The Value of Coaching Tenure in the NBA
Headlines at the 2016 NBA All-Star Weekend were dominated by the flurry of head coach firings from the first half of the season. Much speculation surrounded the conference all-star team coaching assignments—a role traditionally bestowed on the coach with the best record through mid-season—since the Eastern Conference’s winningest head coach, David Blatt, had been fired just two weeks earlier. In an unusual turn of events, management handed Blatt’s replacement, Tyronn Lue—holder of an 8-3 career head coaching record—the reins.
By season’s end, 12 NBA teams (40% of the league) would make head coaching changes. Media outlets questioned the rationale behind the rapid turnover, pointing to the San Antonio Spurs’ unprecedented 20-year run of success under head coach Gregg Popovich. Anecdotally, the link between coaching stability and team success seemed clear, but in-depth analysis of the subject had yet to be undertaken.
What follows is a thorough examination of this link, culminating in identifying a relationship between head coaching tenure and team success, independent of all confounding factors.
To date, Popovich has amassed an extraordinary 1,089-485 regular season record over 20 years with the Spurs. During that time, he has also been fortunate to have a stable roster filled with future Hall of Fame players.
However, to understand how the length of his tenure affected his performance as a coach, one must evaluate each of his seasons individually. Only with distinct season-by-season data can performance trends be identified and studied. Doing so requires analysis of three key areas:
1. Player Talent—Has Popovich’s success been driven more by his talented player pool than by his coaching ability?
2. Roster Continuity—If Popovich’s players had been unexceptional individually, would their stability as a unit have allowed them to thrive even in the absence of a stable coaching situation?
3. Coaching Tenure—Has Popovich’s coaching itself improved over the course of his tenure, so that, over time, the Spurs have been able to achieve more with less?
An analysis of all NBA head coaches since the ABA/NBA merger in 1976 reveals that longer-tenured head coaches have won approximately six more regular season games each year than the 41-win league average. About two-thirds of this differential (~4 wins) is unrelated to coaching performance and, instead, the result of longer-tenured head coaches historically being endowed with superior talent and more stable rosters. But one-third of this differential (~2 wins)—often the difference between a playoff berth and a regular season exit—is a result of head coaching performance.
The Driving Forces
It’s long been visible to the naked eye that teams with longer-tenured head coaches outperform those with newly hired coaches. Teams with first-year head coaches have historically won 36 regular season games, while those with third-year head coaches have averaged 42 wins and those with fifth-year head coaches, 46 wins. Thereafter, improvement plateaus around 47 wins—six wins above the 41-win league average.
The results are unambiguous: Longer-tenured head coaches win more games, on average, than their shorter-tenured counterparts. The question then becomes what drives this trend.
This analysis considers the three primary drivers noted above:
1. Player Talent—The tendency for longer-tenured coaches to have higher-caliber players.
2. Roster Continuity—The tendency for longer-tenured coaches to simultaneously have longer-tenured players accustomed to playing together.
3. Coaching Tenure—The benefits directly attributable to head coaching longevity, be it in the form of player development, improved play calling, organizational stability, or otherwise.
The first two drivers are confounding variables: They inflate the observed outcome through correlation rather than causation. Isolating and examining the third driver, therefore, is the primary objective of this analysis.
Separating a player’s talent level from the influence of the coach is a challenging endeavor. It cannot be done perfectly, but this analysis creates a player performance projection model that seeks to do just that. The projection model takes an adjusted moving average approach to generate season-by-season player projections for each player in NBA history (more on this in the Methodology section).
Once each player’s talent level is isolated, we can explore the correlation between a team’s collective talent level and its head coach’s longevity.
As Table 1 shows, the quality of an NBA head coach’s player pool tends to increase with his tenure. Taken on the whole, the correlation between player talent and coaching longevity explains over 60% of the six-win differential by which longer-tenured head coaches have historically outperformed league averages.
Stable rosters often accompany stable coaching situations. Over the course of NBA history, longer-tenured head coaches have been approximately 50% more likely to enjoy roster stability than shorter-tenured head coaches. (For the purposes of this analysis, roster stability is defined as having completed three or more consecutive seasons with at least three of the same players having played 2,000 minutes during each of the three seasons.)
Over that same period, stable rosters have consistently outperformed unstable rosters. Interestingly, however, this outperformance is more muted among teams helmed by long-tenured head coaches. Roster continuity, it appears, offers far less significant a benefit in the presence of coaching stability. In total, the correlation between roster continuity and coaching longevity explains just 2% of the six-win differential (approximately 0.1 additional wins, as shown in Table 2).
Finally, after removing the confounding factors above, the true value of head coaching tenure can be uncovered. What follows is a rigorous analysis of how to evaluate coaching performance and how to apply this evaluation to coaching tenure. Once an objective measurement of coaching performance is defined, performance trends will be identified vis-à-vis a coach’s longevity with a team.
Because a coach’s day-to-day impact is, by itself, difficult to quantify, this analysis relies upon an expectation-based evaluation system—a head coach’s ability to overachieve what would be expected if the roster is treated as the defining measure of success (as it is more generally in sports and managerial economics, and as it often is among NBA Coach of Year voters as well).
To do so, this analysis relies upon a new, advanced statistic called Coaching Win Shares (CWS), which attributes team wins to coaching performance, much like traditional Win Share metrics do for players. (Other examples of player-level win valuation statistics include Wins Above Replacement Player [WARP], Value Over Replacement Player [VORP], Win Probability Added [WPA], etc.) A CWS of +2, for instance, implies that a coach was able to coax two more wins out of his team than would be expected of a team of comparable talent and continuity. Importantly, those two additional wins could be applied equally to a 20-win team (that has the talent level of an 18-win team) as to a 60-win team (that has the talent level of a 58-win team).
The CWS metric that drives this analysis is built from the ground up. It begins with Player Performance Projections for each player across all seasons of post-merger NBA history (Step 1). These player projections form the foundation for Team Win Projections, answering the key question laid out above: Given the player talent a coach has, how many games can he be expected to win (Step 2)?
The delta between team expectations and actual team wins then provides a preliminary view of head coaching performance (Step 3), before being adjusted for other minor confounding factors and finalized as the Coach’s Win Share (Step 4).
Ultimately, the relationship between the finalized CWS figures and coaching tenure can be analyzed, producing the core findings of this research (Step 5).
Here is a detailed, step-by-step breakdown of how to form and apply CWS figures to coaching tenure.
- Step 1—Player Performance Projections
Create player performance projections for all players based on a moving average of the player’s performance in the years surrounding the season in question. Then, scale projections by empirical inflation/discount factors to account for expected peaks/valleys during a player’s career (i.e., rookie and near-retirement seasons). Note that player performance projections are generated in terms of win contribution per minute of play, forming a normalized foundation on which to base the analysis.
- Step 2—Player Performance Projections Converted to Team Win Projections
Break down each team’s roster into its component parts—the number of minutes each of its players played during the regular season. Multiply these minutes by each player’s performance projection and aggregate them, forming a season-long Team Win Projection.
- Step 3—Team Win Projections vs. Actual Team Wins
Compare team win projections against actual team wins, with the residual representing a preliminary view of head coaching performance relative to expectations.
- Step 4—Residuals Adjusted and CWS Figures Finalized
Adjust residuals to account for roster continuity (discussed above) and strength of schedule (to compensate teams playing in more-difficult divisions and conferences), to produce a final CWS figure.
- Step 5—CWS Figures Applied to Coaching Longevity
With an objective measure of coaching performance established, trends can begin to be uncovered with respect to a head coach’s tenure with a team. This is the focus of the remainder of the analysis.
Analysis of the season-by-season, coach-by-coach CWS figures quickly reveals several interesting patterns. Among the highest CWS figures are a mixture of all-time great teams (Billy Cunningham’s 1983 76ers [+15 CWS], K.C. Jones’ 1986 Celtics [+15]) and unexpected overachievers (Jeff Hornicek’s 2014 Suns [+18], Jason Kidd’s 2015 Bucks [+11]).
Likewise, the lowest CWS figures are a mixture of all-time bad teams (Lawrence Frank’s 2010 Nets [-19], Brett Brown’s 2016 76ers [-13]) and disappointing underachievers (George Karl’s 2000 Bucks [-10], Kevin McHale and J. B. Bickerstaff’s 2016 Houston Rockets [-10]). Of course, the vast majority of CWS figures fall far closer to zero, forming a clean, near-normal distribution peppered with a handful of outliers.
Taken together, the data allow us to chart coaches’ progress over their careers and assess the relationship between tenure and performance. The results are striking. A near-linear relationship between a head coach’s tenure with a team and that coach’s CWS appears. As Figure 2 indicates, first-year head coaches have historically detracted value from their teams (on the order of approximately 1.5 wins), while fifth-year head coaches have added approximately that much value.
Long-tenured head coaches hover at or around a CWS of +2, a figure that explains approximately 33% of the six-win differential identified earlier.
Even more dramatic, however, are the results when one considers only those head coaches defined as long-tenured. This shrinks the sample dramatically from all head coaches in NBA history to those generally regarded as its strongest. We can then normalize the x-axis to reflect how far a coach is into his tenure in percentage terms, rather than absolute terms (i.e., a coach’s second season en route to an eight-year tenure would be categorized as 25% complete). Doing so reveals illuminating trends that are otherwise obscured when viewing coaching tenure in absolute terms.
The parabolic shape in Figure 3 reveals that even among the most successful coaches in NBA history, there is a clear learning curve effect. This elite cohort of coaches underperforms league averages early in their tenure and doesn’t reach full potential until nearly 40% of the way through their tenure. However, upon reaching potential, these head coaches enjoy a period of sustained success, with CWS values of approximately +2. The dip observed in the last 10% of a coach’s tenure reflects an attrition bias in the data set—namely, the tendency for coaching changes to follow shortly after performance begins to decline.
Finally, when we look at how players themselves perform under coaches of varying tenure levels, another interesting trend comes to the fore. Specifically, players who have played under multiple head coaches see an uptick in production when playing under longer-tenured head coaches, independent of all other factors. As Figure 4 shows, among the 948 NBA players who have played under at least five head coaches during their careers, we observe below- or near-average performances when playing under first-, second-, or third-year head coaches and performances of 7% to 17% above average when playing under coaches with at least four years of experience with a team.
The evidence is clear: Long-tenured head coaches have historically outperformed their shorter-tenured counterparts. This analysis has shown that player performance, team performance, and coaching performance all improve as a coach’s tenure grows with a team.
Yet, coaching tenure has changed little over the past 40 years. Average tenure has consistently hovered around three to four years, while the tendency for mid-season firings (the so-called “trigger finger”) has been relatively stable.
The smartest general managers will seek to find the right coaches and let them stay put. History has shown that even the most revered head coaches require a few years to settle in, but once they do, look out—despite only coaching 22% of all NBA seasons, long-tenured head coaches have accounted for 28% of all playoff appearances and 35% of all NBA championships.
Blinebury, Fran. 2016. Popovich Sets ‘Gold Standard of Consistency’ in Volatile Era.
Feigen, Jonathan. 2016. As Firings Mount Across NBA, Coaches Reminded of Job Insecurity. Houston Chronicle.
Oliver, Dean. 2004. Basketball on Paper: Rules and Tools for Performance Analysis. Washington, DC: Potomac Books.
Smith, Sekou. 2016. It’s Open Season on Coaches in NBA.
Windhorst, Brian. 2016. Short on Patience, NBA Owners have Coaches on a Short Leash.
About the Author
Brian Freilich is an MBA candidate at the Wharton School, University of Pennsylvania.
A Statistician Reads the Sports Pages takes a statistical look at sports. If you are interested in submitting an article, please contact Shane Jensen, column editor, at email@example.com.