What Football Can Teach Us
…About Leading Indicators
Our Superbowl LI — Leading Indicators challenge just got started. So far submissions have been a bit light, but we know many of you are deep in research. But what kind of leading indicator are you going to choose?
George Earl and Zack Vella have provided indicators based on thresholds created prior to the start of the post season — winning streak and bench depth respectively. Eric (via Facebook) chose a more event-based indicator — plays greater than 25 yards. Most of the submissions have been more inline with what we anticipated — correlated leading indicators. Each of these three categories has value and drawback. This article will take us a little deeper on each.
Thresholds
Having a number one defense, the top scoring offence, the longest winning streak, or the greatest bench strength is a ranked statistic. It is also a threshold statistic. Essentially a ranked and threshold statistics state that X is higher/highest so some outcome Y is more likely to occur. Ranked stats are compared to a pool of competitor (be they other teams, companies, or products) and threshold stats are compared to a benchmark.
These sorts of statistics are often easy to analyze, easy to find, and require strong judgment in their application. They are analyzed with historic data, typically used to rank order and back test. They can be found through any myriad of statistics. But are only typically successful when they hold a logical and valid connection to the outcome. And best of all, they really do lead — these sorts of stats are often available quite early.
Unfortunately — they lack subtlety,detail, and fidelity. They are discreet events after all. So you are either right or wrong and that is about the only thing you are likely to learn from your prediction.
Events
Turnovers, turnover ratios, 25+ yard plays, and even the coin flip (at least when the old OT rules applied) are all events that create higher probabilities of winning a football game. Similar events likely occur in almost any sort of performance you are tracking.
These events are also easy to analyze and easy to find. Most have a strong logical tie to predicting the outcome. Unfortunately, unless you are allowed to predict the outcome in the middle of the game (or business quarter, or campaign), you need to predict the likelihood of the event. On facebook, Eric projected that New England would have more 25+ yard plays because they have a strong history this season of creating them.
So if Julian Edelman and company can make a few of those plays a reality in Houston, Eric make prove right. If they don’t, he will have a strong hypothesis as to why they lost. But will he be right? What if they don’t but win anyway? Or do, but lose?
Correlations
The list of correlated stats is endless. These stats rely less on arbitrary thresholds, moving ranks, or interesting events. Their popularity seems to come and go — likely inline with their perceived effectiveness, but game strategy plays a role.
Chip Kelly temporarily destroyed the Time of Possession statistic, until it became obvious what it was doing to his defense… Other statistics have come and gone. The West Coast offense changed many of the stats that seemed to work best in the early years. But some teams still see success with through tenacious running games or deep passes…
Correlated statistics have an additional advantage of being continuous more easily measured. They provide better indicators for monitoring and nuance. But like events, this puts the forecaster in a sport where they must forecast their leading indicator prior to the outcome.
Mixing It Up
Football teaches us one final lesson on leading indicators — mix it up or keep it balanced. Teams that pass too much, run too much, blitz too much, or play zone the entire game are not likely to win. Play calling is balanced on offense and both sides try to be unpredictable (not our aim). Quarterbacks are asked to spread the ball around and complete passes at every level.
Forecasters should employ ALL types of Leading Indicators. The best business intelligence would include thresholds, ranks, event-based metrics, and correlated measures. Going a step further — they will also employ statistical models, weight averages, and anything else that provides perspective. Each different angle increase the ease and opportunity of learning. Of course, there is always a trade-off. Too much variety is going to confuse people and unlike most football games, that sort of ‘delay of game’ will be very difficult to overcome.
Thanks for reading! Drop your predictions in the comments below. Tweet them to #SBLIndicators and be sure to add @CorsairsIN. Bragging rights are on the line, but you only get them if you have the courage to post your prediction!
