By John F Murray, PhD
I would like to share some exciting news. I am going to keep it simple and concise, but I think you will realize that this is very powerful.
As many of you know, I wrote a book after developing a new way to analyze football performance that included for mental performance. Both the new statistic and the book were titled “The Mental Performance Index” and the study discussed in the book on all the Super Bowl games showed that this index correlated with winning in the Super Bowl more than any other traditional statistic. It worked because the MPI captures more of reality than more dry statistics that do not include the observable mental aspects of performance and it also works because it measures every moment. It is both a mental measure and a measure of consistency over every play.
On the one hand this was very thrilling and I secured a 4-time Super Bowl winning coach, Tom Flores, to write the forward, and America’s most beloved and successful female sports broadcaster, Lesley Visser, who wrote the epilogue. Many NFL people provided supportive quotes including Don Shula and the late Steve Sabol of NFL Films, but my purpose here is not to promote this book, but to share something much more exciting and new. I aim to just further promote the vitality of mental performance and the need for mental coaching. I think my study has accomplished this. Read on.
I developed the MPI to help football teams by describing overall performance more accurately, because it includes a vital mental component usually ignored, but I never intended to use it to predict future games. Sure, I went on national radio and on television to talk about the Super Bowl over an 8 year stretch to give my fun pick based on MPI data, and I was right against the spread 6 of 8 years, but that was the fun angle and never the point of the MPI. It was intriguing success to get 6 of 8 correct, but an entirely too small sample size of only 8 games does not allow the serious scientist to get too excited about the predictive qualities of the MPI. I needed to do more to wake up the world.
Enter the year 2013. At this point, I realized that many in the sports world, and particularly in football, were still slow in grasping the importance of mental performance and mental coaching, so I endeavored to do something new to help illuminate the importance of mental performance and to also determine on my own how well the MPI could actually predict. I wasn’t sure, but if I could show that the MPI could reliably predict future games, it would add firepower to the notion that since I was measuring an important but often ignored part of the game – the mental game – I would also be able to predict better than most because I was using a tool that others did not have, and a tool that was capturing rich data that was often ignored.
Sure, I had already shown that the mental measure I created correlated best in winning the Super Bowl, but taking it to the level of game prediction was an entirely different animal. I was stuck on game description, but not future game prediction. I did not quit my day job as I have a duty to still see clients out of my office, on the phone, and at client sites, but this side project became a huge passion too, and I am happy to say that I have some very interesting results after having studied the MPI to predict games over an 8 year period of time from 2007 to 2014.
It would be entirely too complicated to discuss in this brief article how I took the MPI and turned it into a prediction machine. It was a great challenge and I tackled it with passion and purpose, starting with the raw data that the MPI produced and tweaking it relentlessly (based on numerous mini studies) for a variety of factors such as home field advantage, the established line on the game, the strength of schedule, and many other factors, but the essence was still a measure that included observable mental performance using the MPI that I discussed in my book.
I even hired professional statisticians to check my work and make sure I was doing everything properly. I have a background in statistics, having taught it at both the undergraduate and graduate levels, but I needed to pay someone to check on my work, and I wanted someone who does this work full time.
In developing my study, I borrowed from a format that the world is very familiar with in the Westgate Super Contest, the largest handicapping contest on the planet. Contestants picks 5 games each week and make their picks against a contest line. So each season contestants picks 85 total games and the most recent contest had over 1700 entries. It is exploding in popularity. The player with the highest win percentage (represented as total points) is the winner. I used their method of selecting 5 games each week, but I did it over an 8 year period of time, and methodically applied the system I developed to select 5 games each week in a totally systematic/objective manner.
The study actually included 4 different composite variations of a multiple regression approach, but the purest multiple regression approach was the clear winner, and boy did it win. The total sample size was very large as there are over 2000 NFL games to choose from over an 8 year stretch, but picking 85 games each year narrows that down to 680 game picks. Since in weeks 1 and 2 there is not enough data, I began each year at week 3, leading to a total of 600 picks. I did not count pushes (ties) in my analysis. If there was a push, I treated it as if it did not exist. In the contest, pushes count as half a point, but I did not give myself that luxury, so my findings conservatively underestimate my true success.
I am only going to share the findings from the most successful approach, the simple linear regression approach that fit the data best. In sum, I used my regression formula to select 5 games each week over an 8 year period of NFL games, and I used this formula in conjunction with the MPI that had been tweaked multiple times into a complex algorithm. The end result was that by using this formula I was able to first identify the 5 best games from which to make my picks, and then the algorithm which had produced an MPI line on the game was used to select a team either above or below the established line to make the actual picks. It was either a win or loss, or it didn’t count as a push. Keeping it simple, I ended up with 5 picks each week from week 3 to week 17 in each of 8 years.
If what I had created was meaningless, we would expect to find close to 50% success rate in an ATS (against the spread) format. The established contest line (or Vegas line) does a very good job of making it virtually a coin flip, so not matter what team you pick, the inexperienced or unsophisticated person making a pick will get closer and closer to 50% over time and since we are starting with close to 2000 games, the statistical power is such that any deviation above a 50% success rate would be interesting. A baby or person with an IQ of 75 making selections would be close to 50%. Professional handicappers who do this regularly and have records on them over an 8 year period of time usually get it right 50, 51 or 52% of the time. Very good ones are at 53% or rarely 54%, and the very best in history are still usually below 57 or 58% over hundreds of games of selections. It is one of the hardest things in life to do to win in an ATS format.
What kind of results did the MPI get? I am thrilled to report that it hit the ball out of the park! Below are the actual records for each of the 8 years of using this system to make picks in this study:
2007: 48 wins, 27 losses (64%) 2008: 44 wins, 31 losses (59%) 2009: 37 wins, 38 losses (49%) 2010: 45 wins, 30 losses (60%) 2011: 46 wins, 29 losses (61%) 2012: 44 wins, 31 losses (59%) 2013: 39 wins, 36 losses (52%) 2014: 44 wins, 31 losses (59%) ________________________________
Overall Average Success Rate Over 8 Years = 58%