Let's talk OTA's

[GTripp0012]

Originally Posted by Schneed10

As much as I love data driven analysis and its applications in so many aspects of life, football is one area where doesn't help all that meaningfully. The biggest reason is one of the fundamentals of statistics - sample size. The NFL season contains only 16 games, any statistician will tell you that a sample size that small can't provide meaningful conclusions that would pass a confidence test.

And while the stats will factor in a probability of injuries, a significant injury at a key position or two is all it takes to completely change a team's approach to game-planning, completely tossing almost everything the stat book would tell us.

Last year, if RG3 were healthy, the stats never would have called for a 3-13 season, the roster was stronger than that. But the fact is he wasn't himself, no calculation can plan for that. For the Eagles - if they lose Jason Peters and Jason Kelce this year like they did in 2012, they could be looking at 5-11 because Foles simply SUCKS under pressure.

There's just too many curve-breaking variables at play in football, and over the course of only 16 games, anything can happen. Stats just aren't that telling - there's a reason they say Any Given Sunday.

The thing about football is that there's so many events in a single game that can be measured that it allows projections on the season level to be based on massive samples: anywhere from 500 to 2,500 events in a season depending on what is being measured. But the trade-off that you touched on is obvious: one week in football is the equivalent of 10 baseball games or five basketball games, and so all analysis in football needs to be inclusive of the idea that a ton can change in one week. Football is just unique in that it packs that many events into a single game.

I think your argument that analytics and data are not going to do a lot to help you predict the outcome of an October matchup between the Redskins and Eagles is pretty accurate, but that's true with most sports. It's always going to be a lot easier to predict the result of 16 games than 1 game.

[Schneed10]

Originally Posted by GTripp0012

The thing about football is that there's so many events in a single game that can be measured that it allows projections on the season level to be based on massive samples: anywhere from 500 to 2,500 events in a season depending on what is being measured. But the trade-off that you touched on is obvious: one week in football is the equivalent of 10 baseball games or five basketball games, and so all analysis in football needs to be inclusive of the idea that a ton can change in one week. Football is just unique in that it packs that many events into a single game.

I think your argument that analytics and data are not going to do a lot to help you predict the outcome of an October matchup between the Redskins and Eagles is pretty accurate, but that's true with most sports. It's always going to be a lot easier to predict the result of 16 games than 1 game.

I meant that it can't even really be used to predict 16 games, not just 1. Because the event argument doesn't hold water - scientists don't measure the number of virus cells eradicated by a vaccine. They measure the number of people who contract the virus.

There are 16 results in a regular season. Yes there are hundreds of events within each that lead up to deciding the W or L on that particular day. But those are all co-varying factors. If you were trying to predict who will win these individual 'events', then yes you have enough data. But you don't have enough to predict a win loss record with any statistical validity.