Overview
Aside from being a huge hockey fan, I am also very much a follower of CFB. I understand what is going on, and can generally interpret models and stats I see online, but I don't know as much about them intuitievly as I do hockey and hockey stats. However, I always see discourse about stating your prior beliefs when creating rankings in CFB. Probably so much moreso in CFB because there is an incredibly arbitrarty ranking system, unlike the rigid standings points or wins in professional leagues. And it's more important in CFB than CBB because basically any team who has a realistic chance to win the championship will make the playoffs. In football that is not necissarily the case, so the rankings are more important. Not saying they aren't important in CBB, as a fan of a bubble team in Pitt, I would have loved if the rankings had fallen our way the past few years, but alas. Because of all of that discourse, I wanted to track various priors in an objective manner throughout the season.
To measure success in as objective way as I could I just wanted to keep it simple with an ELO model. The way ELO works means you "take more" points from teams if you upset them/weren't supposed to win, but loose fewer points if you were already supposed to get blown out in that game and were ranked way below the opponent. Now also, all wins are not created equal, squeaking out a win on a last minute 60 yard field goal (looks at you Steelers) is not as impressive as controlling play on the way to a 25 point win. So I'm stealing a little bit from the incredibly popular world rugby rankings, and taking how much you win by into account. There is a scale factor in the ELO equation that you use to control how much a team takes or looses from a win or loss. For one score games I set it to +/-10, for two score games I set it to +/-25, and for three or more score games I set it to +/-50. Plus if you win, minus if you loose, of course. Those numbers might seem low, or a bit off to those more familiar with ELO. I've never worked with it before, so I was unaware of any standard conventions, so I set the mean to be roughly 500, and the max should end up somewhere around 700 by the end of the season.
Then the last thing I needed were a bunch of prior rankings to look at. I used a few very simple ones, like taking zero information into account and setting every team equal to 500, or only the conferences where I set every SEC and B1G team to 600, every ACC+Notre Dame and Big 12 team to 550 and all G5 teams to 400. Then there were ones where I used outside sources like the AP poll where I gave the top team 650, every team ranked below that 3 points lower all the way through the "others receiving votes" section, then every other team a flat 433. I also looked at Vegas championship odds before the season and translated those into percentages into scores ranging from 306 to 703. I also looked at the ESPN FPI rankings, though these weren't the final preseason ones, I heard they got changed, but I don't think they would change the outcomes *that* much. Finally I looked at roster talent. I'm not sure what the standrd roster talent measure was, but I used 247's recruiting rankings on the roster page for each team. I'm not sure if that is just the HS ranking, or their ranking as a transfer if the team got them as a transfer. I used their rating score (the actual number, not the number of stars, it was easier to scrape) then took the weighted average by positional value for each team, then translated those averages to a scale from 409 to 609. Since then I've just been tracking every game and seeing how the rankings change after each week. Even the coaches poll ones are very different already to the actual coaches poll. I also set every FCS team as a base 250 ranking that won't change. Obviously that is not a perfect solution, NDSU is much better than Duquesne or RMU usually, but I wasn't going to sit here and rank each and every FCS team. Though there is a great guy on Twitter who I see rank every single CFB team each week who I was trying to find for a while earlier today but just couldn't find.
Results
Obviously the season is not over yet, so this project is still a work in process, but I will be posting the updated rankings each week both here and on my twiter with some small comments about what interests me. I will try to update them by Sunday every week, but it may fall to Monday depending on how busy I get on the weekend.
Obviously these are just a list of all the priors I will be tracking throughout the season, so this is kind of the most important part and also the least important part of this project. Obviously then even ranks are just sorted alphabetically, then the conference basead ones are just SEC and B1G teams alphabetically. Roster talent, Texas, OSU, Bama, and Georgia are a clear top 4 favorites, then there is an everyone else of a bunch of solid teams. The discourse around the FPI and coaches poll has also been incredible so far, so I'm just not going to add to that. Finally the Vegas odds seem to be another decent ranking of teams that I think are probably pretty good at football.
Week 0 rankings there was hardly any change, all the winning teams shot up in the even ranks because they were the only teams to play, and Kansas jumped up a bunch in the FPI.
After week 1 there was a lot more change, especially in the conference based and even ranks. Those don't really matter as of now because the cupcake schedules and difference in GP, so I'm going to wait a few weeks for them to settle before discussing agian. Obviously OSU jumped Texas in the rankings where they weren't already above them, but nobody really took that many points because they were all so close to each other, and the teams that won big were expected to win big. Bama dropped the most of the big teams, but even in roster talent they barely dropped, but more importantly they got a lot closer to the rest of the pack, so their room for error now is much smaller. Miami and LSU got punished for winning huge games in some rankings tells me that I need to decrease the difference of the impact of winning 3 score games vs 1 score games, or increase the amount of points between the top and bottom teams, but I can't totally resart now, so I have to hope it will even itself out once conference play starts and there are a lot of big games.
Vegas champ odds barely changed, Michigan dropped because of their loss to OU, and Indianna jumped up after a big win. In the coaches poll, ND got punished this week again by not playing, and Bama gained back a lot of what they lost against FSU. Then there was a bunch of shuffling towards the back of the top 25. Michigan and Florida down of course, then ASU down after their poor performance vs Mississippi State, while A&M jumped up a few spots. Not really anything surprising in the FPI other than USC getting a big jump and BYU going WAY up after a big win over Stanford. Roster talent didn't really change much either, PSU and USC up a bit near the top, and Arkansas and Deion's CU joining the top 25 after wins over Arkansas State and Delaware respectively.
Not much change again this week. Was a pretty chalk week other than Clemson losing to GT. We do see Clemson dropping a few spots, but probably not dropping as far as we will in the actual polls. I've noticed my ELO doesn't punish teams as much for a loss as we do in real life. If you lose a close game to a top 10 team here, you'll still probably be viewed as a top 10 team, but in the real life polls teams might drop way out. I wanted to make wins and losses even, so each win isn't overly rewarding, but that means each loss isn't overly punishing. Maybe I could make something where losses mean more the higher you're ranked, but I'm not sure how that would work. Also, in this the top teams are usually more spread out, so when one does lose, they won't drop that much in terms of rankings. That could be viewed as a positive in certain years, but this year it seems like there aren't any dominant teams that deserve to be ranked much higher than anyone else. Also, I do genuinely think that, if you had ND top 10 preseason they should absolutely still be ranked despite the 0-2 record. They've lost by a combined 4 points to two teams that will probably be top 10, sure you gotts win close games, especially when it comes to playoff time and these are the teams you have to play, but they would still be favored at a neutral site over a lot of teams. Also, respect Diego Pavia, the top DAWG, and Vandy.
Data
My data came from a variety of sources. I just found the preseason coaches poll online. I started this before the preseason AP poll was released, so I just went with the coaches as a generic "media" poll. Then I got FPI from ESPN directly, as an analytical model ranking the strength of these teams. Then of course the Vegas odds telling me what the betting market thinks preseason, I don't remember which book I used, but I don't think that particularly matters because I'm sure they all changed slightly depending on the day you looked at it. Then for roster talent, as I mentioned, I used 247's roster ratings page. I created my own team ranking by looking at the positional value from Over the Cap, because I wan't sure how 247 calculated their positional rankings. Finally I got the game schedules from FBSchedules.com, and I update the game scores every weekend using ESPN.