I did this back in 2017: paper: Miscellaneous Statistics Projects - #30 by Caleb_Sykes
See the post below the linked one as well for significance tests. 2017 had p value of .03, no other years were anywhere close to significant. Part of why I suspect 2017 had a bias was that the field that year had mirrored symmetry and not the more commonly seen rotational symmetry.
Would be interesting to see more recent data, since a lot of the early years in that sample had way fewer matches in the season compared to now