Football season is officially back. With that, here is a breakdown of our model’s predictions for all Week 1 matchups of the 2017 NFL season. We promise we did not make manual adjustments in favor of the Vikings, who the model loves this week.

Note: some elements of these models are built off of each team’s last 7 games. Things change quickly in the NFL, and some teams have seen drastic overhauls, coaching changes, and/or a QB change since those last 7 games were played (e.g. 49ers). For those teams, recognize that these changes are not yet fully incorporated in the predictions below. 

Win Probability and Spread Predictions

For each game, the table below shows our predicted win probability for the home team (284 Home Win Prob), the Vegas Spread, our predicted spread (284 Spread), and our predicted probability that the home team covers the Vegas Spread (284 Home Cover Prob). For the Home Win Prob predictions, figures below 50% predict that the away team will win. Note that all of these figures are from the perspective of the home team (e.g., -9 means the home team is favored by 9 points and +9 means the away team is favored by 9 points). For our 284 Home Cover Prob, figures above 50% point to the home team covering, while figures below 50% point to the away team covering. I would consider > 65%  a high mark (strongly favoring the home team) and < 35% a low mark (strongly favoring the away team). Note that each model is generated separately, so there will be games where the predictions don’t agree with each other. The point of having multiple models is to get different perspectives on the game, so focusing on games where all the models are in agreement is a good place to start.

Over/Under Predictions

For each game, the table below shows the Vegas Total, our predicted total (284 Total), and our probability that the game goes OVER the Vegas Total (284 Over Prob). For those who are not familiar, these “totals” represent the total points scored between both teams (e.g., if one team wins 24-20, the total for the game would be 44 points). For our over probabilities, figures above 50% point to an over, while figures below 50% point to an under. I would consider > 60%  a high mark (strongly favoring the over) and < 40% a low mark (strongly favoring the under). As a quick explanation of how to interpret these numbers: the Vikings/Saints Vegas total is currently 48 points, while our 284 Total is 52 points (favoring the over), and our 284 over probability is 61% (also favoring the over). If the Vikings/Saints combine to score more than 48 points, this game would “go over.” Based on historical predictions, our 284 Total prediction accuracy increases when it is 2-3 points different from Vegas’s Total, and our 284 Over Prob accuracy increases when probabilities are above 55% or below 45% (i.e.,  far enough away from 50/50). Most importantly, focus on games where both models agree with each over (i.e., they both favor the over or they both favor the under). Each model is generated differently, so they do not always agree with each other. If they are saying two different things, I wouldn’t feel too good about hammering the over.