clock menu more-arrow no yes

Filed under:

Intro to Football Analytics: Pre-Season Ratings

New, 37 comments

Taking a look at the factors that influence a pre-season ranking method

Jim Brown-USA TODAY Sports

This offseason I spent some time developing a purely data driven approach to predicting how teams are going to perform in the upcoming season. I wanted this model to be simple enough to be calculated fast (which is a euphemism for me being lazy) and easy to explain. So I looked at three factors; How well you have played in the last three seasons, how well you have recruited in the last three years, and how much draft talent you lost in the past three years. I'll go over the sources of the data in each section below. What is interesting is that for both team strength and recruiting performance the data from 2 years ago had no impact on predicting future team strength when combined with the data from last year and three years ago, so I have left out that data from the model. In each section below I'll explain the metric as well as how the ACC teams stack up on each measure.


I looked at the 247 Composite Rating provided by Compu-Picks here. They provide the rank of each recruit from 2005 to this February. I compiled the average 247 Composite Rank for each team in each year from this data. The scores are on a scale of about .7 - 1. The important information when looking at how teams are going to perform this year is how well a team recruited this past February and three classes ago. This makes sense to me, the information from three classes ago tells you how talented the Juniors and RS-Sophomores on your team are and those are, hopefully, the meat of your starters. The information from the most recent class probably tells you the direction the program is heading, even if those players won't have that big of an impact on the field this year they may show that a program is "on the rise" or not. Here is how the ACC teams stack up:

Average Recruit Ranking
Team 2014 Class 2012 Class
Florida State 0.90 0.93
Clemson 0.89 0.87
Notre Dame 0.90 0.89
Miami (Florida) 0.88 0.88
Louisville 0.84 0.85
Georgia Tech 0.84 0.85
Virginia Tech 0.86 0.86
North Carolina 0.87 0.85
Duke 0.84 0.82
Pittsburgh 0.84 0.85
Boston College 0.82 0.77
Syracuse 0.83 0.81
North Carolina State 0.84 0.84
Wake Forest 0.82 0.83
Virginia 0.86 0.83

Wonder why Florida State is playing lights out? They recruit like crazy. Even Clemson and Notre Dame haven't had the sustained success of FSU on the recruiting trail. Also this helps explain why UNC is a lot of people's "hot pick" for the Coastal this year. Georgia Tech doesn't fare too bad in this metric, placing tied for 8th out of 15 teams from this past February and tied for 6th from three classes ago. in technical terms a one percentage point increase in average recruit ranking (going from .84 to .85) would increase your prediction of team strength this year by .326 points per game. A one percentage point increase from three years ago would make your prediction .388 points per game higher.

Draft Talent Lost

To determine how much talent a team would have to be replacing I used Pro Football Reference to record the picks of players drafted from each team. I then used Football Perspective's Draft Value chart here. This Draft Value Chart measures the relative worth of each draft pick, so losing a first round pick is more important than a 7th rounder. The only significant draft information for predicting next year's team strength is the talent lost from this year's draft. Here is how much draft value each ACC team lost this year:

Draft Value Lost This Year
Team Draft Value
Florida State 36.30
Clemson 27.50
Notre Dame 36.80
Miami (Florida) 5.10
Louisville 33.30
Georgia Tech 7.20
Virginia Tech 15.70
North Carolina 23.60
Duke 3.20
Pittsburgh 16.50
Boston College 6.20
Syracuse 6.20
North Carolina State 2.50
Wake Forest 0.30
Virginia 8.10

Only Notre Dame lost more draft value than FSU, but Louisville, Clemson, and UNC also had a lot of talent leave. Georgia Tech is really boosted by Attaochu. Losing a 7th round pick would hurt your team's prediction by .08 points per game while losing the top pick would hurt your prediction by 2.76 points per game.

Team Strength

The biggest component of the model is simply how well a team has played on the field. I used a team rating system called the Massey Rating. It takes a team's points per game differential and adjusts it for the quality of opponents a team faced using linear algebra. The great thing about the Massey Ratings are that you can just subtract one team's rating from another's to see how much one team would be expected to win by. For instance if Georgia Tech met FSU last year on a neutral field FSU would be expected to win by about 30 points (LOL). Here are the Massey ratings for ACC teams from the 2013 and 2011 seasons:

Team Strength
Team 2013 Season 2011 Season
Florida State 40.23 13.49
Clemson 21.24 5.32
Notre Dame 9.73 13.68
Miami (Florida) 7.12 5.39
Louisville 15.25 1.75
Georgia Tech 10.10 3.33
Virginia Tech 6.78 6.90
North Carolina 6.74 1.57
Duke 6.71 -8.59
Pittsburgh 3.45 3.15
Boston College 1.78 -6.29
Syracuse -2.02 -3.17
North Carolina State -6.02 -0.69
Wake Forest -4.46 -3.11
Virginia -10.72 -5.34

A one point increase in your Massey rating from last year would increase your prediction for this year by .56 points per game, while three years ago would only increase your prediction by .12 points per game.

This year's prediction

Predicted Massey Rating
Team predict
Florida State 30.98
Clemson 16.79
Notre Dame 11.29
Miami (Florida) 10.37
Louisville 9.75
Georgia Tech 9.29
Virginia Tech 8.32
North Carolina 6.60
Duke 5.18
Pittsburgh 4.83
Boston College -0.07
Syracuse -0.09
North Carolina State -0.48
Wake Forest -0.91
Virginia -3.97

Georgia Tech is currently ranked 6th in the ACC by this method, but 5 teams are all within 3 points of our prediction. Home Field Advantage is worth about 3 points in College Football so these games would all be toss ups based on who was at home.

If you want to argue the specific metrics I use so be it, but I'll tell you now I just went for "good enough" metrics and don't think you would get *THAT* much more value from better metrics. But, this does give us an idea of how we can expect team's to preform in a given season with only metrics that we can actually measure. Here is an interactive visual I worked on for every team in FBS (it may not load, sorry).