For long-time followers of BGN, you may remember that I got my start here on this site with "Crunching the Numbers," a power ranking system of sorts that used measurable statistics to rank every team in the NFL. Over the offseason I've made some adjustments to my algorithm that will hopefully make it more reliable.
However, if you've forgotten about last season (I can't say I blame you) or you are new to the site, I'll go through an overview of what Crunching the Numbers actually is. It's kind of a grand experiment of mine, three seasons in the making, that all started when I was having trouble sleeping in Summer 2011.
This might be a strange question to pose to Eagles fans, but do you remember the 2009 Broncos? Josh McDaniels was a rookie head coach and Denver started out 6-0. All of the "pundits" were in love with them... until they completely collapsed in the second half of the season and finished 8-8. And while thinking over football a year and a half later, I asked myself, "What if there was something that could have predicted that? Like a series of small red flags that when alone were insignificant but together could bring a team crashing back to Earth?" And so I set out to find those red flags by compiling a formula of all the metrics that I considered important to success in football.
The idea, in a nutshell, was to rank all the teams in the NFL from best to worst while completely ignoring how many wins each team had. It would be based entirely on numbers - and yes, numbers never tell the whole story, but at the end of the day the winners are those who usually execute the best. And execution can absolutely be measured in numbers: yards per carry, sacks per game, passing yards per completion allowed... it's all execution, right there in black and white. Essentially, I aimed to find out which team was able to perform the best and limit mistakes on a consistent basis. By doing so, I could look at what was causing teams to win rather than just the fact that they were.
The results so far have been mixed, but mostly encouraging. I've been able to make some good predictions; in Week 4 of 2011 my rankings predicted both the Ravens and the 49ers would win their divisions (if you recall San Francisco was still unheralded at that point). I was also able to suggest in Week 4 of 2012 that the Eagles would crash and burn, even though they had started 3-1. I've also had my missteps; according to the rankings both the Eagles and Chargers were supposed to win their divisions in 2011 and the Bears were supposed to be the second seed last year. I've made adjustments to my system each season and this year is no different, so let's dive in.
The formula, for the most part, is pretty simple. I take a collection of statistics, make indices out of them, and add them together. You'll have to excuse me for the extremely generalized naming system; some positions are technically "left out" of the algorithm but their production is included. For example, I don't have an index for running backs, but I do include yards per carry in my system. I stay away from bulk numbers, like rush yards per game and passing yards per game, because those numbers can be misleading if a team has a severe weakness that is constantly exploited, making the other statistics look good (i.e., a team with a horrendous run defense would appear to have a good pass defense because nobody needs to throw to win). So here are the indices:
Offensive Line Index (OL): This is simply the sacks allowed per game subtracted from the yards per carry.
Quarterback Index (QB): This is the quarterback's completion percentage multiplied by his yards per completion.
Defensive Line Index (DL): This is the yards per carry allowed subtracted from the sacks forced per game.
Coverage Index (COV): This is the completion percentage allowed multiplied by the yards per completion allowed.
Adjusted Score Differential (ASD): This is your standard score differential (per game) with a penalty: your defensive scoring rank in the league is divided by four and subtracted from your differential. This is to account for teams who light up the scoreboard and get away with suspect defenses. Those teams usually do not fare well in January, and I'm trying to reflect that.
Miscellaneous Statistics: third-down efficiency for both offense and defense (OTDE, DTDE), expressed as a decimal; and turnover ratio per game (TR/G), which is tacked on at the end as a "bonus." I wanted to adopt the Kelly philosophy and adjust the turnover ratio for the points scored after forcing a turnover and the points allowed after committing one, but it those statistics are surprisingly absent from the internet (at least without paying for it).
When you throw it all together, it looks like this:
OL + QB + OTDE + DL - COV - DTDE + ASD + TR/G = Rank Index (RI)
I use addition and subtraction exclusively because multiplication could lead to sign inconsistencies. For example, if a team has a negative result for OL + QB, multiplying by the third down efficiency would essentially punish teams for having a good third down efficiency, since it would make the number more negative. Adding and subtracting might make the weighting awkward, but it's the cleanest way to do it.
The two biggest changes I've made was scaling down the emphasis on turnovers and reducing the penalty for a bad scoring defense. Yes, those are important, but I think they were skewing my results in a way that wasn't realistic. Last year, I simply used the overall turnover ratio, which ended up dominating the algorithm since all of my other statistics were on a per-game basis. Additionally, the team with the worst scoring defense was slapped with a sixteen-point penalty (I divided the scoring rank by two instead of by four), which brought most of the numbers into the negatives. This year, I'm trying to make the results look a little "nicer" while boosting the reliability of the algorithm.
I need a good sample size for the rankings to really mean anything, which is why I wait through Week 4 before I start publishing the results.
One last thing: keep in mind that these are index numbers, meaning that they are only useful when compared to each other. There is no "benchmark" or number that a team has to reach in order to be considered "good." The standard is set by the team that is ranked first, which is how the league actually works anyway.
I hope I've garnered some interest in Crunching the Numbers, and I look forward to hearing your comments and opinions as the season goes on. I've always considered the BGN community to be the best of the Eagles' fanbase, so I really take your guys' thoughts seriously.
If you have any specific questions (or would like to see the Excel spreadsheet I use), comment with your email address and I'll get back to you. You can also find me on Twitter (@Harks119). Also, almost all of the raw data I use can be found at Team Rankings (www.teamrankings.com).