Picking up where we left off last week ...

Joe,

Thanks for treating my arguments fairly on your banter column, which I just read (http://espn.go.com/ncb/s/bracket/banter/030227.html).

The one point that I don't think you got right, or perhaps I didn't make clearly enough, is that the significance of the I-AA teams showing up right near the top in college football when the RPI is run on it. There must be something fundamentally wrong with an algorithm that could produce those results when another algorithm (ELO CHESS) doesn't produce anything at all like that.

None of the my three ratings (RATING, ELO CHESS or PREDICTOR) have anything crazy like that popping up, and the program treats all 240 (football) teams in the same manner. It does not need to be instructed to rate the I-A teams separately from the I-AA. But the mathematics underneath manage to correctly separate them without being told directly to do so. The program doesn't "know" that there's supposed to be a difference between the two groups.

To the program, all teams are simply code numbers with no special instructions. The key point here is that the 240 teams are actually all "connected," and so it is a valid thing to do to rate them as one universe. Obviously if there's no connectivity between two sets of teams, then one can't validly rate them versus each other.

To make the point, you couldn't validly rate NFL teams versus college football teams just by putting all the data into one data base. There's no games connecting them, so it would be sheer gibberish that would be coming out on the screen. But, in college football, there are many games between the I-A and I-AA teams and so it can easily be shown that all 240 (117 I-A and 123 I-AA) teams are truly connected.

Anyhow, thanks for giving me a very even-handed hearing. I'm really pleased that you're going to be running the RPI, ELO CHESS and your own S-curve for the rest of the season. I think you'll end up being quite impressed by the ELO CHESS.

By the way, ELO is the last name of Arpad Elo, the Hungarian emigrant born in 1903 (died in 1992) who was a physicist at Marquette and Wisconsin and who devised the chess rating system for the United States Chess Federation in the late 1950's and which was adopted worldwide by FIDE (international chess federation) in 1970. I corresponded with him back in the 1970-71 school year and my use of his name is my way of honoring him.

Regards,
Jeff Sagarin

Fair enough, Jeff.

I now also understand the point made by I-AA teams appearing in the sample RPI for college football. And I hope you don't mind if I leave algorithms and the like behind for the time being, as the next few weeks will be taken up a whole lot more by games than by ratings (at least in my world).

However, I would like to suggest something fun and worthwhile for you and Jerry Palm and Jim Sukup and whichever other published bracketologists are out there. One of you come up with a way to numerically "score" each of our final bracket projections on Selection Sunday, with respect to team selection and seeding (and maybe even placement), as it compares to the actual bracket released by the NCAA. I'll post the results in this space after the fact.

To add a little steak to the sizzle, I further suggest that we each kick $100 into a pot. The winner gets his/her $100 back; the rest of the proceeds will be donated to "Coaches vs. Cancer."

What do you say, gang? We've got nothing to lose except our dignity (and, at least in my case, that has long ago left the building!).

The defense of your use of the RPI in predicting the Committee's bracket is exactly right ("When in Rome..."). The defense of the RPI as a valid tool for use by the Committee, or anyone else, is to recognize what it is designed to do and then use it only for that purpose.

The RPI is nothing more than a synthesis of W-L record and SOS. The factors used in picking the at-large berths clearly should include a lot of factors, including W-L (in and out of conference), SOS (in and out of conference), road/neutral play, conference standings/conference difficulty, record vs. various cross-sections of teams (e.g., good wins/bad losses, record vs. Top 50), victory margin (against various types of teams, that is, margin vs. Top 100, Top 50, etc.). And the Committee considers (most) of this information. So why not just list the W-L and SOS for the Committee and leave out the RPI? Because it's very difficult to get an instinctive feel for how SOS effects W-L. Is 17-7 against the toughest SOS better than 22-4 against the 57th toughest SOS? By how much? So the RPI was born.

It was never intended to take into consideration all relevant factors or be considered in a vacuum. And it isn't; the Committee doesn't simply take the top 34 RPI teams. Sagarin was designed to consider most of the relevant factors and so should be a better indicator of which are the better teams.

The final proof of whether the process works should be mathematically available as well: How do the Committee's choices, over the last decade or more, compare to the RPI, ELO Chess or other ratings? That is, does the Committee more closely follow (whether they consider the information or not) the ELO Chess rankings than the RPI?

Keep up the good work.

Peter Schroeder,
Dallas

This seems like a perfectly reasonable extension of last week's discussion, and I am encouraged by the number (and volume!) of people who have contributed their own thoughts on the subject. As to the question posed by Peter at the bottom, I'll not have anywhere near the time to answer it until after the season. If anyone else wants to tackle it in the meantime, feel free.

The most important statement about the RPI remains: "It was never intended to take into consideration all relevant factors or be considered in a vacuum."

Hey, Joe:

Upon seeing the title of your recent article, "RPI Should Be PRI: Probably Right Index," I first thought to myself, how can someone who does this for a living come to such a wrongheaded conclusion?

However, after wading several paragraphs in, I finally understand your frame of reference. As you said, you've invested thousands of hours here. Far too many, apparently, to admit you have been wasting your time on a very silly formula.

The RPI is nothing more than a shill for the big conferences. Your defense of it was less than compelling stuff, amounting to three rather limp points: (1) I spend all this time on it, so it must be accurate; (2) there's no better measurement out there; and (3) Butler isn't the team Alabama is.

I'll give you a pass on the arrogance of you're first point, yet I can't help but think this is what got you in trouble here in the first place. Let's just say that all those folks who toil on behalf of the BCS think it's everything it's cracked up to be, too, but we know differently, don't we, Joe?

Second, while I agree that the polls are unabashedly biased and the Sagarin is flawed, the inadequacies of each in no way makes a case for the RPI. That's just a lazy conclusion on your part.

Your third point is instructive because it so succinctly illustrates your wrong turn in the road. The whole point here is to identify the best team, not the best talent on paper. Why not then just rank teams on the basis of their last four recruiting classes? Clearly, the Crimson Tide are woeful as a basketball TEAM to have so much individual ability, only to find themselves on the verge of an NIT bid.

I'm an SEC fan and I've seen both play. Butler's better. They don't have the better talent, but they are the better team. One that would do quite nicely in a best-of-seven, thank you very much. That's much more "germane" to the discussion than the Vegas line. To suggest otherwise is to ignore the nature of Bama's nine losses while giving far too much weight to an early season win over a rusty and maybe arrogant Oklahoma bunch, not to mention a patsy non-conference schedule worthy of a Jim Boeheim-coached team.

Your first clue that the RPI is a loser should be the formula itself. I mean, 25%-50%-25%? Come on. What kind of tidy coincidence is that? Why not 28%-41%-31%? Sheesh.

Let's not mince words. The RPI rewards teams for merely being members of a select fraternity: a group of schools that snag the most sought-after talent, sometimes playing patsy schedules (all at home), then facing off against each other, all the while raking in additional RPI points for doing nothing more than showing up to complete their conference obligation.

The RPI measures neither quality wins, nor quality losses. The RPI places no value on road games, nor does it penalize teams like Syracuse for playing their first nine games at home, as they did this season and continue to do every season. In fact, schools like UConn and Pitt benefit quite nicely from Syracuse's stay-at-home scheduling, courtesy of that silly "Opponents' Winning Percentage" part of the formula.

The outrageous thing about the RPI is that it doesn't even possess the decency to stop there! The "Opponents' Opponents Winning Percentage" part of the RPI formula allows the entire incestuous cycle to continue unabated. Not only does UConn benefit from Pitt's padded schedule, they'll benefit from it a second time when they play Syracuse, courtesy of -- you guessed it -- Pitt being UConn's opponents' opponent, as well as UConn's opponent. And this goes on in other conferences, not just the Big East.

And while you're giving yourself a hernia patting yourself on the back for being able to project 63 of 64 teams correctly each year courtesy of the RPI, please admit this: 31 of those are pre-determined and you know it, sir. And, frankly, any savvy college basketball fan can predict the next 25 or so who are mortal locks because of the usual prejudices that exist within the NCAA. You're basically picking a handful of teams correctly. Nothing more.

Wanna' take another crack at this, Mr. Lunardi?

Richard Fitch

In no particular order: