I have been monitoring scoring first in association football.
In the 2017-2018 season I used the Worldfootball.net website to record goal scoring timings in six European leagues: EPL; Ligue 1; Bundesliga; Serie A; Eredivisie; and Primera.
I looked at the following events in games played in these leagues:
- score first win (SFW)
- score first draw (SFD)
- score first lose (SFL)
- 0 v 0 game (0goals)
I wondered if these measures could inform a naive Bayes approach to probabilistic behaviours in association football. I am thinking that for the 2018-2019 season I will use these prior probabilities to monitor teams’ progress.
For the six leagues combined, I have these median probabilities:
- SFW 0.64
- SFD 0.19
- SFL 0.12
- 0goals 0.07
I have a measure for scoring first and not losing (SFNL) (a sum SFW + SFD) and my median for the six leagues is 0.82.
My median probability profiles for each of the six leagues is:
| League | SFW | SFD | SFL | 0goals | SFNL |
| EPL | 0.63 | 0.18 | 0.11 | 0.08 | 0.81 |
| Ligue 1 | 0.64 | 0.19 | 0.12 | 0.05 | 0.83 |
| Bundesliga | 0.62 | 0.20 | 0.11 | 0.07 | 0.82 |
| Serie A | 0.65 | 0.15 | 0.13 | 0.07 | 0.80 |
| Eredivisie | 0.58 | 0.20 | 0.18 | 0.05 | 0.78 |
| Primera | 0.69 | 0.15 | 0.08 | 0.08 | 0.84 |
For the champions of each league, my probability profiles are:
| League | SFW | SFD | SFL | 0goals | SFNL |
| Man City | 0.76 | 0.03 | 0.05 | 0.05 | 0.79 |
| PSG | 0.68 | 0.03 | 0.00 | 0.05 | 0.72 |
| Bayern | 0.62 | 0.06 | 0.03 | 0.03 | 0.68 |
| Juventus | 0.68 | 0.05 | 0.03 | 0.08 | 0.73 |
| PSV | 0.59 | 0.09 | 0.00 | 0.03 | 0.68 |
| Barcelona | 0.63 | 0.08 | 0.00 | 0.03 | 0.71 |
For the bottom team in each of these leagues, my probability profiles are:
| League | SFW | SFD | SFL | 0goals | SFNL |
| W Brom |
0.16 | 0.18 | 0.11 | 0.11 | 0.34 |
| FC Metz | 0.11 | 0.13 | 0.08 | 0.03 | 0.24 |
| FC Koln | 0.12 | 0.09 | 0.15 | 0.06 | 0.21 |
| Benveneto | 0.11 | 0.03 | 0.13 | 0.00 | 0.14 |
| FC Twente | 0.18 | 0.09 | 0.18 | 0.06 | 0.27 |
| Malaga | 0.13 | 0.03 | 0.08 | 0.08 | 0.16 |
I am hopeful that these data might be of interest to anyone undertaking a Bayes approach to goal scoring.
In my own work, I am keen to see how early we can confirm the likely trajectory of any team in each of these six leagues.
My next post in this series will share data from the last two FIFA World Cups.
Photo Credit
Supporters FCN (Manuel, CC BY-SA 2.0)
