Before the wonderfully weird 2020 NFL season kicked off, for the second straight year, I picked the winner and loser of every game (you can read my full article here). The logic for doing this was to get the best gauge possible for every team’s true strength of schedule based on how their strengths and weaknesses match up with the opposition opposed to strictly basing a team’s predicted record off of last year’s numbers. That said, if, after filing through every game, a team’s predicted record seemed off, I manufactured some upsets that I didn’t actually think had good odds of happening to compromise their game by game prediction with my general thoughts.
This all goes to say that my record for games I got right to games I got wrong is cool and all, but the ultimate goal of this exercise was to be as right about a team’s final record as possible. This is because, if I was only going for a good “grade” I would have the Jaguars go 0-16 but since teams hardly ever go winless, I had them upsetting the Steelers, placing them at 1-15. Calculating a team’s record differential provides a better assessment of my picks opposed to just how many games I got right. In order to thoroughly analyze my performance, I did three things.
Before things get too crazy, after every week of play I simply marked how many matches I got right and wrong. For example, if I got 10 games right and 6 wrong in a given week, I’d be 10-6 for that week. The one exception to this was the Bengals vs Eagles tie which was excluded from my cumulative record.
As previously mentioned, I wanted to see how close my predicted records were to a team’s real record. To do this I had to make up some stats: my predicted wins (pWin), a team’s real amount of wins (rWin), and win differential (wDiff). I used these values in the formula |pWins − rWins| = wDiff which gave me a value indicating how far off I was from a team’s real record. This was a confusing process so to simplify it, let’s use the Titans and Patriots as examples. I had the Titans projected record at 6-10 and the Pats at 7-9, making their pWins equal to 6 and 7 respectively. Then I used my formula in accordance with their real records of 11-5 and 7-9 (rWin = 11 and 7) to find my wDiff of 5 and 0. If that all didn’t make sense, all you need to know is a higher number, such as 5, is bad and a smaller number, such as 0, is good.
With all these made up stats swirling around, saying my record and my cumulative wDiff means nothing without context. As a comparison for my data, I repeated this entire process with the man I modeled this exercise after, Adam Rank of NFL Network. Rank does the same thing at the beginning of every season so I repeated steps one and two with his predictions.
Now that the process is laid out, let’s talk about the results. My right-wrong record stands at 161-94 and Rank’s at 138-117. I did the same exercise before the 2019 season and my predictions ended at 155-100, just short of Rank’s at 157-98 so, before even calculating the wDiff, I felt good about my 161-94 record for 2020. That said, I can say with certainty that my goal isn’t to get the best record, but best wDiff, and though I can’t say the same for Rank, after studying his work, I believe the same is true for him. Even still, I narrowly beat Rank at his own game, accumulating a wDiff of 81 compared to Rank’s score of 83.
I’m not saying I’m better at NFL predictions than Rank, he’s a great analyst and a headliner at NFL Network, but comparing this to his work, and having better stats than him, was truly stunning.