Analytics

       Brady, Smith Generate Highest Percentage Of Wins For Their NFL Teams   BY ADAM GROSSMAN & ROSS CHUMSKY     

 
   
     
       
         
            
            
         
       
      
       
         
            
            
         
       
      
       
         
            
            
         
       
      
       
         
            
            
         
       
     
   
      Update – The trade reported yesterday of Kansas City Chiefs QB Alex Smith to the Washington Redskins is a good example of how to use our new winning metric. As part of this deal, it is likely that the Redskins will not re-sign QB Kirk Cousins, trade cornerback Kendall Fuller to the Chiefs, and trade a third round draft pick in the 2018 draft to the Chiefs. The Redskins will receive a 2019 third round compensatory pick when another team signs Cousins. Our analysis of the 2017 season shows that Smith generated a 1.67 wins while Cousins and Fuller combined generated 1.14 wins. This demonstrates that Redskins will receive more wins from these transactions if Alex Smith performs at or near his 2017 performance while Cousins remains at or near his 2017 performance. For the Chiefs, providing the opportunity for Patrick Mahomes to become its starting QB at a lower salary may mean fewer wins in the short-term but at a lower cost than keeping Smith  -----------------------------------------------------------------------------------------------------------------------------  What NFL player has the most impact on his team? It is common question, and unlike baseball and basketball, football does not have a consensus, baseline advanced analytic metric to rate players. More specifically, Wins Above Replacement (WAR) is typically used in baseball and Player Efficiency Rating (PER) is typically used in basketball. The idea behind both metrics is to quantify an individual player’s impact on winning above a minimal level of performance. With WAR, for example, a player’s overall contribution is essentially based on how many runs he created and how many runs he prevented as compared to a Triple-A player in the same position.  The reason that it has been more difficult to create this type of metric for football is that the sport relies on the performance of multiple team members on every play. This complexity makes it difficult to identify an individual player’s contribution. For example, the typical successful passing play requires an offensive line to protect a quarterback from a sack, a quarterback to throw the ball, and a wide receiver to catch the pass. Who deserves the most credit if the play is executed successfully?  In the process of creating the  Revenue Above Replacement (RAR ) for the NFL for this season, our team at Block Six Analytics (B6A) identified the need to create a version of this on-field metric. RAR examines how an individual player generates revenue for a team based on his on field, off field, and personal performance. In the past, we relied on WAR or PER to create our models for the MLB and NBA, respectively.  To solve this problem for football, we used multiple data sets and multi-factor regression analysis. We were able to determine two main attributes that accounted for wins. In essence, these factors are how well (Grade) and how much (Snaps) did a player perform during the 2017 season.     

  

    
       
      
         
          
             
                  
             
          

          

         
      
       
    

  


     We found that these two factors identified how many yards over a replacement level player each player contributed to his team (the equivalent of how many runs were created in baseball). We found that summing each individual player’s yards contribution for a team in conjunction with regressing a team’s total wins provided a strong and statistically significant descriptive “winning” statistic.      

  

    
       
      
         
          
             
                  
             
          

          

         
      
       
    

  


      We were also able to attribute yards to every single football player in the NFL this season.  The top-15 players in our analysis are in Table Two. The first area of focus is PlayerWin%. This describes how much that player contributes to a team’s win over a replacement level player (in football think an undrafted free agent) with a 100% meaning the player accounts for all of a team’s wins. Unsurprisingly, Tom Brady leads the NFL in PlayerWin% with the Patriots quarterback accounting for 18.04% of team’s wins. What else is likely not surprising is that quarterbacks makeup all of the top-10 and 11 of the top-15 players in this rating.     

  

    
       
      
         
          
             
                  
             
          

          

         
      
       
    

  


     What is likely more surprising is that Kansas City Chiefs quarterback Alex Smith had the second highest PlayerWin%. How valuable is he? Let’s examine the fbWAR, ExpectedWin% and WinDelta columns. fbWAR calculates the wins a player adds over a sixteen game schedule (fb stands for football in our metric). ExpectedWin% looks at what percent of a team’s wins a player should contribute on an average team. WinDelta looks as the differences between the PlayerWin% and the ExpectedWin%. A positive number means the player is more valuable to that team than to the average team. A negative value means a player is less valuable to the team than the average team.  Not only did Smith have a Pro Bowl year, he is particularly valuable to the Chiefs because he had the highest WinDelta of any player in the analysis. His PlayerWin% is so high because the Chiefs as a whole have fewer players that contribute wins above a replacement level player. In our model, the Chiefs were expected to win 7 games (as seen in Table One) given the performance of the players on its roster (meaning the team won more game that would be expected given the talent on its roster). As a contrast, Brady has a  negative  WinDelta meaning he would be even  more valuable  on a different team than he was to the Patriots.  The three Panthers players followed the same pattern. These players are more valuable to the Panthers than to other teams because the Panthers had an even lower win contribution from its players than the Chiefs. The Panthers were also expected to win 7 games given the performance of its players (and actually won 11).     

  

    
       
      
         
          
             
                  
             
          

          

         
      
       
    

  


     What if you wanted to see a player’s contribution overall (not measured to a specific team)? Table Three sorts players by fbWar instead of PlayerWin%. Once again, Brady leads the way in this metric meaning he generated the most wins and the highest contribution of his teams win total. However, Smith drops to 9th in this analysis. Two other players of note are Jimmy Garopplo and Alvin Kamara. Garoppolo ranked 13th overall even though he only played in five games. Had he performed in this same way over a 16 game season then he would have surpassed Brady on this list. Kamara also had relatively few plays this year as compared to other skilled position players (running backs, wide receivers, and tight ends) yet still finished as the highest non-QB on this list.  Why have we not called Brady the Most Valuable Player? We mentioned that we originally completed this analysis as part of our RAR model. Our approach looks at players’ contributions on and off field. Our next blog post will have the completed RAR analysis where we can determine the overall value of each player. In terms of wins generated, however, Brady is the highest performing player. 
BY ADAM GROSSMAN & ROSS CHUMSKY

Brady, Smith Generate Highest Percentage Of Wins For Their NFL Teams

In the process of creating the Revenue Above Replacement (RAR) for the NFL for this season, our team at Block Six Analytics (B6A) identified the need to create a version of this on-field metric. RAR examines how an individual player generates revenue for a team based on his on field, off field, and personal performance. In the past, we relied on WAR or PER to create our models for the MLB and NBA, respectively.

To solve this problem for football, we used multiple data sets and multi-factor regression analysis. We were able to determine two main attributes that accounted for wins. In essence, these factors are how well (Grade) and how much (Snaps) did a player perform during the 2017 season.

       Interview With NFL Network Analyst and Game Theory and Money Podcast Co-Host Cynthia Frelund   BY ADAM GROSSMAN     

 
   
     
       
         
            
            
         
       
      
       
         
            
            
         
       
      
       
         
            
            
         
       
      
       
         
            
            
         
       
     
   
       Cynthia Frelund joined NFL Network in 2016 as an analytics expert, providing her unique insight into the game on NFL Fantasy LIVE and GameDay Morning. She is also the co-host of the Game Theory and Money podcast. Prior to joining the NFL, Cynthia worked at ESPN and Disney. She graduated with a Masters of Business Administration and Masters in Predicative Analytics from Northwestern University.      This interview has been edited for clarity and length.     Why do you want to pursue a career at the intersection of sports and analytics?   I am huge sports fan, but my path into the NFL was through Anthony Noto who was the CFO of the NFL (now the COO at Twitter). Before he was at the NFL, I used to read his Goldman Sachs equity research reports and I loved the way he structured his thoughts, and he was smart and it was interesting to listen to what he had to say. Given my background in finance and strategy and my passion for sports, working at the NFL was a great opportunity.    Why the NFL?   The impact on analytics in baseball is pretty widely accepted. Use of analytics in the NFL is far more nascent. We are kind of geeks in a corner in the NFL. I own that. I like that I could ask larger strategic questions and have the opportunity to be impactful in the space. It is exciting to be a pioneer in a new market.  I saw that I really had an opportunity to focus on bridging the sports performance and business sides in the NFL. The NFL has unique performance valuations. Unlike the MLB or the NBA, the NFL has a hard salary cap. Looking at the on-field product and what coaches have to do to win a game is also very intriguing.   How do fans interact with what you are doing?   I am really lucky in that the work I did at ESPN prior to the NFL and at the NFL I have great support from my executives. I have really good producers that are able to take complex concepts and distill them into good stories. No one needs to read or listen to a PhD dissertation on a pregame show.  Yet, people want to know the significant information. What I provide is the opposite of the hot take. It is a logical argument. There is no Cynthia-bias in it. They are getting their own [Cleveland Browns Chief Strategy Officer] Paul DePodesta. I have seen fans go deeper into the math throughout the years. It is awesome, and my favorite part of the job.   What technology do you use?   I code in everything from Python to R to relational databases. I am pretty good at using new tools to solve problems and answer questions. I also use Open-Source video tools (most familiar with tensorflow). Computer vision models are also great.  In particular, it is cool to use deep learning and video to look at shapes and angles of players in video. It allows me to create the “waist bender” metric for offensive line play. Offensive linemen that can keep hips parallel and stay low are better at protecting quarterbacks because they do not lose leverage on their defensive counterparts. I was able to map NFL performance back to the NFL combine 40-yard dash results to find elite, above average, average, below average, and well below average waist benders. The lowest waist benders are the best at protecting quarterback. If you can keep your center of gravity low for the first ten yards during the 40-yard dash that is good (or a low waist bender) and if not it is bad (a high waist bender). You can see it as they run the 40 in the video and map the results using video. It is almost a continuous variable which shows that the lower you can stay, the better you will be at protection.   How important is communication in what you do?   I would not be able to get my job done without my knowledge of analytics, but I think it is like 60/40 communication. You would like to think that you can just be amazing at this, but it’s not enough. You must be able to tell the story of your findings, but you really need to be able to communicate in order to figure out what answers someone (a coach, etc) wants to explore. Choosing the right questions to explore is the key and you have to ask the people who would need to execute these findings what they care about and why.     Do you play your own fantasy team? How does it do?   I have a few fantasy teams. The one on NFL.com is difficult to pay attention to because I have to be on-air right before players lock and games begin. It is more important for other people to get the information than to use the information for my own team. In my one team that is very competitive, I have a bot to sub players in and out because my show is on right before the games start.   What the biggest misconception about what you do?   The biggest misconception is that analytics is like a bunch of listed statistics, and I can read them and they are all equally important or not important. The point is to put situations in context using data and take bias out of it.   Where is analytics in the NFL moving to in the future?   Safety is a big deal. Between the sports science technology with biometrics and training metrics there is a lot going on with each team. There is a lot of cool data showing where we can find situations where safety can be improved. For example, rule changes and information gathered and distributed about when and where injuries occur. It is really cool to see how data has and continues to help.   
BY ADAM GROSSMAN

Interview With NFL Network Analyst and Game Theory and Money Podcast Co-Host Cynthia Frelund

Cynthia Frelund joined NFL Network in 2016 as an analytics expert, providing her unique insight into the game on NFL Fantasy LIVE and GameDay Morning. She is also the co-host of the Game Theory and Money podcast. Prior to joining the NFL, Cynthia worked at ESPN and Disney. She graduated with a Masters of Business Administration and Masters in Predicative Analytics from Northwestern University.