Last week, we saw how our extremely simple algorithm managed to produce some interesting results ahead of last season in the Premier League. This week, we will take a brief look at how it rates some of the bigger PL transfers ahead of this season.
To keep things simple, I went through the top 10 Premier League transfers by fee over at transfermarkt.co.uk. This list features a nice mix of players across the spectrum, in both role, ages and score out of ten for expected first year playing time. Remember, a low score on this list does not necessarily mean the player is a dud, just that they may be less likely to hit the ground running right away.
First, the list:
There are no major outliers here that I can see. The very promising Manchester City prospect Leroy Sané remains the most expensive PL transfer this summer, but his age and the relative quality of his previous club Schalke dropped his score a bit. Equally, Granit Xhaka may be something of a surprise disappointment, again because of his relatively young age (though Gladbach ranks fairly high in relatively quality), in addition to Arsenal’s not-so-great record of blooding new signings in their first year. His preseason performances have impressed, but that doesn’t always translate.
Elsewhere, Henrikh Mkhitaryan looks like a decent pickup for Man United, based mostly on Dortmund’s high Elo score and his playing time when healthy. Meanwhile, it’s clear there are one or two higher end speculative bets on this list, including Chelsea transfer Michy Batshuayi and Spurs’ Vincent Janssen. I’ll keep tabs on these this season to see how the algorithm fares, and may make a few tweaks down the road.
Soccer analytics seems to be in something of a competitive phase at the moment, with increasing demands for recompense, less published work to protect intellectual property, and attempts by ‘experts’ to discern their work from ‘charlatans.’
These are all laudable goals in their own right, but in a digital world they come at a cost—outreach, education, experimentation, and healthy, constructive debate. This is one of the chief reasons I decided to start Front Office Report.
To that end, the point of this series was not to develop a working scouting algorithm, but to show what can be done by someone with no expertise in statistical science.
Some will scoff at this—we don’t entrust medicine to amateurs, so why should we do the same in sports analytics?
But there are major differences. For one, physical illness is normally extremely tangible—our bodies are usually good at telling us if something’s wrong. For another, most common medical practices are not proprietary. Most doctors will generally discuss with us exactly what treatment they intend to prescribe and why. Particularly in the age of google, patients are increasingly able to read and understand the peer-reviewed literature which shaped the decision of their practitioner. In medicine, people generally know when they have a problem in need of medical attention, and that a doctor would be the best person to speak to to help.
In sports analytics, particularly in a sport as conservative as soccer, clubs don’t always know that there is anything amiss with how they conduct their business, let alone something a smart statistical analyst could help fix. Moreover, analysts may be happy to provide data to prove their approach ‘works,’ but less inclined to show why, which requires something like a leap of faith from the club.
There is nothing wrong with this state of affairs, but it requires a willingness ability to patiently overcome skepticism, to truly listen, and to communicate statistical ideas effectively to bridge the gap.
I believe one of the most effective ways, espoused by the likes of Daniel Kahneman and Philip Tetlock, is to encourage people to try some of these methods themselves, and to incorporate the themes into their own work. Do they improve odds? Do they add another lay of information? And how could they be improved further, perhaps with the help of an analyst?