Monday, September 27, 2010

regional bias study, second week

If you haven't seen last week's look at the AP poll to study it for regional bias, check that out, to get up to speed and for instructions on reading the charts. Here are this week's:

Chart 1 displays what I referred to as "overvotes" - that is, the number of times a team was ranked in a ballot five places or more than their actual poll ranking. Chart 2 displays the averages. Keep in mind, these are the votes from the poll that came out on September 21, not yesterday.

Some trends from last week continue.

- Marquee teams like Oklahoma, Florida, USC, and Michigan are far more likely to get overvotes than non-name teams.
- Pac-10-area voters are generally more guilty than most of overrating teams from their own region, though not quite as strongly as last week.
- The strongest evidence of regional bias continues to be located in the area of the poll below the top 25 where the simple act of voting for a team at all constitutes overrating them. The four teams ranked from 30th-33rd - Fresno State, Missouri, Oklahoma State, and Clemson - get the lion's share of their votes from their home regions. Particularly Fresno State - good lord the West Coast voters are shameless in their support for the Bulldogs. However, in the top 25 itself, there appears to be little evidence outside the West Coast (and even that is rather weak) that regional bias is manifest.

One trend that disappeared this week was the one-vote teams: last week, the vast majority of them received their lone vote from regional voters. This week that isn't the case.

The averages chart is very interesting, though. Had those numbers shown up last week it would have really perked my ears up for regional biases. Last week the distribution looked about 100% random. This week there's some noticeable upvoting for regional teams. The average "bias" is to rank teams a third of a place higher than the rest of the country - a small number to be sure, but there are just eight teams out of 25 that didn't receive favorable reviews from their home region writers. It'll be interesting to see, going forward, whether last week's trend (basically zero bias in the top 25) or this week's holds true.

No comments: