If you don't want to know the score look away now

Will Wagstaff • 30 April 2021

Using predictive modelling to forecast the Premier League  final table

With the Premier League season down to the last few games, predicting the final table should be easy, right?

While it is true that Manchester City look nailed on for the title and the current bottom three (Sheffield United, West Brom & Fulham) look all but down, there is actually still much to play for. In fact, even at this late stage, there are still many possible permutations depending on how results go.

Many of us have done that thing where you look at your clubs remaining fixtures to predict which ones are winnable, which ones you'll struggle to get anything from and then see where your predictions leave you in the table.  We like doing that too but with a bit of a data science twist. 

We've had a go at predicting the final Premier League table using data modelling. In 1997, Dixon and Coles presented a model for predicting football scores by modelling the number of goals scored in a match as a Poisson distribution (the probability for a team to score one goal, two goals, three goals and so on in a match). We've applied this model to the remaining Premier League fixtures.

To generate the prediction we fed in the results so far from this season's Premier League. The model sets parameters to  represent a team’s attacking strength, defensive strength and home advantage. The model also contains a time decay component so that more recent results are given a higher weighting. Why is this? You may remember a few seasons ago that Crystal Palace lost their first seven games of the season. A change of manager saw a sharp upturn in Palace's form and they finished the season well clear of the drop. Had the model not down weighted the earlier poor form, it would predict worse results for Palace than that which they actually got.

The model generates a range of probabilities for each possible result (based upon each teams probability of scoring one goal, two goals and so on) and assigns 3 points for a win and 1 for a draw to their current points total. 

You can see the range of probabilities generated by the model by looking at the possible scores matrix for the Southampton v Leicester game.

We ran the model multiple times and then took the average of all iterations to produce the final table you see below. The model calculates total points for each team and generates probabilities of finishing in certain places. 


Our table shows teams ranked by predicted points along with the probability of finishing in that position. It also shows the probability of the team finishing higher than the position predicted or indeed lower.


So what does it show?

Campeones, campeones, olé, olé, 0lé

No surprises here. It's Manchester City.


The model suggests that City are 99.99% recurring likely to be top of the league.

These are the champions (League places)

Manchester United has a 90% probability of finishing second and 10% probability of finishing third or fourth. They look certain to join Manchester City in the Champions League next season.


Leicester is predicted to finish third with a 69% probability of doing so. It is possible that they could get second (10% probability). On the other hand, there is a 5% probability that they will finish outside the Champions League places.


The race for the final Champions League place is less clear. The model suggests that Chelsea will snatch the 4th place trophy. Fans of West Ham, Liverpool and Spurs shouldn't give up their Champions League hopes just yet - there is a 53% chance that Chelsea will finish outside the top 4. 


Of course Arsenal, who we predict to finish 10th, could make the Champions League by winning this year's Europa League.

We're gonna win the (Europa) league

The fifth place team goes directly to the Europa league group stages. A place is also given to the FA Cup  winners with the winner of the EFL Cup going into the newly formed Europa Conference. If, however, the FA Cup or EFL cups are won by a team that qualifies for European football due to their league position, then the 6th place team gets a Europa League place and the 7th placed goes into the Europa Conference. As the FA Cup final is between Chelsea and Leicester and Manchester City have already won the EFL cup then this is likely to happen.


The model predicts that West Ham and Liverpool will qualify for the Europa League and Spurs (in 7th place) will qualify for the Europa Conference. But don't give up just yet Everton fans. The model predicts that there is still a 40% chance that the Toffees will qualify for Europe. Game on as they say.

Going down, going down, going down

Fulham and West Brom are set to join already relegated Sheffield United in the Championship next season. Although the model predicts Fulham will be relegated, there is still a 5% probability of them avoiding the drop. It's the hope that gets you.


Big Sam (Allardyce) has never been relegated but the model predicts his West Brom team will finish second bottom.

VAR checking for possible unknowns in the prediction

Our Premier League table is based on a predictive model. With any predictive model there will be things happening that are not taken into account. These include the impact of player availability on form. A side might lose form with the loss of a key player or gain form when a key player comes back into the side.


Sometimes teams with nothing left to play for play with a new found freedom that turns the form book on its head. Similarly, teams whose season is effectively over might be, metaphorically speaking, on the beach. Others, seeing the chance of qualifying for Europe or avoiding relegation play much better than they have recently.


And of course VAR could rule that an armpit hair was in an offside position thus denying a vital goal that could have a profound effect on the table.

Predicted Premier League final placings 2020/21

Position Team Probability of position Probability finish higher Probability finish lower Points
1 Manchester City 99.99% 0 0 88
2 Manchester Utd 90% 0 10% 76
3 Leicester City 69% 10% 21% 70
4 Chelsea 34% 14% 52% 65
5 West Ham 25% 25% 50% 64
6 Liverpool 26% 44% 30% 63
7 Tottenham Hotspur 27% 50% 23% 61
8 Everton 37% 41% 22% 60
9 Leeds 31% 15% 54% 55
10 Arsenal 35% 40% 25% 54
11 Aston Villa 44% 50% 6% 53
12 Wolves 50% 4% 46% 46
13 Crystal Palace 26% 16% 58% 44
14 Burnley 24% 27% 49% 42
15 Southampton 17% 42% 41% 42
16 Newcastle 23% 54% 23% 41
17 Brighton 43% 55% 2% 39
18 Fulham 78% 6% 16% 32
19 West Bromwich Albion 82% 0.2% 17.8% 28
20 Sheffield United 98% 2% 0% 21


For some teams there is a good chance that they could finish higher or lower than their predicted position. The graphic below illustrates just how much things can change for individual teams.



In the graphic we show the range of possible placings. The darker the shading the stronger the probability of finishing in that place. For example, Burnley could finish anywhere between 11th and 17th but 14th is the most likely. Leeds could finish as high as 8th and there is a set of results that see Spurs finish 4th.


For many teams there is still all to play for.

Making Sense of Volume Data

by Michael Wagstaff • 8 April 2025
The huge volume of data available through consumer comments, reviews and surveys can make cutting through the noise difficult. In this article we discuss how text analytics combined with human expertise can uncover the insight.

AI Snake Oil

by Michael Wagstaff • 10 March 2025
Market research agencies are going all in on AI based models to generate next level consumer insight. But are these just more illusion than substance?

Has Data Science Finished Off the Survey?

by Michael Wagstaff • 3 March 2025
With the online survey already on the ropes due to poor quality, has data science finished it off?

Where Do We Go From Here? The Agency Dilemma

by Michael Wagstaff • 24 February 2025
Research agencies are pinning their futures on AI. Are they right to do so or are we missing trick by ditching the human?

Plugging The Online Survey Reliability Gap

by Michael Wagstaff • 12 February 2025
Online surveys suffer from fake, ill considered and unrepresentative responses. What can be done to improve their reliability? Triangulation is the key.

The End Of The Line?

by Michael Wagstaff • 11 February 2025
With so many agency panels riddled with fake respondents resulting in poor quality data, are we witnessing the end of the online survey?

Who'll Win the Premier League 2025?

by Michael Wagstaff • 6 February 2025
With the January transfer window closed, we run our predictive model to work out the probabilities of where each team will finish in the final Premier League table.

Striking The Right Chord

by Michael Wagstaff • 5 February 2025
In this latest article in our series on the power of text analytics we look at how sentiment analysis can be used to really understand what customers think about product offerings.

Beyond The Stars

by Michael Wagstaff • 23 January 2025
The true value of review sites lies in going beyond the stars and analysing what reviewers are actually saying.

Start Making Sense

by Michael Wagstaff • 17 January 2025
Start making sense of customer review and feedback data by using text analytics. In the first of a series of articles we discuss how it can help your business.
Show More