Tuesday, August 14, 2018

The best data categories for early gameweeks

Gameweek 1 has been somewhat less dramatic than last season. With the new Friday kick off and a solid day of Saturday action under our belts I went to bed on August 12th just like any other Saturday night. A few hours later I was woken by my wife who was in labour - 7 weeks before our due date. A few more hours later our twin girls appeared and let's just say that questions around Gabriel Jesus's playing time or Paul Pogba's xG suddenly seemed somewhat trivial. Needless to say, the past year has meant I've had less time to dedicate to fantasy football, although the early starts on the weekend make watching the 7am ET games much easier.

Over the offseason I have been debating how, if at all, this blog can continue to offer value. With the prevalence of Opta's xG, my own version of a similar (but more basic) model becomes somewhat irrelevant and my absence from England means that I will never be as up to date on team news as those who live in the city, read the local press and have Sky Sports New on a loop. Time constraints also make it very hard to individually maintain a broad catalogue of data visualizations trying to monitor fixture strength, expected points, injuries, playing time and form.

My plan then is to try and focus on bigger picture pieces that try to develop a concept and provide some push back to the extreme shifts in mood that tend to fill the discussion boards of popular sites (player X is a "must own", player Y must be sold after his 2 point performance this week).

This first post then is trying to identify (a) what are the metrics we should be looking at in the early weeks to predict future success and (b) how do the current hot/cold transfer targets perform in these areas.

Correlation and stabilisation
I think a majority - or at least a large number - of fantasy managers understand the idea that a goal in gameweek one or two does not mean that player is destined for a great season. Early goals last season for the likes of Jese, Robson-Kanu, Okaka, and Okazaki were not the spark for greatness, with that group all failing to score more than one more goal for the remainder of the campaign. Of course, an early goal for Salah was indeed the first sign of an incredible season on his return to English football, so we don't want to totally ignore the early action and be slow to respond.

The area where analysis seems to typically lie at the moment is to focus on what I will call Level 2 events (with Level 1 being the actual outcome of a goal and/or assist). Level 2 events are the building blocks for goals, most notably shots of various varieties but also key passes and crosses. Over a longer enough period these events are predictive of goals and if you target them over a season you will surely enjoy success. However, here I am going to look at Level 3 events, namely the kind of actions which are indicative of a player who will ultimately create Level 2 events than will lead to goals/assists. Level 3 events include actions such as touches in the penalty box and passes in the final third of the pitch. xG is obviously an incredibly useful metric, but in these early weeks when we're still trying to work out who is likely to be in the right position to even earn/create shots, digging even deeper can be useful.

The reason I want to talk about these Level 3 events is because of the speed at which they stabilize. Taking data from last season I compared the yearly totals for players, on a per minute basis, against their cumulative totals through GW1, GW2, etc. all the way through the first 15 weeks of the season. As expected, the correlation between player totals in GW1 and GW38 is essentially non-existant in all categories: success in GW1 alone is a very poor predictor of season success. Allowing for a bit of noise as players debut at different stages, this correlation improves week-by-week so that by GW15 we have a number of metrics that correlate very well to a player's actual profile. In other words, by GW15 we have a good idea of a player's profile in terms of their ability to generate shots and create chances. The question though is whether we start to understand that profile earlier than GW15.



The interesting point here is that shots inside the box (SiB) - a metric favoured as a predictor of goals by many, including this site - takes a relatively long time to stabilise. Even after 6 gameweeks, the correlation between players' SiB per 90 minutes at that point with the rate they would ultimately post for the season was just 22%. It would take until GW13 until this rate reached the 50% mark. Conversely, the events I have dubbed "Level 3" were much quicker to stabilise, reaching that same 50% correlation level in just six or seven gameweeks. Indeed, passes in the final third eclipsed the 40% correlation rate in just 4 gameweeks. With this in mind, let's take a look at some of the names making early headlines and involved in the early transfer action, and see if this "Level 3" data offers any useful insights:


The first thing I noted here was that fantasy managers are clearly getting more savvy and you don't see a lot of absolutely terrible transfers in this list. Based on this metric alone Neves is a bit of an overreaction although he of course offers value from long range efforts and from set pieces so it's hard to push back too much on a 5.0m purchase price. Diogo Jota did eclipse his teammate in this one metric but it's too close to argue either way and Jota of course comes with a higher price tag.

I am a big fan of Paul Pogba but this table offers another notch against him. Despite playing very well in gameweek one, he still didn't get into really attacking positions and his fitness remains a concern after a long World Cup campaign. Penalty duties are of course a huge bonus but Sanchez may still take some spot kicks in the future, and even then, United scored one penalty all last season with other talented teams like Liverpool and Spurs only notching two a piece, so I'm not sure penalty duties are quite the treasure trove some seem to think (Milivojevic aside).

It's a similar story for Jorginho (who took an even more impressive penalty than Pogba) though he faces a greater threat of losing these responsibilities when Hazard returns. At 5.0m though, he comes at a significant discount so I can see the temptation. I'm surprised his teammate Kante isn't getting more buzz, although he did enter the season with a high ownership rate. His goal is one thing - coming on his only shot - but the presence of Jorginho offers the intriguing chance for the Frenchman to be released to play further upfield in this potentially exciting Chelsea lineup, as evidenced by the number of passes he received in the final third. Pedro tops this list but with Willian also playing well and Hazard close to full fitness, it's very hard to see a scenario in which the former Barcelona man gets reliable minutes. Perhaps if Hazard can take some time in a Fale 9 role in place of Morata and Giroud?

Eriksen stands out here as an outlier among the players being sold, who otherwise did indeed have generally poor GW1 games. If you liked Eriksen going into GW1 there was nothing here to concern you - but for what it's worth his fixtures and teammates' fitness remain worrisome.

The Liverpool trio of Salah, Firmino and Mane all rank highly here, adding weight to the suggestion that doubling up on Klopp's men might be a good idea. Indeed, Keita and Milner would also place highly on this list but haven't yet garnered much transfer attention as yet. If you do want to double up though, it would be a good idea to act now, as bringing in extra Reds before the GW5-8 run of games against TOT, CHE and MCI would seem ill advised.

They aren't shown in the table as they've seen minimal transfer activity to date, but the Fulham duo of Seri and Crainey both did well by this metric with 19 and 15 passes respectively. With Mitrovic racking up 8 of his own, I think we saw enough from Fulham despite their loss to not abandon them too soon.

City looked to hit the ground running despite a few absences with Aguero (9), Bernardo Silva (11), Mahrez (10) and Sterling (17) all placing well. I am very surprised that Sterling has received so little transfer interest as he's clearly an elite player who's only drawbacks are his fitness after the World Cup - which seems fine - and I suppose rotation risk. Like Pogba above, a rest day could be in his future but this game seems to suggest he is first choice when fit and thus there remains significant value despite the high price tag.

Finally, the hype around the Bournemouth group of King, Wilson and Fraser seems well deserved based on this single game with all players heavily involved. Defoe's presence on the bench is still a slight concern but in the medium term you have to like the value here given Bournemouth's reasonable fixture list.

3 comments:

Unknown said...

Hello Chris,

just wanted to say thank you. Your work before really opened up a new fantasy dimension for me (shots on target and shots in the box were like a discovery of the atom for me) and changed the way I play the game.

Keep up the good work, and best of luck to you and your family.

Zouboulzeng said...

Congrats on the twins!

Fantastic article and very interesting finding regarding Kante. How does Will Hughes compare?

I too learned about underlying stats and strategy here in my early days of fantasy football and have been coming back ever since. I love your insights & hope you keep posting even if it is irregular.

All the best

2ndMan said...

Great to see you posting again Chris, love the idea of articles like this! Don't worry about updating models or providing week-to-week updates, focus on the more in-depth stuff because that's where you really add value over other sites (and I suspect what you enjoy doing more as well).

All the best for the season and congrats on the new family!