I don't think it would surprise many of us if there was a significant amount of bias in the two human components of the BCS rankings: The Harris and USA Today Coaches polls. Much internet ink has been spilled on how each of these are bias, but I wanted to take a quick look at the current iteration of the USA Today Coaches Poll in one small facet to see if there is evidence of bias: Is there a regional bias behind a coach's vote for Oregon or Auburn. To do this I first grabbed all of the votes by coach for the latest poll (Dec. 5th) and also found lattitude-longitude for the school of each coach. Here mapped out is whether each coach voted for Oregon over Auburn or Auburn over Oregon by location. I also have marked Oregon and Auburn
The first thing that struck me is the stark east-west different in voting. The only coach west of Texas to vote Auburn over Oregon was Mike MacIntyre of San Jose State. Oregon also had a much better showing on the east coast than I would have initially expected.
To see if there was regional bias in voting I decided to use a modified form of the Ripley's K statistic. In breif that this statistic calculates is what portion some type of event (in this case a vote for a given school) are within a given radius to another type of event (in this case, I am using the location of that school. So for instance within 5 miles of the Oregon campus K would be equal to 0 (no votes in that radius. That closest is Mike Riley at Oregon State at about 38 miles away.) At a radius of 3000 miles around Oregon the K would be equal to 1, because all of the votes for Oregon are within this radius.
OK, now bear with me here for just a moment. That K stat doesn't tell us anything by itself, we need something to compare it to. What I chose to do here was a Monte Carlo randomization method to generate a new random distribution of votes and recalculate my K statistic. I took the locations of the school for grant, as well as the numbers of votes for Oregon (35) and Auburn (24), and simply shuffled them. In other words given a 35/24 ratio of votes randomly distributed around the schools, how likely would it be to see the pattern of votes that we actually got.
After running 10,000 randomizations of the data we can then plot the 95% confidence interval of the K statistic by radius (the gray range in the graphs below), as well as the observed K statistic (the red line). If the red line rises above the gray, there are more votes for the school than would be predicted by chance (perhaps bias for that school within that distance), and if the red line falls below the gray range there are few votes than would be expected at that radius (perhaps bias against that school at that distance) .
This is what the analysis seems to suggest: There seems to be a slight bias for Oregon within 1300 mi, but a pretty pronounced bias against Oregon at distances over 2200 mi. There seems to be a pretty strong bias for Auburn at intermediate ranges (500-1000 mi).
Now from the map it looks like there should be a pretty strong bias against Auburn on the west coast, but the analysis didn't bear this out. Let me explain why this may be: First off, I am taking the number of votes as granted, so at any given location there is a greater chance the vote will be for Oregon over Auburn. Additionally there are far fewer coaches on the west coast voting than on the east coast. Together this makes it much more likely to get a large geographic range with no votes for Auburn simply by chance. Never-the-less, there still was a small amount of bias for Oregon in that region. Either way, there seems to be ample evidence of regional biasing in the coaches poll, which seems like just another reason to scrap the BCS once and for all.