Therefore, we can reasonably expect that, as the 2023 season unfolds, team performances are likely to gravitate closer to their Pythagorean expectations. This mean reversion could result in a tightening of the performance spectrum, where teams currently outperforming or underperforming their P_PCT may see their outcomes align more closely with these statistical predictions by the end of the season.

As a result, we can expect teams like Baltimore and San Francisco or Dallas (as opposed to Eagles and Lions) to top their respective conference by the end of the season.

Enhancing Predictive Accuracy with Remaining Opponent Strength

In a bid to refine our predictive model for NFL teams' end-of-season winning percentages, we engineered a new component: the Remaining Opponent Strength (RO Strength). This metric is calculated by summing the Pythagorean Winning Percentages of each team's remaining opponents and then dividing by the number of remaining games. The rationale is that the strength of a team's future opponents should provide valuable insight into their potential performance in the remaining season.

The scatter plot analysis reveals a nuanced relationship between teams' Remaining Opponent Strength (RO Strength) and their performance deviations, termed Delta (Δ), from the Pythagorean Winning Percentage. The trend, marked by a descending red line, (logically) implies that teams facing stronger future opponents tend to slightly underperform against their P_PCT. Yet, the broad spread of the data points suggests that other factors are at play, influencing the outcomes beyond the RO Strength alone.

Incorporating our discovery of a negative correlation between Delta (the difference between Pythagorean and actual winning percentages) and RO Strength, we've constructed a linear regression model to predict end-of-season winning percentages. This model integrates both P_PCT and RO Strength, aiming to adjust our forecasts to account for upcoming challenges faced by the teams.

Our regression results, showing an r-squared value of 0.655, demonstrate a substantial explanatory power, with both P_PCT and RO Strength emerging as significant predictors (p-value < 0.05). The positive coefficient for P_PCT confirms its strong predictive value, while the negative coefficient for RO Strength aligns with our earlier correlation findings, suggesting that a tougher schedule could lead to underperformance.

Predicted 2023 NFL Playoff Picture

Current NFL Playoff Picture through Week 12

Our Predictions

Starting with the American Football Conference, division champions are projected to be Buffalo in the East, Baltimore in the North, Houston in the South, and Kansas City in the West with Baltimore clinching the number one spot in the conference. Despite current struggles, Buffalo P_PCT is above 0.73, well above the just above average current winning percentage, hence justifying their top position in the AFC East in front of the best offense in the league, the Miami Dolphins. More interestingly, if the season was ending today, the Bills would not even make the playoffs at the benefit of the Steelers.

For wild card spots, Miami, Cleveland, and Jacksonville emerge as top contenders, indicating their robust season play. Even with a 0.7 current winning percentage and a quasi similar P_PCT with Houston, Jacksonville is expected to give away the first place of its division due to a higher RO Strength number with for example the Texans, Browns, and Ravens yet to play…

Shifting focus to the National Football Conference, the playoff landscape is set with Dallas leading the NFC East, Detroit narrowly claiming the NFC North, New Orleans reigning in the South, and San Francisco asserting dominance in the West and in the conference in general with an adjusted winning percentage of slightly more than 0.8 due to the combination of a high P_PCT and facilitating remaining schedule. These teams, with their impressive adjusted win percentages, are poised to become the division champions.

For the coveted wild card spots, Philadelphia, despite being 9-1 after 11 weeks, the Eagles seem to outperform this season as seen earlier and hence losing their division first place to the Cowboys as per out predictions. Minnesota from the North, and Tampa Bay from the South, are the top contenders, edging out their rivals and demonstrating their prowess in what promises to be a thrilling playoff race in the NFC.

In conclusion: our article delved into predicting NFL teams' end-of-season performances using the Pythagorean expectation, enhanced by the innovative metric of Remaining Opponent Strength. Through rigorous data analysis, we observed notable correlations and trends, leading to predictions of division champions and wild card contenders for both the AFC and NFC. While our statistical approach provided insightful forecasts, we acknowledged the inherent unpredictability of football. Factors like injuries, coaching strategies, and many others crucially impact outcomes, reminding us that the true essence and excitement of football lie in its dynamic and unpredictable nature, beyond what numbers alone can capture.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

The core of the Moneyball theory posited that On-base Percentage (OBP) was an undervalued metric in determining a player's contribution to a team's success, more so than the traditionally favored batting averages and runs batted in.

In their groundbreaking study, "An Economic Evaluation of the Moneyball Hypothesis", researchers delved into the relationship between team success—measured by winning percentage—and players' OBP and SLG, shedding light on the economic efficiencies gained from this approach. Crucially, they also examined the relationship between player salaries and OBP before and after the book was published, offering insights into the market's valuation of player contributions.

Two decades on, this post aims to re-evaluate the Moneyball hypothesis in light of the last 23 seasons (2000-2022). We will examine whether OBP continues to hold its place as a more crucial indicator of team success compared to SLG, and introduce Isolated Power (ISO) into our analysis as a modernized slugging metric that specifically measures a hitter's raw power.

Furthermore, with the advent of advanced analytics like Statcast, which offers a treasure trove of player performance data, we are poised to assess the validity of Moneyball's principles in today's game. Lastly, we'll conduct a correlation analysis between team payroll and OBP alongside more advanced metrics to determine what is currently valued by MLB franchises.

Reassessment of Moneyball Metrics (2000-2022)

Utilizing publicly available game logs from Retrosheet, we transformed raw text files into a structured dataframe (note: The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org").

Once the raw data was structured, we created binary variables to reflect home and away wins. Given that teams can be either the home or the away team, it was necessary to aggregate their statistics separately for each context. This dual aggregation approach allowed us to prepare a dataset for further analysis, capturing all home and away records for each team in every season.

The key statistics for On-base Percentage (OBP) and Slugging Percentage (SLG) were computed using the following formulas:

• OBP is calculated as

  where:

  • H is Hits

  • BB is Bases on Balls (Walks)

  • HBP is Hit By Pitch

  • AB is At Bats

  • SF is Sacrifice Flies

• SLG is calculated as

  where:

  • 1B, 2B, 3B, and HR represent singles, doubles, triples, and home runs, respectively.

By merging the home and away statistics, we constructed a dataset to analyze each team's performance over multiple seasons, incorporating these vital batting statistics. The calculated team OBP and SLG, both for and against, were pivotal in recreating the analytical framework used by Hakes and Sauer in their study.

When contrasting contemporary analysis using data from the last 23 seasons with research conducted between 1999 and 2003 (above), the importance of OBP over SLG in determining a team's winning percentage persists, albeit with nuanced differences. The more recent findings continue to validate OBP as a stronger contributor than SLG to winning games, evidenced by its larger coefficient in the models. However, there has been a slight decline in the coefficient values for OBP as well as a marginal reduction in the R-squared values, indicating a somewhat diminished but still significant predictive power for winning. Even after including a metric like Isolated Power (ISO), the raw power of a hitter by taking only extra-base hits and the type of extra-base hit into account, the combination of the three metrics accounts for about 82.7% of the variation in win percentage (no improvement) where ISO’s coefficient is 0.47.

The advent of advanced metrics like xSLG and wOBA may shed light on the observed decrease in R-squared and the coefficients for traditional stats. In the following section, we'll investigate how these modern analytical tools might offer explanations for these changes and provide fresh insights into baseball's winning formula.

Moneyball in the Statcast Era (2015-2022)

As baseball analytics have evolved, so too have the metrics used to dissect and understand the game's underlying factors that contribute to winning. The incorporation of cutting-edge Statcast metrics such as expected batting average (xBA), expected slugging percentage (xSLG), expected weighted on-base average (xwOBA), and expected wOBA on contact (xwOBACON) has opened up new avenues for analysis.

Yet, after trying multiple combinations and having introduced interaction terms (difference between the metric and its expected value, e.g. SLG - xSLG), these sophisticated metrics have not eclipsed the explanatory power of OBP in the context of winning percentage. It appears that the ability of OBP to signify a player's proficiency in reaching base still plays a pivotal role in shaping a team's success. As we continue to refine our regression models and integrate these advanced stats alongside traditional ones, it is becoming apparent that the search for a metric with as much predictive power as OBP is an ongoing quest in the era of modern baseball analytics.

The “Launch Angle” Revolution

In recent years, baseball has become interestingly obsessed with launch angles, with hitters optimizing their swings to achieve a higher trajectory on batted balls, aiming for more home runs and extra-base hits. This approach has altered traditional hitting philosophies and has been linked to the surge in home run rates across the major leagues. By leveraging sophisticated tracking technology from Statcast, metrics such as launch angle, exit velocity, and barrel percentage have become crucial in evaluating a player's ability to produce high-value hits.

The subsequent section will delve into the correlation between these modern metrics and winning percentage, examining if and how they contribute to a team's success. By analyzing these relationships, we aim to uncover potential advantages in player evaluation that extend beyond the traditional reliance on OBP. To do so, we used data from Baseball Savant (Statcast) over the last 9 seasons including the most recent regular season.

The launch angle determines the initial trajectory of the ball after being hit. A balanced launch angle is crucial; too low could lead to groundouts, and too high might produce easy flyouts. Its correlation coefficient with winning percentage is 0.264750, which is a weak to moderate positive correlation. This suggests that while an optimized launch angle does have a positive impact on winning, it is not as strong a predictor as exit velocity or barrel percentage, especially given the fact that too high angles (resulting in flyouts) detract from a team winning percentage.

Exit velocity gauges how hard a ball is hit, with higher exit velocities generally leading to more successful hitting outcomes. The correlation coefficient of 0.415538 with winning percentage is a notable positive correlation, implying that teams whose players consistently hit the ball with high exit velocities are more likely to have higher winning percentages, perhaps due to increased chances of hits and home runs.

Barrel percentage measures the frequency of a batter's ideal contact—hard and at an optimal angle, a combination of the above metrics. A higher barrel percentage indicates a batter who regularly makes the type of contact that leads to extra-base hits. The correlation coefficient of 0.356148 with winning percentage suggests a moderate positive relationship, indicating that teams with players who achieve higher barrel percentages tend to have better winning records.

The negative correlation between Ground Ball Percentage (GB%) and winning percentage further emphasise the above. With a correlation coefficient highlighting an inverse relationship (-0.277288), it is evident that teams with a lower GB%, implying a lesser frequency of ground balls, tend to win more games. As teams and players adapt, it becomes increasingly clear that those who master the art of the launch angle combined with high exit velocity, not only excite with soaring home runs but also significantly contribute to their team's success.

Financial Analysis: Payroll vs. Performance

In the evolving landscape of baseball's analytical era, a discernible trend has emerged in the correlation between team payrolls and key performance metrics. The data from 2015 to 2023 illuminates a progressive valuation of OBP and barrel percentage among MLB teams. While OBP consistently signified a player's aptitude for reaching base, its correlation with team payrolls saw a noteworthy climb up to 2022, suggesting that teams have been increasingly willing to invest in players proficient at avoiding outs.

Meanwhile, barrel percentage has also risen in correlation with payrolls, peaking in 2022. This underscores a strategic shift towards players capable of generating impactful hits. The marked dip in OBP correlation in 2023 juxtaposed with the still robust, albeit reduced, link with barrel percentage could reflect a nuanced balance in team investments, marrying on-base skills with the power and precision heralded by the launch angle revolution.

As teams navigate through the data-driven waves of modern baseball, these correlations with payroll expenditures reveal a compelling narrative of evolving priorities, with the 2020 season standing out as an anomaly likely influenced by the unprecedented circumstances of the pandemic.

On average, teams have spent approximately $132.36 million in total payroll for a near-even win percentage of 49.82% over the last 9 seasons, and a positive correlation between winning percentage and total payroll is observable.

Yet, the Tampa Bay Rays have notably defied this trend. Despite having the lowest average payroll in the league, a modest near $70 million, the Rays have consistently outperformed expectations, securing an average win percentage of close to 55% well above the break-even point of 50%, reminiscent of the early 2000s Oakland Athletics narrative that gained widespread attention with the release of "Moneyball".

Like the A's, the Rays have demonstrated that a team doesn't need the financial firepower to outscore their competitors; they can also win through intelligent management and a keen eye for undervalued talent. By prioritizing metrics that other teams may overlook, Tampa Bay has consistently achieved a winning percentage that defies the expectations set by their modest payroll, the lowest in the league over the period.

In the two decades since "Moneyball" popularized the use of OBP as an undervalued metric for building competitive baseball teams, the landscape of talent acquisition in MLB has significantly evolved. High OBP players are now widely recognized for their value and are consequently well-compensated. This shift has particularly impacted teams with lower budgets, which can no longer easily compete for such talent without exceeding their financial constraints.

However, these financially constrained teams sometimes show proficiency in metrics that could indicate a focus on efficiency or specific skill sets that are undervalued in the market. For instance:

• The Tampa Bay Rays (TBA), with the lowest payroll ranking at 30, show a relatively strong performance in the wOBA and wOBACON at rank 11 and 5 respectively, which are considered more sophisticated metrics that estimate a player's overall offensive value. TBA ranks amongst the best in quality of hit metrics like barrel percentage and exit velocity.

• The Oakland Athletics (OAK), with the second-lowest payroll at rank 29, have the best ranking in launch angle, suggesting a strategic approach to hitting that emphasizes optimal ball flight paths for better results, despite not leading in traditional hitting metrics.

In summary, as more sophisticated and advanced metrics become available, they open new avenues for low-budget teams to compete. By shifting focus from the now highly valued OBP to under appreciated aspects of hitting like launch angle and barrel percentage, these teams continue to challenge richer clubs, finding novel ways to assemble competitive rosters in the ever-evolving game of baseball sabermetrics.

In conclusion: the Moneyball revolution that took root two decades ago continues to shape the landscape of Major League Baseball. While the importance of OBP over SLG in determining a team's success remains evident, the modern era of advanced analytics has introduced a new layer of complexity to player evaluation. Metrics like launch angle, exit velocity, and barrel percentage are now new factors in understanding a player's ability to contribute to their team's winning percentage.

The correlation between these modern metrics and winning percentage highlights the evolving nature of the game, where a balance between on-base skills and power-hitting precision is becoming increasingly important. Teams are reevaluating their investments, and the correlation between player performance and payroll expenditure reveals a dynamic narrative of shifting priorities.

Notably, the Tampa Bay Rays, much like the Oakland Athletics in the Moneyball era, have shown that a strategic focus on undervalued metrics can lead to success even with a lower payroll. In this ever-evolving world of baseball sabermetrics, low-budget teams are finding innovative ways to compete by harnessing the power of advanced analytics.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

To explore this question, we turn to data from the NBA's 2016-2017 season. The "hot hand" phenomenon isn't a novel area of interest. Academic giants such as Gilovich, Vallone, and Tversky (1985) have critically examined it and challenged its existence, whereas other researchers like Miller and Sanjurjo (2018) and Yaari and Eisenmann (2011) have presented more nuanced perspectives, but mainly in free throw situations.

While our exploration might not delve as deeply as some academic works, we intend to explore the data using renowned analytical techniques and see what the narrative embedded within the numbers reveals. Specifically, our investigation will leverage:

• Conditional Probability: Assessing the probability of a successful shot based on previous successes or failures.

• Autocorrelation: A method that gauges the similarity of a player's current shot success to their past performance.

• Regression Analysis: This technique will help us incorporate multiple influencing factors, revealing their potential impact on shot outcomes.

Our analysis aims to offer a clear and concise exploration of the data, allowing us to engage with the perennial debate around the "hot hand" from an informed perspective.

From Shot to Shot: Using Conditional Probability to Gauge the 'Hot Hand' Effect

The first step in our basketball analysis dives into the idea of "conditional probability". Put simply, conditional probability helps us figure out how likely it is for something to happen based on a previous event.

Imagine this in a basketball scenario: We want to know the chances of a player making their current shot if they've just made their previous shot. If the hot hand theory holds true, a player who made their last shot should have better odds of making their next one, compared to their usual shooting rate.

To test the “hot hand” theory using conditional probability, we looked at our dataset and performed the following steps for each player. We first identified instances where two consecutive shots were successful, and by averaging it, we determined the joint probability - the likelihood a player makes both the current and previous shots.

Similarly, we calculated the average of successful previous shots. From these, we computed the conditional probability - the chance a player sinks the current shot given they made the previous one.

The scatter plot visually demonstrates the relationship between a player's conditional probability of making a shot given the fact they made the previous shot, and their overall average shot percentage. The diagonal red line represents the line of perfect equality - i.e., where the conditional probability and average shot hit percentage are identical.

Linear Relationship: A striking majority of the data points closely follow the red diagonal line, indicating a strong linear relationship between the two metrics. This suggests that for most players, the conditional probability of making a shot after making the previous one is roughly equal to their overall shooting percentage. Essentially, for these players, making or missing a shot doesn't seem to have a significant influence on the outcome of their next shot.

Notable Outliers: While most players hover around the line of equality, there are a few who demonstrate a notably higher conditional probability compared to their average shot percentage. At first glance, this might be taken as evidence in support of the "hot hand" phenomenon. However, a closer inspection reveals an important caveat: these outliers predominantly represent players with a relatively low shot count for the season. It's possible that their elevated conditional probabilities might be artifacts of the smaller sample size, making these figures less reliable. Statistically, smaller sample sizes are more prone to large deviations due to randomness.

When looking at players with an average number of shot per game higher than 10, the linear relationship appears even stronger.

To understand the significance of our findings, we compared the conditional probability with a player's overall shot success rate. The difference between these two values gave us an indicator of the hot hand effect – a positive difference might suggest a "hot hand", while a value around zero would not support the theory. Based in the distribution of the probabilities’ difference below, it confirms what was previously perceived and does not support the theory.

To ensure the robustness of our analysis, we applied a t-test. This test aimed to discern if there's a statistically significant difference between the conditional probability of making a shot after a successful attempt and the general shot-making average.

The outcome? A p-value of 0.1386, which is higher than the typical threshold of 0.05 used for statistical significance.

This suggests that there's no compelling evidence to support the existence of the hot hand phenomenon based on our dataset. Further visual inspection of the difference between conditional and random probabilities showcased a distribution centered around zero, echoing our statistical findings.

Linking the Shots: Understanding the 'Hot Hand' through Autocorrelation

As we dive deeper into the investigation of the "hot hand" phenomenon in basketball, we wanted to employ another method to substantiate or refute our previous findings based on conditional probability.

Autocorrelation, often referred to as serial correlation, describes the degree of similarity between a given time series and a lagged version of itself over successive time intervals. Simply put, it reflects how well the current value of a series can be predicted by its previous values. In terms of basketball, think of it as examining the relationship between a player's current shot outcome and their past shot outcomes.

The beauty of autocorrelation in this context is its ability to provide insight into whether a player's current performance is influenced by their immediate past performances. If there's a significant positive autocorrelation, it might hint that a successful shot (or series of shots) could increase the chances of the subsequent shot(s) being successful too. This would be a compelling piece of evidence in favor of the "hot hand" theory.

The above density plot, which represents the autocorrelation of players with an average shot per game greater than 5, is tightly centered around zero. The mean of the shot autocorrelation is -0.011133. This central tendency is further emphasized when considering that 75% of the players have an autocorrelation value below 0.021804. Furthermore, 96.37% of the filtered players have an autocorrelation lower than 0.10, which further underscores the proximity of the bulk of the data to zero. This suggests that, for the majority of players with a substantial number of shots per game, there's minimal evidence of the "hot hand" phenomenon—most players' shot outcomes are not strongly predictable based on their previous shot outcome, according to the above autocorrelation analysis.

Regression Analysis: The Pursuit of the "Hot Hand" Continues

After exploring our basketball data, we're now shifting gears to regression analysis. Think of regression as our special detective tool that helps us understand the story behind the numbers.

Why regression? At its heart, regression is about relationships. It allows us to see how one thing (like a player's previous shot) might influence another (like their next shot). So, when we talk about regression, we're basically asking: "If this happens, what's the chance that that will happen?"

And the great part about regression? We can throw multiple clues into the mix. So, apart from just looking at a player's previous shot, we can also consider things like whether they're playing a home game, how long it's been since their last shot, and more. This helps us get a clearer picture of what really goes on during a game.

In addition to our primary model, we will zoom in on some notable NBA players. Names like the season's MVP Russell Westbrook, and Steph Curry, often hailed as one of the best shooter in NBA's history, will take the spotlight. By analyzing such elite players individually, we hope to discern both visual and statistical indicators of the "hot hand" phenomenon. Does an MVP or a record-breaking shooter like Curry experience hot streaks differently than an average player? We're about to find out.

The first model we used using the all players is an ordinary least squares regression. While the model is statistically significant, it does not explain at all the variability in the prediction error (R-squared is 0.000). This implies that using the previous shot's error to predict the current shot's error is not highly effective for the dataset.

The negative coefficient (lagerror) for previous shot's error suggests even the opposite of what we thought - a small negative relationship. In other words, if a player performed better than average on the previous shot (i.e., scored), they are slightly more likely to underperform on the next shot (miss). However, with a coefficient of only -0.0123, the impact is likely to be very very small. It is obvious that there might be other variables not considered in this model that could help better explain the 'hot hand' phenomenon, given the R-squared value of 0.

Basketball, like many sports, is a multifaceted game with various dynamic factors influencing every moment. While the outcome of a player's shot is of central importance, it's also a result of various conditions that can influence a player's performance. Hence, it's essential to recognize and account for these factors to get a more holistic understanding of the "hot hand" phenomenon.

For a more comprehensive analysis, we are now extending our regression model to include several other variables: player position, home game, outcome of opponent's previous shot, and time from last shot.

Not every player has an equal opportunity to shoot in every game. Some players, due to their role in the team, might take more shots than others. Recognizing this, we decided to refine our analysis further.

To address this disparity and to ensure that our model is representative of the actual shooting dynamics, we introduced a weighting mechanism. We used the Weighted Least Squares regression, where each observation is weighted by the number of shots taken per game. This approach emphasizes the influence of players who shoot more frequently, thus giving a more balanced perspective on the "hot hand" phenomenon across different types of players.

Analyzing the WLS regression results:

• R-squared value stands at around 0, which means our model explains only 0.2% of the variability in the error term. This is a rather low value, implying that our predictors don't strongly determine the outcome.

• The coefficients for different player positions are not statistically significant at the conventional 0.05 level, which means that the player's position might not have a major influence on the shot's error when other factors are accounted for.

• The coefficient for home game is positive and statistically significant, suggesting that players are more likely to deviate from their average performance when playing at home. This could be due to the home-court advantage, but the magnitude of the effect is still minimal.

• Interestingly, both when the opponent missed or scored their last shot, the coefficients are negative and significant. This suggests that a player is more likely to miss after either scenario, possibly due to the psychological factors associated with the game's flow.

• Coefficient accounting for the previous shot outcome and the time between two shots are negative and statistically significant, albeit closer to zero. This suggests that the previous shot's outcome or the time between the current and previous shot have a very tiny to no influence on the current shot's outcome.

This suggests that the "hot hand" phenomenon is somehow a fallacy, and while intriguing, might be overshadowed by other game dynamics or might not be as influential as popularly believed. Further, this could also suggest that many other unaccounted factors play a role in a player's shooting performance, highlighting the complex nature of the sport.

Analyzing the "Hot Hand" Phenomenon in Stephen Curry

Stephen Curry, widely regarded as one of the greatest shooters in basketball history, provides an intriguing case for the "hot hand" investigation. With a shooting prowess that has left both fans and experts in awe, one might expect Curry to exhibit clear streaks of consecutive successful shots. To delve into this, we plotted all his shots over multiple games during the season. However, even for a player of Curry's caliber, the visualization did not show a distinct pattern of streaks.

While there were games where he had multiple consecutive successful shots, these were not consistent enough across the season, at least visually.

In our weighted least squares (WLS) regression analysis for Stephen Curry's shooting performance in the 2016-17 season, the R-squared value was nearly zero (0.003), suggesting that the model does not explain the variability in the dependent variable.

Of all the predictor variables, only home game appears to have any substantive relationship with the error rate, where playing at a home game is associated with a slight increase in the chance of making the following shot (coefficient of 0.0528). However, this relationship is only marginally significant, with a p-value of 0.052, just above the standard significance threshold of 0.05.

In simpler terms, the outcome of Curry's previous shot does not provide any predictive power for his subsequent shot. Furthermore, the autocorrelation value, at just 0.0012, further emphasizes the lack of a "hot hand" effect for Curry. Despite his reputation as an elite shooter, the data implies that each of his shots is essentially an independent event, unaffected by the success or failure of the previous one, aligned with the analysis on the entire player’s population earlier.

What about the MVP?

Russell Westbrook, the season's MVP with the most shots attempted, serves as our next focus in the "hot hand" inquiry.

Upon examining Russell Westbrook's shooting performance from the 16-17 season through regression analysis, the model's R-squared value of 0.017 suggests a limited explanation of the variability in his shooting error. Moreover, given the high p-values and low coefficients of the variables, none of them appear to have a significant influence on Westbrook's shooting performance during this season.

This heat map below of the coefficients and their respective p-value highlights the lack of significance from the variables in explaining Westbrook's shooting performance.

In summary, for Westbrook, even after accounting for various game-related factors and using weights for shots per game to improve the model's representativeness, there's limited evidence to support the "hot hand" phenomenon. Combined with the Curry observation, it reinforces the idea that even elite players, known for their impeccable shooting skills, may not necessarily manifest this supposed pattern. It underscores the complex interplay of factors that influence each shot and the inherent randomness in basketball shooting outcomes, even for the very best in the game.

In conclusion: In our investigation of the "hot hand" phenomenon in the 2016-2017 NBA season, we used various analytical methods to explore whether a player's success in making consecutive shots supports the idea of a "hot hand." Our findings suggest that the concept of the "hot hand" is not well-supported by the data.

1. Conditional Probability Analysis: Most players showed a strong linear relationship between making or missing shots, suggesting that a player's previous shot does not significantly influence their next shot.

2. Autocorrelation Analysis: For players with a substantial number of shots per game, there was minimal evidence of the "hot hand" phenomenon. Shot outcomes were not strongly predictable based on past performance.

3. Regression Analysis: Factors such as player position, home game advantage, and previous shot outcomes had little to no impact on a player's current shot success, indicating that the "hot hand" effect is not statistically significant. Even top players like Stephen Curry and Russell Westbrook did not exhibit a "hot hand" effect in their shooting performance.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.

To realize this project, team-specific data (100+ features, including internally engineered features) covering shooting, passing, pass types, shot and goal creation, defensive actions, and finally possession, was collected from FBREF.

Playing Style Clustering

We used the KMeans clustering algorithm to group teams. Clustering works by segregating the dataset into subsets or 'clusters' based on similarity. In our case, teams with similar statistical profiles are grouped together (4 groups).

Since visualizing multi-dimensional data is challenging, Principal Component Analysis (PCA) simplifies this by transforming the original features into a set of linearly uncorrelated variables called Principal Components (PC). These PCs capture the most variance in the data in decreasing order, allowing us to plot teams on a 2D plane using the first two PCs.

Taking the most recent complete season as an example, we'll begin by examining two distinct clusters that emerged from the data.

Cluster 2 (The Blue Brigade) - This cluster stands out with three top-tier teams, including the likes of Manchester City and Arsenal. Remarkably, Manchester City secured the crown, with Arsenal closely following in second place. It's noteworthy that all three teams in this cluster finished in the top 5. What's even more captivating is Cluster 1 (The Red Ensemble) on the other end of the chart, which consists of five teams, and despite their differing playing styles, these teams managed to secure their spots in the league.

But what exactly makes teams gravitate towards specific clusters?

To answer this question, we turn to the "loadings" of principal components. These loadings shed light on the features that influence the most influence in determining where teams position themselves along these principal components.

Loadings in PCA help in understanding the weight or importance of each feature in the dataset concerning the PCs. They give insights into which features are most influential in distinguishing between the clusters. By examining the loadings, we can interpret the most defining characteristics of each cluster.

Principal Component 1 (PC1) - It comes as no surprise that the team led by the esteemed Pep Guardiola, Manchester City, shines in this category. PC1 is dominated by metrics such as controlled possession (possession multiplied by pass completion percentage), progressive passes, and passes completed. Manchester City shares this space with Arsenal and Liverpool, both boasting impressive possession statistics of approximately 59.3% and 60.8%, respectively. But possession-based football isn't exclusive to Cluster 2. Teams like Chelsea and Brighton, with around 60% possession, find themselves in different clusters.

So, what sets Cluster 2 apart?

For answers, we delve into PC2 and focus on features with negative values. Here, we encounter metrics like pass efficiency (goals divided by passes completed), goals per shots on target, and the difference between actual goals and expected goals.

Manchester City's Mastery - Manchester City's style combines possession-based football with formidable attacking prowess. Meanwhile, Arsenal, despite having slightly lower possession compared to Liverpool and Brighton, extracted the utmost efficiency from their possession. They possess the league's highest conversion rate of shots on target to goals, resulting in one of the most impressive goal differentials over expected goals.

The Outliers: Brentford and Chelsea - Brentford sits at the other end of Cluster 2, with a negative PC1. Negative PC1 values primarily relate to defensive actions like clearances and blocks, portraying a team that engages in numerous tackles in their defensive third and relies heavily on long passes (high percentage of long balls over total passes). This style leans towards a more defensive and direct, counter-attacking approach.

Interestingly, among all teams with a negative PC1, only two finished in the top 10, implying that a defensive and direct style may not be the optimal path to EPL success. But why did Brentford finish three places higher than Chelsea?

The answer lies in our above analysis. While Chelsea posted one of the highest possession figures in the league, their PC1 value is somewhat offset by an incredibly high number of tackles, particularly in the defensive third. This can be interpreted as despite having ‘possession’, Chelsea tend to often lose the ball, and do not recover it straight away. Only relegated Leeds United had more tackles than Chelsea. The critical differentiator lies in both teams' attacking efficiency. Football is simple—score more than your opponent. Chelsea struggles in this department, with low pass efficiency (it takes Chelsea a lot of passes before being able to score, potential sign of sterile possession) and goals per shot on target ratios, resulting in a negative goals difference over expected goals. In contrast, Brentford's offensive efficiency shines, topping the pass efficiency category and scoring 19 more goals than their London counterparts.

In Conclusion - Last season's relegated clubs, including Leicester, Leeds United, and Southampton, all found themselves hovering around neutral values for both PC1 and PC2. Chelsea's offensive struggles and Brentford's attacking prowess illustrate how these factors can significantly impact a team's performance.

From Patterns to Progression: Tracing the Evolution of Playing Styles

Having discerned the clustering of teams based on their playing styles for each season, it becomes palpable that football, much like any other dynamic system, isn't static. Teams evolve, adapt, and sometimes, revolutionize their approach to the game. While the previous section provided a snapshot of teams' styles for individual seasons, the real intrigue lies in understanding the shifts over time. How have traditional powerhouses like Manchester City changed the landscape? And how have underdogs like Brighton adapted to these changes?

This next segment delves deep into the ebb and flow of playing styles in the English Premier League over the seasons. Through a lens focused on trends, we'll chart the transformative journey of teams.

Over the years, the most influential loadings for PC1 have consistently highlighted "possession" as a defining theme in team play styles. This trend was primarily characterized by data related to controlled possession, touches, progressive passes, and pass completion rates. However, a distinct shift was observed in 2019 and 2020 when "carries" also became heavily weighted, indicating a complementary or alternative approach to maintaining ball control. On the flip side, negative values for PC1 have traditionally been associated with defensive actions and more direct attacking strategies, manifested through metrics like tackles and long balls.

For PC2, the landscape of influential loadings has been relatively steady. Defensive actions consistently dominate the positive side, while attack efficiency metrics, as previously elaborated upon, hold strong positions in the negative spectrum of the most impactful loadings.

In 2017, Manchester City, under the tactical expertise of Pep Guardiola, presented a style of football that was distinguishably unique within the Premier League landscape. This distinctiveness was reflected in our clustering analysis where City emerged as a standalone cluster. Positioned significantly to the right on our PCA chart, their high positive PC1 value epitomized the quintessence of possession-based football.

Guardiola's philosophy, deep-rooted in controlled ball possession, progressive passes, and an intricate play style, set City apart. They were not just keeping the ball; they were strategically dominating spaces, making every touch count, and ensuring the opposition played largely as spectators.

However, as with all evolutionary phases in football, City's groundbreaking style didn't remain unchallenged or unemulated. And despite already having important possession, top clubs like Arsenal, Liverpool, and Chelsea were far in terms of controlled possession and pass completed. In 2017, the Citizens completed almost 6,000 more passes than Liverpool. Last season the gap was reduced by almost 50%. These teams’ PC1 values increased noticeably, indicating a greater emphasis on possession and ball control. At the same time, reductions in their PC2 values revealed modifications in other tactical areas, likely stemming from a blend of enhanced defensive solidity and attacking efficiency. The appointment of Mikel Arteta in late 2019, a former assistant coach at Manchester City, marked the start of a transformative phase for the Gunners. Yet, change takes time, especially in football. Adapting to a new tactical philosophy often requires not just adjustments in play but also targeted player recruitments to fit the evolving system. While Arteta's influence was evident from the outset, it wasn't until 2022 that Arsenal's stylistic shift was profound enough for them to share a cluster with City in our analysis.

By the time the next few seasons unfolded, the Premier League witnessed a convergence of its top teams towards the 'City blueprint'. The solitary Manchester City cluster of 2017 evolved into a more populated cluster, as Arsenal, Liverpool, and Chelsea nudged closer, reflecting the homogenization of top-tier tactical approaches.

This trend underlines the football’s adaptive nature. It showcases how successful strategies, once considered outliers, can become the benchmark, influencing and reshaping the tactical foundations of other elite teams. And what a better example than Brighton & Hove Albion Football Club?

Brighton's evolution in the English Premier League over the past years showcases a transformative journey in their style of play. Back in 2017, with a PC1 score of approximately -6, their tactics leaned heavily towards a non-possession based more direct approach, indicating a more defensive stance in matches. Such a strategy, although common for teams aiming to secure their position in top-flight football, saw them dangling perilously close to relegation, finishing a mere 7 points clear of the drop zone.

In 2017, Brighton prominently positioned itself in the top-right quadrant of the below chart, indicative of a team that relied heavily on long balls and frequent blocks. This spatial orientation aligns with a footballing philosophy that's both defensive and direct, hinting at a predominant counter-attacking style of play.

Fast forward to 2022, and Brighton's tactical metamorphosis is evident. With a PC1 score exceeding +8, their playstyle has shifted drastically towards a possession-oriented game, placing them amongst the league's elite in this metric, trailing only powerhouses like Manchester City and Liverpool. Such a drastic shift in strategy not only underlines their ambition but also brings tangible results as partially explained by a correlation of 0.93 between the PC1 value and points. As a testament to their evolution, Brighton secured a coveted European competition spot for the 2023 season, finishing an impressive 6th in the league. This ascent from near-relegation to European contenders within a span of a few years is a compelling narrative of strategic transformation, ambition, and successful execution on the pitch.

As shown in the above chart, only one team (Manchester City) had both higher controlled possession and passes completed last season.

To delve deeper into the intricate dynamics of team performance, we extended the analysis by adding a third dimension to our visualizations: pass efficiency (key PC2 loading). By juxtaposing teams in this 3D space, it's striking to observe Brighton prominently positioned alongside football giants like Liverpool last year. This positioning represents not only a testament to Brighton's significant improvement but also underscores a radical shift in their playing style over the past five seasons. Such dramatic transitions don't occur in a vacuum. In our upcoming section, we will peel back the layers to understand the catalysts behind this metamorphosis, giving special attention to key player contributions and the pivotal role of the coaching change.

Deciphering the Role of Individual Brilliance: A Deep Dive into Player Impact Analysis

The arrival of manager Potter in 2019 initiated whispers of change, yet when we dig into the numbers, the transition was not statistically remarkable until the 2021/2022 season. With the appointment of coach De Zerbi at the beginning of last season, Brighton embarked on a quantifiable transformation in their style of play.

The evidence is clear when observing key possession-based metrics of the midfield maestros: metrics like pass completion percentage gained almost 10% and when combined with increased possession, it naturally elevates the level of controlled possession (PC1). Key passes (leading to a shot), and touches/passes in the attacking third witnessed a more than 60% improvement.

However, while a manager's philosophy plays an undeniable role in steering a team's performance, individual prowess cannot be sidelined. Several players posted extraordinary figures last season, suggesting that individual brilliance may have amplified the effects of De Zerbi's managerial approach. In the ensuing sections, we shall delve into these standout performances to ascertain their role in Brighton's metamorphosis.

In the world of football, possession isn't just about having the ball; it's also about the players' ability to be available when off the ball. Being consistently present, ready, and offering a viable passing option is a skill in itself. When we dive deep into statistics regarding touches, entries into the attacking third, and passes received, two names from Brighton emerge distinctively: Groß and Caicedo, and we can tell they understood the assignment.

To ensure a holistic and comparable view of their performance metrics, we normalized the values. The results? Both Groß and Caicedo posted numbers that not only eclipsed Brighton's historical average but also bested most of their peers. Last season, only midfielders Rodri and Hojbjerg registered more touches and received more passes.

Diving deeper into their distributional acumen, Caicedo's pass completion rate of 88.5% is an accolade in itself. Only Manchester City's Rodri and John Stones (tactically repositioned at midfield) boasted higher numbers. Moreover, when it comes to driving the ball forward, all three Brighton midfielders under the scope of this analysis, found their names amongst the top 20 progressive passers in the league.

To complete our analysis, we decided to also look at defensive attributes. The role of a defensive midfielder, especially one playing in a double-pivot, usually focuses on breaking up the opposition's plays, protecting the backline, and also initiating the team's own plays. The chart below reveals an undeniable truth: Caicedo excelled phenomenally. He didn't just register commendable ball possession and play-creation metrics, but his defensive numbers were equally, if not more, impressive.

By juxtaposing Caicedo's figures against Brighton's historical averages and the elite midfielders of the Premier League, it becomes evident that he was pivotal in elevating the team's overall quality of play. With such immense talent and potential, it's hardly a shocker that Chelsea deemed him worth every penny of the whopping €116m they shelled out for his signature.

While the above obviously shows the enormous impact Mac Allister, Groß, and Caicedo had last season, it's essential to shine a spotlight on Kaoru Mitoma. While he's categorized as a forward in our dataset, thus not included in this specific analysis, his influence in Brighton's laudable success last season can't be understated.

In conclusion, our analysis unveiled the dynamic evolution of playing styles in the EPL over the past five seasons. It showcased how traditional powerhouses like the Manchester City of Pep Guardiola influenced the league's approach to football, with possession-based play becoming a benchmark. Notably, Brighton's transformation from a defensive-counter-attacking stance to a possession-oriented game, driven by key players like Moisés Caicedo, exemplified the adaptability and ambition of EPL teams. This study demonstrates in part the ever-changing nature of football and the pivotal role of key players in shaping success.

Disclaimer: The essence of data analysis is not to capture the entirety of the story but to piece together a narrative from the data fragments at hand. While the data utilized herein is robust and comprehensive, it's not exhaustive. Every statistical representation has its constraints. Our aim is to utilize accessible data judiciously, aiming for an honest and insightful interpretation, recognizing that there are always more layers to the story.