Bad football analysis

i just read an article on ESPN, which I think is a perfect example of bad football analysis.
http://www.espn.com/nfl/story/_/id/18008483/how-jay-ajayi-adding-value-via-expected-points-added
it happens to be about the Phins, but for what I want to discuss it could be about any team

Looking at expected points is a fundamental misunderstanding of football. in football it is the team that has scored the most points after 60 minutes that is the winner, Time efficiency is a very important part of football, and you can easily see it in common terminology like 'running out the clock', 'garbage time' and 'two minute offense'. in a sport like baseball where time is not a constraint and teams simply have to have the highest score after a set amount of resources this type of analysis is perfect. We see in advanced basketball and soccer that stats like points per minute/shots per minute. are more useful.
I am in no way saying EPA is useless. Simply that you shouldn't use it if there is a better tool available. In football we can calculate win probability added (BTW thanks to cbrad for showing me the usefulness of this as a predictive measure in one of our off season threads).

This might be true as a statistical analysis, but it is fundamentally wrong when applied to football. Football is where your opponent adjustments matter. Football is a game and game theory analysis should be applied. I posted this in the off season, but just to recap:
Firstly comparing yards/carry, passer rating. And rush/pass split

NFL AVERAGE: 4.1 ypc; 88.4;
40.9/59.1%

1st and 10: 4.2 ypc; 85.6
50.2/49.8%
1st and 10+: 4.4 ypc; 84.7
36.2/63.8%

2nd and 1-4: 3.5 ypc; 112.3
61.6/38.4%
2nd and 10+: 4.4 ypc ; 89.4
33.3/66.7%

3rd and 1-4: 3.5ypc; 98.0
43.3/56.7%
3rd and 10+: 5.4yps; 77.4
14.4/85.6%

In situations where a rush is expected ypc goes down and passer rating goes up and vice versa. The only exceptions is 2nd and long.

If we look at splits by snap type.
Shotgun: 4.9 ypc; 85.8
23.6/76.4%
Under Center: 3.7 ypc, 95.5

Again where the defense predicts run there is lower ypc and higher passer rating and in passing situation there is a higher ypc and a lower rushing average

Looking at splits by game win probability
0-19%: 4.0 ypc; 65.1 (4.1% interception rate)
27.7/72.3%
20-39%: 4.0 ypc; 76.4 (2.8% interception rate)
39.0/61.0%
40-59%: 4.0 ypc; 87.1 (1.7% interception rate)
40.7/59.3%
60-79%; 4.3 ypc; 101.2 (1.5% interception rate)
42.6%/57.4%
80-99%: 4.2 ypc; 120.0 (1.0% interception rate)
53.8/46.2%

In this case rushing efficiency doesn’t change significantly. But as passing becomes more predictable defenses are able to shoot down opposing passers.

Looking at splits by score differential
Leading 3.9ypc; 92.1
50.3/49.7%
Tied 4.1ypc; 90.0
43.2/56.8%
Trailing 4.3ypc/84.7
32.9/67.1%
Again we see the same pattern, as passes become more predictable, defenses get better at defending passes.

If the person who had written the article had bothered to think of how defenses react for more than 1 or 2 nanoseconds they could have done this type of research to see that rushing does perform a very useful role in making the passing game more efficient. Essentially a defense has resources it needs to split into defending the pass and defending the run. the more they put into defending the run the easier passing gets, the more they put into passing the harder passing gets.

Another reason that offenses run is to take time off the clock. Win probability added will measure the value of that time, but EPA won't.

Finally there are additional benefits to the team for rushing that are not measured by EPA, which we have seen in Miami this year. (1) It gives our defense time to rest and be effective [Seattle game for example where our tired defense allowed the late comeback] and (2) Conversely it tires out the opposing defense [Ajayis rushing against the Jests in the 4th quarter]. I'm not sure exactly how to do the analysis, but 4th quarter performance when there is a discrepancy in TOP would be a place to start.

Ajayi has 40.0 (#1) EPA against Ezekial elliot's 35.6 (#2)

Which comes from Ajayi converting a higher percentage of his runs into first downs, or second and short. This part I have no particular problem with, but if win probability was used Elliot would get more of a benefit because the Cowboys have been using him to close out games. Whether it would take him above Ajayi is an open question. But it is incredibly impressive for Ajayi to do add similar value in essentially 3 games what Elliot has added over 8 games.

 

Can't disagree with any of that. Looking at their article maybe their definition of expected points added has a different meaning to what the normal reading of those words would indicate.
My main points of disagreement with a lot of football analysis I see are:
1) They are more concerned about yardage efficiency than time efficiency. Since WPA is available it should be used more.
2) Opponents adjusting their strategy according to your strategy is not taken into account.

I am not saying things like EPA or yardage efficiency lack usefulness. They are very valuable tools, but that their usefulness is diminished greatly if time and opponent's strategy are not included.

 

I suppose if I were to use an analogy.
Baseball and Basketball are more like a casino game like Baccarat. Because of the long seasons you are essentially playing against the environment in the regular season. [NB post season is different]. Knowing the odds allows you to adjust your strategy based on known information.

Football is more like poker. Knowing the odds is very useful in becoming a good player, but it is reading and adjusting to your opponents that is the difference maker.