Stammers
Legend
Posts: 38,187
Joined: Feb 2004
Reputation: 1739
I Root For: Memphis
Location: Montreal, Canada
|
RE: Why the RPI is a poor metric by which pick/seed teams
(03-01-2013 02:20 PM)MemphisCanes Wrote: (03-01-2013 12:00 PM)Stammers Wrote: (03-01-2013 10:39 AM)MemphisCanes Wrote: (03-01-2013 10:20 AM)Stammers Wrote: (03-01-2013 09:09 AM)MemphisCanes Wrote: Well, Sagarin's ELO model measures SOS via wins and losses. Everyone starts off tied for first before any games are played, and then teams are spread over the spectrum of 1-344 as they win/lose.
Effectively, both ELO and RPI give "greater value to wins against tough competition", so you can't really mean that. A win against a high RPI team (a team with a good winning pct, good O winning pct and good OO winning pct) jumps your RPI score significantly, thus conveying greater value for defeating a greater opponent. As opposed to playing a very low RPI opponent, winning handily and possibly losing ground in the RPI rankings (this is one of my biggest concerns over the RPI). This is the crux of any SOS weighted system.
This is not true. The RPI gives the same value to a win or a loss with the only variable being the location of the game. Math then determines the value of the win/loss based on data from wins and losses.
ELO assigns an arbitrary value in addition to the mathematical calculation, based on the opponent's strength; which is disproportionally weighted based on the strength of the opponent. In addition, the predictor is biased and is used in tandem with ELO.
If I am wrong in assuming this, please explain why in detail.
Does ELO or a similar weighted system have its place? I think it does. A great example would be with professional golf. The 4 majors are bigger tournaments and victories carry a greater importance. In NCAA basketball, none of the games are more or less important, when considering a team's overall body of work.
Ah, but that's only part of the calculation. Your RPI, inherently, takes into account each opponent's win pct and each of their opponent's win pct. Therefore, playing a tougher opponent (a higher RPI team) is of more value than playing a weak opponent. Each win is not treated equally in the RPI's eyes, when looking past each respective team's win pct.
Just think, if the Tigers, at RPI 18 or so as of today, played and beat Miami tomorrow, our RPI would improve significantly because we just beat the RPI #2 team, who has a great win pct and whose opponents have a great win pct. Conversely, if we play and beat Kennesaw State RPI #341, our RPI score barely increases and because other teams grouped around us in the RPI probably played better teams, our RPI rank will slip.
The RPI is weighted by SOS just like all other metrics.
Also, Sagarin's ELO and Predictor are two separate metrics, the predictor doesn't care what your wins/losses are, it only cares about the scores of your games. He combines both to get his "Sagarin rankings".
The ELO rankings assign no more arbitrary numbers than the RPI with its .25s, .5s, .6s and 1.4s.
First, with the RPI, SOS is weighted solely based on data derived from wins and losses. Second, the values you describe for RPI are not attached to any specific team or any component of performance.
Well, the RPI is weighted on wins and losses, and arbitrary multipliers like .25 for your own win pct, .5 for your opponents, .25 for your opponents opponents, a "road win" is worth 1.4 and a "home win" is worth 0.6. Those numbers are arbitrary, right?
Also, the values in the RPI ARE attached to specific teams. Miami has an RPI score of .6602. Kennesaw State has an RPI score of .3673. Miami essentially has a higher score because of better performance against a stronger schedule. Beating Miami is thus better for an opposing team's RPI score than beating Kennesaw State.
The values are arbitrary but they are identical for each team. The data is identical because it is based on wins and losses. Sagarin, Kenpom etc are different because they include things that I don't consider important. Yes you can say that if Team A beats Team B by 40 points, it is probably better than Team C if it beats Team B by 4 points...over 1 game with common opponents. The reason why I don't think that Sagarin, Kenpom are accurate; is because based on data from thousands of games; margin of victory doesn't matter. Each team is being "protected" by the fact that every team in the NCAA is measured against the same criteria...wins and losses.
When you input so many other variables outside of wins and losses, you are not giving enough credit for wins and losses. Other stuff is given too much importance, when it really has little or no importance.
The other major problem with Sagarin and Kenpom is that it fails to take game in situations into account. A player is 16 minutes late for practice so he sits 16 minutes of a game. A game is out of hand one way or the other so a coach experiments with different lineups. A team is bored and comes out flat. One team scores the last 10 points in a game to turn a tight game into a laugher or a blowout into a close final deficit.
The bottom line is that a win is a win and a loss is a loss. Nothing else really matters. If we were 20-8 with 19 blowout wins and 7 close losses and Sagarin/Kenpom were telling us that we were a great team with a top 25 power ranking, who would really care if you have an RPI of 50 and won't make the tournament?
|
|