The use of analytics in sports is quite well known. The Oakland A’s popularized analytics by using it to build winning teams with a minimal budget— we mentioned and extended this case in our book. Recently, ESPN Magazine ran a special “Analytics Issue.”
Late last year we were approached by Roger Goodell of the NFL to use analytics and Big Data to solve a vexing problem.
This project was bigger in scope than most sports analytics projects. As the head of the NFL explained:
“Many teams have used analytics to help them win more games. But, we wanted to use it across the league, to make every team win more games.”
In survey after survey, what the NFL fans wanted more than anything else was to see their team win. But, an analysis by the NFL, and later confirmed by our team, showed that the overall winning percentage was just barely 50%. Even worse, the fans wanted to see their team win in the playoffs. The results were similar in the playoffs and Super Bowl— we found the overall winning percent very close to 50%.
The goal of this analytics project was to use data and advanced algorithms to find ways to get the percentage up to 70% or 75%, if not higher. Or, as explained by the head of analytics at the NFL,
“50% is about the same as flipping a coin. What kind of business only delivers 50% customer satisfaction? We knew we could do better with all the data we are collecting.”
Since the NFL had collected a significant amount of data. Our job was to uncover new patterns in the data or identify key actions the teams could take– all with the goal that each team’s winning percentage increased to at least 70%.
We are just in the initial phases of the project, but under a lot of pressure. The NFL would like to implement for start of the 2014 season.
So far, we’ve built a model to validate our findings and have tested some initial assumptions. For example:
- We noticed that there was a strong correlation between scoring more points and winning. So, we applied a neural network algorithm to see if both teams in a game scored more points, would the winning percentage go up. Initial results were OK– we saw that the winning percent went up to a pinch over 50%.
- We noticed that teams that prepared the most (like Peyton Manning is known for) tended to win more. So, we tested a logistic regression model to see if both teams prepared more, could both teams walk away with a win. Again, initial results aren’t promising (we are still hovering around 50%), but we are confident that this will show something with some more research.
- We would like to take credit for it, but it was a bright analyst at the NFL that noticed a subtle correlation within each game. The analyst noticed that in a given game, if one team wins, there is a strong correlation showing that the other team is likely to lose. If we can break this correlation, we may be on to something. We are working right now on creating a supervised machine learning algorithm that will allow us to find the root cause of the correlation, and hopefully break it. If we can break the correlation, there is a better chance that both teams could walk away with a win. And, this would certainly help improve overall win percentages.
We are just getting started. Big Data and Analytics have made a lot of progress in many fields. We expect the same here. So, for next year’s Super Bowl, if things go right, both teams may win. And, for that you will be able to thank Big Data.