In statistics, you rarely care about the outliers. If the data set is big enough, these are naturally occurring, but generally we want information about trending in the population as a whole. Outliers are something to be discarded.
In sports, outliers are everything.
In summer 2015, I was lucky enough to head up recruitment for Brentford football club in West London. We had to rebuild everything that McParland and Warburton took with them, and we had to do it from scratch, which meant scouting, market knowledge, player fit, etc. It was a monumental task, but we ended up with a really good recruitment team of Ricardo Larrandart, Nikos Overheul, Mark Andrews, and Robert Rowan, and a couple of part-time scouts including tactical superstar Rene Maric.
From the point we knew Warbs was leaving until the close of the summer transfer window was one of the craziest and most exciting times of my life. We were both researching and applying statistical football theory to the transfer market on the fly.
How well would players from various leagues translate to the English Championship?
What was the lowest price we could pay for players and still get them?
Could we rebuild an ageing squad into something that could potentially challenge for a promotion place again while playing an attractive, positive style?
This is one of the stories from that summer…
We knew we definitely weren’t getting Alex Pritchard back on loan. After finishing in the Championship Team of the Season in 14-15, Spurs wanted to keep him in training camp and then likely loan him out another rung up the ladder. There was the briefest chance we could get Dele Alli, but that quickly dissipated as he wowed Poch in training. This left a big hole for us in the 8/10 position.
Our first choice was to get Arsenal’s Jon Toral back on loan. Toral was tremendous in limited minutes for Brentford in the playoff season, and his profile was unlike anyone else we could get in our price range. I sat next to him and talked him through what I saw from the numbers and what his age corollaries were in the data set. He seemed smart and interested.
Unfortunately, somehow Marinus dragged his feet on whether Toral was the right fit. He was slow to make up his mind or get in touch with the player. Jon apparently was guaranteed starter minutes at Birmingham, and POOF! What seemed like a great fit flew right out the window, leaving us without a first-choice AMC. Owner Matthew Benham had negotiated to bring in Andy Gogia from Bundesliga 3’s Hallescher on a free in the spring. He could fill the role, but a bit like Alan Judge, we thought he would be better as a creative passer and dribbler out wide. (We also had Judge as a potential 8 because his defensive numbers were so good, but that never quite worked out.)
We could not get Pascal Gross or Ziyech, and no one else was super exciting.
Faced with a ticking clock and a very low budget that we would prefer to spend elsewhere, I put this Austrian guy no one had ever heard of back into the scouting queue.
The data suggested he was a solid attacking midfielder who could dribble and had the great ability to create shots for teammates. He also had reasonable tackling stats for a guy who primarily attacked, and scouting agreed that he was decent at pressing.
Now this was clearly a risk. At no time did we ever think, “Yes, this guy will be great in the Championship.” Instead we thought, “For the right price and in the right role, he certainly shows enough potential to be a solid performer in England.”
Everything in transfers comes down to money. Are you paying the right price for the talent and the risk involved? In Brentford’s budget, half a million pounds is a big deal, and a difference of £500K in valuation will kill a deal. In a Premier League budget, half a million pounds is chump change, and you’d be an idiot for missing out on a player for that small an amount.
The numbers lined up and scouting was positive, so we needed to get in touch with his club and his agent to find out if we could afford him. That’s where the Chris Palmer story came from. [Scroll to the bottom here.]
An eventual deal was sealed for low six-figures, and we had ourselves a low-cost wildcard of a 10 with potential upside. Even if Kersch was a bust, he was still probably cheaper than anyone we could have signed from League One, and for a club like Brentford, that mattered.
The Real World
Kerschbaumer showed up at training camp in amazing shape, and tested for the highest vO2 max in the group. Dude could run for days. It was all very exciting back then.
Unfortunately, things in football go weird sometimes. Brentford went through three head coaches that season and by the end of it no one really knew he was supposed to play 10 except the recruitment guys. He basically never played at AMC until the dead end of the season in 15-16.
Brentford had a horrible winter run, and things looked very grim. The club announced the closing of the academy and also the Football Analytics Team – my group – was made redundant as part of cost-cutting efforts. We had already finished most of the recruitment workload for the 16-17 season, and the perception was that the squad we had recruited was struggling mightily.
Now the truth was that we intentionally built a youngish squad with the blessing of the owner because that is what we could afford, and also so that they could potentially grow and improve together. As long as your recruitment is good, this is a good plan.
Then a funny thing happened. Brentford had an amazing run-in. From April 2nd at Nottingham Forest until the close of the season, they only lost one match, against eventual promoted side Hull. They also won six and drew two, most of which was without player of the season Alan Judge, who broke his leg in a nasty tackle at Ipswich. Scott Hogan finally came back from two different ACL injuries to be the hottest scorer in the league. Yoann Barbet started regularly with Harlee Dean in central defense, displaying an impressive passing range from his left boot, and a team that could not win a match from Christmas through February suddenly could not lose.
Brentford finished 9th. Without the poor start from the Dijkhuizen era, they might have been right back in the playoff mix. Additionally, they did it with a massive surplus of transfer fees. Worst case scenario, performance suffered a little but the club was now making big money in the transfer market.
Lost in this was Kerschbaumer’s performance. He subbed on when Judge broke his leg at Ipswich and set up Sam Saunders for the first Brentford goal. He also created an early goal for Hogan against Fulham, and two more in the final match of the season at Huddersfield.
Then the summer came and seemingly Brentford once again forgot about Kerschbaumer. This wasn’t unfair – Brentford had a lot of competition for the midfielder roles, and Romaine Sawyers, Ryan Woods, Nico Yennaris, and Josh McEachran shared the bulk of the minutes. Injuries bit throughout the season though, and Kerschbaumer finally started to see more playing time, once again in the spring. Since March 18th, Brentford have lost once, drawn twice, and won five times. And once again, KK is out there racking up assists.
Why the long story about a bit player in a small Championship team?
The answer is because Konstantin Kerschbaumer is a major outlier. Combine his minutes across two seasons and you get the following:
2320 minutes, 1 goal, 12 assists.
That’s an assist rate of about .47 per 90, which is in the top 3% of footballers. Kersch also doesn’t take set pieces, meaning nearly all of his assists come from open play. To give you an idea of how unusual this is, in the last four seasons in the Championship nine players have posted 12 assists or more, all with more minutes and nearly all of them taking set pieces.
Assists are really valuable – I view them basically the same as goals. Fans still have a very different perspective if a player scores half a goal a game than if he creates half an assist a game, there’s a decent case to say they shouldn’t. The bulk of Kerschbaumer’s minutes also came during that first year, many of which were not in his natural position. That’s a tough situation to succeed in, but his numbers in this one particularly valuable area continue to be crazy.
Is Kerschbaumer a success? I have no idea. It would be hard for Brentford to lose money on his transfer should he leave the club, so if that’s how you grade success, I guess it’s a check mark. He’s also produced exactly what I thought he could when we recruited him. But… there are questions about whether he does enough on the pitch when he plays, and I can certainly see why those exist. I think he’s still learning, and I hope he ends up with starter minutes next season, preferably in a system that plays him in his natural AMC spot. Like most data scientists, I want more data and preferably a lot of it.
Part of me roots for the players we recruited like they are my children. I want them to succeed no matter what. There’s also a part of me that is scientifically evaluating their successes and failures to see what worked, and what I need to do better the next time I have a chance to dabble in the transfer market.
Anyway, the combination of Kersch’s crazy assist rate in the run-in and Fabregas’s continued creative skills for Chelsea made me think back to four years ago, when I first started writing about player stats. So much has changed in my approach, but remarkably, so much is still similar. I think a lot of the early ideas I latched on to as mattering ended up being very valuable. That said, I have made plenty of mistakes along the way, both inside and outside of football.
Making mistakes – and learning from them – is most of the fun.
*Thanks again to Matthew Benham for the chance to do all of this while learning on the fly. Looking at the quality in the squad right now, I think we did pretty well.