How did AFL HGA change in 2020?

April 11, 2021, 12:35 p.m.

It's been widely observed that HGA effects reduced in 2020 in many sports due to smaller crowd sizes. Many people have done analysis and found the impact is fairly significant in sports like soccer. Nothing has been done with respect to AFL so I thought I'd do a small investigation and see what the numbers suggested.

To do this I've used my own model as a starting point. This uses travel distance, venue familiarity and historical venue performance to estimate home ground advantage. The only change I made during the 2020 season was to reduce it by 20%, in line with the shortened match length. I will try and minimise my MAE by reducing the HGA I've applied to matches. If I can reduce my MAE by reducing HGA this suggests the impact of reduced crowds could have an impact on HGA. I already topped the Squiggle MAE leaderboard in 2020 (flex), so reducing MAE might be hard, but if there's any way to improve it for 2020 it would be reduced HGA.

The situation in AFL is complicated as we didn't have no crowds for the entire season. A lot of matches had no crowds (about 20%), some had small crowds (about 60% had crowds less than 10% of venue capacity), some had large crowds (about 10% had crowds over 50% capacity). To quantify the impact I've tried to estimate a new HGA using the original value, multiplied by some constant to reduce it depending on how small the crowd capacity is. I'll try and fit these parameters to minimise my MAE across 2020.

The reduction factor will be linear. An example set of parameters might look like:

  • HGA Factor when capacity zero = 0.5
  • Capacity when HGA unchanged = 0.5

This would mean for matches with 0 attendance we'd reduce HGA by 50%. The reduction would become smaller as capacity increases, until capacity reaches 50%. Matches with capacity over 50% will have no reduction to HGA.

By fitting these parameters I should be able to reduce the MAE in my margin predictions. Running an optimisation to minimise MAE through adjusting these parameters resulted in the following output:

  • HGA Factor when capacity zero = 0.985
  • Capacity when HGA unchanged = 0.13
  • MAE improvement = 0.001 points

This suggests for no crowds, HGA should be reduced by about 1.5%, and full HGA should be applied when attendance is over 13% of capacity. This impact seems very small relative to what I was expecting and suggests the impact of no crowds was fairly minimal.

What this is missing is my model updates team rankings based on performance relative to expectation. If I changed my expectation by changing HGA, team rankings should update to reflect this. Team rankings weren't updated when fitting the parameters above so this could be a source of error. Additionally I've done no statistical significance test here, just some quick optimisation to see the effect size.

I'm not sure how many matches of data other sports analysed, but 162 for AFL in 2020 is probably too small to draw too big of a conclusion from. The results for AFL in 2020 probably needs to be considered among all the other work done in estimating change in HGA with no crowds in other sports in determining the true impact.

Let me know your thoughts on twitter @AFLalytics.

Powered by Froala Editor