BCS Bias
8th December 2011
Here’s a graphic, but not vulgar, take on Bias. Begin with Objective Processing.
Start with a simple 2 way ANOVA experimental design with WATTage manipulations on the X axis and Argument Quality in the body of the table. The Y axis holds the attitude scores. Dual Process Models predict that lovely fan shaped interaction between WATTage and Argument. Under Low WATT, the Other Guys lack willingness or ability to think, so they don’t use Argument Quality to determine attitude. But, under High WATT, Other Guys seek Args, then elaborate over them in that Long Conversation in the Head along the Central Route. Argument Quality is decisive with Strong Args producing positive change and Weak Args producing negative change.
Now. Let’s Bias this experiment. Make the attitude object self-relevant, self-defining, self-loving. What happens with our 2 way ANOVA now?
See the fan shift up with this (positive) example. See that Weak Arguments produce more positive change. That fan shift between the two experiments is the DNA, the hominid bone, the crucial difference that marks two breeds in the same species. Both show the impact of High WATTage. Both show the impact of Argument Quality. But see how Bias, biases that Long Conversation in the Head with Weak Arguments in this example.
Now realize that this is exactly what’s going on with college football coaches who vote each week on the rankings of teams that determines BCS bowls and championships. They see the weaker argument as the stronger under the stimulus of a biasing treatment like financial incentive. Consider this press report on the study.
Research conducted by Yale University economist Matthew Kotchen and University of Calif.-Santa Barbara political scientist Matthew Potoski, which covers the USA Today coaches poll administered by the American Football Coaches Association from 2005 to 2010, shows that coaches rank their own teams, teams in their own conference, and teams that they’ve defeated more favorably than merited. The researchers argue those biases skew the results of the poll, which is one of the components in the system used to determine which teams get to play in major bowl games, and what two teams go to the national championship game.
The abstract of the study admirably explains itself.
This paper provides a study on conflicts of interest among college football coaches participating in the USA Today Coaches Poll of top 25 teams. The Poll provides a unique empirical setting that overcomes many of the challenges inherent in conflict of interest studies, because many agents are evaluating the same thing, private incentives to distort evaluations are clearly defined and measurable, and there exists an alternative source of computer rankings that is bias free. Using individual coach ballots between 2005 and 2010, we find that coaches distort their rankings to reflect their own team’s reputation and financial interests. On average, coaches rank teams from their own athletic conference nearly a full position more favorably and boost their own team’s ranking more than two full positions. Coaches also rank teams they defeated more favorably, thereby making their own team look better. When it comes to ranking teams contending for one of the high-profile Bowl Championship Series (BCS) games, coaches favor those teams that generate higher financial payoffs for their own team. Reflecting the structure of payoff disbursements, coaches from non-BCS conferences band together, while those from BCS conferences more narrowly favor teams in their own conference. Among all coaches an additional payoff between $3.3 and $5 million induces a more favorable ranking of one position. Moreover, for each increase in a contending team’s payoff equal to 10 percent of a coach’s football budget, coaches respond with more favorable rankings of half a position, and this effect is more than twice as large when coaches rank teams outside the top 10.
You can read the gory details in the extended paper (pdf) if you like, but that Abstract gives it up. Financial incentives bias coaches and you see it in how they express their attitudes on Argument Quality (their ratings of college football teams). Every way you can identify self relevance – whether My team, My conference, My opponent versus Their team, conference, or opponent – the researchers document the Biased outcome. When it is Mine, it is Better and when it is Yours is it Worse.
In this instance we can interpret the computer rankings of football teams as the Objective Processing version of this experiment. Given no self interest, where do the different Arguments take you? Now. Compare the computer Objective side of the experiment to the human Biased side. If, indeed, there is no Biased Processing in coaches with financial incentives then there should be no differences between the attitudes for “computer voters” versus human voters.
While the careful quantitative analysis reveals the markers, it is the persuasion theory that predicts, describes, and explains the results. Bias alters how you process Arguments, making the Weak seem Strong, when it favors you, and the Strong seem Weak, when it doesn’t.
P.S. One of the oldest attitude experiments ever published (pdf) was entitled, They Saw A Game, and it studied the attitude differences between fans at a college football game. Here’s a nice description of the study if you don’t want to read the pdf. The more things change, the more they stay the same, right?
P.P.S. Of course, since this was published in 1954 and only available as a weird looking scanned PDF, you know it’s not true. We need an MRI replication to prove it, right?





