02-11-2022, 05:45 PM
(02-11-2022, 05:00 PM)casear2727 Wrote: Great points. How do these algorithms account for variables such as injuries, home or away, weather.
So, the one that I am most familiar with (NFLFastR) was written in part by a buy named Ben Baldwin. Essentially, Ben and other statisticians have combed through NFL data and analyzed performances in various types of environments - snow, rain, sunshine, domes and everything else not included here. You can figure out performance trends and how the weather affects those performance trends across large datasets. For instance, say you have 2000 games; 1000 of those games are in sunshine and the teams average 25 points per game and the other 1000 games are snow and teams average 17 points per game. You have a fairly clear indicator that snow is a detriment to offensive performance. This is a super simple example and isn't exactly how they go about analyzing this stuff, but it gives an idea. From there, functions are built that can account for weather and modify whatever metrics you are looking at.
The same example goes for home/away. Analyzing performances between the two in large datasets and gathering more useful data to be used in calculations. Injuries can be a little more difficult, because you are wanting to know the impact of a particular player and how his absence will affect the group. In basketball and baseball, this is easier to define because the impact of one single player is very well known. In the NFL, it is SIGNIFICANTLY harder. For instance, how do you judge the impact of Jonah Williams and how a replacement player would perform in his absence? It's really tough. The best way that I know of is just to compare datasets of performance metrics before and after a player is injured, but you run into a bunch of noise like opponents played, further injuries, weather etc. There probably isn't a good way to account for injury, at least not that I know of.
