SCROLL
Former NFL coach Jerry Glanville is attributed with the quote "Not For Long" when referring to an official's career, but the phrase is most often used to reference the turbulent change that occurs in the NFL.
When analysts and fans speak of a "championship window," they are referring to a short number of years in which they believe a team has the opportunity to win it all. These teams are constructed so that a core group of players peak in their careers simultaneously, strengthening the team beyond the norm. Roster continuity correlates with winning. But most players' careers are short. Their peak years are even shorter. Championship windows do not last long. Which teams are currently set up for roster longevity?
If we look at historical data by position we can see how many games a typical player starts at each age, often called an aging curve. We can use that data to estimate individual players' future games and visualize their peak years. For further details, see the methodology section.
This curve shows the average games started per season by age among quarterbacks who started at least one game in their career. The height of the chart represents 100% of games possible to start (16 for individual players). Tap on the chart to reveal the actual values.
Most quarterbacks don't start any games before age 23. They peak at age 25. This does not mean they are best at that age. It only means they are most likely to start games at age 25. This runs contrary to our intuition, because the players we think of are the prominent ones, who have played for many years. But they are rare. Many players are out of the league within a few seasons.
After that we see a long tail where quarterbacks slowly become less likely to continue playing.
As an example, let's look at the fictional quarterback, Steamin' Willie Beamen. Each darker bar represents the number of games started for a single season. Beamen didn't start at first, but successfully replaced an injured veteran and started a majority of the time in his two most recent seasons. The light grey bar indicates the unplayed current season, with no stats.
If the average curve is shifted over to align its age with Beamen's age, it becomes clear that he has been starting far more games than the average quarterback does at his age.
This circle marks the average starts for quarterbacks at Beamen's age last season. The curve creates a projection of how many games we can expect Beamen to start in future seasons, assuming he matches the QB average for his age. But this seems to underestimate Beamen, given his recent history...
Let's briefly ignore the positional average and instead find the weighted average of Beamen's actual starts. His first season is not as relevant as his most recent season, so we weight them accordingly. The earlier a season was, the less it factors into the weighted average.
To project future starts we can amplify the average starts curve so that it crosses through the weighted average we just calculated. This also shows how many games we expect Beamen to have started in the past. In reality, of course, individuals don't perfectly align with the average, but this does provide a more reasonable expectation of future seasons.
It may be helpful to think of this arc as the likelihood that Beamen will start on any given Sunday in a specific season. In the team charts below, expected starts are labelled as a percentage (out of the total number of games possible to start in a single season).
Projecting individual players relies on a small sample size of historic starts. However, by stacking multiple players on top of each other, we gain a larger sample size and can estimate starts for entire teams. This allows us to project how much of the current roster will remain on the team in the future. Don't confuse this for skill or value. It is simply likelihood to start games, which leads to roster continuity, potentially opening a championship window. Logically, we expect teams with young players who have started many games in their early careers to have many future starts. While we expect veteran teams with less historic starts to have few future starts.
The charts below are composed of the 22 starters for each team (11 offensive and 11 defensive). Starters are determined on the first game of the season (when they are least likely to be injured) and are assumed to be the starters for the remainder of the season. Regardless of which team a player was on earlier in his career, all starts count towards his current team's projections.
Use any team's button for a view of that team's individual players, sorted by tenure (number of consecutive seasons with at least one start with their current team), then by age. The table further below provides links to navigate to specific teams.
Most teams share a similar curve shape with a peak near the current year due to the brevity of the average player's career and the nature of injuries in a brutal sport. Also, the age a player is most likely to start games occurs surprisingly early in their career. Most players are thus on a downward trajectory. Players and team rosters simply don't last for long. The teams at the top of the list below are set up for the most roster continuity and, perhaps, a championship window.
| Rank | Team
▲
▼
|
Next season
▲
▼
|
Next five seasons
▲
▼
|
Next ten seasons
▲
▼
|
|---|
Positional averages are calculated from all retired players who began their career after 1980 and started at least one game.
An interesting aside is that if all players are included in the calculations (rather than limiting to players who started at least one game), the curve shape of the average career is nearly identical.
Team rosters are determined by who started in the first game of the current season.
Means (averages) are used rather than medians, because the median games started for each position, at virtually every age, is zero.
Weights are assigned such that the most recent season's coefficient is 0.7, the next is 0.2, and the third most recent is 0.1.
Players who have played for less than three years have their weights adjusted so that the most recent season has more weight. Players with no starting history, such as rookies, are automatically assigned 6 expected games started for the prior season. Future seasons are projected from that point.
All projections are, of course, 100% accurate predictions of the future.