Wednesday, August 31, 2016

Three Likes Three Games In

It's a new season, and it's a brand new City. This is at once surprising, as give or take a couple players it's been the same team that was so disappointing last season, and unsurprising, as having a top echelon manager is bound to make a difference. The first international break (of too freaking many, let's be honest) gives us a chance to consider the early changes to the team, what has worked, and what needs improvement. In an effort to be a more optimistic person, let's start today with three things I've liked so far from the team, because there is an awful lot to like.


Pep's base 2-3-5 formation has gotten a lot of attention, but I'm going to have to gush some more over its brilliant simplicity. The amazing thing about it is how easily it makes it to move the ball vertically forward. Consider this well-known Mourinho quote about the 4-3-3:
Look, if I have a triangle in midfield – Claude Makelele behind and two others just in front – I will always have an advantage against a pure 4-4-2 where the central midfielders are side by side. That’s because I will always have an extra man. It starts with Makelele, who is between the lines. If nobody comes to him he can see the whole pitch and has time. If he gets closed down it means one of the two other central midfielders is open. If they are closed down and the other team’s wingers come inside to help, it means there is space now for us on the flank, either for our own wingers or for our full-backs. There is nothing a pure 4-4-2 can do to stop things.
In Chelsea's case, the formation worked because Makelele always had two easy forward options. The genius of Pep's system is that EVERY PLAYER in defense or midfield has two easy forward options. The goalkeeper can pass to the two central defenders. The central defenders can pass to Fernandinho or their respective fullback. Fernandinho can pass to Silva or De Bruyne, the left-back can hit Nolito or Silva, and the right-back can hit Sterling or De Bruyne. That leads to passing charts like this, clearly emphasizing the verticality of the formation. It really shouldn't be that surprising since the numeric representing the formation is a partial Fibonacci sequence, and the latter's connection with Pascal's triangle underlines that there are always multiple options the further you go. Indeed, in the first few games, where City have had the most trouble is where the sequence ends in the final third of the pitch.

Game by Game Progress

I've been looking a lot at a stat I call Usage Rate, which shows the percentage of a team's possessions that a given player is responsible for the end result (i.e. a shot, key pass, assist, or turnover). It helps to show how involved each player is in a team's attack, and can be very useful in assessing a change in role (I've posted a brief explanation and a link to 2015-16 data for all players here). With Pep coming in, I wanted to see how the usage rates of the players looked in the first three games of the new season (starters only):

Game 1
Game 2
Game 3
Player Usage Rate
Player Usage Rate
Player Usage Rate
Clichy 14.94%
De Bruyne 15.19%
Silva 21.70%
Nolito 13.99%
Sterling 11.93%
Sterling 16.11%
Aguero 13.12%
Aguero 11.08%
De Bruyne 15.08%
Sterling 12.28%
Otamendi 11.01%
Aguero 12.67%
Silva 11.71%
Navas 10.93%
Nolito 11.46%
De Bruyne 10.66%
Caballero 10.09%
Fernandinho 8.48%
Fernandinho 10.66%
Fernandinho 8.26%
Zabaleta 8.48%
Sagna 9.02%
Silva 8.26%
Clichy 6.60%
Kolarov 7.38%
Zabaleta 7.34%
Otamendi 4.71%
Stones 5.74%
Kolarov 6.42%
Caballero 3.77%
Caballero 2.46%
Stones 3.67%
Stones 2.92%

As you can see, there were some kinks in the first game, as Clichy is not a player you want in a high usage role. The second game looks better, but Caballero's usage rate is way too high for a keeper (this was probably due to the fact that Stoke pressed high a bit more) and Otamendi's similarly high for a CB. But in the West Ham match, we see a much better distribution of possessions. All the attacking players are using more possessions, all the defensive ones are using fewer, which is exactly how it should be in a functioning offense. The team metrics are better too, with turnovers (unsuccessful passes + unsuccessful take-ons) dropping from 106 to 97 to 84, and shots rising (albeit with a slight dip against Stoke) from 16 to 12 to 22. That there is such a clear progression in the data makes me pretty confident of continued improvement, where I was really just hoping performance levels would stay the same under Pellegrini.


One of the biggest differences so far has been the play of Raheem Sterling. This is not to say he played badly last year, as I think his season last year was very underrated. Still, he was never really a good fit for the role Pellegrini tried to shoehorn him into. In my diagnosis of City's problems last year, I wrote the following about Sterling in Pellegrini's 4-2-2-2:
Sterling too doesn't really fit. His best attributes are his intelligent forward runs, pace, and trickery, all of which are neutered in a role that prioritizes pinpoint passing. Due to Pellegrini's influence, his take-ons are down from 3 per 90 last year to 2 this year, and his limited passing (at least compared to Silva and Nasri) results in tons of back passes. His usage rate is way too low for the position, posting just 10.9%, 8th on the team, and both left backs have higher usage rates.
It's been completely different under Pep. Many have noted the fullbacks underlapping rather than overlapping has given Sterling the space to attack his man, which is completely true. Just as important though is that he doesn't have that safety net of the fullback right next to him, so he's incentivized to beat his man himself rather than pass it tamely backwards to keep possession. His take-ons are up to almost 4 a game (in the top 10 in the league for players with at least 90 minutes) and his Usage Rate so far has increased from 10.76% last season to 13.37% this, ranking fourth on the team behind Silva, Nolito, and De Bruyne. Somewhat ironically given the suddenly positive media attention, the possessions he has used have resulted in positive outcomes (shots, key passes, assists) at a lower rate than last year (24% compared to 35%). Still, that is (a) partly to be expected given a larger role in the attack and (b) far less dangerous when the fullback is not bombing on beyond him so much. Given the consistent improvement the team has shown so far under Pep, I expect Sterling will be able to get that rate back to last year's levels and really start to fulfill his potential.

Saturday, August 6, 2016

City's Signings and Pep's Plans

One of the things I've been hoping to do with my usage rate stat is see how it relates to potential transfer targets. By showing the percentage of possessions used, it gives an idea of how involved a player is in a team's attack. It also correlates very strongly from season to season. I took a look at all players who were in the Premier League in both 14-15 and 15-16 and compared their respective usage rates in each season. The r-squared is .75, suggesting that players generally don't change their style of play. However, that r-squared drops to .59 among the subset of players who changed teams between those seasons. If we know a player has a high usage rate, when he transfers to another team there will need to be an adjustment: either he takes possessions away from players currently on the roster or his own usage rate drops. If the former, the team may suffer if the player is taking away possessions from more skilled players (the mooted Deeney to Leicester deal would have been a good example of this) and if the latter, the player's individual counting and per 90 stats will likely drop since he is using fewer possessions. Knowing this, I wanted to see how City's signings (at least those likely to feature in the first half of this coming PL season) looked, both to see how they would fit into the team and to also potentially glean what Pep is looking for from the players he recruited.

Let's take a look on Gundogan first. Below are his stats last year for Borussia Dortmund:

Player Successful Passes Total Passes Unsuccessful Passes Key Passes Assists Goals Shots Take-Ons Take-On Success % Unsuccessful Take-Ons Minutes Possessions Used Usage Rate Positive Outcomes Negative Outcomes % Positive
Gundogan 1727 1955 228 23 3 1 50 56 69 25 1990 329 12.57% 76 253 23.09%

Ilkay took on a very large role for a central midfielder in their offense. His usage rate was second on the team to Mkhitaryan, though at 12.6% it would just be 6th at City. His stats are probably most similar to Toure's, who posted a 13.5% usage rate, with fewer shots per 90 (to be expected given Toure's more prominent role in free-kicks) and more take-ons. I think he can slot in a similar role, and he appears to have a defensive quickness that Toure doesn't at this point in his career which could allow the pair to play together. Of course the question mark over Gundogan is his injured knee, which is what allowed City to pick him up for such a reasonable price. I am hopeful that he makes a full recovery, but that is by no means certain and City's medical department has a less than stellar track record.

Next up is Nolito. His stats for Celta Vigo are below:

Player Successful Passes Total Passes Unsuccessful Passes Key Passes Assists Goals Shots Take-Ons Take-On Success % Unsuccessful Take-Ons Minutes Possessions Used Usage Rate Positive Outcomes Negative Outcomes % Positive
Nolito 736 1091 355 53 7 12 81 58 41 83 2471 579 18.38% 141 438 24.33%

You'll note the usage rate is high. Very high. So high in fact that not only would it have been the highest on City last season, it would have been second-highest in the Premier League last season behind only Alexis Sanchez. That is simply not going to continue on a team with Kevin De Bruyne, David Silva, and Yaya Toure. He was the focus of Celta Vigo's attack in a way he just won't be with City. The problem is his counting stats (i.e. goals and assists) are artificially inflated by the amount of possessions he used. He scored 12 goals and had 7 assists, very respectable season totals for someone playing primarily on the wing. However, those totals came from him using a whopping 579 possessions. De Bruyne, probably the most similar player currently on the City roster, used just 400 possessions to obtain 7 goals and 9 assists (and 20 more key passes than Nolito). As a result, the percentage of possessions Nolito used with a positive outcome (i.e. a shot, key pass, or assist) is just 24% (KDB's is at 32%). Even though it's certainly possible efficiency can be improved when you are not the primary focus of a defense the way Nolito was at Celta (though I haven't seen much evidence of this in the data), I just don't see how his goal and assist totals are repeatable. He's also 30, so he's probably in the down slope of his career at this point.

Finally, the signing I'm most excited about: Leroy Sane.

Player Successful Passes Total Passes Unsuccessful Passes Key Passes Assists Goals Shots Take-Ons Take-On Success % Unsuccessful Take-Ons Minutes Possessions Used Usage Rate Positive Outcomes Negative Outcomes % Positive
Sane 479 615 136 22 6 8 68 92 46 108 2318 340 11.76% 96 244 28.24%

As you can see, a much lower usage rate here. In this regard, he's most similar to the speedy wingers already on the City roster in Sterling and Navas. While he doesn't rate particularly high in the total numbers of key passes and shots, the locations he takes shots from and makes passes to are excellent, which is reflected in his excellent xG/xA numbers. The lower usage rate means he'll also be easier to integrate into the side since he can achieve production without needing the play to run through him all the time. He's also much younger than the previous two signings, so he has significant potential to improve.

Do the players Pep recruited tell us much about his plans for the squad next season? He certainly seems to be targeting players with a high usage rate for their position, and (relatedly) players with a large number of take-ons. That would indicate he wants players who are comfortable and dynamic on the ball, but I haven't seen a focus on retention of possession (Sane's passing percentage is 78%, Nolito's just 67%). This might also explain why Nasri and Fernando, two players with a low number of take-ons per 90 but strong passing percentages, might be less utilized by Pep than Pellegrini. To me, this suggests Pep is looking to build an offense that is capable of attacking from anywhere, rather than be reliant on a single talent the way City were last year on KDB. However, given Nolito's indifferent stats, Gundogan's injury, and Gabriel Jesus' late arrival, he'll largely need to do it with the players already here. I think this might mean a bigger role for Delph and Sterling, both of whom had a higher usage rates on their previous teams and were under-utilized by Pellegrini. No matter how it unfolds, it will be fascinating to see how he pulls it all together.