Topic: Statistics and Analytics - insight into our performance

[Trendisnotdestiny]

First 13 games last season yilded 9W 3D 1L. Last 12 games yilded 8W 3D 1L.

So that right there is proof we can string good winning runs together.

Maybe, we are laying in the weeds for the Mancs?

Its a six pointer --- and oh would the Kop go Klopp crazy were we to punk these bastards.

Maybe the start and the end of the Int'l breaks will put us back where we belong --- up top? 

[U-238A]

Being new to the statistical analyses side of the game, I may have missed it... Can anyone point me to a similar analysis or tool that calculates Expected Goals Against which is independent of the chances created for xG? Does one exist?

Thanks

[Groundskeeper Willie]

What it says to me is we do a lot better when we've only got one game per week, which we haven't this season

This season is way to small a sample size to draw any conclusion whatsoever.

[McrRed]

I think we've discussed it previously but the Anfield Index Under Pressure podcast really is an excellent look at LFC stats, game by game.

[JCB]

xG and Efficiency

This week I've decided to see if I can replicate the analysis from this article https://statsbomb.com/2013/08/goal-expectation-and-efficiency/ and see if it's possible to use my data to determine how efficient/clinical we are in relation to the other teams this season.

It's a mixed bag of results that doesn't paint a good picture for us but I'm taking it with a pinch of salt. The analysis doesn't really account for the frequency of chances (either for or against) and this can play a huge factor if using this solely to draw conclusions. For example, Burnley might be clinical from their chances but if they don't have many to begin with then it stands to reason they won't score many; a team such as ourselves however will have plenty of chances. In other words we don't have to be clinical - unlike burnley - to win the game (but it sure as hell would help).

So here's the Expected vs Actual plots, what's interesting to note is the deviation from the diagonal line. The further away you are from this line, the better/worse (depending on which side of the line) you are than you should be. Looking at the first table for example (xG for) we can see that the biggest deviations are Chelsea (over performing) and Crystal Palace (under performing). Unfortunately we're under performing both offensively and defensively.



Using this data we can also try and calculate some 'efficiency scores'. Here's the tables based on this season's data:



And here is the data plotted visually:



It's a small sample of games to draw conclusions from and time will tell if this will change over the season (come on Liverpool), but clearly there's a lot of work for us to do. The point to remember is that unlike a lot of other teams our defensive and offensive performances are intrinsically linked, once one starts improving I suspect the other will as well.

[killer-heels]

So we can't attack and can't defend. Nice.

[19th Nervous Title]

So we can't attack and can't defend. Nice.

Where's your diagram KH?

So, today is world statistics day. There seriously is such a thing. Hope to see a lot of diagrams and charts from the usual suspects in here.

[Funky_Gibbons]

So, today is world statistics day. There seriously is such a thing.

What are the chances of that?

[JCB]

What are the chances of that?

1 in 365 

Wasn't even aware myself but it doesn't surprise me, they'll make a day of anything these days.

In honour of today I'll contribute something I'm working on at the moment. It's somewhat unfinished but it's the only thing I have to offer at the moment. Basically I got the idea last season on the top 6 mini league thread. We all work on the assumption that we need to average 2 points per game (ppg) to get a top 4 spot, and that 2.5 ppg will win you the title. The question I had at the time was how did that translate if you broke the league into smaller groups. In other words how much on average was needed against the top half as opposed to the bottom half. So I grabbed the data from 2000 all the way up to last season so that I could have a big enough sample size. I broke the teams into quarters, thirds and halfs as I was unsure which method would yield better results and came up with this:

 

By looking at the PPG (No Groups) column you can see the results are actually lower than our initial assumptions of 2 ppg for Cl and 2.5 for title winners. Because of this, I decided to have a look at the median values instead so that I could ignore any outliers that may have thrown the dataset but I still got very similar results (differing from the average by 0.05). I then had a look at the last 3 seasons and realised that my average results seemed about right.

Just to be on the safe side however, for one last check I decided to plot the data on a graph and added some trendlines to them to see if the historic data had anything to say and came up with this:



Basically for the top three positions there's a minimal trend with regards to Final Points over the course of the last 17 years, however the next three positions shows a more pronounced upwards trend, pretty much reinforcing the fact that the EPL has become more competitive over the last few years, especially for that CL spot.

I'll need to do some thinking/analysis to see if this needs to be factored in or not but if it all pans out I might use the results for benchmarking or possibly forecasting. In the meantime if we take the results at face value as it is (and I'm not quite there yet as it's not fully finished) it means that aprox. 1.85 ppg per game might be enough for 4th spot or seen another way 1.5 ppg against the top half and 2.15 ppg  against the bottom half.

[BabuYagu]

/snip

Happy Stats Day mate!

This is very interesting. Out of curiosity, if you shit out of the matches against the top 6 and average say 0.5 points per game, how many points would you need from the other games to win the league. Also I remember reading last season that Man City getting 2.5 points per game against the bottom 10 was something of a record in the league. Amazing with that record they still needed a result on the final day just to get Champions League football. But it showed me that the old adage that you can win the league by being a flat track bully might not be accurate. You still need to pick up a reasonable amount of points against top sides somehow, somewhere.

[Achilles Heel]

Happy Stats Day mate!

This is very interesting. Out of curiosity, if you shit out of the matches against the top 6 and average say 0.5 points per game, how many points would you need from the other games to win the league. Also I remember reading last season that Man City getting 2.5 points per game against the bottom 10 was something of a record in the league. Amazing with that record they still needed a result on the final day just to get Champions League football. But it showed me that the old adage that you can win the league by being a flat track bully might not be accurate. You still need to pick up a reasonable amount of points against top sides somehow, somewhere.

City were 2.55 ppg against the bottom 10 but no record. See below for that.

All of Chelsea, Spurs, City and United lost just 1 game against bottom half teams, Arsenal 2 and Liverpool 5.

Chelsea 2.65
Spurs 2.65
City 2.55
Arsenal 2.50
United 2.25
Liverpool 2.05

Then against the bottom 14 (all non-Big 6) Chelsea's record was even better with 2.75ppg and a record of W25D2L1. Spurs and United only lost 1 as well, City 2 and Liverpool 6.

Chelsea 2.75
Spurs 2.54
City 2.43
Arsenal 2.36
United 2.11
Liverpool 2.00

Liverpool were, of course, top dog in the head to heads with the Big 6 matches.

Under Guardiola's stewardship City have a true "flat-track bully" record in away games  in the league and domestic cups against non Big 6 EPL teams - incredibly  Won 21 Drawn 1 Lost 2.

[JCB]

Hiya Babs,

If you only got 0.5 ppg against the other 5 teams in the top 6 that would equate to 10 points.

The average Points total for winning the league was 87 points over the 17 years.
That means that you would have to get 77 points from the remaining 28 games : ie 2.75 ppg

Altenatively the lowest points total in that period was 80 points (happened twice: 2001 & 2011)
That would mean 70 points from the remaining 28 games: 2.5 ppg

I'm not sure if you ever saw my grouped PPG analysis back in April but with about 6-8 games remaining it looked like this. Tottenham were giving them a run for their money against the bottom half of the table near the end.

[JCB]

http://www.oulala.com/en/community/blog/item/premier-league-conversion-rate

They try and paint it as doom and gloom in the article but I'm a glass half full kinda guy:
We have the 2nd most shots created and that means we're creating a lot of chances... and as long as we keep doing that then there is scope for our conversion rate to increase as a result (Of course this article doesn't account for the quality of chances created...that's where xG comes in).

[BassTunedToRed]

Of course this article doesn't account for the quality of chances created...that's where xG comes in.

Indeed, but unfortunately it paints a worse picture...

[Yiannis]

^^Christ. What's going on with Chelsea? They were like that last year too. Pedro's goal last Saturday is a good reason why this keeps up maybe.

Quick question because this keeps bugging me a bit. Richarlison's glorious missed chance vs Chelsea. Why is it only valued at 0.57 xG? Shouldn't be a lot higher as in close to 1? Did the proximity of Azpilicueta play its part?

https://understat.com/match/7199

[BassTunedToRed]

Quick question because this keeps bugging me a bit. Richarlison's glorious missed chance vs Chelsea. Why is it only valued at 0.57 xG? Shouldn't be a lot higher as in close to 1? Did the proximity of Azpilicueta play its part?

https://understat.com/match/7199

I expect the defensive pressure played a part, but also it was a header and they always have a lower xG. Take a look here:

https://torvaney.github.io/projects/xG.html

As an example, a shot from about three yards out centrally has an xG of 0.78, but that drops to 0.61 if it's a header.

Irrespective of that, it's a cool thing to play around with!

[clinical]

In our first 11 only 3 players were signed by Klopp. Matip, Mane and Salah. Great players especially the latter two. Amazing stat actually after 2 years.

[BabuYagu]

In our first 11 only 3 players were signed by Klopp. Matip, Mane and Salah. Great players especially the latter two. Amazing stat actually after 2 years.

Slightly misleading as Gini would start if fit.

Agree though. I suspect by 1/9/18 we will have a new CB, GK, DM(Can replacement), & CM (Keita) all starting too. Maybe more if Coutinho leaves. It's a credit to Moreno's performances so far this season that he is keeping Robertson out of the team too. I'd written him off after Sevilla up until maybe end of July this year when I noticed some changes in his game that made me believe he would start defending with some tactical intelligence instead of just running and diving into every tackle.

[Yiannis]

I expect the defensive pressure played a part, but also it was a header and they always have a lower xG. Take a look here:

https://torvaney.github.io/projects/xG.html

As an example, a shot from about three yards out centrally has an xG of 0.78, but that drops to 0.61 if it's a header.

Irrespective of that, it's a cool thing to play around with!

Thanks mate. Made a mistake actually. I'm talking about the 47th minute chance that Richarlison missed with his left foot in what was a virtually open net. They rated it as 0.13 which is mad.

https://youtu.be/ZXROAJeeci8?t=3m36s

[BabuYagu]

Thanks mate. Made a mistake actually. I'm talking about the 47th minute chance that Richarlison missed with his left foot in what was a virtually open net. They rated it as 0.13 which is mad.

https://youtu.be/ZXROAJeeci8?t=3m36s

That will be down to the pressure of the defender. I agree it still seems low though considering Mane's back post chance he missed, which was under more pressure and the ball was at a more awkward height was rated at 0,56 I believe. I would score this likewise.

[BassTunedToRed]

Thanks mate. Made a mistake actually. I'm talking about the 47th minute chance that Richarlison missed with his left foot in what was a virtually open net. They rated it as 0.13 which is mad.

https://youtu.be/ZXROAJeeci8?t=3m36s

Ah right, yeah that does seem very low. All models are different though, so there'll always be discrepancies.

[JCB]

That will be down to the pressure of the defender. I agree it still seems low though considering Mane's back post chance he missed, which was under more pressure and the ball was at a more awkward height was rated at 0,56 I believe. I would score this likewise.

I'm actually going to be controversial here and go against the grain:

I've long suspected that xG doesn't really use the defensive positions of the opponents when calculating their values. It's not that it's not relevant, it's just that it's not readily available in the data feeds and would therefore be a nightmare to include in the calcs. What they do offer however is:

a) 'Big chance' qualifier which is denoted as: 'Shot was deemed by Opta analysts an excellent opportunity to score – clear cut chance eg one on one'.
b) Goalkeeper position when shot is taken (this is not as useful/detailed as you would expect)

There's lots of other 'qualifiers' such as shot position, assist type, body part (which bassturnedtored has covered), type of play (ie. open play, setplay etc **) and a lot more which I won't go into detail about but none that I can see that accounts for the defenders when the shot is taken, just the two I've listed above.

The thing is I've just had a look at both Richarlison's chances on the opta feed and they are both listed by opta as 'Big Chances' so this  doesn't add to the disparity that you've highlighted. I'm still trying to wrap my head around how these models are calculated based on what data opta provides and there's always a chance that they're getting defender positional data from somewhere I'm not privy to yet but my gut says that's not the case. You can extract whether the shot event type was blocked, (thus denoting players in the way), but not much more than that and it doesn't apply to both of those chances anyway. If somebody has evidence to the contrary I'm more than happy to be proved wrong on this, until then I'm crying foul on that stat - something's not quite right there.

** - btw  Opta have the header listed as coming from a set piece play. I've watched the chance and the the ball is cleared out and delivered straight back in again. As a result understat seem to have categorised it as open play. Considering all the players are still in the box for the second attempt I'm more than likely to lean on Opta's definition of that play rather than understats', but I can definitely understand the ambiguity.

[Klippity Klopp]

Saw this on Twitter, not sure what it means really.

[SamAteTheRedAcid]

Saw this on Twitter, not sure what it means really.

Who the hell is IronBorn, Woody, and Ghost?

Must have missed it when we signed Theon Greyjoy, a talking cowboy and Patrick Swayze?

[Klippity Klopp]

Who the hell is IronBorn, Woody, and Ghost?

That I do know.

Milner, Woodburn & Wijnaldum.

[Lastrador]

Had no idea Milly had lost his dong, wonder if that was the reason his corners were so much better yesterday.

[Djozer]

Had no idea Milly had lost his dong, wonder if that was the reason his corners were so much better yesterday.

The majority of this thread goes waaay over my head, but this is the type of statistical analysis I can really get behind.

[Klippity Klopp]

[Klippity Klopp]

[idontknow]

In our first 11 only 3 players were signed by Klopp. Matip, Mane and Salah. Great players especially the latter two. Amazing stat actually after 2 years.

Agree with Babu just down the page that Gini would have been in too.
But also Klopp signed Karius, and that hasn't worked out so far, we've got Keita, but not till next season, and the whole summer was spent waiting for Virgil.

So that could have been 7 first-teamers in 2 years, which would have been an interesting turnover.

Agree with you though, all the intentions aside, only Mane and Salah are indisputably good enough for us.

[Klippity Klopp]

[Bend It Like Aurelio]

Surprised Fraser Forster is up there, considering how many mistakes he’s made.

[harryc]

No great surprise those keeper stats, as LFC fans we have probably suspected as much.

[JCB]

Interesting. Nice graphic.

The fact that Pope comes out top makes me suspect that the saves are more than likely due to Burnleys defensive block restricting the quality of shots (ie  taken from further out). I'll see if I can get some time over the next couple of days and do a similar analysis based on xG. It might tell a more insightful story.

[BabuYagu]

Surprised Fraser Forster is up there, considering how many mistakes he’s made.

I think this is pretty much the problem with how lay people like us assess goalkeepers. We don't really have a clue on their technique really, if they are over compensating near post, poorly positioned for shots, reacting time etc. And a difference of 1/2 a yard either way is the difference between easy save and needing to make world class or getting nowhere near shots. We seem to solely value keepers on world class saves and mistakes pretty much.

So Forster, at the moment, makes mistakes but also is a great shot stopper. Is that preferable to someone who makes less mistakes but is a worse shot stopper?

[Chris~]

Doesn't that sort of thing depend massively on shot quality though. Which is why Pope is at the top in the same way Heaton was near the top last year, because Burnley force lots of shots from distance with lots of bodies in the way. Where as we give up a low amount of shots but they tend to be high quality. The chart would be interesting with xg per shot added in somehow

Or I should have read down and jcb is already looking at that. 

[Bend It Like Aurelio]

I think this is pretty much the problem with how lay people like us assess goalkeepers. We don't really have a clue on their technique really, if they are over compensating near post, poorly positioned for shots, reacting time etc. And a difference of 1/2 a yard either way is the difference between easy save and needing to make world class or getting nowhere near shots. We seem to solely value keepers on world class saves and mistakes pretty much.

So Forster, at the moment, makes mistakes but also is a great shot stopper. Is that preferable to someone who makes less mistakes but is a worse shot stopper?

I think in terms of stats, someone like Fraser Forster is always preferable to someone who is a worse shot stopper but makes very little mistakes. But in terms of confidence inspiring goalkeepers, you would probably prefer the one that makes less mistakes.

[Klippity Klopp]

No correlation between a team's height and results, says study

There is no correlation between the average height of a team's players and its success, says a new report.
A study by CIES Football Observatory found the average height of a player in 31 European top flights was 5ft 9.7in.

Manchester City and Everton were found to be the shortest teams on average in the Premier League, with West Brom and Huddersfield the tallest.

"The gaps observed rather reflect different approaches to the game," said the report's authors.

"While some coaches prefer tall players, other give priority to shorter ones. However, very short players will struggle to establish themselves as professionals."

Barcelona, Lyon, Nice, Sevilla and Real Madrid are also among the teams to field shorter-than-average players, with Bulgaria's Ludogorets the shortest.

FC Copenhagen were named as the tallest team in Europe, while Schalke, Atalanta and Roma also have notably tall players on average in their squads.



The English top flight ranked 11th tallest among Europe's top 31 divisions, with Jose Mourinho's Manchester United the third tallest team on average in the Premier League.

Joining Manchester City and Everton among the shortest teams in the league are Bournemouth, Liverpool and Watford, while champions Chelsea rank slightly taller than average.



Following Scotland's failure to qualify for the 2018 Russia World Cup, former head coach Gordon Strachan pointed to his team's lack of height and physicality after a 2-2 draw with Slovenia.

"Genetically, we are behind," he said. "In the last campaign we were the second smallest, apart from Spain."

While clearly it does not contain solely Scottish players, the country's Premiership is the 13th smallest on average, while Spain's La Liga came in as the second shortest behind the Bulgarian top flight.

Meanwhile, Swansea City's Sam Clucas, who is 5ft 8in, told Football Focus' Mark Clemmit last month how his height hindered him through academy football and nearly prevented him from turning professional.

Clucas was told by coaches he was not tall enough to make it at the highest level.

The 27-year-old moved from Hull City to Swansea City in the summer of 2017 for an undisclosed fee.

http://www.bbc.co.uk/sport/football/41818468

[BassTunedToRed]

I'd put more faith in this sort of thing, which is based on xG:

https://twitter.com/footballfactman/status/923632562861412355

[JCB]

This is what I got (data from understat.com)