Sunday, October 17, 2010

Initial BCS Rankings Computer Analysis; Big-Televen Bias and Margin of Victory, Vote in a Poll!

The initial BCS rankings were released this evening and were remarkably close to what my earlier analysis predicted.  I had all of the top-10, nailed the top-3, #6, #9, and #10 and had #4 and #5 swapped and #7 and #8 swapped.  Brad Edwards, look out!

But now that they are out we can do a deeper analysis and see what makes sense and what doesn't.

First, the polls and computers are way out of whack on a few teams.  Comparing the computer after with the polls we see that:
  • Oregon is #1 in both polls but #8 in the computers.  Only Sagarin has them rated close to their poll ranking (#2) so the computers are pretty consistent.  This is likely because the computers cannot take margin of victory (MOV) into account against their (so far) weak schedule.  My ratings have their schedule #69 in FBS.
  • Boise State is #2 in the polls versus #7 in the computers.  Not quite as bad as Oregon but the same reason as their schedule is #59.
  • Alabama doesn't get respect from the computers being #7 in the polls and #12 in the computers.  The computers seem to ding the loss more than the pollsters.
  • On the other hand, Missouri is #16 in the polls and #6 in the computers.  But oddly, Billingsley doesn't have Missouri in the top-25!  Hard to explain that one as my computer has their schedule #55, but Billingsley seems to not use where a team is ranked when the game occurs and doesn't adjust if a team improves, and so Missouri doesn't get credit for Illinois and San Diego State being better than expected.  Seems to be a fatal flaw of the algorithm, just that's just IMHO.
  • The computers don't like Wisconsin having them #16 while the polls have them #11.  This seems to be pollster bias and not a MOV issue though as my computer has the #20 and their schedule is #76.
  • The computers do like Oklahoma State at #9 but the polls don't as much at #15.  I agree more with the polls having them #22 and their schedule strength is #77 so hard to figure why the computers have them so high.  Note that Sagarin agrees with me having them #19.
  • That pollster bias appears again for Iowa having them #12 while the the computer have them #17.  With a schedule rated #84 the bias seems the issue.  Note the two bias cases thus far are in the Big-Televen.
  • Nebraska at #13/14 in the polls is higher than the computer's #20.  My computer has them at #11 so this seems to be a case of the BCS computers not being able to take MOV into account.
  • Kansas State is #25 and NR in the polls but #19 in the computers.  I agree with the polls more as my computer has them #36.  They have a decent schedule at #43.  Here the computer's inability to take into account MOV in their big loss to Nebraska seems to be the issue.
A couple things stand out in the above.  First, there seems to be a bit of Big-Televen bias (Ohio State is 4 spots higher in polls than computers too).  Second, BCS handcuffing the computers not allowing MOV causes some problems.

The subject of MOV is controversial and the BCS got rid of the computers that used it after Nebraska snuck into the championship game a few years ago.  The issue is that without it you punish teams unnecessarily when they play their weaker opponents.  For example, Boise State would have been #1 in the BCS last week but because they played a weak opponent (whom they beat 48-0 after calling off the dogs) they dropped in the computers this week and are #3 in the BCS.

Now, you can't reward running up the score (RUTS) but an algorithm that gives diminishing returns against strong opponents so as to not reward it, but gives suitable credit against weak teams to not punish it what is needed.  And that is exactly what my algorithm does and (IMHO) results in rankings that pass the smell test much better than most of the BCS computers and also performs well picking games.

What do you think?  Vote in the poll on the right to let your voice be heard.