Date: Tue, 29 Jun 2004 17:18:20 -0700 From: "Peter R. Wolfe, MD" Subject: Rothman Hello David, A friend of Dave Rothman's called today to tell me that he had died. His funeral will be on Thursday in Long Beach. He was an opinionated guy, but in my view, he was entitled to that because he was a real smart guy as well. I have his system coded in c++ and it would not be much work for me to send you his football results, if you wish to continue to post them. PRW Date: Tue, 6 Jan 2004 04:57:52 -0800 Subject: Commentary From: David Rothman There has been a great deal of misunderstanding in the media and in sports about the role of computers. They do what humans tell them to do. In other words, there is some set of instructions, or formula, which they're following. Just because the public can't see what the computers do at each step is an irrelevancy. Suppose, for example, one lived at a time when the first genius introduced the formula Pct. = (games won)/(games played) as a ranking mechanism. The jocks and scribes of that era probably couldn't do any mathematics, so a formula with long division would have been mystifying, just as much of the computer output seems today. Indeed, many charlatans would have proposed their own methodologies to their royal sponsors, and there were probably quite a few frauds perpetrated before the simple ratio above became the standard for tournaments which are moderate departures from round robins. Now we just don't argue when professional sports like football or baseball use this formula. In our own era we have many college sports with highly unbalanced schedules for which the above formula isn't applicable. The formulae required for such complexity usually require computers for implementation, but they are no different in concept from the simple rule given above. How are the jocks and scribes of our era going to determine which formulae to use? The difference in our era is that we have developed institutions of peer review called professional societies to which some new formula can be presented. Indeed, the method which is used by FACT was presented at the 1972 meetings of the ASA in Montreal, and recapitulated in 2002 at the JSM, the Joint Statistical Meetings in New York City, in an invited paper. Any other techniques can be presented, in person or submitted for publication, if proprietors would be willing to go public. So the next time you read about a guy hating computers because of what they've done to the BCS, you can think about the power of the modern scribes to influence the modern businessman. After all, a general consensus of computer systems agreed that the BCS has, in every one of the years of operation, produced the teams with the two best average records over the past season, exactly what is indicated by the above formula in professional sports. If the BCS wants something a little different for college football, they can always add bonus points for winning post season conference championship games. To some extent, the opinions of our modern scribes are corrupted by other factors, such as past performances. For example, FACT placed USC 3rd and LSU 28th in final rankings for 2002, the previous season. This is known to affect opinion, especially when actual performances are almost indistinguishable. LSU beat both opponents the two had in common in 2003 by only one extra point. FACT's final standings for LSU and USC were only 2.637 LOGIT points apart (these resemble football points), a difference which, if true, would require over a decade of repeated play to demonstrate superiority. Don't be surprised that this small difference could be reversed by a small unconscious bias towards USC because of the previous season's results. But this is behavior which we can instruct our computers to avoid, if we want to. Note that the "Pct." above does exactly that. There may be other factors affecting opinions. USC is in a major city, for example. LSU isn't. One expects more sportswriters in a metropolis, and lots more fans who can write nasty letters and flood talk shows. I'm guessing that the number of USC alumni in the media is also a help in getting increased coverage. Finally, why should anybody give a damn about opinions from guys who can't get a reasonable opinion on a team like TCU? Are there any readers who think that TCU is in the top 25 for 2003? Yet the sportswriters you are accustomed to relying on made this their consensus for 2003! Any computer permitted to look at all the data, and that means margins of victory, would put TCU a lot lower. Apparently our modern scribes saw that TCU had only the two losses, and pretty much quit thinking. A computer wouldn't do that unless deliberately hobbled! Date: Tue, 16 Dec 2003 05:53:35 -0800 Subject: Commentary From: David Rothman Now is the time to reflect constructively on the polls and the BCS. Certainly the BCS would be improved by a policy of symmetric trimming of the computer rankings. Removing both highest and lowest data is recognized by professional statisticians as a valuable technique for increased robustness of estimators. ------------------------------------------------------ A long time ago I recommended that the BCS formula can be improved significantly by using value = 26 - (points)/(panel size) instead of ordinal for the poll entry. As an example, here are the major polls after the 2003 regular season (they have 65 sportswriters and 63 coaches): Ordinal Team Sportswriters Coaches Mean Points Value Points Value Value ------- ---- ------ ----- ------ ----- ----- 1 USC 1595 1.462 1542 1.524 1.49 2 LSU 1580 1.692 1516 1.937 1.81 3 Okla 1491 3.062 1449 3.000 3.03 4 Mich 1437 3.892 1393 3.889 3.89 This would seem to give more realistic results, and is going to prevent the tipping effect (when voting is so close that a strategic change by a voter can change an ordinal). Note that the USC-LSU difference is reduced considerably by this procedure. I believe there is one coach who appears to be marking down LSU a lot, under the cloak of anonymity. ------------------------------------------------------ But there are those who still wonder why the polls and the computers don't agree, particularly this year. We know that humans are marking down Oklahoma because the conference championship game was the most recent game, and because of the huge margin. I'd guess that all or most of the BCS computers try to weigh results equally during the season, and they aren't allowed to refer to margin. Dropping the latter requirement is one way we could improve the overall BCS product. (Those systems which completely ignored margins produced the shameful rankings of TCU in the past year.) And if we want the end-of-season conference championship to get increased weight, we could award bonus points for those winners, without tampering with computer rankings. The USC-LSU conflict is hardest to understand. Humans have a behavior (called stickiness in economics) which might explain why they persist in ranking USC slightly higher, even when a slow accumulation of evidence made most computers switch to LSU. We human beings like to stick to our opinions unless there is an extraordinary piece of evidence to the contrary. This phenomenon is why lobsters don't leave a pot of water that is slowly heated. In this case people stuck to their opinion on LSU as long as nothing extraordinary happened, but the computers assessed evidence impartially, without being wedded to previous estimates. Reflexive impartiality, if I may coin a phrase, is a desirable characteristic, and one that people just don't have. From drothman1@juno.com Tue Dec 9 17:29:12 2003 Subject: Commentary Those who do not learn the lessons of history are con- demned to repeat them. The BCS learned nothing from a situation in 2001 where a team was not the champion in a conference after being clobbered late in the season, yet that team's average performance in the season gave a high enough result to send them to the championship, where they were also clobbered. Scapegoats were found (computer systems using margins), but the real problem was that the BCS should have given a bonus to any team winning an end-of-season conference championship game. If that bonus were large enough, Colorado (rather than Nebraska) would have been selected to play in the Rose Bowl after the 2001 season, and USC would be scheduled for the Sugar Bowl this year (instead of Oklahoma). Of course the Sugar Bowl might well be a decent match, and the two teams with the best average performance in the season are again going to play, but we are running a risk of a repeat of the 2001 debacle. (The best fix would be to expand the playoffs to at least four teams as soon as possible, but this could take time.) I at- tend many fine seminars at USC, but cannot explain why the polls this year repeatedly preferred USC over LSU. FACT's standings had LSU slightly above USC for weeks. Oklahoma clearly was demoted to #3 in both major polls because voters weighed more recent information (a rout by Kansas State) more heavily than if such a loss were in the first game. (A computer system usually assigns equal weight to all games.) Note that I'm not criticizing the final BCS standings. Once again, the top teams are in the correct order for average performance purposes. But which teams get the invitation to the championship is a different issue. ------------------------------------------------------ If Oklahoma or LSU should win the Sugar Bowl by a huge margin, there wouldn't be much doubt that this team is the undisputed national champion. But if the decision in the Sugar Bowl is very close, and USC wins very big in the Rose Bowl, then FACT may award a cochampionship this year. ------------------------------------------------------ The polls and the BCS ended the 2003 regular season in a blaze of ignominy, giving us absurd rankings of TCU. They really don't have any shame now, do they! But at least the BCS did one good deed by ameliorating a poor ranking that both polls gave Miami (Ohio). There, I'm ending on a positive note. From drothman1@juno.com Tue Dec 2 12:33:40 2003 Subject: Commentary Is there no end this year to the abominations apparent in the gross overranking of TCU by the polls (both the polls have them 19th this week) and by the BCS (17th)? To paraphrase a public question I heard about 50 years ago: GENTLEMEN, HAVE YOU NO SHAME? TCU struggled last week to beat a winless SMU by one touchdown, but these authorities must have missed the game. Decent respect for the opinions of mankind require that we once again analyze TCU's record for the 2003 season (11-1), while updating the rankings of the opponents (who knows when this seemingly futile exercise may eventually register on the minds of sportswriters and coaches): Score Opponent Fact's Rank ----- ---------- ----------- 38-35 Tulane 100 17- 3 Navy 72 30-14 Vanderbilt 103 13-10 Arizona 87 27- 0 Army 168 13-10 South Fla. 78 27-24 Ala.-Birm. 82 62-55 Houston 79 31-28 Louisville 60 43-10 Cincinnati 94 28-40 So. Miss. 45 20-13 SMU 157 TCU had seven close wins, and lost to Southern Missis- sippi. FACT's ranking for TCU is now only 44th. I am biased, but that seems eminently reasonable to me. ------------------------------------------------------ By the way, the polls both gave poor results this week for West Virginia (8-4), which most recently had a win over Temple (now 1-11). This seems to corroborate how human voters cannot comprehend a really weak opponent. But the overranking might instead be due to having all their losses in the first five games, and then winning seven in a row. Those computers trying to emulate the standings in professional sports, where any outcome is weighed equally, wouldn't behave that way. From drothman1@juno.com Tue Nov 25 10:32:18 2003 Subject: Commentary This week the annual playoff invitations were made for NCAA Division I-AA. Ordinarily, we wouldn't want more than two or three representatives from a conference in a 16 team playoff. But 108 Idaho State 36.24 394 423 8 4 0 111 Weber State 35.66 325 241 8 4 0 both uninvited, were so close to the top teams in I-AA and in their own conference, while so much better than 206 No. Carolina A&T 11.35 284 205 10 2 0 208 Bethune-Cookman 10.96 352 220 9 2 0 both invited, that an exception would have been useful this year. Both of the latter teams are set to play a first round game against a school roughly 3 touchdowns above them in FACT's standings, not likely to generate much information. From drothman1@juno.com Tue Nov 18 16:56:27 2003 Subject: Commentary This week the annual playoff invitations were made for the NAIA and for NCAA Divisions II and III. That's my cue for my annual rant and second-guessing: In NCAA Division II, Bentley may have been undefeated, but North Dakota State (8-3) was over 18 points higher in FACT's standings and should have been invited (even though one other team from their conference is also in the playoffs). The methodology for Division II failed to analyze the difficulty of games against Montana and UC Davis, where North Dakota State was 1-1. In NCAA Division III, uninvited teams 250 UW Stevens Point IIIW 4.82 394 186 8 2 0 297 MaryHardin-Baylor IIIS -4.13 419 120 9 1 0 315 Pacific Lutheran IIIW -6.26 258 169 6 3 0 318 Menlo IIIS -6.64 397 241 7 3 0 319 Willamette IIIW -6.77 323 191 7 3 0 324 Concordia(Moorhd) IIIW -7.18 353 202 7 3 0 were all over two touchdowns higher than invited teams 421 Redlands IIIS -22.30 173 203 6 3 0 422 Hope IIIN -22.61 318 230 7 3 0 425 Trinity (Texas) IIIS -22.99 351 138 8 2 0 446 Allegheny IIIN -25.52 245 170 7 3 0 486 Hanover IIIN -30.27 386 261 8 2 0 506 Curry IIIE -31.92 371 140 11 0 0 One consequence of inviting such weak teams is that we will see a first round pairing of Hanover and Baldwin- Wallace, which is 5 TDs higher. Fans deserve more. And giving Lycoming (#412, 8-1) a bye when they hardly deserve a playoff invitation also seems strange. In the NAIA, Southern Oregon was only 5-4, but over 24 points above Tabor (9-1), which got an invitation. All things considered, this isn't as bad as last year, but there's still much room for improvement. There is still no good reason for automatic qualification. Not one. Dave, 310-676-4032 David Rothman 14125 Doty Avenue, #23 Hawthorne, CA 90250-8042