Expected points & goal difference- How accurate can predicting a final EPL league table be?

Points per game is boring

Trying to predict the outcome of a team’s position at the end of the season is always going to be tricky and probably, in most people’s eyes, near on impossible and in some people’s opinions pointless however there are ways in which we can get reasonably close in achieving this. Given it’s the international break (yawn!) I thought this was a good time to publish what I have been working on.

Goals per game and goals against per game, points per game have all been used in the past to calculate the final league positions of clubs across various leagues but the future is nigh and when expected goals came along in giving us a better measure of a team’s performance across not only a single game but also longer term across a whole season we also gained a significant and more reliable way of assessing the way a league table may look at the end of the season.

I am going to be using two metrics. Expected goal difference and xP (expected points) which I believe are sustainable as they have been used in articles published before and also they are fairly accurate. In calculating expected goal difference I will use the difference between xG and xGA multiplied by the remaining games to get our goal difference totals and league position finishes and for the xP component I will use a more intricate method detailed below.

For this article I will be focusing on the English Premier League and in a few days publish another article covering the Championship and Leagues 1 and 2.

Expected Goals (make a) Difference

Using goal difference is probably not the most thought about way of trying to predict a final league table and work has been published in the past using simple goal difference and it has been proven to be a good indicator of the longer term sustainability of a team’s results but in this article I will not only put forward the case for what I believe are the benefits of using this measure but also why it may not be the best indicator.

So let’s start with the methodology used in creating the tables within the goal difference using some examples from a previous season.

The best teams earn the most points and it’s fair to say they will end up scoring more goals and in theory will end up conceding the least amount of goals over the course of the season but will this dictate their final league position? In the mind’s eye of course but in reality is it the same?

In an excellent, albeit brief, article published in 2012 on 5addedminutes.com Newcastle were used as an example of this so rather than trying to search through seasons and seasons worth of data to try and find something similar I will use them as an example.

The EPL does not seem to change in regards to competitiveness from year to year and it is often spoken about that there are three ‘mini leagues’ within the EPL. In the 2011-12 season Newcastle United finished the season on a goal difference of +5 and 5th in the table on 65 points (normally this would equate to 55 points and only be good enough for around 7th or 8th position) so in theory they overachieved by 11 points and with only eight wins by a single goal and a trio of big defeats. These slender one goal margin victories can inflate a team’s final league position and although results fluctuate this may in part explain the Magpie’s final position in the table and of course there is no reason to believe this is sustainable winning by these small margins. The following year Newcastle finished in 16th position with a -23 goal difference with a 41 points total proving that it certainly wasn’t a sustainable way to play over the course of a season.

The model above, as I explained above is merely using standard goal difference so what happens if we use expected goal difference to produce an ‘alternate’ final goal difference table.

The graphic below shows the EPL table at the end of the season when we apply this model.

EPL xG diff
EPL 18/19 final league table using Goal difference as the measure.

The method for this is simply xGF per game – xGA per game = goal diff / 36 (games remaining) = final goal difference. Then we can assign a final league position based on the given clubs goal difference.

This is where I have some doubts about the reliability of this method with some of the numbers a little low on the eye test but we shall see in time. When referring and compiling anything along these types of lines looking at previous data sets is vital and it is imperative to go back and look at a previous season or seasons. For simplicity, we will look back to last season’s final EPL table.

Manchester City topped the goal difference charts with a +79 goal difference and finished as champions and in my table above this year they could finish with a +69 GD, a Liverpool (4th last year with +46 GD) are expected to have a +40 at the end of this year so all looking pretty rosy so far in comparison.

Things get a tad shady when we go a bit further down my table. Watford are expected to end up with a GD of +26 and a possibility of finishing 4th when last year they finished on -20 and ended up finishing in 13th  on 41 points, quite a large margin and unlikely I think it is fair to say. Bournemouth, with an expected GD of +19 and a final league position of 5th   (it should be noted I have placed them above Spurs for alphabetical order purposes only in this table), performed significantly worse last season with a -16 GD and a points finish of 44 points,  one would assume they are unlikely-ish to finish in the Europa League positions this year.

Watford celebration.JPG
Watford- Can Troy Deeney and co beat the usual suspects into a top 5 finish?

But is it so inconceivable that Watford could finish above the likes of Arsenal, Spurs and Manchester United? In truth it probably is but it is interestingly comparable in that last season The Cherries finished only 2 places and 3 points in front of The Hornets with only a -4 GD in the favor of Eddie Howe’s men. Judging by my table they will be closely matched again this year only higher up the table. However unlikely that maybe this is a useful comparison when identifying which teams in the league are consistently  going to be your closest rivals in terms of gaining points throughout the season and I’ve only compared one season to another of which one is predicted but the trend is already (slightly) evident.

The bottom four look a fair bit more ‘believable’ in regards to their final league positions but I still think this table is slightly skewed somewhat given Cardiff and Fulham are playing in a different league after being promoted last year so their data is less reliable even though Neil Warnock’s side look nailed on in finding it tough going getting out of the relegation places.

Now we have identified a kind of mini trend in regards to comparing two team’s possible similarities in scoring and conceding goals to end up with similar goal differences and final league positions come the end of the season I now want to look at my alternative method of calculating a points finish.

xP (Expected Points)

I will start with the methodology again to briefly get a gist of what to expect and to give some clarity.

I will use Burnley as a quick example. Using an adapted Poisson distribution calculation we input Burnley’s xG per game in the first column and their xGA into the second column this then spits out the expected points per game they should achieve we then multiply that by the remaining games (36) to give us Burnley’s expected points finish from there we can assign them a final league position. This is similar to what betting agencies would use in a type of Monte-Carlo type calculation. Simple, right? My math symbols are poor at best so I will leave them up to smarter people than me but the methodology is the same regardless. On to the table.

EPL xP finish 1819

Although this is my personal preferred method of calculating a team’s expected points finish it, like any predictive model, is never going to be 100% perfect as Manchester City’s 103 xP finish in my table may show… but hang on a minute didn’t they did finish on 100 points in winning the title last year? So perhaps I am being too harsh on the model, myself and Pep Guardiola’s rampant City squad. And as with the Bournemouth/Watford goal difference situation over the last two seasons we see a bit of consistency edging in, as we should of course if the workings are correct.

Throughout the process of compiling these two sets of equation’s my mind swung back and forth in regards to one being a better method than the other and its clear to see why. There are many similarities throughout when comparing the two tables. The top three are the same and the top five in general look very familiar, 7th to 11th are fairly similar as are the bottom four.

Nuno Santo
Wolves manager Nuno Esporito Santo

One team sticks out above the rest though. Wolverhampton Wanderers.

Nuno Esporito  Santo has carried on the form and style of play that has seen his Wolves team comfortably go up as champions in the Championship last season with their infamous 3-4-3 shape. Is it possible that they could continue as they have started this season and end up in 4th spot come the end of the season? Of course if we go by the goal difference table most definitely not as they would finish 7 places below in 11th, something to ponder whichever way your opinion might sway.

Arsenal are starting to come round to Unai Emery’s way of thinking so I would expect them to finish higher than the predicted 8th but my confidence in that happening isn’t helped by the fact in both tables they are only a place or two apart and this would also, obviously, depend a fair bit on Bournemouth, Watford and Wolves faltering which they may well end up doing of course but in my opinion a top four, or even top five finish for the Gunners is not looking good either way.

unai emry
Unai Emery’s Arsenal side are showing signs of improvement but can they finish high enough come the end of the season? 

How many teams will be involved in the fight for the last of (in my opinion) the automatic champions league spot after City, Liverpool and Chelsea remains to be seen but The Gunners have a fight on their hands from some unlikely teams at this point in the season and the end result will be interesting come the 19th of May next year.

Take it or leave it (depending on who you support)

So there we have it and make of it what you will and take from it what you want but you will probably want to disregard all of it if you are a Huddersfield United, Cardiff City or a Newcastle United fan and you may want to think about crossing your fingers and toes if you’re a supporter of the teams in the red half of Manchester and North London. Only kidding (maybe).

Gareth Cooper

GC Analytics

xG and xA: Some (really) early numbers

Oh not this again, isnt it to early for this??!

The critics of xG (expected goals) and xA (expected assists) will be so annoyed that I have brought this up so early on in the season. In fact some of the main supporters of the two metrics will probably frown upon it as well.

Personally, I like to start as early as possible in collecting the data and also to publish some early tables, I have my reasons and they are the following.

  • Have last seasons top players according to these metrics started the season as they left off in 17/18?
  • Are there any new players that have started the season well?
  • Given relegations and promotions are they any players who have performed well with their clubs?
    Transfers. Have any new signings started well with their new clubs?

The EPL

This is one reason why I like to get a look in early on.

Mo Salah has indeed it seems carried on his form from last season in to this one. Okay i hear you, and I cant push it any further than what it is and of course xG is a minor factor in performance statistics but Salah looks to have returned from an unsettling time with Egypt at the World Cup with confidence and his performances so far have backed that up. His xA numbers arent quite their but they will be, we all know it.

Another Liverpool man Andy Robertson has provided an assist so far and a question to be asked is does Jurgen Klopp want more of this type of attacking threat from the Scotsman this year?

Sergio Aguero is already exceeding his xG numbers and still looking like a world class striker season after season. It should be noted the xG stat is probably irrelevant at this stage (no player will get near .90 per 90 mins all season).

When we talk about assists in the EPL we tend to think of De Bruyne, Salah and Hazard and with KDB out for a few months it will be up to Salah and Hazard (I also expect David Silva, Lucus Moura and James Maddison to have decent numbers xA wise) to have excellent seasons in providing assists for their respective clubs.

SkyBet Championship

According to the pundits and a lot of professional people within English football, this is probably the toughest of the four leagues to get promoted from that I have written about in this piece.

Dean Smith’s Brentford side impressed many people last year and they have began this season well again. With his preferred 4-3-3 shape, with Josh McEacharan in the heart of the midfield, Brentford play a quick, free flowing brand of attacking football and with Neal Maupay at centre forward and having scored 5 goals in as many games and 3 assists to boot Brentford have a striker in form and after his 17 goal haul last term Bees fans will hope it continues.

One of the reasons I gave earlier for wanting to look at some xG numbers this early was to see if some goalscorers would adapt to their leagues after relegation. Step forward West Brom’s Jay Rodriguez. Injury hit Rodriguez looks to have come into some early form and the Championship is suited to him theres no doubt.

Former Villa and Derby striker Andy Weimann is also playing well after his transfer to Bristol City with 5 goals exceeding his xG by 2 goals.

Can Norwich get a good return out of the much travelled Jordan Rhodes? Can Marco Bielsa do the same with the strikers he has at Leeds United until he can get new faces in in January? Both cases are in the balance in my opinion.

Sheffield Wednesday’s Barry Bannan is player who has the ability to change a game in the Championship but needs to improve on his assist tally from last year (3 assists) and would do well to repeat the 8 he achieved in the 16/17 season.

Another of Bielsa’s charges at Leeds, Pablo Hernandez, is another man who is very good on the ball and has 2 assists to his name already contributing to Leeds top of the table form.

I expect Leeds United’s Samu Saiz, QPR’ s Luke Freeman and Derby County’s Tom Lawrence to top the assist and xA tables come the end of the year especially Freeman who is more than capable of beating his 12 assists from last year.

SkyBet League 1

Walsall’s Andy Cook has had some very good chances to have at least 3 more goals than the two he has so far. His xG total of 5 suggests the former Carlisle man isnt finishing as well as he should do.

On loan Nottingham Forest hitman Jason Cummings has the opposite issue he cant stop scoring. Having scored 6 goals and with an xG of 3.93 the Scotsman hasn’t had any trouble in finding the net and in doing so propelling Posh up up the table.

Shaun Whalley’s delivery from set pieces doesnt come as a surprise to some of us hence him always having pretty decent numbers when it comes to asssits and he will pick up as the season progresses.

Sunderland’s Liam Gooch looks to be finally finding some form and long may it continue. Gooch has 3 assists already this season and with an xA of 1.95 the signs are looking good for him if he can stay injury free.

SkyBet League 2

In the table above are two reasons why I like to do these tables early.

MK Dons were relegated from League 1 last year but have began the season in League 2 unbeaten but they lack serious potentcy upfront. Kieron Agard has had some very good clear cut chances to add to his 2 goals, trust me ive seen it with my own eyes, hence his xG of 5.3. But his finishing needs a lot to desired and Paul Tisdale will need to think about this area of his squad if the Dons are to mount a serious promotion push.

On the flip side newly promoted Tranmere have a promising goalscorer in James Norwood who has scored 5 times this season and exceeding his xG of 4.07.

Assist wise bottom of League 1 Notts County have Dutch winger Enzio Boldewijn. The former Utrecht and Crawley Town man looks impressive early on and although mangerless after the sacking of Kevin Nolan, County shouldn’t be in the current situation come the end of the season given the squad they have to compete in League 2.

In summary

As I have said a great deal in this article it is still very early days in the season but I hope I have given you some insight as to what you can expect from your team in the coming season.

xG and xA numbers are always open for criticism for different reasons but they are a good marker when looking at taking goalscoring chances and also the creation of chances by the main players who try to provide those opportunities.

Gareth Cooper

GC Analytics