speedy & elusive. credit: tim hawk/south jersey times

   Last Week 

Week 3

Veering course wildly from last week, the games this week exhibited a large dosage of blowouts, defined somewhat arbitrarily at HTO as games with victory margins of at least ±20 points. There were eight (50%) such games in week 3, contrasting with only a total of three during the first two weeks of the season. The blowout rate for all of 2012 in the NFL was 23%, or 59 out of 256 games, making 2013 now officially exactly on pace.

fingertip defense. credit: bob donnan/USA today

premature td celebration. credit: streeter lecka/Getty

Stats Viz 2013

The season is far enough along now to allow some visualizations. These results will still show some anomalies which will normalize over the course of 16 games, but the evolution of the data is still of interest. The graph below compares teams in terms of points allowed and achieved, but it also constructs a power ranking based upon offensive and defensive comparisons with the entire league. Brighter blue indicates the highest power ranking while brighter red indicates the lowest. As the season progresses, I will add at least one other factor into the power ranking, involving performance (winning percentage) in close games. Generally, there are not enough of these yet in a team’s record to reveal anything meaningful, but after more games this measure can give a good indication of competitive toughness and also luckiness. Number of wins are also visualized by the size of the point. Perpendicular distance from the diagonal green slope line tells how well or poorly a team is doing in terms of points scored versus points allowed.

The R code has some unwieldy annotation placement coding in it for positioning the team names. I need to automate this in some sort of function to save time (it will need to take into consideration things like proximity of other teams’ points as well as point size), and will include this improvement next time.

                                                                                                                                     view SQL and R code 
w h i t e s p a c e

nfl2013 week3 power

combined rankings for points scored and points allowed

Two factors currently figure into the power factor: offensive points ranking and defensive points ranking. I favor these measures over more detailed things like yardage gained and allowed, as they seem to me to have a more direct effect upon winning. Two more will be added in coming weeks as the sample data fills out: margin ranking (offensive points minus defensive points), and a ranking based upon winning percentage within close games, as already mentioned. I like this last item as a kind of measure of playoff toughness. There is also a potential for attaching weighting coefficients to each component ranking factor in the future.
w h i t e s p a c e

credit: andy lyons/getty photo

1) I use 2.37 as the exponent for computing Pythagorean Wins instead of the vanilla 2.00. The reason has to do with the fact that the usual exponent was originally developed, by Bill James, for major league baseball which has a sample size of 162 regular season games. In the NFL, there are only 16, so the hand of chance plays a more visible role and must be compensated for. This link will justify the math if you’re interested.

2) I count ties as half-wins (and also half-losses). The statistical reason is obvious. A 7-8-1 team, such as last year’s St. Louis Rams, has to be distinguishable from both an 8-8 team and a 7-9 team (for example, last year’s Dallas Cowboys and Carolina Panthers, respectively). Note that the average number of ties during an NFL season, league-wide, is in recent decades less than 1.

3) Pythagorean Wins, for those unfamiliar, is a measure of expected wins based upon comparing a team’s points scored (Offense) and points against (Defense) over the course of a season. Read more about it here.

4) The basic philosophy I am following is that professional football is a team sport, and therein all the drama lies; so for the present at least, I am eschewing individual player stats. Also, there is an effort to see how much can be revealed, in terms of data visualization, with the least raw data. More with less. The three main graphics, comparing team offense and team defense, actual wins and Pythagorean wins, and blowouts versus close game results — all can be driven off a very simple table of weekly results during the season. Perhaps other visualization ideas, employing other datasets, will occur to me later on as time permits, or I can consider requests.

5) Much of this stuff has been pioneered in non-football posts at Hearing the Oracle, especially the graphics for offense vs. defense and actual vs. Pythagorean wins. To further orient yourself, you can look at relevant posts about Oracle SQL, R, and the NFL 2012 season below:

         • GGPLOT graphics for the NFL 2012 Season

         • Introduction to using R with Oracle databases

         • Accessing the Original 2012 Spreadsheet

         • SQL queries and results for NFL 2012
w h i t e s p a c e

6) This Week’s Credits: Sometimes you just can’t choose, so why suffer? The four NFL Nerds Fotos of the Week are by (1): Tim Hawk for The South Jersey Times, and (2): Bob Donnan for USA Today Sports, and (3): Streeter Lecka and finally (4): Andy Lyons, of a soon-to-be Bengal touchdown score, selected because of it’s blend of great resolution, bright autumn color, and captured airborne derring-do.

~RS