Can a person really predict the outcome of the biggest game of the National Football League season?
North Dakota State Ph.D. student in the Department of Statistics Joe Roith gave a Super Bowl Seminar Friday to predict the winner of the Vince Lombardi Trophy.
Using the season average statistics, Roith said for the point-spread model the New England Patriots should win by about 1.21 points. “The logistic regression model gives the Patriots about a 65 percent chance (they) will win,” Roith said.
However, using the 2014 playoff averages, “we would expect the Patriots to win by almost two touchdowns … so this is just based on how they performed in the playoffs … we give (New England) a 99 (percent) chance of winning.”
The models that Roith presented are over 70 percent accurate, but you never really know what is going to happen before the game actually starts.
“So, in this case, both of our models are using playoffs and seasonal averages kind of suggest the Patriots winning,” Roith said. “The big question is are both teams going to play like that, are the Seattle Seahawks going to play like they did against Green Bay, (by) giving up lots of turnovers, (and) are the Patriots going to play like they did against the Colts, really rubbing it in and deflating balls?”
Roith explained that if one team has more first downs than their opponent, they should score 1.2 points more at the end of the game. He went on to explain this logic for turnovers:
“ … For every extra turnover that a team commits, (they) are basically giving the other team almost four points at the end of the game.”
Out of 33 key variables that determine the outcome of a NFL game, Roith found that only seven or eight variables are needed to be plugged into the program. For example, first down margin, yards per rush, turnover margin and yards per sack.
“The largest coefficient was to the turnover margin,” Roith said. “Lots of turnovers means you’re giving up a lot of points.”
Jenn Johnson, a graduate student, who is part of Rhonda Magel’s Stat 775 class — using statistics in sports — also presented a similar presentation to Roith’s, but it was a little less complex.
To conclude the event, Scot Jones presented a different model by trying to predict the postseason performances of the 10 NFL teams based on their regular season averages.
No one knows the outcome of NFL games until they are played, but these three students tried to predict the outcomes of games by digging up statistics and plugging them into a program that spits out the final outcome of the games.