Philosophy Behind the Daniel Ratings

This page lists some of the rationale behind the Daniel Ratings. This is especially useful for prospective critics of the ratings.
The Daniel Ratings rate past performance; they do not predict future results.

This distinction is definitely the thing most often overlooked by people commenting on this system (and probably every other system as well). Being aware of the distinction between these two purposes is crucial to understanding any ratings system.

Here's a synopsis of the two main types of system:

  1. A power index, also called a predictive system. It's purpose is to predict the future. A high ranking in a power index says that a team is strong, and likely to beat a team ranked lower. The highest ranked team in a power index is the "strongest".
  2. A performance index, also called a retrodictive system. It's purpose is to rate past performance. A high ranking in a performance index is akin to saying that team is in "high in the standings"; the highest ranked team in a performace index would be in "first place".

The Daniel Ratings are performance index.

Therefore, you will often see one team ranked above another team it's obviously going to lose to next week, especially early in the season. The reason is simply that the ratings are not a measure of strength, but of performance. They say what the team's accomplished, not what they're capable of.

And, I'm sorry, but the strongest team in the world hasn't accomplished much if they've only beaten a bunch of chumps. A weaker team that's defeated some nonchumps is going to be ranked higher.

Margin of victory is not a factor; only wins and losses matter.

I get an awful lot of comments saying, "How can you rank X so low? They killed everyone on their schedule!" The reason, simply, is the Daniel Ratings only judge a team based on wins and losses. Scores do not factor into the system at all.

But why only wins and losses?

Well, as I mentioned above, the Daniel Ratings are a performance index. But what is performance? There are many, many aspects of the game that you could take as an indicator of peformance: wins, road wins, scoring margins, offesive yards, etc. A good performance is really just whatever you want to reward. Well, it turns out that the perforamce I wanted to reward was wins. I wanted to say who did the best job at winning, not who did the best job at winning by large margins.

But it's not quite as simple as me just waking up one day and deciding to only use wins and losses; there's a reason for it. When it comes time to choose or seed teams for the postseason, I feel that it is wrong to use anything other than wins or losses, except as a tie-breaker. Winning is the name of the game in football. I feel that a system that uses any other factor goes against the tradition and culture of football, and field sports in general. It's almost universal in such sports that winning is the only factor that matters when determining who won the "round" (be it a regular season, a playoff series, a match, or whatever).

To illustrate:

  • To make it to the NFL playoffs, the only aspect of performance that matters is wins (except as a tiebreaker, and deep one at that).
  • To win the World Series, the only aspect of performance that matters is the wins. You can outscore your opponent in the World Series 55-27, and still lose. (In fact that happened once: in 1960.)
  • To win a volleball match, the only thing that matters is to win games. You can outscore your opponent 132-75 in a match yet still lose the match three games to two. Likewise, in a tennis match, the only thing that matters is to win sets. And in a set the only thing that matters is to win a game.

The point is, in a lot of the sports we deal with every day, we only care about wins. Winning is the name of the game.

I realize that some sports, like auto racing, do use points. It could be possible to win the This-Year's-Corporate-Sponsor Cup in some some racing league without winnnig a single race. So it's not a moral travesty to use margin of victory to choose postseason teams, but for team field sports like football, the tradition is that winning is the only thing that matters.

An early game counts just as much as a late one.

For much the same reason that I didn't want to use margin of victory, I also didn't want to consider the date when the game was played.

It's pretty much the same argument: traditionally, sports don't weight recent games higher when deciding who wins the round and advances. An early season win in the NFL counts just as much as a late season win; Game 1 of the World Series counts just as much as Game 7, etc. If we want a system that chooses or seeds teams, it should weight all games equally, regardless of date.

This can sometimes confuse people who think rating systems ought to be a computer version of the human polls. Human polls nearly always reward recent wins and penalize recent losses disproportionately more than earlier ones. The Daniel Ratings don't do this, so sometimes you'll see a team with a large change in the polls but hardly any change in the ratings.

The Daniel Ratings uses no a priori information about teams.

The Daniel Ratings are completely based on wins and losses from the current season. They use no prior knowledge of how good teams are.

This confuses people sometimes, especially early in the season, where there isn't a lot of data. Sometimes, a generally-accepted strong team will play a weak schedule early. The Daniel Ratings has no prior knowledge of the team, so all it knows is that they beat a bunch of patsies, which certainly this doesn't qualify that team for a high rating.

As the season goes on, and more information comes in, and the supposedly strong teams play other supposedly strong teams, these sorts of oddities happily fade away somewhat.

The Daniel Ratings rates each team's peformance over their entire schedule.

Oftentimes, I hear comments like, "X beat Y, so why do you have Y ranked above X? Your rankings suck!" There's a good reason for it: the Daniel Ratings never look narrowly at one or two games to determine a team's rating; it always rates a team's peformance against it's entire schedule. (More to the point, no game result can trump any other result. No single game result can constrain the order of the rankings.)

The fact is, teams have vastly different strengths of schedule. And sometimes, the disparity is strong enough that even if X beat Y, and has a better record than Y, Y could still be ranked above X. The Daniel Ratings will look at a team's whole schedule, and will reward Y for beating hard teams, and if it's enough Y will be ranked higher. The fact that X beat Y cannot trump those other games on Y's schedule; it cannot force X to be ranked above Y.

And if you think this is retarded, think again: it's manifestly impossible to do this consistently for all teams without producing absurdities. I'll repeat myself. It's impossible to do what you ask consistently for all teams.

This whole issue is one of my pet peeves, actually. It never ceases to amaze me how much people believe that The Better Team Always Wins. Sure, almost everyone would say, "I know the better team does not always win." Yet they often turn around and make arguments (like the one above) that are predicated on it, at least in part.

They're called the Daniel Ratings, and not the Daniel Rankings, for a reason.

Ranking is just a side effect of rating. Let me explain.

In the Daniel Ratings, every team gets a rating. The rating of the team is it's measure of performance. Winning a game can never result in a team's rating going down. Losing a game can never result in a team's rating going up. That's as it should be: you're measuring performance, and your overall performance can't get worse if you win a game (and remember, winning is only aspect of performance in this system).

Ranking, on the other hand, is not a measure of performance. Even if you win a game, your ranking can still fall. Why? Because some other team previously ranked below you might have won a tougher game, thus gaining more rating points than you, thus overtaking you.

So if your team wins, but still goes down in the rankings, you shouldn't be crying "I won, but my ranking went down." I assure you, according to the Daniel Ratings, your peformance has improved. Your rating has gone up. It's just that some other team's rating has gone up more.

These ratings are for fun.

Really. I do these rankings because the amuse me. I find these ratings amusing because it takes a seemingly silly concept (transitivity) and produces something that's surprisingly reasonable.

I introduced these ratings into a culture where it was common to taunt good teams who lost an early game with long chains of transitivity. These ratings are actually a sort of parody of that thinking, and people who took part in those sorts of arguments should find these ratings interesting.

However, these ratings have flaws, so they shouldn't be taken too seriously.

Having said that, if someone comes in and attacks the Daniel Ratings from completely baseless grounds, I'm going to defend them. I don't care much if you disagree with details, but I won't stand for unfair criticism based on misguided thinking. "Your rankings suck because they have Y ranked above X," is anecdotal evidence, not a well-reasoned argument, and oftentimes isn't even valid because of the complainer's tunnel vision.

A reasonable argument like, "I can see how Y's schedule is stronger, but it looks like they might be getting a little too much credit for beating this team here," is likely to get a "probably" from me.

Copyright (C) 2001-2005, Carl Banks. All rights reserved.