This report is on the use of a simple elo ratings model to predict the 2024 AFL Finals Series.
library(fitzRoy)
library(tidyverse)
library(PlayerRatings)
library(elo)
library(gtools)
afl2023_raw <- fetch_results_afltables(season = 2022:2023)
I first ran an elo model based on the 2022 and 2023 AFL seasons to provide some context for the initial elos of the 2024 season. The final elo ratings based on these seasons were:
afl2023_final_elos
## rowname .
## 1 Adelaide 1630.532
## 2 Brisbane Lions 1877.970
## 3 Carlton 1803.385
## 4 Collingwood 1928.664
## 5 Essendon 1569.117
## 6 Footscray 1638.432
## 7 Fremantle 1632.905
## 8 Geelong 1657.408
## 9 Gold Coast 1550.572
## 10 GWS 1770.360
## 11 Hawthorn 1524.609
## 12 Melbourne 1764.529
## 13 North Melbourne 1296.943
## 14 Port Adelaide 1748.786
## 15 Richmond 1619.431
## 16 St Kilda 1631.307
## 17 Sydney 1732.853
## 18 West Coast 1322.196
Using the elo ratings calculated from the 2022 and 2023 seasons as a basis, I calculated the elo ratings from the 2024 regular season from games only against the teams that finished top 10 in the ladder.
Setting the k-factor high at 60 to account for form, the elo simulation produced the following results:
final_elos_top8
## rowname .
## 1 BRI 1831.647
## 2 CAR 1635.590
## 3 WBD 1738.677
## 4 GEL 1738.571
## 5 GWS 1733.091
## 6 HAW 1691.670
## 7 PTA 1763.909
## 8 SYD 1806.749
Using these elo ratings, running 1,000 simulations of the AFL Finals to create a win probability model produced these results:
final_results
## # A tibble: 8 × 6
## Team `Eliminated EF` `Eliminated PF` `Eliminated SF` RunnerUp Winner
## <chr> <int> <int> <int> <int> <int>
## 1 BRI 254 179 291 112 164
## 2 CAR 746 55 168 21 10
## 3 GEL 0 422 249 188 141
## 4 GWS 0 320 363 171 146
## 5 HAW 549 113 265 47 26
## 6 PTA 0 420 199 199 182
## 7 SYD 0 346 178 201 275
## 8 WBD 451 145 287 61 56
According to the Elo model, the team most likely to win the 2024 AFL Grand Final was Sydney, winning it 210 times, the actual premiers, the Brisbane Lions, ended up winning 117 times out of 1,000 despite their Elo rating being the highest, which is due to the fact they finished outside of the top 4, making their path harder.