Not too long ago I wrote a thing to try to build a portal to the world of data nerds. One of the concepts contained in that article was that top-ten lists generally aren’t incredibly useful in declaring an NFL team “good” at something. The plea was to avoid that type of analysis. There’s a major issue with that request, though. As a football community, we really do need some shortcut language to make discussions easier. Naturally, I have a suggestion.
The Rule of Four
Before we begin this exercise in me being a know-it-all, I’d like to point out that there is no actual magic number. What I’m suggesting (and personally use already) isn’t perfect and I can bet it won’t take long for someone to demonstrate a stat the rule of four doesn’t capture cleanly. That’s okay though. Top [insert number here] lists are simply to start a conversation. It’s my assertion that top-ten lists start bad conversations and rule-of-four conversations should start better ones (at least when stats are part of the deal). What’s the “rule of four?” Time for a GIF!
The rule of four is based off of population distributions and personal experience with NFL data. About half the league is generally very tightly clustered around the perfect average. The eight teams on either side of the average make up the 16 “pretty average” group. The next few teams (I use four) on each side of that large cluster are starting to separate from the pack and can be called “bad” or “good.” The last four on each end have distinguished themselves and get the word “really” added. Let’s see how well it works in practice.
Points per game (PPG)
:format(webp):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18299991/Rule_of_four_ppg.jpg)
The last time I discussed this stat I highlighted the Atlanta Falcons as they’re the cut off from a traditional top-ten look. Here we have things color coded by the rule of four to see how it applies. As a reminder, perfect average for this metric was 23.4 ppg. Through sheer luck we get some fun coincidences using the rule of four.
All of the teams in black are within one field goal of the mean. So, if we’re considering this metric a measurement of how consistently a team is better or worse than average with scoring, the field-goal distinction is easily understood and meaningful. So far the new lens is working beautifully.
Working toward the left, Washington is 5.8 ppg less than average and begins our look at the really bad group. That’s pretty close to a touchdown (without XP) or two field goals. Conceptualized, that means Washington should be expected to lose by two field goals in a random game against a perfect average team. That’s not a bad cut off for a really bad team. It’s still technically a one-score game, but remember that our hypothetical team is perfectly ordinary. Sadly, our Buffalo Bills are in this grouping for 2018 at 6.6 ppg less than average.
Working in the other direction, the New England Patriots are only 3.9 ppg above average. I told you it wouldn’t take long for someone to find an example where the rule of four isn’t perfect. The top three on this end really illustrate where there’s separation.
Yards per game
:format(webp):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18299992/Rule_of_four_yards_per_game.jpg)
We’ll go to another common stat regarding offensive output and see if the rule of four works out. In the black range the bottom team is less than 25 yards away from the perfect average of 352.2 yards per game. At the top it’s less than 23. There’s less than half a field separating the top and bottom performers in our average group.
In the really bad group (which Buffalo is still in) we have 53 yards away from the mean. Could you make an argument that the bottom should be a bottom-five or even bottom-six list? Yeah, you could. Jacksonville is the first team that’s 50 yards away from the mean, which sounds meaningful.
The top end works out better with the fourth-best team being the first one to break the 50-yards-away-from-the-mean barrier. Feel free to dissect the numbers more, but a quick glance at this chart makes the rule of four look good.
Interception rate
:format(webp):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18299993/Rule_of_four_interception_percentage.jpg)
The quick glance doesn’t work so well here. Most of this is due to something you’re always gonna have to deal with from us chart people. The idea of “scale” is important. Because this chart only tracks from 0-5, a difference of 0.5 looks gigantic. On the last scale that shows from 0-450, the difference of 0.5 is minuscule.
Tangent aside, what we’re looking at is how often teams came away with an interception based on the number of passes they dealt with. I chose this stat because I like to end on a high note. I won’t continue boring everyone with numerical breakdowns. If you’d like to do it yourself, the mean for this metric is 2.24%.
Summary
I’m not vain enough to ask everyone to start using this as their new thing (though you’re welcome to). But since you’re stuck with me and it’s peak offseason time around here you may as well have a window into how I look at things.
Is the rule of four perfect? Not by any means. In ppg, New England makes a good case for top three being the most relevant separation. For yards per game, the Jaguars make a good case for a bottom six. Interception rate shows that a top-five list can be the best number to sit at. So why do I use four? Four is pretty safe and the idea is to have a rule of thumb. If we use four as that quick reference, then the anomalies noted above only necessitates us paring away a team or two and suddenly we have an accurate and statistically meaningful metric.
And even if we don’t pare down to increase accuracy beyond four, let’s look at the outcome. For ppg, we introduce the Patriots as a “really good” team that’s on the fringe of only being “good.” For yards per game, Washington and Jacksonville are “only” bad rather than really bad. Yeah, the rule of four didn’t get a ringer for those but to me it looks like a leaner.