A: Since high school, I've been interested in mathematically rating and ranking teams of various sports. It's been an off and on hobby of mine since then. Stronger computing power and cheaper web hosting now allows me to publish my results.
Q: So why is it called Sports Ratings III?
A: This is the 3rd major iteration of my rankings. Well, if you don't count the one I did on pen and paper for the NFL in high school. The first true iteration was in Microsoft Access, about 20 years ago. Then I had a second iteration in C# with a SQL Server backend on a self-hosted website about 15 years ago. This is the 3rd, and hopefully the best. I stil have all the code for all three iterations. That mght make me a little weird.
Q: Your rankings have weird names. What's that about?
A: Call it a desire for something unique and original, combined with a total lack of imagination.
The Davis ratings only look at wins and losses and who you played. They're named after legendary Raiders owner Al Davis, who famously said: Just win, baby! This sounds like a ridiculous way to rate teams, but for sports with long seasons, like NCAA College Basketball, a win-based rating is probably going to be the best by the end of the season. For other sports, others may be better. Or not. 😊
The Samuel ratings are a bit more complex, using more team statistics, and a sprinkle of AI.They're named after Arthur Samuel, a pioneer in the field of artificial intelligence and machine learning.
The Composite ratings are a combination of the Davis and Samuel ratings, to give a more balanced view. This rating most closely matches what pollsters do when ranking teams.
The Sanders ratings are just for fun and have no numerical validity whatsoever. They take the Davis ratings and add an even greater emphasis on winning. They're named after USC coach Red Sanders, who said: Winning isn't everything. It's the only thing.
Q: I remember the computers in BCS rankings. Are these like that? They sucked.
A: Yes, they are similar. But it wasn't the computers that sucked in the BCS rankings. It was in how they were used. First, the stupid people running the BCS changed the rules every year. Always in reaction to what happened the previous year. Second, they were far too concerned about promoting the idea of "running up the score" that they didn't allow computer algorithims that took scores and other factors into account. Third, yes, a couple of the computer algorithms were crappy. But that shouldn't have mattered, because the real fault was that the BCS didn't have nearly enough computers in it. By the end, they were down to 5 and they dropped the biggest outlier for each team. An ideal solution like the BCS would have 50 human pollsters and 50 computer rankings of various types. Frankly, this would be far better than the Committee that we have now. Kenneth Massey keeps a computer composite on his site, that is far better than what we had with the BCS.
Q: What do the numbers in your ratings mean?
A: Not much, and quite a bit.
For Davis, it's essentialy a number between 0 and 1. It is possible to get a number slightly over 1, as road wins and upsets count a little more than predicted and home wins.
The numbers for Samuel are a bit harder to describe. If you look at the Samuel ratings, there is an offensive factor and a defensive factor. The offense number tells you how much that team would be likely to score against an "average" team. And the defense number tells you how much better or worse the team's defense is compared to an "average" team. Doing it this way allows me to add the two numbers together to get a rating that makes sense. It also allows a rough score predictur. If team A has an offense rating of 27, and a defense rating of 6, and they're playing team B who has an offense rating of 23, and a defense rating of 10, you can calculate the score thusly: Team A should score 27(AO) - 10(BD), and Team B should score 23(BO) - 6(AD), giving us a 17-17 tie.
Composite is again basically 0 to 1. It plots the Davis and Samuel ratings on a bell curve scaled to [0:1] and combines them.
Q: How good are your ratings?
A: Well, I think they're pretty good. Are the ones on ESPN and other sites better? Maybe. They have easier access to more data than I. Getting detailed game data for every team is either expensive or difficult or both. That is part of why I only do three sports. What I like about mine is that I know how they work. I have no idea how ESPN's SP+ ratings work, so I don't know whether to believe them. Of course, that doesn't help you...
Q: Can I use them to place a few bets?
A: Sure, but don't blame me if you lose money. This is not Gray's Sports Almanac. I provide no guarantees or warranties about the accuracy of these ratings. They're just for fun. Honestly, if you want to bet against the spread, right now you're probaby better off betting the opposite of what the computer says. 🤣 If you want to gamble, do so responsibly. I do have plans which should hopefully make my ratings better as predictors, but that's going to take a little time, and it's hard to beat those people out in the desert.
Q: So where do your predictions come from?
A: I do what is called a Monte Carlo simulation of each game. Based on the stats I have, I simulate each game 10,000 times. The prediction is based upon the results of these simulations.
Q: Some of your numbers and rankings look a little crazy.
A: There may be a couple of reasons for that. Early in the season, the ratings can be highly volatile. There are things I do to mitigate that, but it's not perfect. Also, one thing computers do much better than humans is that they look at every single game. Humans can't do that. Any reasonable computer algorithm for ranking teams is going to be far better than almost any human. That's a polite way of saying, "My computer is smarter than you." Don't be offended. It's smarter than I, also.
Q: This seems interesting, but I want to create my own. How hard is that?
A: There are a number of papers on the subject and various starting points. Elo ratings were originaly created for chess, and have expanded to many other sports. I find Elo to be a poor choice for sports like football and basketball, but I know some use it. Colley and Massey are fairly simple and basically the same. (I'm not sure that either Colley or Massey would agree with that statement, but they are.) The linked article on Massey's method is a great place to start. My approach is somewhat similar, although not nearly identical.