/* ]]> */
Dec 212009

Fan Pick Distributions:  Consistent Across Sites
Several websites sponsor college bowl confidence pool contests on all 34 bowl games.  The two most popular probably are those on ESPN and Yahoo.  It is interesting to compare how the contestants in the two sites’ confidence pool contests allocated their picks.

The first observation is how similar the percentages of contestant picks are. They’re so close they resemble polling results taken a few days apart. The correlation coefficient of the two distributions is 0.99.  Here’s a plot of the two series:

Fan Confidence Distributions:  Clear Example of Anchoring Bias
The second thing that’s obvious is how different the contestants’ average confidence weights are for each pick.  In this 34 bowl game contest, contestants are supposed to predict the game winners and rank their predicted winners according to how confident they are each team will win, assigning numbers from 34 (highest confidence) down to 1 (lowest confidence).  Given the similarity of the percentages of fans that picked each team, it is reasonable to expect their average confidence on each team might be similar as well.  But as you skim down the columns labeled “Conf.,” you’ll notice on Yahoo the later games get more confidence weight, while on ESPN the later games get less weight.  That’s no mere illusion:  the correlation coefficient of Confidence weights between Yahoo and ESPN is -.62.  What’s going on here?  How could contestants with nearly identical views on who will win have such different degrees of confidence about their expectations?

A simple answer lies in the two columns labeled “Default.”  While some online confidence pools require the user to enter or pull-down a confidence weight, both Yahoo’s and ESPN’s entry pages begin with pre-assigned confidence weights.  It’s up to the user to deviate from these default assigments.  In Yahoo’s contest, the assignments are in ascending order, while in ESPN’s they’re in descending order.  Clearly the contestants on both sites were anchored by the starting confidence weights.  Multiple regression analysis reveals that in both contests over 50% of the variation in Average Confidence Weight is explained by the Default weighting.
 Posted by on December 21, 2009 at 1:21 pm