NFL Scoring is Up, An All-Time High, But is it a “Signal” Worth Explaining?
I do enjoy watching football, so this headline caught my eye:
If you know me and my book Measures of Success, you know my first question was:
“Is it a statistical signal worth explaining?”
From the article:
“N.F.L. teams have been scoring points at an unprecedented level in the season's opening four weeks, an augmented efficacy that if sustained will rewrite the league's 100-year-old record book.”
That's a long-winded and sophisticated way of saying “scoring is at an all-time high.”
Well, every data set (including business metrics) has a “highest number ever” but that doesn't mean it's a statistical signal worth explaining.
We need more data. We need more context, as Don Wheeler says.
The article continues:
“Through four weeks of the 2020 season, the average combined score of a game is 51.3 points, an increase of 16 percent over the same period a year ago…”
OK, those are facts. The math is right. Is 2020 a signal worth explaining?
Also, as the article does point out, we are only four games into the 2020 season, so it's not “apples to oranges” compared to other seasons (we aren't into the bad weather of early winter when scoring probably drops — I'd want to see the data that confirms that dispels that assumption I am making).
As always, we have to be careful with two-data-point comparisons. Scoring is up 16 percent from the year before. How much does scoring normally fluctuate from season to season? Show me the data! #PlotTheDots, as my NHS England friends say.
The Times adds another two-data-point comparison, saying:
“…roughly 20 percent increase in the average score of games since 2000.”
That's a fact, but is 2020 a signal??
The Times does show a chart (and it's, annoyingly, a column chart when I much prefer a run chart for time series data):
So is that last data point, 51.3 points (for both teams), a statistical signal? It's hard to eyeball it. There does appear to be an upward shift from scoring of the '70s and '80s.
I also went to pro-football-reference.com, but I started out simply creating a run chart… going back just as far as 1996 to start. 25 years, 25 data points.
Is my chart better than the column chart? I think so. It uses less ink, well that is, if we are still printing newspapers.
What's more helpful is a Process Behavior Chart… where we calculate an average (in green), and two other lines called the lower and upper limits (in red). Learn more about how to do this.
Here is what's called the “X Chart” (not to be confused with the XFL, but it sounds cool like that):
So 2020, so far, IS a signal. Often, when I do this analysis, the data point is within the limits, so I end up writing about how we shouldn't overreact to “noise” in a system or a metric.
There are three rules, per Wheeler, that I document in Measures of Success:
- Any data point outside the limits
- 8 or more consecutive data points on the same side of the average
- 3 of 3 or (3 of 4) consecutive data points that are closer to the limit than they are to the average
We can see that 2013 was also a signal (above the upper limit). I'll change the Y-axis to make some of this easier to see (and be careful to not make the Y-axis misleading in the workplace):
While 2013 was barely above the upper limit (creating the risk that it's a “false positive” signal, 2020 is a very strong signal.
Beyond that, the last three years are a “Rule 3” signal. The upward shift in scoring didn't really start this year.
Look at the totality of this chart… we have many years below the average followed by many years above the average (2017 scoring is 21.7 while the average is 21.71, so technically it's below average, but I think the trend is clear.
There's not a single “system” over time. It looks like two systems and maybe that new system starts in 2010. I can calculate averages and limits for two different time ranges… then the chart looks like this:
This chart shows how scoring had been fluctuating around an average… then it shifts upward… sort of fluctuating around an average and now 2020 is a signal (and these stats were not including the Monday Night Game).
Had I gone back further in time (maybe I will and I'll update the post), we'd likely see additional shifts in scoring — caused by rule changes or other factors.
There are many possible reasons for this “signal” — what we would call a “special cause.” As with any business metric, a signal in the Process Behavior Chart tells you the system changed… but it doesn't tell you WHAT changed. You have to figure out the cause-and-effect.
Possible causes of this signal include:
- No fans (or limited fans) reduces some of the home field advantage, so scoring is higher for road teams (we could look at the data on that)
- Lack of a pre-season means that defenses are at a disadvantage compared to offenses
And, from the Times:
“Theories on the cause of the scoring surge have proliferated, including the advantages afforded road teams since they are no longer tormented by raucous home crowds, a curious drop in the number of penalties called and a decline in tackling skills resulting from the cancellation of preseason games.”
I'd say this is all a bit premature, given that we are just four games into the season. A better comparison would be a chart that shows “points per game through the first four weeks of the season.”
Either way, the point for business metrics is the same… don't react to every up and down in a metric. Look for signals instead of reacting to all of the noise. Don't just compare two data points… instead look for the context that we get from additional data points.post a comment and join the discussion. Subscribe to get notified about posts via email daily or weekly.