Football Fantasy Football Cricket Rugby League Other Sports eSportsZone Entertainment About Jobs Contact
22 Aug 2016 by TSZ

Projecting the 2016 NFL Season


With the NFL season fast approaching, we decided to make some early season projections using a variety of statistical methods.

Before we go any further we should say that, due to there only being 16 games for each team during an NFL season, you will usually find good teams that get on the wrong side of the variance and bad teams that get on the right side of the variance, which often means you can find some outliers that fall far from their predicted spots. You will also get players that unexpectedly have breakout years and big stars that will get injured. The Baltimore Ravens are a good example of this; they lost their starting quarterback, running back, wide receiver, pass rusher and a talented rookie wide receiver to season ending injuries last season, meaning they fell well below where most people had predicted.

How we did it

A well-established method for determining the win/loss records for a team over the season is to use a team’s points difference (points scored minus points allowed). Generally, a points difference of zero equates to around eight wins and eight losses. An additional 34 points would gain you an extra win and similarly, that many fewer points would result in an extra loss. It’s not a perfect solution but with predictions nothing ever is.

We looked at a number of variables to project teams’ performance for 2016. These were calculated from performing linear regression to determine which factors had large or highly correlated effects on the points difference of a team (and thus their winning record). Below is listed the coefficients of each of the factors used to determine the points difference of a team.

Offensively, we graphed the progression of a quarterback as they age, and then used this to project their 2016 performance based on their past three years’ statistics. We used the statistics available for any players who have played for less than three years, and in the case of the newly established LA Rams, we used an average for first-year starters. In most cases, the projected starting quarterbacks are with the teams they played the 2015 season with. For these quarterbacks we were able to simply use the age coefficient to adjust their 2015 numbers. There are two instances where teams will be starting existing quarterbacks but not ones that have played for their teams before. The Houston Texans' likely starter will be Brock Osweiler and the Denver Broncos' likely starter will be Mark Sanchez. For these players, we used the age coefficient to project their average yards per attempt, touchdowns per attempt and interceptions per attempt, but used their new teams’ passing attempts from 2015 rather than the quarterbacks’ previous passing attempts. Both of these teams are highly likely to play the same or very similar offensive systems to 2015, meaning using the passing attempts from 2015 is a reasonable estimate. There will be some teams that are playing new offensive systems. For example, the San Francisco 49ers are likely to be playing a much faster paced offense as new head coach Chip Kelly prefers to play that way. This means that, for the 49ers, we increased the number of offensive plays to mirror what Chip Kelly did whilst he was head coach of the Philadelphia Eagles.

We went through the same process for the running game too, but in this situation we had to do a lot more manual adjustment due to a number of factors; teams using multiple running backs; running backs returning from injury; running backs changing teams.

Finally, on the offensive side we adjusted the run/pass ratio from 2015 based on three things; coaching changes, player changes, and players coming back from injury. We chose not to include receiver statistics as these are defined by the quarterback statistics. Previous analysis showed that losing the average league starting receiver had only a very small effect on a quarterback’s performance; only losing a top-ten receiver had any noticeable effect and even that wasn’t significant.

Defensively the projections are a little more difficult and involved slightly more human intervention. We assigned a multiplier based on the players a team has lost and gained defensively. This was then used to multiply with a team’s defensive statistics from 2015.

Next, we adjusted for the strength of schedule. We calculated the change between the 2016 theoretical strength of schedule and the 2015 strength of schedule and multiplied it by 16 to get the overall difference over a 16 game season, and then by the average points difference per win that we calculated earlier. This allowed us to project the total points difference a team’s 2016 opponents would score against them in 2015 compared to 2016. We added/subtracted this to the projected points that came from using the fumbles, touchdowns and yards per attempt to get projected points difference based on the strength of schedule.

Finally, we turned all of this back into wins and losses to get a prediction for the season. This is what we got.

We decided that we’d compare these to the bookmakers’ odds to assess whether there was anything that particularly stood out as being different. Below are the William Hill season win totals at the time of writing compared against our projections.

The largest difference we found from the William Hill win totals was 3.5 wins. The interesting part was when we compared the profit from a £10 bet based on those odds and then applied a safety factor. The safety factor is the product of the profit from a £10 bet and the number of wins our projections are away from the bet losing. The safety factor should indicate what the best options are by maximising returns but minimising risk at the same time. From this, our favoured picks for the season would be for the New York Giants to be under 8 wins, the Jacksonville Jaguars to be under 7.5 wins and the Cincinnati Bengals to be above 9.5 wins.

We weren’t able to incorporate the effect of moving to a new city and possibly having a weak home field advantage for a while as there wasn’t any particularly useful data for this. The reason for avoiding the Tampa Bay Buccaneers is because they are in a division with two other poor teams, which means it’s entirely possible they could get four wins from those games alone. Finally, we’d stay clear of the Baltimore Ravens. Most players usually perform below their career average the season after suffering a major injury. However, they do have some very talented players and if they perform back to their career average levels then we could see them surpassing that 8.5 game win total.