The Dunkel Index
By Woody Brammel, 11/9/99

Is it just me, or is anyone else wondering what sort of foo-foo dust got lodged in the Dunkel Index computer over the weekend? The Dunkel Index, of course, is one of the eight computer polls used in the BCS rankings. But what it spewed out this week is so bizarre, so unlike the other seven computer polls, so unlike itself actually, that it's hard to see the Dunkel as "one of" anything.

I should say up front that I have worked with computers for my entire professional career. I even have written a few programs to handle competitive ranking problems which were similar in many ways to the problem of ranking competing sports teams. All of this has made me a pretty strong supporter of the computer rankings. They may be a bit hard to understand, but generally the computer rankings do a more objective, more consistent and more accurate job of ranking than do the subjective polls of writers or coaches or even Joe Paterno.

The value of any computer ranking system is that it works from a defined, objective, and consistent mathematical model and is able to fit an enormous amount of data to that model in computing the final "index" or "power rating" or whatever. You can criticize the model - accept it or reject it - but it's out there for all the world to see (Well, OK, in this case the systems are proprietary, but the model is out there for the BCS Committee to see). The point is: Objective; Consistent; No funny business. And the data considered by ranking programs include the thousands of direct and indirect ties between teams in their competitive encounters - far too much information for any human brain to assimilate and evaluate properly.

The knock on computer ranking programs, other than disagreements with the model, is that they tend to be quite unstable early-on when there are relatively few ties between teams. But at this point in the season when most teams have played eight or nine games, there are lots of ties and the systems are much more stable. Also, the different polls tend either to yield fairly similar results, or to differ in ways expected from differences in their models. At least, that's the theory.

Not so, however with this week's Dunkel results, especially as they concern Virginia Tech. The most recent Dunkel Index has the Hokies rated sixth, and this in turn has a major effect on Tech's BCS ranking released on Monday. In fact, relative to what might have been reasonably expected, the Dunkel rank is solely responsible for Tech's third place position behind Tennessee rather than second behind Florida State.

Of course, it's easy to see self-interest surfacing in all this. It happens frequently. For example, Ryan Hokensmith, a student writing for the Penn State Collegian, last week strongly criticized the Dunkel index. In his view, the major flaw in the Dunkel Index is the system's failure to consistently rank Penn State #1 and Virginia Tech somewhere below all of Penn State's opponents (like Akron). That's not the argument here. The argument here involves consistency of the Dunkel, especially this late in the season.

Last week, the Dunkel had Tech rated #1, which means the system dropped Tech five positions this week, following a close win at West Virginia. Penn State, in contrast, dropped two places - from sixth to eighth - with a close LOSS AT HOME! Granted, the numbers say Minnesota is a better team than West Virginia, but losing to an unranked 5-3 team is three rating positions better than beating an unranked 3-5 team? Gimme a break!

The Dunkel results are even more bizarre when compared with those of the other seven computer polls. Tech dropped one position in four of these polls, increased one position in one poll, and was unchanged in the remaining two - an average decline of .38 positions. This compares with a drop of five positions in the Dunkel. Penn State, on the other hand, dropped just two positions in the Dunkel but an average of 3.8 positions in the other seven polls.

So, someone help me out here. With more than three quarters of the ties between competitors established, Team A loses to an unranked team at home. Seven computer ranking systems translate that into a drop of almost four positions while our featured program says it's a drop of half that. Team B wins a close road game. Again, seven computers equate that to a drop of less than 1/2 a position while our featured computer says "down five". I obviously don't know what happened but I can tell you this. Similar outlandish results from one of my programs would have me checking data entry and/or searching for the programming bug. End of story.

Since we don't know the details of the Dunkel Index system, the mystery likely will remain just that. The only clue I can find from the Dunkel Index ratings is the fact that seven teams from the Big-We-Can't-Count-To-Ten conference are ranked in the top 21. This includes in the system's Top-10, Wisconsin (4), Michigan State (9) and Michigan (10), ALL WITH TWO LOSSES, along with Ohio State at number 21 with a sterling 6-4 record.

On second thought, I think I’ve figured it out. Maybe the problem here is the line of programming code that reads: "If conference = B10 Then Index=Index +20".

          

HC Voice of the Fan Archives

HC Home

HokieCentral.com is an independent publication and is not affiliated with or endorsed by Virginia Tech or the Virginia Tech Athletic Department. All material is Copyright ©1996-2000 by HokieCentral.com, all rights reserved.