With the onset of the Allianz Football League, I’m back to rating & modeling the Senior County Football teams for this 2024 season. This year, I’d like to continue my ratings and predictions all the way through the All-Ireland championship, time and other commitments permitting.
I previously wrote about my methodology for developing my ratings here, but last week I went ahead and revamped them for the upcoming season. The ratings are still a bit crude, but I think they’re a decent reflection of the quality of the county teams heading into the 2024 season. I used 538’s methodologies for developing their NFL Elo Ratings and Soccer Power Index as guides throughout my process. These are super interesting reads and provide a lot of context into how much goes into developing their ratings. I’d love to get to the point where I’m incorporating a lot of similar data points for these ratings in the future. If anyone wants to help make these ratings more robust and encompassing, reach out and I’d love to hear your ideas.
Ok, let’s get into how these ratings work.
The Data
I collected all match results from the main GAA Fixtures & Results page (sidebar: It looks like the Fixtures & Results page is either broken or the website is acting up as of this writing, which is frustrating). When I started collecting the match data, the earliest available matches dated back to the beginning of 2016. The match results are now up-to-date from 2016-2023 and the data is available here.
The match results are pretty basic – it includes the Date, competition, Team 1, Team 2, the points, goals, and total score for each team.
Calculating the Ratings
To start the ratings, I initialized every team’s rating at the start of 2016 to 1500, which represents an average rating. The rating will be used to calculate the expected score and win probability of the match. This means that the win probability for each team at the start of the 2016 season is 50%, as they all have the same Elo rating.
In a slight change from my previous rating attempts, I decided to add a Margin of Victory (MoV) multiplier into the model. Because there are potential ties in games of Gaelic Football, this MoV multiplier gives teams credit for how they win – do they dominate or barely beat their opponent. The MoV multiplier takes the natural logarithm of the point differential and adds one point. Ideally, this will help with auto-correlation problems, which sometimes means that blowout wins by good teams could see their ratings swell disproportionately to the quality of the opponent.
Another factor of the Elo model is the k-factor. The k-factor tells the model how quickly to react to recent events. A high k-factor tends to over-emphasize recent results while a lower one may not accurately reflect the strengths and wins of a team. For now, I settled on a k-factor of 20, which is large enough to properly rate teams but not so high that ratings fluctuate too much from week to week. This means that the most Elo points any team can gain or lose from one game is 20.
Next, I defined a function to calculate the amount of Elo points the teams will gain or lose depending on the outcome. There are several factors that the function takes into account – the difference in Elo rating between the 2 teams; the MoV; the expected score/outcome. I then looped through the entire data set using the function to determine new ratings for each team after every match.
To try and account for player turnover from season to season, I rotated each team’s starting Elo rating for the next season a third of the way back to 1505 from their ending rating of the previous season. Because I currently don’t account for new players coming into the squad and players leaving, or a new coach coming in, this is an attempt to adjust for all the offseason moves that may happen. For example, Dublin ended the 2024 season as All-Ireland champions and had an Elo rating of 1770 to end the campaign. Heading into the 2024 campaign, Dublin starts with an Elo rating of 1688, which is still tops among all the county teams.
After the loop finishes working its way through the data, the result is that for every Senior County Football match from 2016 onwards we have the starting and ending Elo rating for every club. That rating gets carried over to their next match and is used to predict that outcome and so on and so forth.
To use a visual example, let’s look at the journeys Dublin’s and Kerry’s Elo rating took throughout the 2023 season. These teams played each other in the All-Ireland Final in July 2023, but one can argue that Kerry played against tougher competition during the run-up to the All-Ireland, as Dublin was in the second division of the AFL, whereas Kerry was in the top division. Dublin started the 2023 campaign with an Elo of 1619, while Kerry began the season rated at 1673. Dublin tore through the 2023 season, while Kerry had a rougher go of it. You can see that in the chart below, which shows Kerry’s and Dublin’s Elo rating after each match.
Kerry’s flatter Elo rating throughout the course of 2023 indicates the better competition they faced, though they still increased their Elo rating nearly 30 points overall. Dublin, however, increased their overall Elo rating dramatically, winning nearly every game and often winning by a fairly sizable margin. Going in the championship match for the All-Ireland, Dublin had a 1770 rating and Kerry was at 1708, with the Elo model giving Dublin a 59% chance of coming out with a victory. Dublin did end up winning the All-Ireland, ending the 2023 season with an Elo of 1779. Kerry ended the 2023 season with a rating of 1700.
The 2024 campaign kicks off today, and I’ll be working on a season-long simulation to try and project out towards the summer how the clubs will finish. Stay tuned!
Leave a Reply