"Statistical thinking will one day be as necessary for efficient citizenship as the ability to read and write"
H G Wells

Why Statistics?

I believe statistical thinking will soon become essential for efficient football management. Intelligent analysis is already necessary for running baseball and basketball teams.

The book Moneyball popularized a new approach to baseball taken by Billy Beane and the Oakland A's. Beane believed that the traditional analysis of baseball was flawed: stats such as batting average were poor predictors of how many runs a batter would score. He also believed the collective wisdom of managers, coaches and scouts to be subjective and prone to bias.

Beane understood that a statistical record of a player's past performance was a much better predictor of future performance than expert opinion based on watching only a few performances. Beane became manager of the A's in 1998: by 2002 they were consistently performing above average with a wage bill 20% as big as the New York Yankees'.

Since 2006 the A's have not been as successful: Beane's secret is out and most Major League teams now have their own stats departments. We are currently witnessing a similar revolution in basketball: The general manager of the NBA's Houston Rockets is a computer science graduate, and Beane's approach has been replicated by many NBA teams.

Football is not Baseball

Most of baseball is a series of one-on-one encounters between pitcher and batter. Many pitches are seen during a match and each ball tells you something about the quality of the players involved. Football is a highly collective game where each player's performance depends on the performance of his team mates. There may only be five saves and two goals in a match - not much on which to base player analysis compared to baseball.

Does statistical analysis work in football? Yes. Since 2002, Decision Technology's football predictions, published in the Times every Saturday, have routinely outperformed tipsters and experts and in five seasons out of six outperformed bookmakers.

I would like to give some examples of research that Decision Technology has done for the Times Fink Tank column to illustrate how statistical analysis can be applied to teams and players.

How Do You Win A Game?

By scoring more goals than the opposition. One method of measuring a player's contribution to a match is to work out which actions correlate with scoring and conceding goals. If a player generates lots of shots during a match we are likely to see a goal. If a keeper has a high save percentage, his team is unlikely to concede.

Using data from Premier League matches between 2000 and 2009 I asked the question "How many points did a team win given that they had more corners than the opposition?" Likewise, we can ask the same question for passes, shots or anything we like.

The results? Teams with more shots won 1.93 points per match - much better than the league average 1.37 points per match. How valuable is a corner? Teams with more corners won 1.48 points per match.

As usual with football research, there are some counter-intuitive results and some thought is needed to explain them. Teams with more offside passes are significantly more successful - winning 1.53 points per match. Does this mean players should be coached to make offside passes? No; an offside pass is simply a consequence of a team playing well - offsides are also correlated with goals and shots.

Conversely, red cards are very damaging - the team with more red cards wins only 0.80 points per match. However, even this seemingly simple result hides a great deal of complexity: The cause and effect of red cards is surprising. We found that in league matches, teams who suffered a red card could expect 1.25 points at the start of the game. Just before the red card was shown, they could only expect 1.02 points from the game. In other words, teams that suffer red cards are already doing significantly worse than expected before the red card is shown.

The intuitive effect of a red card - that teams do worse with 10 men - is correct but only statistical analysis can reveal that teams tend to be losing even before the sending-off occurs.

Do Imported Players Need Time To Settle?

Collective wisdom suggests that overseas players new to the Premier League need a season to settle in. Poor performance in a player's first season is accepted on the understanding that he will improve once he has become used to the particular style of the league.

Is this true? To find out, we rated each player by recording all of their actions and relating their performances to goals scored and conceded. This gives an "extra league points added per season of playing time" rating with the average player adding no extra points.

We compared the 05/06 ratings of 21 new overseas players to their 06/07 ratings. In their first season the new overseas players added 0.1 extra league points per season of playing time on average. In their second season, that improved to 1.9 points.

But that is not the whole story. In 06/07 new overseas players added 2.0 points per season, but only 0.6 points in their second season. In 07/08 the new players added 0.5 points but only 0.1 points in 08/09.

Taking all three seasons, the new players added 0.88 points in their first season and 0.86 points in their second: New overseas players don't improve in their second season (the same is true of new British and Irish players).

In this case the "collective wisdom" is wrong: It was probably true in the past, but knowing that it is not true today may be of immense value to a team.

Predicting the Future

In 1998, basing a baseball team's recruitment policy on statistical analysis that shunned traditional approaches seemed outlandish. In twelve years a more quantitative approach has become the norm.

In football, analysis is harder to do because of the nature of the game, harder to explain because of the extra work required in generating meaningful results and harder to sell because there is no great tradition of statistical analysis in the game.

But the potential is enormous: Creating tools that rate players and teams based on objective data can identify talent, save scouting resources, validate subjective opinion and inform player valuation. Statistical analysis is only a tool, but currently one that is sorely under-used in football. Any professionals willing to spend time understanding its strengths and weaknesses can only improve their understanding of the game. Further, their expert knowledge and feedback will inform and improve the analysis.

The statistical revolution has already happened in baseball and basketball. Football? It's just a matter of time.