| Player Count 2014-2024 | |
|---|---|
| count | 352.0 |
| mean | 67.2 |
| std | 6.5 |
| min | 54.0 |
| 25% | 63.0 |
| 50% | 67.0 |
| 75% | 71.0 |
| max | 91.0 |
[Whitepaper] Homegrown Picks: The Relationship Between Player Origin and Winning
Abstract
Building an NFL roster through the draft is a narrative of yore [It’s from yore. Like the days of yore, you know.]. Here we evaluate player origin and quantify its relationship to winning. Analysis of the 2014-2024 NFL seasons reveals several key findings. First, when we setup contribution of homegrown players and winning percentage we see only moderate predictive ability but there is indeed a string relationship between the two variables. Second, when we bin teams by winning percentage, winning teams consistently show higher contribution from homegrown players than losing teams. This disparity is more pronounced between the top and bottom quartiles. Third, when we bin teams by the top and bottom quartiles of contribution of homegrown players, there is a clear difference in team success as measured by winning percentage, playoff appearances, and playoff advancement. Additionally, teams in the top quartile have significantly longer tenures for both head coaches and general managers compared to the bottom quartile. Lastly, when this unlabeled data is processed by a basic clustering algorithm, the procedure clearly aligns higher percentages of homegrown players with more successful team outcomes. These findings suggest the contribution of homegrown players is a foundational element to building winning teams and that teams must dedicate resources to developing players and increasing collective intelligence.
Background
Players on an NFL roster are acquired with draft picks or cash. For the latter, those players are either undrafted or drafted elsewhere. There is the special case of the traded player, which usually costs some blend of draft picks and cash. But ultimately this player is either undrafted or drafted elsewhere. And so in the dimension of player origin there are only three types of players on an NFL roster: homegrown pick, elsewhere pick, or undrafted.
It doesn’t take a George Young to grasp draft success means team success. In the extreme, if a team could draft nothing but all-pros then a team would do nothing but win. Additionally, the NFL’s collective bargaining agreement restricts drafted-player salaries. Thus drafted players are not subject to bidding and are cheaper. In a hard salary cap league, the more cheap players you have on your roster the more resources you have to increase your talent with overpriced free agents, disproportionate trades, and extending existing players.
Historical draft hit rates suggest this is not an easy solve. The are multiple criteria by which to evaluate draft success rates. But across any dimension, these rates are at best a coinflip. As it relates to the historical ranking ability of the NFL the probability that the player drafted before another player will start more games is 52% (Thaler 2015). The percentage of first round draft picks that sign a second contract with their drafted team is 43% (Risdon 2024). Below we show that the average amount of homegrown picks who simply play in a game is 40%.
While there exists a material number of NFL owners who prioritize their revenue share over winning, the seven-figure salaries and short tenures of losing general managers (GM) indicate winning remains the incentive. And that solving the draft would come with substantial reward. If not at the GM’s current team then certainly another. Like not even the most Neanderthal of turf wars or fog of groupthink could prevent draft solutions from being known. Since most draft programs are failures, we must assume a closed-form solution does not exist.
Instead, populating the roster with as many homegrown picks as possible is likely a multivariate problem. The solution for which requires incremental gains across many factors. Picking the right players matters. But to assess draft success in this binary fashion implies that prospects are either made for the NFL or they are not. Getting the correct player in the door is the first step. From there, the player should be considered a human element in a talent pipeline, so that other factors might come into play and affect the success of the pick. Certainly, the existence of these factors should be identified and evaluated so as to optimize their effect.
Methods
We collect games-played and snaps-taken for each player for every team for the seasons 2014-2014. We compute total and percentage of games-played and snaps-taken for each player type: homegrown picks, elsewhere picks, and undrafted. This includes the postseason. A team’s season is summarized by a single row. For the seasons 2014-2024 we have 352 rows.
For games-played, we collect the list of any player who played in a game for a team during a season. This includes players that played for multiple teams. We know the origin of these players and simply count the three player types then compute a percentage.
For snaps-taken, this of course must match the games-played length. But here we assess contribution by totaling snaps. A snap is a single play. Note there are 11 positions to fill on each play. So all percentages are not complimentary. We simply sum the total offensive, defensive, and special teams snaps to gets total snaps for a season. We sum the total snaps for each player. We partition by player origin, sum all snaps in each partition, then divide by total snaps for a season. Snaps ends up being tricky. There are in fact 11 opportunities to get a single snap so none of these percentages sum to 1. Additionally special teams snaps adds another layer. We try to simplify by dividing snaps-taken by all snaps available, the end.
This snaps approach likely penalizes injuries. Consider a bad-luck season where several high draft picks are injured and did not play a game. Using the games-played dimension might solve for partial seasons lost to injury, but not entire seasons. And snaps-taken will underweight the existence of valued draft picks unable to play for extended periods. An alternative could be counting the origin of all players who were simply on the roster. But this penalizes for roster churn. Roster churn is the lower end of the roster is turned over due to underperformance or injury. The positive or negative value-add not evaluated here. But we could imagine any number of scenarios where a percentage of the roster is continuously turned over and we would have zero knowledge as to why. In this approach we would count the valued but injured homegrown picks but also the murky roster churn. We consider this is a suboptimal tradeoff and use games-played and snaps-taken.
We collect the winning percentage for each team’s row. We compute winning percentage in the typical fashion which is the sum of total wins plus half the total ties, divide that sum by total games. Where total games is 16 regular season games preceding 2021 and 17 thereafter. Total games include the postseason.
We collect playoff depth for each team’s row. For the seasons 2014-2024 there were four rounds in the playoffs: wildcard, divisional, conference championship, and league championship. Each round counts as one. For seeds with a first-round bye we count the bye as a game played. If a bye-team makes it to the league championship its playoff depth is four. If a team loses in the wildcard round its playoff depth is one.
We collect head coach and general manager tenure for each team’s row. For both head coach and general manager, we assign to the season the tenure that existed for that season. Kyle Shanahan and John Lynch were hired as head coach and general manager of the 49ers in 2017. For the 2017 season these tenures are listed as one in the team’s row. By the 2024 season those tenures were eight in the team’s row.
Head coach tenure is easily-defined except for years when the head coach was fired during the season. These instances include an interim head coach. That year is credited to the fired coach i.e. the interim is ignored. Additionally, repeat coaches with the same team are not a continuum. Jon Gruden coached the raiders 1998-2001 and 2018-2021. These are considered two separate tenures of four tears.
There is likely some qualitative effect with our general manager tenures. Public data may show concurrent executive tenures that overlap with the official GM. This executive my even be the head coach. Whenever we programmatically list the GM by year and we get data with more rows than years, we evaluate. For example in 2015 Howie Roseman’s tenure as the official GM for the Eagles was interrupted by head coach Chip Kelly. Kelly gets credit for GM in 2015 and his tenure for that season is one. But the next season we pick up Roseman’s tenure in a continuum because Kelly was fired and Roseman hasn’t left the building since he became official GM in 2010. There are similar discrepancies between official GM titles and other executives between 2014-2024 with the Panthers, Browns, Commanders, and Seahawks.
From here we observe distributions, use ordinary least squares regression, partition by various dimensions and obtain intra-partition statistics. All of which can be accomplished in any software environment.
We also deploy a basic k-means clustering algorithm. We use the “KMeans” module in the scikit-learn library (Pedregosa et al. 2012). We first standardize the data by z-score. We assess the inertia of one through 20 clusters. Using the elbow method, inertia appears to flatten at five clusters. We reverse standardization to express results in native units.
Results
In Figure 1 and Table 1 below we show the distribution of the number of players who played games and took snaps for an NFL team during a season ( datasets are the same length.) From 2011-2019 there were 46 active players allowed on gameday. This increased to 48 in 2020. The average and median number of players is 67 and 67.2, respectively. So on average there is this 20-or-so player churn across a season. This means not even a well-drafted healthy roster core can cover all games (let alone snaps) across a season. And when we do consider injuries and perhaps the fungible nature of “replacement players” i.e. the bottom three at the respective postion (James 1988) teams appear to move players in and out of this tail of the roster quite frequently.
From a player-origin perspective, Figures X and X and their corresponding tables show who is playing games and who is getting snaps. For games, we can see homegrown picks play the most games followed by undrafted players and then elsewhere picks. These are somewhat bunched distributions with visible overlap.
Much different story for snaps. Figure 3 below shows homegrown picks get the most snaps followed by elsewhere picks. On average this is a 2:1 ratio. These two distributions show visible dispersion i.e. beyond the averages this ratio can vary. Undrafted players get the least snaps and we can see this is a bit tighter distribution i.e. undrafted players get the same amount snaps across the league. Clearly, drafted players do most of the work in the NFL. There is a material amount of undrafted players that dress and play on gameday. But when it comes to actual contribution some form of the drafted player averages nearly 80% of the snaps.
| Homegrown Pick Games Percent | Elsewhere Pick Games Percent | Undrafted Games Percent | Homegrown Pick Snap Percent | Elsewhere Pick Snap Percent | Undrafted Snap Percent | |
|---|---|---|---|---|---|---|
| count | 352.0 | 352.0 | 352.0 | 352.0 | 352.0 | 352.0 |
| mean | 40.1 | 27.6 | 32.3 | 51.3 | 27.5 | 21.3 |
| std | 8.4 | 7.3 | 6.4 | 9.5 | 8.8 | 6.1 |
| min | 18.8 | 3.4 | 13.0 | 28.1 | 3.9 | 6.4 |
| 25% | 34.3 | 23.0 | 27.8 | 45.0 | 21.8 | 17.3 |
| 50% | 40.0 | 27.7 | 32.4 | 50.9 | 27.1 | 20.8 |
| 75% | 45.2 | 32.4 | 36.5 | 57.5 | 33.7 | 25.4 |
| max | 70.4 | 47.3 | 52.9 | 80.1 | 56.6 | 39.0 |
Now we begin to formulate winning percentage as the dependent variable. In Figure 4 below we show the correlation matrix of player origin variables and winning percentage. We can look at the winning percentage row/column and see its relation to player origin variables are moderate, with percentage of homegrown picks games played being the strongest. We can also see that the game-played and snaps-taken variables are about the same correlation. And they have the same relationship pattern to winning percentage: homegrown picks are positive, elsewhere picks and undrafted are negative with a bout the same magnitude. Additionally we can see the indirect relationship homegrown picks has to elsewhere picks and undrafted, since these would be approximate compliments.
Now we formulate winning percentage more formally as a response variable to player origin. We can see in the correlation matrix there is likely multicollinearity among the player origin variables and so we compute pairwise ordinary least squares (OLS) regressions. We show selected regression statistics in Table 3 below.
| R-Squared | Beta Estimate | Beta t-Stat. | Beta p-Value | |
|---|---|---|---|---|
| Homegrown Pick Games Percent | 0.113 | 0.0076 | 6.6648 | 0.0000 |
| Elsewhere Pick Games Percent | 0.033 | -0.0047 | -3.4516 | 0.0006 |
| Undrafted Games Percent | 0.054 | -0.0069 | -4.4667 | 0.0000 |
| Homegrown Pick Snap Percent | 0.088 | 0.0059 | 5.8016 | 0.0000 |
| Elsewhere Pick Snap Percent | 0.030 | -0.0037 | -3.2702 | 0.0012 |
| Undrafted Snap Percent | 0.045 | -0.0066 | -4.0494 | 0.0001 |
The direction and magnitude of the betas reflect the same pattern as the correlation matrix: more homegrown players more winning; more elsewhere picks and undrafted players more losing. But the r-squared suggests a cloudy model. The proportion of variance in winning percentage is not well-explained by player origin. However all betas are statically significant. Although the model doesn’t explain a large portion of the total variance in the dependent variable, the independent variables are significantly related to the dependent variable, even if the model’s overall predictive power is limited. This situation can occur when the model captures a significant portion of the relationship between the predictors and the outcome, but other factors also influence the outcome. This is certainly something we would expect in a simple OLS model that attempts to predict the entire goal of a sport: winning is a multivariate problem. An example of a strong relationship but low explanatory power is illustrated in Figure 5 below.
The scatter has an up and to the right pattern, but we can see plenty of counterexamples dragging the regression metrics. Of note are the 2015 Panthers who played a low percentage of homegrown picks but won a lot of games and ultimately lost Super Bowl 50 to the Broncos. This speaks to head coach Ron Rivera doing more with less and he won coach of the year in 2015. The 2020 Bills and Chiefs are similar outliers. It’s notable that 2020 was the COVID season when players were deactivated in swaths. Conversely, we see a lot of the Bengals seasons in the high percentage of homegrown picks but low winning record. Ever terminal, this of course jibes with the Bengals generational reputation. Owned and operated by people who inherited the team from their parents, the Bengals are known for its frugal player contracts. It is interesting to see how this might express itself in Figure 5 as a team clinging to drafted players simply because they wont pay for the alternative. If they’ve been drafting well, they field good players and win. If they’ve been drafting poorly, they field bad players and lose.
Next we bin by various dimensions. We first sort by winning percentage. We partition by winning and losing teams and top and bottom quintiles. Table 4 below shows the average value for each player origin variable in in each bin.
| Winning Teams | Losing Teams | Top Quintile | Bottom Quintile | |
|---|---|---|---|---|
| Homegrown Pick Games Percent | 42.5 | 37.4 | 43.5 | 35.8 |
| Elsewhere Pick Games Percent | 26.5 | 28.8 | 25.9 | 29.1 |
| Undrafted Games Percent | 31.1 | 33.8 | 30.6 | 35.0 |
| Homegrown Pick Snap Percent | 53.6 | 48.7 | 54.5 | 46.5 |
| Elsewhere Pick Snap Percent | 26.2 | 28.9 | 25.6 | 29.9 |
| Undrafted Snap Percent | 20.2 | 22.5 | 19.8 | 23.6 |
Again, homegrown picks appears to be the most relevant variable, as the differences between these partitions are the most stark. The difference between the percentage of homegrown picks that play in games between winning teams and losing teams is 5.1% or about 3 players. The difference between the top and bottom quartiles is 7.7% or about 5 players. This estimate is derived from 1 divided by 67.2, the average number of players to play in at least one game. And so one player is about 1.5%.
For snaps, the difference between the percentage of homegrown pick snaps between winning teams and losing teams is 4.9% of available snaps. The difference between the top and bottom quartiles is 9% of snaps. Our percentage of snaps-taken variable is based all available snaps for the season. The average by teams across 2014-2024 is 2,629. And so 4.9% translates to 129 more snaps and 9% is 237 snaps. Average snaps per game across the period is about 160. Divide that by two to geta rough estimate (no one plays both ways, special teams is an unknown) of game units. And so 4.9% translates to about 1.5 more games worth of snaps and 9% is about three more games worth of snaps.
Next we sort by the homegrown picks variables, both games-played and snaps-taken. We partition by the top and bottom quintiles. We want to summarize the success of each bin. The results in Table 5 below show winning percentage, playoff appearances, semi-final appearances, and finals appearances. Success is lopsided in favor of the top quintiles.
| Count | Winning Teams | Playoff Teams | Conference Title Games | League Title Games | |
|---|---|---|---|---|---|
| Top Quintile Percent Homegrown Pick Games | 71 | 50 | 41 | 10 | 5 |
| Bottom Quintile Percent Homegrown Pick Games | 70 | 18 | 17 | 4 | 1 |
| Top Quintile Percent Homegrown Pick Snaps | 71 | 46 | 40 | 10 | 6 |
| Bottom Quintile Percent Homegrown Pick Snaps | 70 | 16 | 15 | 3 | 1 |
Next we show two program characteristics of the same quintiles. In Table 6 below we show the average and median tenure of the general manager and head coach when those percentage of homegrown pick seasons occurred. Like team success, head coach and general manager tenure overwhelmingly favors the top quintiles.
| Count | Average Head Coach Tenure | Median Head Coach Tenure | Average General Manager Tenure | Median General Manager Tenure | |
|---|---|---|---|---|---|
| Top Quintile Percent Homegrown Pick Games | 71 | 7 | 7 | 13 | 13 |
| Bottom Quintile Percent Homegrown Pick Games | 70 | 3 | 2 | 4 | 3 |
| Top Quintile Percent Homegrown Pick Snaps | 71 | 7 | 6 | 12 | 11 |
| Bottom Quintile Percent Homegrown Pick Snaps | 70 | 2 | 2 | 4 | 3 |
Lastly we do a quick k-means cluster analysis. The procedure partitions five clusters from unlabeled data. This matters because we are asking the procedure to idetify clusters from unlabeled data. Wheares our above partitions are labeled by win/loss and quintiles. The entries in Table 7 below show the average value of constituents in each cluster (excluding cluster count) as detrmined by the \(k\)-means algorithm.
| Cluster 1 | Cluster 2 | Cluster 3 | Cluster 4 | Cluster 5 | |
|---|---|---|---|---|---|
| Homegrown Pick Snap Percent | 52.5 | 62.0 | 51.6 | 49.7 | 38.8 |
| Elsewhere Pick Snap Percent | 27.8 | 19.2 | 31.0 | 22.6 | 38.7 |
| Undrafted Snap Percent | 19.7 | 18.8 | 17.4 | 27.6 | 22.5 |
| Win Percentage | 74.4 | 56.6 | 50.9 | 38.6 | 32.1 |
| Playoff Success (Round) | 3.1 | 0.9 | 0.4 | 0.2 | 0.1 |
| Tenure Head Coach | 6.0 | 10.0 | 2.7 | 3.3 | 2.8 |
| Tenure General Manager | 7.3 | 17.4 | 5.0 | 5.5 | 3.8 |
| Cluster Count | 82.0 | 88.0 | 56.0 | 58.0 | 68.0 |
From Table 7 we note the following:
Winning percentage. The algorithm clearly sorts winning percentage. In Table 7 we ordered the columns in descending order of winning percentage and there appear to be five kinds of winners. One average team, one above average team, one below average team, an extreme winner, and an extreme loser.
Cluster 1, the extreme winner. Winning over 70% translates to over 12 games. Cluster 1 also has by far the most playoff success. However it shows an unexpected player origin and tenure pattern: it has lower homegrown picks and tenure than the next winningest cluster and a higher percentage of elsewhere picks than the next winningest cluster. This contrasts with the somewhat descending order of the other clusters.
Clusters 2-5. These clusters show a pattern more consistent with the partition tests above: homegrown picks and tenure descend with winning percentage. Note there are varying approaches to allocating to other player origins: cluster 2 splits those allocations evenly, cluster 3 descends like clusters 1 and 2 (but has less winning), and cluster 4 overweights to undrafted players.
Cluster 5, the extreme loser. Winning 32% translates to just over five wins a season. Indeed these teams are cellar dwellers. Bolshevik bridge trolls whose primary directive is collecting the revenue share. Thet can’t develop their picks to play more than 40% of snaps so they make up for it by signing elsewhere picks. So much so that homegrown and elsewhere picks are equal allocations. Considering the public antics of the NFL’s perennial losers, this picture of discombobulation comes as no surprise.
Comparing clusters 1 and 2. Cluster 2 appears to be playing by all the rules and it shows in its moderate success. It wins more than nine games and nearly averages a wildcard appearance. This cluster has by far the longest tenures, which suggests consistency and established culture. It allocates the highest percentage of snaps to homegrown picks. Cluster 1 has these characteristics too, but to a lesser extent. Which contrasts with the somewhat descending order of clusters 2-5. Cluster 1 suggests teams with bunches of elite talent—likely at quarterback—that can tilt the field of play. Cluster 1 likely doesn’t cling to doubtful draft picks because it doesn’t have to. Perhaps it can lowball elite free agent talent that just wants to win. Perhaps it can elevate mediocre free agent talent. Perhaps it scheme to its elite talent knowing it will make up for known deficiencies. Cluster 2 suggests a team that is missing this elite talent and instead must try to win by not doing dumb things. And this approach trumps all clusters but one.
Discussion
Developing, retaining, and getting homegrown picks to contribute to gameday production is a key factor in NFL team success. When we set up seasonal winning percentage as a response variable to homegrown picks in a linear model, there is a strong relationship but limited linear explanatory power. More importantly we find that higher rates of homegrown picks are associated with higher winning percentages, more playoff success, and longer tenures for both head coaches and general managers. The rarity of counterexamples in these partitions suggests that emphasizing homegrown picks is a critical ingredient to success.
There is an expression among MBA types that says, “you can’t manage what you don’t measure.” We must find value in the actual metrics that express heuristics. In 2024 the bottom five teams in percentage of homegrown picks snaps-taken won 8, 5, 3, 5, and 13 (Vikings) games. The top five won 10, 12, 11, 8, and 15 (Chiefs) games. Super Bowl winner Eagles were number nine on the homegrown picks list. Here we can measure and monitor a factor that fuels success. Here we can identify the organizations that value winning and those that value being professional barnacles. We can spot high performing front offices and we can spot the millionaire gym teachers.
There are many characteristics we can observe about human beings in teams and teams in competition. These do not require quantitative screeds and are available upon inspection. Two are relevant here.
The first characteristic is individual greatness breaks the numbers. The greatest of all time distort broad analysis. They literally change the average. We can see this when inspecting clusters 1 and 2 above. Cluster 1 clearly benefits from an elite factor that exempts it from some of the challenges faced by cluster 2. It is overly successful but doesn’t perfectly align with the broad pattern. And so as we process our results we should expect idiosyncratic human factors to affect the systematic quantitative analysis.
The second known characteristic is the magic of team. It’s literally the motto of the United States Marines: Esprit de corps (Stallard 2018) the spirit of the body, the spirit of one . We can thumb through the pages of sports history and collect repeated examples when the best team–not the best collection of talent–won. The magical team factor might be difficult to quantify but there is extensive study on complex adaptive systems and their properties (Mitchell 2009).
Teams of human beings can behave as complex adaptive systems. A property of complex adaptive systems is emergent behavior. Emergent properties are characteristics that materialize only when system components interact, transcending the capabilities of individual parts. Such properties demonstrate that interconnected behavior can be fundamentally different from nodes in isolation.
An example of emergent behavior is the collective phenomenon of starling murmuration. When thousands of starlings fly together, they create patterned, fluid shapes in the sky without central coordination. Each bird follows simple rules of maintaining a certain distance from its neighbors and matching their velocity. From these individual interactions emerges a complex, seemingly choreographed aerial display that can change direction and aesthetic instantaneously. Other well-known examples include ant colony behavior from individual interactions, economic market behaviors from individual transactions, and white blood cells in the human immune system. That an individual may perform better in high functioning teams is likely no surprise to coaches and people leaders.
With the above in mind our results suggest the following two imperatives.
The first imperative is teams must commit to drafted player development. GOAT draft picks happen, and we have shown true talent can break the numbers, but solving the draft means having more homegrown picks on the roster. Drafted player development should be holistic. Teams should believe that NFL players can be made. To some extent that’s what every team says every year. But it doesn’t happen. Teams need to fully assess the value they add to player development.
At minimum, teams need to consider the opportunity cost of failing to develop a drafted player. Teams get seven draft picks a year. It’s possible to materially increase this amount but the moves required to do so are too extreme. Instead, the number of rounds, length of contracts, and salary cap imply a talent pipeline that highly sensitive to even a single failure.
A baseline scenario would be a team adds seven draft picks a year and signs them for four years. It does this every year for four years. Best case scenario it now has 28 homegrown picks on the roster before contracts start to roll off. We might multiply this number by failure, injury, development, or contract-extension rates. But without modifying it we could shoot for a ratio like 28 players divided by the league average of 69 players that get in games equals 41%. Table 4 above shows the average number of homegrown picks in the top and bottom winning quintiles is 44% and 36%, respectively. If we wanted to shoot for 44% that’s 30 players. If we wanted to bump it to 50% that’s 35 players. The difference between 44% and 36% is about six players.
And so every pick matters. Players are drafted, developed, retained or roll off, and the next cohort comes in behind. Teams must stay on schedule. Every jettisoned draft pick must be replaced with an expensive alternative: another undeveloped draft pick that requires time and resources to develop, an expensive free agent, or an inexpensive untalented free agent. And as we have shown, each alternative adds to an aggregate associated with less winning.
The second imperative is teams must commit to building football programs with stable culture. Here it is easy to borrow from business and corporate studies to confirm the effects of stable culture. A stable corporate culture means an enduring set of organizational values, beliefs, norms, and practices that persist over time. Some defining characteristics of stable culture are institutional memory, leadership consistency, and reliable expectations (Schein and Schein 2016; Cameron and Quinn 2011). Some benefits of stable culture include operational consistency (Chatman and O’Reilly 2016), knowledge retention and transfer (De Long and Fahey 2000), reduced transaction costs (Guiso, Herrera, and Morelli 2016; Kreps 1990), role clarity and efficiency(Cameron and Quinn 2011), coordination advantages (March and Simon 1993; Cremer 1993), and predictable decision making (Tushman and O’ReillyIII 1996; Bettenhausen and Murnighan 1985). It is unclear what healthy versus unhealthy corporate culture might be. Here we are simply discussing stable corporate culture.
The prevalence of homegrown picks on successful NFL rosters serves as both an indicator and reinforcing catalyst of stable cultures, potentially facilitating emergent team behaviors. To us this reads as not only more homegrown picks on the roster, but homegrown picks that outperform projected ability.
Teams maintaining higher percentages of drafted talent create environments where players develop within consistent systems over multiple seasons, enabling complex interactions and shared understanding. This stability–evidenced by the significantly longer leadership tenures we observed in high-retention teams–creates conditions where collective intelligence can emerge organically through thousands of repetitions and interactions. Just as complex adaptive systems research demonstrates that stable rule sets and consistent component interactions enable emergent properties, teams with stable cultures and retained homegrown talent may develop sophisticated capabilities that emerge only through shared time in a consistent program. The relationship between homegrown picks and winning suggests these teams benefit from emergent cooperative behaviors that materialize when players internalize systems, develop implicit communication patterns, and construct shared mental models. Abilities that cannot be derived through mindless talent acquisition.
References
Bettenhausen, Kenneth, and J. Murnighan. 1985. “The Emergence of Norms in Competitive Decision-Making Groups.” Administrative Science Quarterly 30 (September).
Cameron, K. S., and R. E. Quinn. 2011. Diagnosing and Changing Organizational Culture: Based on the Competing Values Framework. BusinessPro Collection. Hoboken: Wiley.
Chatman, Jennifer A., and Charles A. O’Reilly. 2016. “Paradigm Lost: Reinvigorating the Study of Organizational Culture.” Research in Organizational Behavior 36: 199–224.
Cremer, Jacques. 1993. “Corporate Culture and Shared Knowledge.” Industrial and Corporate Change 2 (3): 351–86.
De Long, David W, and Liam Fahey. 2000. “Diagnosing Cultural Barriers to Knowledge Management.” Academy of Management Perspectives 14 (4): 113–27.
Guiso, Luigi, Helios Herrera, and Massimo Morelli. 2016. “Cultural Differences and Institutional Integration.” Journal of International Economics 99: S97–113.
James, Bill. 1988. The Bill James Baseball Abstract. New York: Ballantine Books.
Kreps, D. 1990. “Corporate Culture and Economic Theory.” In Perspectives on Positive Political Economy, 90–143. Cambridge University Press.
March, James G, and Herbert A Simon. 1993. Organizations. Hoboken: John wiley & sons.
Mitchell, Melanie. 2009. Complexity: A Guided Tour. New York: Oxford University Press, Inc.
Pedregosa, Fabian, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, et al. 2012. “Scikit-Learn: Machine Learning in Python.” CoRR abs/1201.0490. http://arxiv.org/abs/1201.0490.
Risdon, J. 2024. “First-Round Draft Hit Rates: Tracking Rates from the Best (Centers) to the Worst (WRs).” 2024. https://draftwire.usatoday.com/story/sports/nfl/draft/2024/04/25/first-round-draft-hit-rates-tracking-rates-from-the-best-centers-to-the-worst-wrs/73449048007/.
Schein, E. H., and P. A. Schein. 2016. Organizational Culture and Leadership. 5th ed. The Jossey-Bass Business & Management Series. Hoboken: Wiley.
Stallard, Capt. D. 2018. “Esprit d Corps: Morale and Force Preservation.” Marine Corps Gazette, 78–82.
Thaler, R. H. 2015. Misbehaving: The Story of Behavioral Economics. New York: WW Norton.
Tushman, Michael L., and Charles A. O’ReillyIII. 1996. “Ambidextrous Organizations: Managing Evolutionary and Revolutionary Change.” California Management Review 38 (4): 8–29.