D.C. (Away) - Postmatch

[Ulrich]

it depends how you feel about stats. It will certainly paint some kind of a picture for you, then you need to look at and interpret that picture based on whats actually happening on the field. No stats are perfect.

ETA: But you can look at is as well we have a very high xG but we arent scoring very many goals. More than likely our strikers/fowards are fucking up

Or the team is taking a lot of low percentage shots (that add up) from outside the box.

 

[Kjbert]

We take a lot of shots. Not a lot go in. Based on the number of shots we get, we should score some of them.

For some reason, we don’t score them.

Am I catching on

 

[mgarbowski]

So it’s a trailing measurement?

i.e. “so far this season, NYCFC have had an XG of “ whatever?

Correct. Every shot is assigned an expected goal value based on where it was taken, plus other factors that vary from one xG system to another. Those factors include whether the shot is from a set piece (and what kind), was it a breakaway, how many defenders are in the box, etc. PKs are usually worth about 0.7 xG. All other shots fall somewhere between there and 0. The idea is that for every 5 shots that are worth 0.2 xG you should expect to make 1 goal on average.

 

[Kjbert]

So we have underperformed our XG under Dome.

But that doesn’t answer the question of whether or not that is on the manager or players? So we’re back to square one?

 

[Kangaroo Jack]

Correct. Every shot is assigned an expected goal value based on where it was taken, plus other factors that vary from one xG system to another. Those factors include whether the shot is from a set piece (and what kind), was it a breakaway, how many defenders are in the box, etc. PKs are usually worth about 0.7 xG. All other shots fall somewhere between there and 0. The idea is that for every 5 shots that are worth 0.2 xG you should expect to make 1 goal on average.

just to expand slightly an xG of .7 means if you took a shot 10 times u should expect it to go in 7 #math (just felt like using units of 10 makes it easier for people typically to understand)

 

[Ulrich]

So we have underperformed our XG under Dome.

But that doesn’t answer the question of whether or not that is on the manager or players? So we’re back to square one?

Forward has ball outside the box at its corner, looks up and sees 4’8 midfielder playing CF attacking the net trailed by 6’2 CB. Forward elects to take shot from there rather than crossing to teammate. Shot goes into stands.

Forward was shitty at shooting yet manager’s tactics failed to give a viable target that would crash the net.

 

[Ulrich]

just to expand slightly an xG of .7 means if you took a shot 10 times u should expect it to go in 7 #math (just felt like using units of 10 makes it easier for people typically to understand)

Or you’re using Vela as the example.

 

[mgarbowski]

The team xG under Torrent actually tracks results pretty well after you split his fast start with everything since.
Let's look at xG differential (xG For minus xG Against) per game.

NYCFC under Vieira in 2017-18: 0.27 xGD and 1.73 PPG
NYCFC Torrent's first 9 games: 1.04 xGD and 2.11 PPG
NYCFC last 16 games (excluding DC): -0.05 xGD and 0.88 PPG
NYCFC 2019 (excluding DC): -0.28 xGD and 0.83 PPG

DC is excluded only because I don't have the xG data yet.

If anything we probably were over-performing under Vieira, and you can argue whether that was luck or residue of design.

 

[Christopher Jee]

Very serious.

I’m not a nerd like Christopher Jee

Well I guess I'll wear the "nerd" label, but you don't need to be a nerd to get xG!

 

[Christopher Jee]

The team xG under Torrent actually tracks results pretty well after you split his fast start with everything since.
Let's look at xG differential (xG For minus xG Against) per game.

NYCFC under Vieira in 2017-18: 0.27 xGD and 1.73 PPG
NYCFC Torrent's first 9 games: 1.04 xGD and 2.11 PPG
NYCFC last 16 games (excluding DC): -0.05 xGD and 0.88 PPG
NYCFC 2019 (excluding DC): -0.28 xGD and 0.83 PPG

DC is excluded only because I don't have the xG data yet.

If anything we probably were over-performing under Vieira, and you can argue whether that was luck or residue of design.

I love this.

Opta's xG data you can get pretty quickly after the game (pretty sure it's within 24 hours) here: http://widget.cloud.opta.net/helper/v3/#/football/expected_goals

You don't need a subscription, you can just click the link below the "Subscription ID". It's showing NYCFC 1.51 xG vs. 1.06 xG DCU.

I believe xG data is compiled at the league and season level but I'm not entirely sure. That kind of knowledge (and more) varies from entity to entity which is why you see xG scores varying from entity to entity (e.g. American Soccer Analysis vs. Opta).

Like Kangaroo Jack says, it's just one tool. I wouldn't say it's either "use xG or just trust your eyes". You should use both. You can use xG to direct your eyes (e.g. look at a high xG shot and then go back and watch the game around that time to see how it got there), for example. Ironically, some models even include a "big chance" modifier which is hand-coded, so it's already internalizing "eye tests". Ulrich is correct that it doesn't track the position of defenders, but there is a "phase / type of play" modifier in most models now that captures counter attacks vs. breaking down a low block, for example. That too, is hand-coded and internalizes the power of "eye tests".

mgarbowski segmenting the xG analysis according to inflection points in the data or in the process that leads to our players playing in specific ways on the pitch is another great example of combining human intelligence with groomed data.

One of the great things about xG is that it's a model that can be improved in predictable ways. There are several organizations looking to add the appropriate equipment and infrastructure to track player positioning (vs. touch data). The more data the xG models can ingest, the more accurate they become.

Another benefit to xG is that those improvements scale - since these models are written in code (and in some cases open source), the improvements in xG models are both repeatable and theoretically shared by everyone. Or at least everyone who cares to believe in the value of xG data. You can watch more games to improve the analytical power of your "eye tests" but that's arguably harder to share with a broader community. And it's easier for people to say "like that's just your opinion, man", whereas it's harder to disagree with data which is produced in predictable, somewhat transparent / objective ways.

 

[canchon]

So, what should our PK taking order be going forward?

1. Heber
2. Mitrita
3. Tajouri-Shradi
4. Ring
5. Not Maxi

Or we can do an order of “cover your eyes” penalty order - as in those who would probably blow it.

2019 edition:
1. Maxi
2. Sweat
3. Matarrita
4. Medina
5. Rocha

All time edition:
1. Lord Kwame
2. Nemec
3. Calle
4. Mendoza (dude didn’t even show up so of course he’d blow it)
5. Saunders

 

[Keith Putnam]

So it’s a waste of time then Ulrich ?

An xGD table (D for "difference"... xG minus xGA [A for "allowed"]) for a league tends to correlate very strongly with the actual table, and where it doesn't it's instructive to look at which side of the ball is being overperformed or underperformed to see why the correlation has broken down.

As mentioned above, xG is only as useful as stats can be, but so far it seems to be the most statistically rich way of describing a team's performance over a period of time vs. other teams.

 

[SoupInNYC]

It also doesn’t take into consideration the location of defenders on a particular play. A tap in on the back post from 2ft is probably a high percentage shot, but it means little if the goalkeeper has followed the cross/play and laid himself out with the back post completely covered. xG may be a 0.99/1.0 on that (making the number up) but the goalkeeper creamed the play and it’s a 0.0. Pretty much what Johnson did to DC on that one set piece cross that statistically was a sure goal until it wasn’t. Voodoo numbers.

Just to clarify, the one set piece cross you're referring to had an xG of 0.4211

 

[Annagram]

So, what should our PK taking order be going forward?

1. Heber
2. Mitrita
3. Tajouri-Shradi
4. Ring
5. Not Maxi

Somehow I have the feeling that Sands might be a pretty good penalty taker because of his coolness on the pitch. Heber and Mitrita could be also good. Not sure about Ring.

 

[dummyrun]

What is XG

ಠ_ಠ

 

[LionNYC]

Somehow I have the feeling that Sands might be a pretty good penalty taker because of his coolness on the pitch. Heber and Mitrita could be also good. Not sure about Ring.

His PK won a national championship for the NYCFC Academy. He’s as cool as the other side of the pillow.

 

[S1ARK5]

Just Dome finally solving the puzzle.

Let's hope the puzzle isn't too hard the rest of the year.