English Premier League
Predicting the outcome of the 2016/17 Premier League season
The FT’s model simulates the final standings using teams’ underlying performance data

DECEMBER 16, 2016 by: John Burn-Murdoch
“The Premier League table never lies,” goes the aphorism. The claim seems simple enough — the team at the top of the table is the team at the top of the table, and when the final whistle goes on the last day of the season, that’s all that matters.

But the implied corollary that we need not look beyond the standings to determine which sides are better than others at any given time is much less robust.

At the time of writing, for example, Liverpool have played two fewer matches at home than away — they swapped venues with Burnley early in the season while their new main stand was being built. The home advantage in football is well documented, so does this mean Jürgen Klopp’s side might be better than their points and goals totals would suggest?

Then we have the matter of schedule difficulty. Table-toppers Chelsea have already played all other members of the current top six. Manchester City are still yet to play Arsenal and Liverpool. Surely this should be factored in to our interpretation of the standings?

And what of the results themselves? Goals in football are rare relThen we have the matter of schedule difficulty. Table-toppers Chelsea have already played all other members of the current top six. Manchester City are still yet to play Arsenal and Liverpool. Surely this should be factored in to our interpretation of the standings?

And what of the results themselves? Goals in football are rare relative to points in tennis, pitches in baseball and so on, meaning results in football are more subject to random variation — in other words, luck — than many other sports. We need to look beyond the scoreboard to get the true measure of a team’s performance level.

The FT’s Premier League predictor, the first iteration of which you see here, aims to deal with all three of these issues, and provide an early glimpse of how the table might end up looking based on teams’ underlying qualities and fixture schedules.ative to points in tennis, pitches in baseball and so on, meaning results in football are more subject to random variation — in other words, luck — than many other sports. We need to look beyond the scoreboard to get the true measure of a team’s performance level.

GRAPHIC

At the time of writing, current leaders Chelsea win the title in 62 per cent of the 50,000 simulated seasons, ahead of Liverpool and Manchester City, with Arsenal and Tottenham fighting for the fourth Champions League qualification berth.

The most substantive difference between the predicted outcomes and the purportedly infallible current table is at the bottom, where the model expects Burnley to be sucked down into the fight for survival, competing with Swansea to survive the drop.

Elsewhere, Southampton are projected to move up the table into a head-to-head with Everton for 7th spot, driven largely by their excellent defensive record. Saints have conceded fewer shots-on-target per match than any other side in the league this season, according to the FT’s data.

In order to predict the likelihood of each team finishing in a given position, the model calculates a distribution of how many points each team is likely to accrue by the end of the season.

GRAPHIC
Methodology

The model is based on an approach developed by Dutch football statistician Sander Ijtsma, of 11tegen11.com. Its inputs are various no-pitch metrics including numbers, balances and ratios of points, goals, shots and shots on target, as well as shooting accuracy and time spent at different score differentials.

The specific steps are as follows:

Using a historical database of more than 6,000 Premier League matches, these statistics are then regressed against the numbers of goals a team scores and concedes in its next match, respectively, using a separate linear model for each.
The results can then be used to assign attacking and defending ratings — more simply, the expected numbers of goals scored and conceded in the next match — to any team at any time, given its underlying performance data. Ratings are generated for each team in every historical match in the database, generating expected score-lines to sit alongside the known results.
The impact of home-field advantage is then assessed using a new pair of linear models — one for the home team, one for the visitors. Each regresses the expected numbers of goals scored and conceded by a team in a given match against the observed number. The results allow the home team’s scoring and conceding ratings to be adjusted in its favour, and the away team’s ratings downgraded, based on historical patterns.

At this point, matches can be simulated in full. Having adjusted for home advantage, the expected average numbers of goals scored and conceded are used to generate the likelihood of a team scoring specific, integer numbers of goals, based on the knowledge that goal-scoring in football closely follows the poisson distribution*. For example, a team projected to score 1.75 goals in a given match (based on its underlying performance data and the venue) will score one goal in that match 30 per cent of the time, two goals 27 per cent of the time, zero 17 per cent of the time, and so on. To simulate a match, random draws are made from these goal probabilities to determine how many goals each team scores.
To create the end-of-season projections as they appear on this page, every remaining match of the season is simulated in this way based on teams’ underlying performance statistics as they currently stand. The points and goal-difference from these results are added to existing totals to produce final standings after every team has “played” 38 matches. The aforementioned random draws mean that, each season, simulation produces a slightly different set of results. These ranges of outcomes are transformed into probabilities of different points totals and placings by running each season sim 50,000 times and taking the average number of times each team finishes on a given points total, or in a given position.
* Tied results are more likely in reality than the true poisson distribution would suggest, so a small adjustment is made to ensure that the ratios of winsraws:losses is consistent with historical patterns.

Edited by Aray (18 Dec 2016 10.00am)

Attachment: IMG_5846.PNG (405.39Kb)

Missing a key graphic. Here it is.

Edited by Aray (18 Dec 2016 10.05am)

Attachment: IMG_5847.PNG (227.74Kb)

It's not personal, which is why Parish shouldn't have extended his chum's contract and should end it right about God knows how many weeks ago.

No need to play charades this year. Just listen to super duper he'll turn it around eventually Al. Yeah, Barnsley at home next November.

COYP

I would say Benteke has , he has scored goals and carries a threat , something we have lacked for a couple of seasons now , without his goals we would be bottom , certainly in the bottom already.
The problem is with the way we are attempting to play,
unfortunately the team can't do it,and it is causing the him to become more and more isolated. He is a forward and cannot influence his own supply line , especially against teams like Man U and Chelsea when he is one against three or four.

Stay up ?

Last Season we were 6th at this time and barely managed to stay up , not done the maths I would guess that we would have to improve on our after Christmas form of last year by about three times.
Bearing in mind we are still in poor form a year after it began things are not looking promising.
We can only stay up in current form if Hull , Swansea and Sunderland never go on a mini run of wins.

Someone needs to find a way to play to the players strengths using a system they understand and can play to end to bad form.

I like this reply and my Sunday is getting a whole lot better.

100% STAY UP

It's ridiculous that we are even discussing this possibility with the calibre of players we have at our disposal, but that highlights how poor a manager Pardew (aka Mildew) really is.

We should be in a better position than WBA, Watford and Bournemouth, rather than 1 point above the bottom 3 at Christmas.

Defeat at WAtford and Arsenal away will see us start off 2017 actually in the bottom 3 and will make it difficult to attract the new players we need.

The longer we keep Pardew as manager, the more likely we are to go down.