Premier League fantasy preview: Statistics and more statistics

Good individual player value stats remain the great dream of football statheads. The problem, of course, is that so much of a player's value isn't identifiable from top line statistics like goals or even big chances or passes in the final third. The great thing about fantasy football is, that doesn't matter! We already know what a player's value is. It's how many goals he scored and assists he provides, plus how many clean sheets he notches if he's a defender. That's most of it. I can work with that. That cuts out all the hard problems!

The problems that remain are still significant, but they're problems external to the data and thus not my problem. I see fantasy football success as resting on three legs. You have to (1) identify the best players, (2) identify the players with the greatest likely playing time, and (3) manage your budget and plan your transfers strategically. I can do very little with regard to (2) and (3). So good luck with that. What I'm going to talk about today is problem (1), using the data to identify the best players for your fantasy football team.

As I have argued elsewhere, the correlations week-to-week of shot on target conversion rates are very small at the team level. This does not mean that teams and players don't have real finishing skill, but it means that real finishing skill is nearly impossible to identify from the statistics that we have. The correlations at the player level are stronger, but still quite weak. So when I look at how many fantasy points a player should notch, I focus much more on underlying stats like shots on target and big chances rather than top line numbers like goals and assists.

My method, basically, is to use everything. I create expected goals and expected assists numbers starting with shot on target and chance created numbers. I create a player-specific expected G/SoT number using (a) their number of big chances, (b) their number of shots from inside and outside the box, and (c) their actual G/SoT. I regress player big chances and shots on target to a team mean, adjusted for position, since getting a shot on target or a big chance is partly a function of one's teammates rather than one's own play. The expected number of assists per chance created is built off the team's expected conversion rates and number of big chances.

I also have an expected clean sheets number based on my Monte Carlo sims. With my team defense ratings, I can project how many clean sheets a club should have based on the quality of their defense. (To the best of my knowledge, no other fantasy football site includes a clean sheet prediction engine like this. So, yay. Enjoy.)

I should say, though, these are something short of real projections. I only have data I like for the 2012-2013 season. So what I have are regressed 2012-2013 numbers. I think they're a lot more useful for projecting 2013-2014 performance than the raw top-line numbers, but hopefully by next summer I can have something that looks a lot more like a real projection engine.

I am also only using EPL data, since that's what I have access to. So players coming up from the Championship or over from La Liga or the Eredivisie don't make the charts. In general, my (not deeply studied) opinion is that numbers from the top four leagues in Europe should translate well to England. Michu, I will note, led La Liga in goal scoring by a midfielder in 2011-2012.

Top Fantasy Players by the Numbers: Top 20

I have listed here in the tables, my fantasy points per 90 min numbers, FPL prices, and positions for outfield players who played over 600 minutes last year.

So what have we got here? Mostly you see the game's most expensive players toward the top, plus some goal-scoring substitutes with uncertain playing time. And Frank Lampard. He was a beast last year for goal-scoring. But given the midfield situation in Chelsea and Lampard's increasing defensive frailties, I don't think that's a punt I want.

You might have noticed Michu just barely making the top 20 despite his ridiculous points figures last year. That's because Michu was a bit of a conversion rate fluke. He scored just slightly under half his shots on target, and no one maintains a rate like that. Michu also enjoyed the benefit of an attack structured entirely around himself, as he picked up about 1/3 of the big chances of the entire Swansea club. With Wilfried Bony now likely to join in on the goal-scoring, Michu's opportunities will probably decrease a little along with his conversion rate.

Also, Jonjo Shelvey. He really did put a lot of shots on target for Liverpool and made some good passes. We'll see.

The other comparison I like is between the three big Spanish attacking midfielders. Santi Cazorla, David Silva and Juan Mata. These are their underlying stats:

David Silva last year both finished poorly and suffered poor finishing from his teammates on the end of his passes. You can see how Santi Cazorla, despite much better SoT numbers, was not taking quite as high quality shots as Juan Mata. And Juan Mata's 50% SoT conversion rate is due for a regression. I don't think any of these three is a fantasy must-have necessarily, but I definitely prefer David Silva to his likewise undersized countrymen.

Top Fantasy Players by the Numbers: 21-40

So, I'll have less to say about this set. The main thing I notice is that defenders from good clubs are showing up here interspersed with mid-price attackers. If you can identify a likely starter for the best few defensive sides, you can probably get better value at 6.0M than you'd get from an 8.0M striker or midfielder. You need to pick the right player and club, though.

Adel Taarabt doesn't have a price yet in the FPL game. This is just a warning. Adel Taarabt is the football player that many stats-haters think all football players are. He just shoots randomly at the target, and once his team is losing, he finds a way to shoot even more randomly in the vague direction of the target. SoT numbers that seem to "work" for most players do not work for Adel Taarabt. I carried him on my fantasy team last year for weeks, and to show for it I got a small handful of assists and a missed penalty. But those are the numbers, so I published them.

Jan Vertonghen is overpriced. You can see from the placement of Leighton Baines that attacking points for defenders are meaningful, but don't pay for last year's goal scoring. There's too much randomness there.

That's kind of all I've got. I always like my pieces to end on a sort of "eh" note.