README.md

In dafyddhowells/randy: Data and Functions for Modelling Goals Scored in Premier League Football Matches

randy

The purpose of this package is primarily to generate historical data from Premier League football matches suitable for fitting goal prediction models. It is my first package and the name ‘randy’ was the working title of the project that this originated from.

I was inspired to build this to help with a long running game I have played with friends.

Predicting the Scores in Premier League football matches

• 1 point is assigned to that person for predicting the correct result
• 3 points are assigned to that person for predicting the correct score
• Points accumulated each week and the person with the most points at the end of the season wins
• The BBC Football pundit Mark Lawrenson plays the same game each season challenging a different set of celebrities each week
• We include him in our league and his predictions for ths 2021/22 season can be found: here

An example of our completed points league from the 2020/21 season can be found here

Using Poisson models to do my own score predictions

Previously I predicted scores using largely gut instinct which had returned modest results over the years. I would typically get 45-50% of results correct with about 40 / 380 correct per season. Thinking that I may do better with a more methodical approach I started researching methods developed by countless others to achieve the same aim. I will not document all of that here but see below a couple of excellent packages that really helped my learning in this area:

• https://github.com/Torvaney/regista
• https://github.com/opisthokonta/goalmodel

Whether I used the approaches documented above or worked on building my own Poisson models, the first step was to pull together the historical premier league data sets to fit models to.

Installation

You can install the development version of randy from GitHub with:

# install.packages("devtools")
devtools::install_github("dafyddhowells/randy")

library(randy)

Pulling in historical Premier League results

This function pulls in the results from all fixtures in the entire history of the Premier League including the current season:

head(get_prem_history())

| date | home_team | away_team | fthg | ftag | fixture | |------------|-------------|----------------|------|------|----------------------------| | 1993-08-14 | Arsenal | Coventry | 0 | 3 | Arsenal v Coventry | | 1993-08-14 | Aston Villa | QPR | 4 | 1 | Aston Villa v QPR | | 1993-08-14 | Chelsea | Blackburn | 1 | 2 | Chelsea v Blackburn | | 1993-08-14 | Liverpool | Sheffield Weds | 2 | 0 | Liverpool v Sheffield Weds | | 1993-08-14 | Man City | Leeds | 1 | 1 | Man City v Leeds | | 1993-08-14 | Newcastle | Spurs | 0 | 1 | Newcastle v Spurs |

Create form data

To create a data set for fitting models to, each team in a fixture needs to appear as it’s own observation whether it is home or away to allow us to predict the goal outcome for that team in a fixture. We also need to calculate the following variables to show a team’s attacking and defending strength either home or away:

• Running averages for goals scored at home
• Running averages for goals conceded at home
• Running averages for goals scored away
• Running averages for goals conceded away
• Form: Running average of points accumulated over past n games (home and away combined)

prem_history <- get_prem_history()
prem_form_df <- get_prem_form(prem_history)

head(prem_form_df)

| team | fixture | date | fthg | ftag | form | scored | conceded | home | win | draw | |---------|-----------------------|------------|------|------|------|--------|----------|------|-----|------| | Arsenal | Arsenal v Coventry | 1993-08-14 | 0 | 3 | 0 | 0 | 0 | TRUE | 0 | 0 | | Arsenal | Arsenal v Leeds | 1993-08-24 | 2 | 1 | 0 | 0 | 0 | TRUE | 1 | 0 | | Arsenal | Arsenal v Everton | 1993-08-28 | 2 | 0 | 0 | 0 | 0 | TRUE | 1 | 0 | | Arsenal | Arsenal v Ipswich | 1993-09-11 | 4 | 0 | 2.25 | 2 | 1 | TRUE | 1 | 0 | | Arsenal | Arsenal v Southampton | 1993-09-25 | 1 | 0 | 3 | 2.25 | 0.25 | TRUE | 1 | 0 | | Arsenal | Arsenal v Man City | 1993-10-16 | 0 | 0 | 2.5 | 1.75 | 0 | TRUE | 0 | 1 |

Create view of the history of a team in the same fixture over time

We may also want to consider the average goals scored and conceded, home and away in the same instance of a fixture over a number of years. For e.g. how many goals on average do Everton score at home vs Arsenal, how many goals on average do West Ham concede away vs Man Utd.

prem_history <- get_prem_history()
prem_fixture_history_df <- get_prem_fixture_history(prem_history)

head(prem_fixture_history_df)

| team | fixture | date | fthg | ftag | sf_scored | sf_conceded | home | |---------|-----------------------|------------|------|------|-----------|-------------|------| | Arsenal | Arsenal v Aston Villa | 1993-11-06 | 1 | 2 | 0 | 0 | TRUE | | Arsenal | Arsenal v Aston Villa | 1994-12-26 | 0 | 0 | 0 | 0 | TRUE | | Arsenal | Arsenal v Aston Villa | 1995-10-21 | 2 | 0 | 0 | 0 | TRUE | | Arsenal | Arsenal v Aston Villa | 1996-12-28 | 2 | 2 | 1.25 | 1 | TRUE | | Arsenal | Arsenal v Aston Villa | 1997-10-26 | 0 | 0 | 1 | 0.5 | TRUE | | Arsenal | Arsenal v Aston Villa | 1999-05-16 | 1 | 0 | 1.25 | 0.5 | TRUE |

Running averages for xG for and against in recent games

Sourced from https://fbref.com, similar to calculating running average goal form provided by get_prem_form(), we can calculate the same for a team, either home or away for their expected goals (xG) to provide additional variables for a team’s attacking and defending strength.

xg_fixture_history_df <- get_xg_fixture_history()

head(xg_fixture_history_df)

| date | team | fixture | actual_score | xg1 | xg2 | home | xg_for | xg_against | |------------|---------|-----------------------|--------------|-----|-----|------|--------|------------| | 2017-08-11 | Arsenal | Arsenal v Leicester | 4–3 | 2.3 | 1.3 | TRUE | 0 | 0 | | 2017-09-09 | Arsenal | Arsenal v Bournemouth | 3–0 | 2 | 0.9 | TRUE | 0 | 0 | | 2017-09-25 | Arsenal | Arsenal v West Brom | 2–0 | 2.4 | 0.7 | TRUE | 0 | 0 | | 2017-10-01 | Arsenal | Arsenal v Brighton | 2–0 | 3 | 0.4 | TRUE | 2.425 | 0.825 | | 2017-10-28 | Arsenal | Arsenal v Swansea | 2–1 | 1.6 | 0.6 | TRUE | 2.25 | 0.65 | | 2017-11-18 | Arsenal | Arsenal v Spurs | 2–0 | 1.4 | 1.4 | TRUE | 2.1 | 0.775 |

Combining form, xG history and past fixture history

We can combine all 3 of these sets of variables into 1 data frame using get_model_data() with some additional variables including:

• The season the fixture was played in
• The position of the team in the league when the fixture was played
• The cumulative number of points accumulated by the team at that point in the season

model_data <- get_model_data()

head(model_data)

| team | fixture | date | fthg | ftag | form | scored | conceded | home | win | draw | sf_scored | sf_conceded | season | points | cum_points | game | league_pos | actual_score | xg1 | xg2 | xg_for | xg_against | |----------------|---------------------------|------------|------|------|------|--------|----------|-------|-----|------|-----------|-------------|-----------|--------|------------|------|------------|--------------|-----|-----|--------|------------| | Man Utd | Man Utd v Wolves | 2022-01-03 | 0 | 1 | 2.25 | 1.75 | 1 | TRUE | 0 | 0 | 1.25 | 0.75 | 2021/2022 | 0 | 31 | 19 | 4 | 0–1 | 0.8 | 0.7 | 1.45 | 0.975 | | Wolves | Man Utd v Wolves | 2022-01-03 | 0 | 1 | 1.25 | 0.25 | 0.25 | FALSE | 1 | 0 | 0.75 | 0.75 | 2021/2022 | 3 | 28 | 19 | 5 | 0–1 | 0.8 | 0.7 | 0.625 | 1.725 | | Aston Villa | Brentford v Aston Villa | 2022-01-02 | 2 | 1 | 2.25 | 2 | 1.25 | FALSE | 0 | 0 | 1.5 | 1.5 | 2021/2022 | 0 | 22 | 19 | 8 | 2–1 | 0.6 | 1.2 | 0.925 | 1.125 | | Brentford | Brentford v Aston Villa | 2022-01-02 | 2 | 1 | 2.25 | 1.25 | 0.75 | TRUE | 1 | 0 | 2 | 0.75 | 2021/2022 | 3 | 23 | 19 | 7 | 2–1 | 0.6 | 1.2 | 1 | 0.875 | | Brighton | Everton v Brighton | 2022-01-02 | 2 | 3 | 1 | 0.25 | 0.75 | FALSE | 1 | 0 | 1 | 1 | 2021/2022 | 3 | 27 | 19 | 6 | 2–3 | 1.7 | 1.5 | 1.225 | 1.45 | | Burnley | Leeds v Burnley | 2022-01-02 | 3 | 1 | 1.75 | 1.25 | 1 | FALSE | 0 | 0 | 1 | 1 | 2021/2022 | 0 | 11 | 17 | 15 | 3–1 | 1.6 | 1 | 0.675 | 1.4 | | Chelsea | Chelsea v Liverpool | 2022-01-02 | 2 | 2 | 1.5 | 1.75 | 1.5 | TRUE | 0 | 1 | 1.25 | 1.25 | 2021/2022 | 1 | 43 | 21 | 2 | 2–2 | 1.3 | 1.3 | 1.975 | 1 | | Everton | Everton v Brighton | 2022-01-02 | 2 | 3 | 1 | 1.25 | 2 | TRUE | 0 | 0 | 2 | 1 | 2021/2022 | 0 | 19 | 18 | 12 | 2–3 | 1.7 | 1.5 | 0.95 | 1.4 | | Leeds | Leeds v Burnley | 2022-01-02 | 3 | 1 | 1.75 | 1.75 | 1.75 | TRUE | 1 | 0 | 3.25 | 1.25 | 2021/2022 | 3 | 19 | 19 | 10 | 3–1 | 1.6 | 1 | 1.525 | 1.55 | | Liverpool | Chelsea v Liverpool | 2022-01-02 | 2 | 2 | 1 | 1 | 1.25 | FALSE | 0 | 1 | 2 | 2 | 2021/2022 | 1 | 42 | 20 | 2 | 2–2 | 1.3 | 1.3 | 1.95 | 1.375 | | Arsenal | Arsenal v Man City | 2022-01-01 | 1 | 2 | 2.25 | 2 | 0.5 | TRUE | 0 | 0 | 0.25 | 2.5 | 2021/2022 | 0 | 35 | 20 | 3 | 1–2 | 1 | 1.8 | 1.8 | 0.725 | | Crystal Palace | Crystal Palace v West Ham | 2022-01-01 | 2 | 3 | 1.75 | 2.5 | 1.5 | TRUE | 0 | 0 | 1.75 | 1.75 | 2021/2022 | 0 | 23 | 20 | 5 | 2–3 | 2.2 | 2 | 1.725 | 1.3 | | Man City | Arsenal v Man City | 2022-01-01 | 1 | 2 | 0.25 | 1 | 3.25 | FALSE | 1 | 0 | 3.25 | 3.25 | 2021/2022 | 3 | 53 | 21 | 1 | 1–2 | 1 | 1.8 | 2.325 | 0.6 |

dafyddhowells/randy documentation built on Nov. 1, 2022, 4:01 p.m.