Australian Rules Football and ELO Ratings


The ELO rating system is widely used in professional individual and team sports to compute relative rankings of teams or individuals. I had some free time on the weekend and wrote a fairly simple MATLAB script to compute the ELO rating for each team in the Australian Rules Football League (AFL). I used the data files from AFL Tables as input to the MATLAB code. The table below lists all the ELO ratings as well as some basic statistics, such as the total number of wins, losses and draws for each team.

TEAM ELO WINS LOSSES DRAWS TOTAL
1 Hawthorn 1475 869 951 10 1830
2 Geelong 1469 1224 1042 21 2287
3 Collingwood 1468 1460 910 26 2396
4 Sydney 1428 1061 1199 23 2283
5 West Coast 1360 339 269 5 613
6 Brisbane Bears 1319 72 148 2 222
7 Adelaide 1316 269 241 1 511
8 St Kilda 1311 877 1338 25 2240
9 North Melbourne 1293 803 1005 17 1825
10 Fremantle 1292 168 236 0 404
11 Carlton 1253 1385 939 33 2357
12 Essendon 1194 1315 973 34 2322
13 Richmond 1170 1058 1035 22 2115
14 Brisbane Lions 1147 192 175 6 373
15 Western Bulldogs 1099 803 976 22 1801
16 Melbourne 1032 1034 1209 21 2264
17 Port Adelaide 1032 183 181 5 369
18 GW Sydney 1016 2 20 0 22
19 Gold Coast 992 6 38 0 44
20 Fitzroy 871 869 1034 25 1928
21 University 781 27 97 2 126

Note:

  • I used the results of all 14,166 games played in the VFL/AFL since 1897 to create the table.
  • The ranking scores of the top three teams are really close, but Hawthorn has edged out Collingwood and Geelong.
  • The ELO of each team was initialised to 1200; the only exception to this rule were the Brisbane Lions which inherited the final ELO score from the Brisbane Bears. This seemed reasonable given the details of the “merger” between the Brisbane Bears and Fitzroy Lions.
  • The team University withdrew from the competition for World War 1 and never came back.
  • All ratings were updated using a constant K-factor of 32.
  • My MATLAB script and the data file used to create this table are freely available for download here.

It would be of interest to use this data for prediction of the AFL 2013 season, perhaps using a methodology similar to the ELO++ work.

  1. No comments yet.
(will not be published)