51动漫

It鈥檚 a debate that rages in pubs throughout the world: which is the best football team of all time? Brazil in 1970, Hungary in the early 1950s, the Johan Cruyff-inspired Netherlands team of the 1970s? The Barcelona team built by Pep Guardiola? Or Liverpool during the 1980s?

鈥淭he whole point of sport is to work out who the best is,鈥� says statistician of the University of Salford, UK. 鈥淚t seems like a natural question to me to ask 鈥榃ho has been the best over all time?'鈥�

So McHale and his colleague Rose Baker waded in. They might not have an answer to the big question 鈥� yet 鈥� but they can reveal the best team in the history of the English league.

Fans of beautiful football look away now. The greatest ever team is apparently the Chelsea team of the 2005/06 season.

Defensive skills

To figure out which team from which era tops the rankings, McHale and Baker created a statistical model of team strengths. They used this to analyse goal data from 200,000 matches in England and Wales that occurred between 1888, when the Football League was founded, and 2012 in the top four English leagues, the FA Cup and the League Cup.

In the model, each team scores goals at a rate determined by its attack ability divided by the opposite team鈥檚 defensive skills, with an extra boost for the home team to model home advantage. Using the model, McHale and Baker could work out the most likely outcome of any English football match played over the 124 year study period 鈥� and use the results to calculate which was the all-time greatest team.

Doing so revealed that the Chelsea team of 2005/06 comes out top, followed by the Manchester United team of 2007/08 and Arsenal鈥檚 2003/04 squad of 鈥渋nvincibles鈥�, which went unbeaten in the league. The similarly invincible Preston North End team from 1888 came just tenth.

鈥淎s a Liverpool fan, I was convinced they would at some point be the best, so it was a bit of a surprise,鈥� says McHale 鈥� the club鈥檚 1987/88 squad is ranked fourth.

Best of the best

An alternative measure looks at a team鈥檚 strongest average performance over a 10 year period, and gives the same top three teams.

A third measure, dominance, gives the probability of a team winning against the second-best team over a 10-year period. The Manchester United of 1992-2002 come out on top with a 37.7 per cent chance of beating their next best rival, closely followed by Liverpool during 1979-89 and Wolverhampton Wanderers in 1951-61.

As well as coming up with all-time rankings, the pair鈥檚 model lets them set teams from different eras against each other. 鈥淓ven the guys in the physics department can鈥檛 time travel, but with this model you can,鈥� says McHale.

For example, a match between Manchester United鈥檚 1998 team would have a 28.9 per cent chance of beating Liverpool鈥檚 1980 team, but a 39.7 per cent chance of losing the game. There鈥檚 a 31.4 per cent chance the two teams would draw.

Out of time

鈥淚 think it is interesting, but it just focuses on what happens on the pitch,鈥� says sports historian of De Montfort University in Leicester, UK. 鈥淚t鈥檚 very difficult to wrench them out of the social and cultural context of the period and imagine how they would compete against each other.鈥�

鈥淵ou need to be careful how literally you take this,鈥� agrees McHale, because the model can鈥檛 account for wide-ranging shifts in the training and diets of all players over a number of decades. 鈥淚 cannot accept any modern day team wouldn鈥檛 absolutely destroy Preston North End 1888.鈥�

Sanjit Atwal, who runs football stats site , says the model鈥檚 top picks are in line with what he would have chosen, but he鈥檚 not sure the model will work as well in other countries where football is played according to different tactics. 鈥淭hey are weighing attack and defence equally. In other styles of play, possession of the ball is very different.鈥�

As for whether this will settle the argument of the greatest team ever, these statistics are just more fuel for the debate, says Atwal. 鈥淔ans will take whatever they can out of the data to win an argument.鈥�

Journal reference: Journal of the Royal Statistical Society: Series A, DOI: