The Math Speaks: Georgia Has Been Less Likely to Win With GameDay In Town
Lucas Puente, a 2010 UGA alumnus and an Interdisciplinary Graduate Fellow at Stanford, stops by to dispute a claim I made earlier in the week. This is absolutely incredible work, and I sincerely appreciate him sharing. You can follow Lucas on Twitter right here.
Since 1995, ESPN’s College Gameday has broadcast from the scene of 16 UGA games, though the crew has only been to Athens twice. The last of which was the infamous blackout game against Alabama, when the Tide raced to a 31-0 lead and never looked back. With that disastrous game in the back of many fans’ minds, the news that Gameday would be returning to the Classic City this weekend was met with a fair degree of trepidation. Those buying into the “Gameday Curse” were convinced Lee Corso coming to campus could only mean one things: the Dawgs would lose.
Andrew blogged about this earlier in the week, writing why he thought this theory was bogus. In that piece, he eloquently argued that the reason Georgia had only won 3 of these 16 games was not because of some superstitious curse, but rather just because they were the outmanned team on many of these occasions. Among other evidence, he points to UGA being the lower ranked team in 7 of its 13 losses. The point is a good one: we shouldn’t compare UGA’s record in College Gameday games (winning percentage of .1875) to its performance in all other games (.7402). In many ways, the two sets are apples and oranges. After all, Chris Fowler normally isn’t there to witness a blowout against Kentucky, but often is at the Georgia Dome on the first Saturday of December.
While I agreed with the gist of Andrew’s point and really wanted him to be right, I wasn’t ready to fully accept that he was until I could test the theory myself. After getting in touch with him (we first met as freshmen living in Myers Hall), I began this fun little project. He shared the list of College Gameday appearances at UGA games as well as the results from every UGA game from 1995 to today. I then added to this data set by looking up the Vegas line for each of the 232 games in the sample.
Once the data were collected and organized, I set about investigating the theory. As a first cut, I first ran a series of logistic and OLS regressions. Through theses tests, I examined whether the appearance of College Gameday could predict UGA’s result after controlling for other relevant variables (editorial note: I used a logit model when measuring this outcome dichotomously as either a win or a loss and an OLS model when measuring it as the point differential; both models produce substantively similar results). The controls included the following: Vegas’s line on the game, where UGA and the opponent were ranked (1-5, 6-10, 11-25, or unranked; I also used interactions of these), and the location of the game. I tested a few specifications of the empirical model (i.e. what I controlled for), but the unfortunate result that there is indeed a Gameday Curse indeed emerged. That is, holding these other variables constant, the appearance of College Gameday at a UGA game is associated with a statistically significant (i.e. non-zero) decrease in the probability of winning (I’ll hold off on a discussion of the magnitude until the end).
Still, as some of you are likely to point out, College Gameday doesn’t randomly choose where to go every Saturday (though I still don’t know why they went to Fargo, ND last week), so we can’t interpret the results of the regressions as confidently as we would a well-executed experiment. In statistical parlance, the findings discussed in the preceding paragraph do not necessarily represent the average treatment effect of the College Gameday crew coming to town (we have to make a series of additional assumptions for that to be true). Thus, I proceeded to test it again, this time using propensity score matching (for fellow stats geeks, I also tried Mahalanobis and Euclidean matching; the results produce the same conclusions). Basically, this technique finds the non-Gameday game that is closest to each Gameday game, based on the same same observed characteristics listed above, and then compares the outcomes. There are two ways to do this, matching with replacement or without it. The former enables a non-Gameday game to be the relevant comparison for more than one Gameday game (thereby minimizing bias), while the latter does not (minimizing the variance).
The results from this matching exercise are displayed in the table below (remember that this is done through my statistical software, R, not by my own ad hoc guesses). These two sets of games are basically identical to the games that the College Gameday crew came to, with the obvious difference being no ESPN Gameday around pre-game. Demonstrating this, in, non-Gameday games, the average line is -7.7034, but this rises to +3.0938 after matching, a figure indistinguishable from Gameday games’ average line of +3.0938.
Unfortunately, the finding that College Gameday’s appearance is associated with the Dawgs being more likely to lose holds even after matching on these observable characteristics. In terms of magnitude, the “treatment” of Fowler, Corso, Herbsteit, and company showing up is associated with a 37.5–50% decrease in the likelihood of winning.
The conclusion here is that traditionally Georgia hasn’t played well following the on-site broadcast of College Gameday as compared to otherwise (seemingly) identical games. On the surface, this does not bode well for the Dawgs on Saturday. However, these findings do not mean that we should go ahead and write off our chances against the Tigers. Remember that my statistical analysis is backward looking and therefore not designed to predict what will happen after the first whistle blows. My point is that this year’s team is different in many ways than the teams that came before it and there’s really no reason the exceedingly poor performance of UGA in Gameday-games in the past should cause the Dawgs to lose on Saturday. So let’s help them reverse this trend by wearing red and making Sanford as loud as possible. Go Dawgs!
Thanks for this, Lucas. Please come back often.