Last Thursday’s NFL game between the Patriots and the Jets was hardly the marquee product the league wants to be putting out. For the first time since 2009, New England QB Tom Brady completed fewer than 50% of his passes, and in the fourth quarter, his New York counterpart, Geno Smith, managed to toss three interceptions. The teams combined for a total of 23 points, which was 20 below the Las Vegas total.
If you’re like me – or one of several radio and TV personalities – one plausible explanation for Thursday’s yuckfest was the limited preparation time each franchise had. Specifically, the thought was that relative to games played on Sunday or Monday, Thursday contests would be associated with lower points, lower total yards, and higher turnovers.
Last night, I did a bit of research into this idea. I used every regular season from the last eight years of data at pro-footballreference.com, tossing away the first week of the season, where, of course, teams have several months of time to prepare. My three outcomes were (i) Total points (ii) Total yards and (iii) Total turnovers. My “exposure” of interest was Thursday games, where I used an indicator for whether or not the game was played on a Thursday (this includes Thanksgiving games). I adjusted for three game variables: the home and away teams, and the week in which the game was played. This was done because certain teams generally play higher or lower scoring games, and certain weeks of the season may be more conducive to points, yards, and turnovers. Lastly, I used traditional multivariate linear regression, although I also tried a generalized linear model for Poisson outcomes with the turnover data, finding similar results.
The table below presents estimates for the models of points, yards, and turnover outcomes. Estimated changes on Thursday games, along with their standard errors and p-values, are shown. While there appears to be a slightly lower number of points, there are actually more yards and fewer turnovers on Thursday games…none of which fall anywhere close to levels of statistical significance.
|Change in points, Thursday games||-0.637||1.749||0.716|
|Change in yards, Thursday games||11.738||14.386||0.415|
|Change in TO’s, Thursday games||-0.170||0.233||0.464|
One of a few things could be happening.
First, there could truly be no difference in our outcomes on Thursday and non-Thursday games, in which case the Patriots and Jets have no excuses for that game. Second, while I adjusted for team type to account for style of play, ideally I would’ve been able to adjust for the Las Vegas totals in these games. If anyone knows of sites that have that information easily accessible, I’d love to know. My inclination is because the the NFL tends to place their top teams in these Thursday games, our model may be effected. Lastly, the sample size for early season Thursday games is quite small, and while I tried to check for differences based on early season versus late season games, until there are more years of data, this is impossible.
PS- In case you were wondering, this was roughly my expression when I ran the regression and found no evidence of any differences in Thursday game outcomes.