The college football season is underway, and with it, the 28th season of ESPN's pregame show, College Gameday. With it also, the College Gameday curse, or at least rumors of its existence.


This week's show will be broadcast from Eugene, Oregon where #3 Oregon will face the #7 Michigan State. It will be the 16th time that Gameday will broadcast from the location of an Oregon game and the eighth time the show has broadcast from a Michigan State game. Oregon fans should feel good about the national spotlight, since the Ducks have won 11 of their prior 15 Gameday games (73 percent). Michigan State fans, on the other hand, might be more concerned because their team has won only 3 times out of 7 tries (43 percent).

The question we try to answer here is whether the concern some Spartan fans may have when Lee Corso and crew show up for their games is grounded in a broader pattern. In other words, do any of the regularly-featured Gameday teams suffer from a "Gameday curse" and do some of them even potentially benefit from a "Gameday bump"? To answer this question, we collected data for the 30 schools that had Gameday broadcast from their games at least five times. Our dataset includes every non-bowl game played by an FBS team since the first full season of College Gameday in 1995.


Gameday win percentage versus non-Gameday win percentage:

School Gameday Appearances Gameday Win% Non-gameday Win% Difference
Oklahoma 24 75% 73.70% 1.30%
Oregon 15 73.30% 74.60% -1.30%
Miami-Florida 15 66.70% 69.80% -3.10%
Southern Cal 16 68.80% 72% -3.30%
Michigan 18 66.70% 70.90% -4.30%
Colorado 5 40% 47.50% -7.50%
Auburn 12 58.30% 67% -8.60%
Louisiana State 20 65% 74% -9%
Florida 31 67.70% 77.90% -10.20%
South Carolina 7 42.90% 54.90% -12.10%
Texas 13 61.50% 76.10% -14.50%
Alabama 25 56% 70.60% -14.60%
Ohio State 26 69.20% 84.30% -15.10%
Michigan State 7 42.90% 59.70% -16.90%
Missouri 5 40% 57.10% -17.10%
Notre Dame 21 47.60% 66% -18.40%
Wisconsin 8 50% 70.30% -20.30%
Florida State 21 57.10% 78.50% -21.40%
Stanford 6 33.30% 55.40% -22.10%
UCLA 6 33.30% 58.10% -24.80%
Nebraska 11 45.50% 75.20% -29.80%
Clemson 6 33.30% 65.10% -31.80%
Tennessee 16 37.50% 70.20% -32.70%
Virginia Tech 9 44.40% 78.80% -34.40%
Penn State 12 33.30% 68.60% -35.20%
Iowa 5 20% 58.60% -38.60%
Oklahoma State 5 20% 59.50% -39.50%
Kansas State 5 20% 69.40% -49.40%
Georgia 17 23.50% 75.40% -51.80%
Texas A&M 5 0% 62.80% -62.80%

As the table shows, there is only one school, Oklahoma, which has won a higher percentage of the time that they're on Gameday than when they aren't. Every other school that has appeared more than four times has performed worse on Gameday than they do otherwise. Some schools, like Texas A&M and Georgia, have a drop in win percentage of more than 50 points.

So should we take these results as evidence that every school except the Sooners has a Gameday curse? Of course not. The overall Gameday schedule has a heavy, heavy selection bias, full of matchups between the best teams in the country—and very far removed from the rest of the schedule, which features matchups like Alabama nuking Florida Atlantic from orbit.


So how can we try to correct for this? The best way of doing so is to employ a statistical technique called matching. The idea behind matching is that you use a computer algorithm to pair "treated" cases to the "control" cases which are most similar in terms of measurable attributes. In this context, we take each team's Gameday games and find their non-Gameday games which are most similar in terms of their and their opponent's AP rankings before the game and whether or not it was a home game. Ideally we would match on other factors, like offensive and defensive stats, but small sample sizes make it difficult to find good matches.

To further improve the validity of our analysis, we account for the Vegas line for the game by using whether the team covered the spread, rather than whether they won the game, as the outcome variable. If we did not do so, teams that tend to be featured on Gameday when they're underdogs and then lose would still unfairly appear to suffer from a curse. Thus, by focusing instead on a team's performance against the spread, we can correct for any bias that results from teams being either the favorite or underdog in most of their Gameday appearances.

The graph above shows the estimates of the Gameday effect for each team, based on logistic regression using the matched data. As with the previous table, Miami is at the top of the chart. The model estimates that they are 31 percent more likely to cover the spread when they're featured on Gameday, compared to similar games that are not featured.


Clemson is next, with a 27 percent jump in the probability that they outperform expectations. While some fans of the Tigers may be quick to point out that their team has only won a third of these games, that ignores the fact that they weren't often expected to do so in the first place (hence our focus on the performance against the spread instead of win percentage).

An even more extreme example is that of Texas A&M, which is 0-5 in Gameday games, but has one of the highest Gameday bumps (about 16 percent). The reason for this is that they have covered the spread in 60 percent of their Gameday games, but only 34 percent of the time in similar non-Gameday games.

At the other end of the graph is UCLA, which has an estimated Gameday curse of 39 percent. Virginia Tech (35 percent) and Stanford (31 percent) fans should also hope that their teams do not get picked for Gameday any time soon.


Breaking down a non-random schedule like Gameday's certainly provides lots of challenges. The show's schedule-makers want to pick the most compelling matchup possible based on how hot each team is and the national implications of the game. Other areas to look at would include observing how win percentages change by week, or how teams grouped by relative strength through a metric like ESPN's Football Power Index fared. But for now, history shows that by these broad measures, Oregon stands a very good chance of covering its -12.5 line this Saturday, and Michigan State doesn't.

Stephen Pettigrew is a PhD candidate at Harvard University, where he studies political science and American politics. He also has a master's degree in statistics from Harvard. In his spare time, he writes about sports analytics, particularly in hockey and football.

Lucas Puente is a PhD candidate in political science and Interdisciplinary Graduate Fellow at Stanford University. He is a passionate fan of the Georgia Bulldogs, having completed his undergraduate education in Athens.