The Big 5 UEFA teams that qualified for the 2018 World Cup arguably contained some of the very best players in Europe. Their average age was 26.9 years; in the 2014 World Cup it was 27.2. and it was 27.5 and 27.4 in 2006 and 2002 respectively.
In recent seasons, player ages in the corresponding domestic leagues have been hovering around these values. The chart below show the average age within the starting elevens for teams in their respective domestic competitions, and we can see that player ages have stayed pretty close to 27 at least since since 2009.
Kalen and his coworkers found that the market value of Champions League players – and by implication their performance – reaches a maximum for players in the age band 26-30. We can conclude that 27 year-old players would be at or near their athletic peak and at their most sought after and expensive.
Recently Sefir Dendir looked at the performance-age relationship for football players using a number of different methods; using a random effects model (perhaps the most appropriate), he concluded that forwards peak around just over 25 years, midfielders around 26.7 years and defenders around 27 years.
If there are well-defined peak age bands for players, it is reasonable to ask if there are similar peak ages for teams. So in this post I will be looking at the performance-age profiles of teams rather than individuals. The data is drawn from matches in the the Big 5 European leagues between 2009 and 2017. The chart below shows the distribution of starting 11 ages.
Is There a Peak Age for Teams?
Given that individual players peak around 27 years old, we might suppose there is some kind of peak age for teams. I explored this possibility by regressing team performance indicators on various functions of team age (i.e. the average age of the starting 11). There were two levels of analysis, match and season. The dependent variable for the match was the goals scored by home team minus the goals conceded by the home team which I called the home team goal advantage, and the dependent variable for the season was the points.The regression equations are shown below.
In (1) ij represents a match between the home team (i) and the away team (j). This is the match level of analysis. In (2), is represents squad i in season s. The quadratic terms for age are included to capture the peak age if it exists;
It turned out there is no discernible effect of team age on performance. Young, middling and old teams perform at the same level. I also tried regressing the performance indicators on the number of players between 26 and 28 in the starting 11. This was intended to identify the number of 27 year-olds. But once again there was no discernible effect on performance.
Effects of Age profile on Performance
There was however, one age-related variable that did seem to affect performance, and that was the spread of ages within the team. The effect was small to moderate in size, but statistically highly significant.
The regression equation I used to estimate the effects of spread on performance was :
where the right-hand terms represent the standard deviations of the ages of the home and away teams respectively. Table 1 shows the regression results.
Table 1. Regression results
|SD Home team ages||-0.087||0.017||-5.13||< .0001|
|SD Away team ages||0.068||0.017||4.00||< .0001|
The negative coefficient for the spread of home team ages means that when there is a wide disparity in player ages, the goal difference in favour of the home team is reduced. Conversely the positive coefficient for the away team means that a wide disparity in player ages increases the goal difference. Put another way, the home team performs best when it has a narrow spread of ages within the team, and when the opposition has a wide spread of ages.
The chart below shows the effects for teams with low, medium and high spreads.
Looking for example at the first panel, we can see that when a high spread home team plays a low spread away team, the goal difference in favour of the home team is 0.25. Looking at the third panel however we see that when a low spread home team plays a high spread away team, the goal difference increases to 0.55. Calculations show that over a season, the goal difference of a team with a low age spread will be about 6-7 goals better than the goal difference of a high age spread team, other things being equal.
Another way to look at the same thing is in terms of the youngest and oldest players in the team. The regression model for this analysis is
Table 2 shows the regression results.
|Age of Youngest player home team||0.025||0.008||2.998||0.003|
|Age of Oldest player home team||-7.635||1.836||-4.158||<. 0001|
|Age of Oldest player home team (squared)||-5.358||1.807||-2.965||0.003|
|Age of Youngest player away team||-0.021||0.008||-2.462||0.014|
|Age of Oldest player away team||6.368||1.833||3.475||< .001|
|Age of Oldest player away team (squared)||3.419||1.807||1.892||0.058|
The coefficient for the age of the youngest player in the home team is positive. This means that as the age of the youngest player increases, the home team goal advantage increases. Similarly, the coefficient for the age of the youngest player in the away team is negative. This means the home team advantage increases as the youngest player in the away team gets younger. In other words fielding young players puts teams at a disadvantage.
The significant coefficients for the quadratic terms indicate that the age of the oldest player in a team has a non-linear effect on the home team goal advantage.
The effects are illustrated in the chart below, where goal differences are plotted for 3 selected values of youngest player. We can see that the home team goal advantage is highest when the youngest player is 24, and the oldest player around 30. Conversely, the goal advantage is smallest when the youngest player is 18, and the oldest player 26.
Unlikely as it might seem, although individual players have a fairly well-defined peak age band, there seems to be no peak age for teams. However, the age profile within the team – as indicated by the standard deviation – does seem to matter. It is likely though that the standard deviation is not the best metric for quantifying the age profile, and other ways of quantifying it may give more insight into what is going on.