In this model, you simply pick the home team as the winner. The 'home team advantage' which people talk about is a small, but very real effect - when averaged over many games, an NFL team playing at home can expect to score about 3 more points than if they were playing as visitors. Although one can theorize about differences in weather, stadium, the cheering (or jeering) fans, the best theory I have heard is simply that home teams get preferential calls by referees.
For the NFL 2014 season, here is how week-by-week (Week 5-20) predictions would have gone if you simply called the home team on each game:
One thing you might notice from the above graph is that there is a high variability in the weekly accuracy - from 27% correct in week 5 to 100% correct in week 20 (although week 20 only had 2 games, so this high accuracy was just a matter of 'getting lucky'). This is a general feature of all of the models I show - the prediction algorithm might make some absurdly bad calls, but over a large number of games the overall performance may not be so bad. In the large scheme of things, however, if you consider the accuracy over the 2012, 2013, and 214 NFL seasons, calling the home team as winner would have been correct about 57% of the time - which is just a little better than a coin toss (50% correct).
You may be wondering why I started the forecasts at week 5. The reason is that the particular data set I used for input did not have the pre-season games, so early-season forecasts for most techniques would have little supporting data, and in the case of Week 1, would be based on no data whatsoever (basically the coin-toss case).
Professional odds-makers have better predictive methods and algorithms than this one. This algorithm will NOT let you beat the odds in a consistent manner in Las Vegas-style gambling.
Next Page - Win Percentage