Topic: Statistics and Analytics - insight into our performance

[Nessy76]

Come on, own up which Muppet said VVD was slow. He clocked faster speed than Mbappe in our match. Was fastest player.

Dunno. Trent burned off Neymar a couple of times.

[stevieheighway]

Really good article assessing the season so far using xG for and against for all teams..

https://www.rte.ie/sport/soccer/2018/1009/1002020-8-games-in-expect-the-unexpected-as-season-takes-shape/

[Larse]

Really good article assessing the season so far using xG for and against for all teams..

https://www.rte.ie/sport/soccer/2018/1009/1002020-8-games-in-expect-the-unexpected-as-season-takes-shape/

Shows that Chelsea and especially Arsenal are overperforming. Watched quiet a lot of their (Arsenal) games this year as they never played the same time we did, and they they got quiet lucky in most of their wins. Watford, Everton & Fulham had loads of chances against them. If they keep on playing like this a drop-off is inevitable.

City’s other wins have all come against teams struggling at the wrong end of the table, while Liverpool have notched up results away at trickier foes such as Leicester and Crystal Palace.

+ Tottenham and Chelsea. City on the other hand seem to be playing Cardiff every week. 4 of their 6 wins were agains the bottom four of the league (Fulham[H], Newcastle[H], Huddersfield[H], Cardiff[A]).Other wins were Brighton and Arsenal in their first game with a new manager.

[stevieheighway]

City on the other hand seem to be playing Cardiff every week.

Yep, they have had a very gentle start to the season which may well prove valuable to them, especially has they had KDB out for a good chunk of the period. My hope is that the intensity of some of their upcoming games might catch them on the hop a bit...

[BabuYagu]

Working on an article at the moment to hopefully post tomorrow before the game. But wanted to share this with you all as I thought it was pretty interesting.

I'm looking into big chances, their importance to top sides, why we had a slump in creating them, how we can create more of them, etc.

Made some interesting discoveries when digging into how we are creating those chances (crosses, set pieces, through balls, gegenpressing, etc). Was curious though, without looking into this for yourselves, what do you instinctively believe to be true just from watching us this season & looking at the above?

[rise of the rest]

Where are people sourcing the raw data from?

[BabuYagu]

Where are people sourcing the raw data from?

Technically, all raw data comes from opta.

In this instance, we are using all chances with an xG value above 0,28 to indicate big chances. The best site to verify this publicly would be understat

[RogerTheRed]

Would it relate to the number of misplaced passes about 30 yards out, inversely proportionate ? In most of our low chance games (apart from City home iirc) we had a frustrating knack to have promising breaks into the opponent half that fizzled out. Imagine Salah had passed to a Cardiff defender just before the fourth against Cardiff, mo chance would result.

[kloppismydad]

Played with a hand-brake on during the tough run, especially with the impeding international break. We'll go a bit more gung-ho in our next 2-3 games, then I expect the hand-brake again with another international break on the horizon.

[rise of the rest]

Technically, all raw data comes from opta.

In this instance, we are using all chances with an xG value above 0,28 to indicate big chances. The best site to verify this publicly would be understat

Is it paid or free?

[Yiannis]

Is it paid or free?

I know it must be dead intimidating but don't be afraid to google it. Sod that, here it is

https://understat.com/team/Liverpool/2018

[BabuYagu]

Would it relate to the number of misplaced passes about 30 yards out, inversely proportionate ? In most of our low chance games (apart from City home iirc) we had a frustrating knack to have promising breaks into the opponent half that fizzled out. Imagine Salah had passed to a Cardiff defender just before the fourth against Cardiff, mo chance would result.

Article here if anybody is interested.



This was the most interesting thing for me. While it's only 10 games - that is a strong correlation between throughballs and xG. And no correlation really at all with crosses.

For me it's the difference between focusing our buildup play down the flanks = crosses / through the middle = through balls. Plus when you play centrally, you get a lot of passes that aren't quite through balls but are effectively. Those short, incisive little passes around the box like the one for Shaqs goal, or the ball Shaqiri played to Salah for his. They aren't through balls, but are happening in the same positions.

If we are focused on the flanks, it is more of a lottery. You are playing in percentage balls and for forwards who aren't ideally suited to such service perhaps.

[Yiannis]

https://understat.com/match/9330

Paints certainly a fair picture of the game going by what teams created last night. Oh and Mina's header was offside.

I like xG and is the only stat I go about searching after the games are finished but of course they aren't perfect. Like Mina's header worthy of 0.31 xG and Shaq's great chance only 0.39.

Oh well, at least not that bad as infogol's chance grading in Firmino v Spurs and Salah v Soton 

[groove]

Yeah, I don't find Understat's xG as accurate as Caley's but it's still good as an accessible xG table. At full time I thought the game had been fairly even, but after looking at the xG chart it did remind me of a couple decent chances that maybe did indicate we edged the game.

[Ray K]

https://understat.com/match/9330

Paints certainly a fair picture of the game going by what teams created last night. Oh and Mina's header was offside.

I like xG and is the only stat I go about searching after the games are finished but of course they aren't perfect. Like Mina's header worthy of 0.31 xG and Shaq's great chance only 0.39.

Oh well, at least not that bad as infogol's chance grading in Firmino v Spurs and Salah v Soton 

I love that Divvy's goal was 0.90 xG.  I mean, technically that's true, but doesn't *exactly* describe the goal as such... 

[Yiannis]

I love that Divvy's goal was 0.90 xG.  I mean, technically that's true, but doesn't *exactly* describe the goal as such... 

Yeah. I can certainly understand it though cause they count ( I guess) what happens after the ball is bounced off the post. So considering where Origi is, where Pickford is (lol) and where the ball lands, it's a gimme.

[PhaseOfPlay]

Yeah, I don't find Understat's xG as accurate as Caley's but it's still good as an accessible xG table. At full time I thought the game had been fairly even, but after looking at the xG chart it did remind me of a couple decent chances that maybe did indicate we edged the game.

How do you define "accurate"?

All expected goal models are necessarily inaccurate, in my opinion. That doesn't mean they are useless, because they are quite useful. But they are all imperfect models.

[AmanShah21]

Is there any metric that measures the runs made by forwards in behind (It doesn't have to specifically be ones where they are actually being passed to, just runs in general). It feels like we're having fewer of those this season because we dont often have players who can play that final ball. We had loads of those last season to just open up the defences and force them into decisions. It can sort of be seen in game, but I wonder if there's a stat/metric that actually captures something like that.

[PhaseOfPlay]

Is there any metric that measures the runs made by forwards in behind (It doesn't have to specifically be ones where they are actually being passed to, just runs in general). It feels like we're having fewer of those this season because we dont often have players who can play that final ball. We had loads of those last season to just open up the defences and force them into decisions. It can sort of be seen in game, but I wonder if there's a stat/metric that actually captures something like that.

There should be. Forward runs ahead of the ball, and penetrating runs behind defenders are the cornerstone of a functioning attack

[AmanShah21]

I thought so. Do you know what they call it?

I am looking for numbers that show how many empty runs are being made. Cause that's the number that seems to have dropped.

[BassTunedToRed]

I thought so. Do you know what they call it?

I am looking for numbers that show how many empty runs are being made. Cause that's the number that seems to have dropped.

I doubt you'll find that anywhere. Data in the public domain usually focuses on events on the ball. Clubs will have the data you're after but probably nobody who will give it to you!

[BrandoLFC]

Caley's model is good because he documents everything he did to make it and why.  Understat, Infogol and the rest are all understandable for me because of what Caley did though I don't trust them per se as I have no idea how they came up with their numbers aside from assuming it must be similar to Caley.

https://cartilagefreecaptain.sbnation.com/2017/8/17/16164068/explaining-expected-goals-soccers-newest-star
https://cartilagefreecaptain.sbnation.com/2015/4/10/8381071/football-statistics-expected-goals-michael-caley-deadspin
https://cartilagefreecaptain.sbnation.com/2015/10/19/9295905/premier-league-projections-and-new-expected-goals

There was also a perfect example of xG being much more useful than e.g. shots/SoT from this weekend.

Shots: 16 - 7
SoT: 5 - 4

In looking at this you'd expect the team with 16 shots and 5 on target to be the better team?  This was from the Newcastle vs. West Ham, West Ham's xG was 3.1 vs. Newcastle's 1.2 xG for a 3-0 West Ham win.

[BassTunedToRed]

Caley is great, particularly with his pitch maps. But if you want easily downloadable xG for multiple leagues, FiveThirtyEight is the place to go: https://data.fivethirtyeight.com/#soccer-spi There's a spreadsheet with over 20,000 matches there.

[groove]

How do you define "accurate"?

All expected goal models are necessarily inaccurate, in my opinion. That doesn't mean they are useless, because they are quite useful. But they are all imperfect models.

Subjectively, Caley's look closer when viewing individual chances and comparing his vs Understat.

A bit more objectively, Caley's model was found to be the best predictor of actual future goals scored. That was a bit ago though so not sure if that's still true.

[rscanderlech]

A big swing in our chances, according to this: https://projects.fivethirtyeight.com/soccer-predictions/premier-league/

[farawayred]

A big swing in our chances, according to this: https://projects.fivethirtyeight.com/soccer-predictions/premier-league/

Hard to believe. It shows too much of a swing in chances based on two games.

[JackWard33]

Hard to believe. It shows too much of a swing in chances based on two games.

Don’t have to believe it it’s just a model

We just won one of our harder remaining games, City just lost one of their easiest remaining games ... that’s probably about a 4 point swing in the model and it was already close (50/50 before today)

Really these things are only a guide but what it shows you is that was a terrible loss for City

[riismeister]

8 games in 27 days in December.

Lowest xG difference: +1.4

[Mr_Shane]

Hard to believe. It shows too much of a swing in chances based on two games.

Because we are 3-0-0 , while city are 0-0-2 . Lets see what the prediction is after the city game.

[farawayred]

Because we are 3-0-0 , while city are 0-0-2 . Lets see what the prediction is after the city game.

On Dec 12, City were 60%, we were 35%. Next week, it's 53%-to-42%, and the week after that, Dec 26, we were leading City 72%-20%. This change was in the space of two games. I know that City should have won those games and they were not considered "difficult", but that change in the forecast is huge! Statistics are described in probabilities and confidence levels. The probabilities are in the table, but I can't see this model having high confidence levels. I'm not even sure if it would be possible to create a high-confidence-level model.

[DanA]

On Dec 12, City were 60%, we were 35%. Next week, it's 53%-to-42%, and the week after that, Dec 26, we were leading City 72%-20%. This change was in the space of two games. I know that City should have won those games and they were not considered "difficult", but that change in the forecast is huge! Statistics are described in probabilities and confidence levels. The probabilities are in the table, but I can't see this model having high confidence levels. I'm not even sure if it would be possible to create a high-confidence-level model.

City dropped as many points in two games as we dropped in the first half of the season. Of course that's going to have a pretty significant effect on odds.

[BrandoLFC]

On Dec 12, City were 60%, we were 35%. Next week, it's 53%-to-42%, and the week after that, Dec 26, we were leading City 72%-20%. This change was in the space of two games. I know that City should have won those games and they were not considered "difficult", but that change in the forecast is huge! Statistics are described in probabilities and confidence levels. The probabilities are in the table, but I can't see this model having high confidence levels. I'm not even sure if it would be possible to create a high-confidence-level model.

Yet the betting markets shadow this pretty closely.  And why?  Because making up 7 points on a team that has dropped 6 in half a season probably isn't likely.

[BrandoLFC]

Thought this was pretty interesting, especially the team by team breakdown towards the end.

https://karun.in/blog/expected-threat.html

[Byrneand]

Why would they deteriorate? Champion contenders usually convert at a rate of below 3:1. Do you think they're going to get that bad?

Just seeing that they're now running at 3.04 now after 26 games. Given the first 5 games were at a rate of 2.21 then it means that the 21 since have been at a rate of 3.24.

[PhaseOfPlay]

Just seeing that they're now running at 3.04 now after 26 games. Given the first 5 games were at a rate of 2.21 then it means that the 21 since have been at a rate of 3.24.

Just for context of that discussion:

Good points, but just to be clear, I was very careful to add the context of the numbers being "right now" and for performances in general to be incomplete "just yet", so I'm not making any predictions off the numbers - I'm just saying that right now, Chelsea have earned their spot in the top three, City and us are not performing as expected but we're not too far off, and to add now - the lower clubs that are running good numbers will probably drop off within the next 5 games - although there's always one that manages to be an outlier.

But you called it right. They have been in free fall attack-wise since then.

[vagabond]

Thought this was pretty interesting, especially the team by team breakdown towards the end.

https://karun.in/blog/expected-threat.html

This is very interesting. Thanks for sharing.

[Bob Sacamano]

This is very interesting. Thanks for sharing.

Does 538 have something similar? Non-shot based xG?

[BrandoLFC]

Does 538 have something similar? Non-shot based xG?

Yes, and it accounts for 25% of their model I believe.

[Geppvindh's]

Thought this was pretty interesting, especially the team by team breakdown towards the end.

https://karun.in/blog/expected-threat.html

Have to hand it to the guy, he's got a very neat blog and twitter profile.

[McrRed]

Excellent stuff.

Thought this was pretty interesting, especially the team by team breakdown towards the end.

https://karun.in/blog/expected-threat.html

You can see the merging of this model with the developing defense and goalkeeper models. Soon we won't need to play the actual matches!