Given the lull before the draft, I thought I’d put together another article. I hope that this will be the first of a series of articles looking at Madden Ratings. For some time now, I’ve been using a simple algorithm I developed to better understand the value of my roster as it pertained to Madden. I’d like to think I’m building a team based on players I would value in real life, and for the most part I am. At the end of the day, however, this team is tested on the Madden platform, for better or for worse. Therefore, it would be unwise to completely neglect Madden when making roster decisions. Rather, I feel the Madden Ratings should be one piece of the puzzle of 3GML roster construction, along with other pieces such as cap value, positional value, age, injury history, etc. You can’t justify keeping an overpaid, low Madden ranked player, just as real teams can’t keep an overpaid, underperforming player. In an ideal world, the Madden Rating would perfectly reflect this example. Unfortunately, Madden has tendencies we’ve all come to know, be it overvaluing “big name” players or undervaluing non-skill player positions.
In an attempt to make Madden Ratings better reflect real player value, I’ve been toying with a number of ideas. Early on in my 3GML career, I experimented with a composite value score based on Madden Rating and PFF grades, amongst other items, with the idea that I could “forecast” a player’s development as a positive or negative trend and be proactive in signing and cutting players (the Tre Boston signing was an early signing based on this idea). Unfortunately, just as I began to make progress on this concept, PFF changed their grading system to what is the statistical equivalent of crap and I scrapped the idea for lack of a better all-encompassing metric.
Instead, I decided to focus on another metric: salary.
The “Algorithm”
The idea was simple: if a player has a high Madden Rating and a low salary, they should in theory be more valuable to a team than a player that has a high Madden Rating and a high salary. Why? Well, because the high salary player could prevent you from signing another highly rated player, and thus reduce your overall team rating. I thought a player’s value could reflect this concept; indeed, we see evidence that real organizations recognize this concept in some fashion. We see it as front offices trading away players for picks before massive extensions. In theory, the draft picks allow a team to get players at lower salary (who could potentially become highly-rated players) as well as provide cap space to sign other high-impact players. Now, we can argue over the validity of this concept in real life, as many organizations value picks too highly and players too little. Further, the salary cap is basically a “myth” to real teams. But the inescapable fact is real teams have more cap flexibility that 3GML front offices. We can’t structure or restructure contracts how we want to achieve a favorable cap; we are dependent upon whatever happens in real life. Each front office has different methods for contract structure and cap distribution. Some prefer to front-load cap hits, others to push them off, and even that may change year-to-year depending on a team’s current salary gap. The end result is the same balancing act for 3GML managers as those GMs in real life, but only we are blindfolded and have one-hand tied behind our backs.
Thus, I felt I needed a better way to value players in Madden that reflected some aspect of the salary cap. Nothing fancy, just a simple formula to plug a players Madden Rating and salary cap hit into and return an adjusted value – a “Cap-Adjusted Madden Rating” if you will. After trial and much error, I came up with a simple logarithmic equation based on the ratio of a players cap hit to their Madden OVR rating. The equation is founded on the assumption of a “perfect player”: a player with the best possible Madden rating (99 OVR, although Madden once gave Barry Sanders a 100 rating) at the lowest possible salary (NFL minimum salary, $610,000 cap hit in 2020). Such a player would achieve a perfect “100” score using this equation. Why 100 and not 99? Just as you will never see a perfect 100 OVR in Madden, you will never see a player with a 99 OVR who is paid a NFL minimum salary.
The tendencies of the algorithm are as follows:
1) A player with a low cap hit and high Madden rating will be rated higher than their Madden rating
2) A player with a high cap hit and high Madden rating will be rated lower than their Madden rating
3) A player with a modest cap hit and an average Madden rating will be rated around their Madden rating.
Everything else falls in between these tendencies.
Examining the Cap-Adjusted Madden Rating (CAM Rating)
While I could see the CAM rating of my players, I didn’t have a contextual understanding of what the CAM ratings actually meant. Yes an expensive player at an average Madden rating is bad, but just how bad? I needed context, and that context could only come with more observations. Now, as much as I’d like to sit and go through every players Madden Rating and salary cap hit, I don’t have the time, and the EA and OTC / Spotrac websites aren’t set up in such a way that I could easily scrape that data with script. However, the 3GML spreadsheet solved part of the problem: it provided me with the salary cap hits for over 160 players. I just had to provide the Madden 20 ratings for those players, and I had a subset of observations wider than my own team.
Using the 3GML subset, I calculated CAM ratings for every player and plotted them against their original Madden 20 OVR (Figure 1). I then found a best fit function to describe the relationship (in this case, I found that a simple quadratic equation had the lowest residuals). The trend line (red) produced can be thought of as a “league average” CAM rating for a given Madden Rating. Because our distribution is fairly normal (skewness -0.4, kurtosis 3.8), we can assume the values within 1σ and 2σ are within ~ 68% and ~ 95% respectively of the 3GML league distribution. Anything beyond 2σ would thus be exceptional. If we ascribe value to these ranges, then we could assume a CAM rating above our trend line would be above average value, while a CAM rating below would be below average. The natural extension of this would be that anything beyond 1σ of our trend line is either very good or poor value, and beyond 2σ of our trend line would either be exceptionally good or exceptionally poor value.
Figure 1: Cap-Adjusted Madden rating plotted against EA Madden 20 player OVR ratings. A quadratic equation (best-fit) is plotted in red, showing in essence the “league average” CAM rating at a given Madden Rating.
With this quantitative relationship, we can now examine each 3GML team’s rosters, and better understand the strengths and weaknesses of the CAM rating. By splitting the CAM ratings out by team, we also have enough space to label each rating with their respective player. As you will see below, there is a clear pattern where most players above the trend are players still on rookie deals, while most players below the trend are on their 2nd or 3rd deals. This should come as no surprise – good players get paid (and often bad ones too). Because of this bias, I decided it was best to ascribe less harsh value to those players falling within 1σ below our trend line. The categories I came up with are:
CAM > +2σ → “Exceptional Value”
+1σ < Cam < +2 σ → “Excellent Value”
μ < CAM < +1σ → “Good Value”
CAM = μ → “Average Value”
μ > CAM > -1σ → “Fair Value”
-1σ > CAM > -2σ → “Poor Value”
CAM < -2σ → “Very Poor Value”
Keep these categories in mind as we discuss the CAM ratings for each team below. You will note that players on your roster without a 2020 cap hit (pending contracts, etc.) are not listed on the charts.
Green Bay Packers
Figure 2: CAM rating for 3GML Green Bay Packers.
Coming off of a Super Bowl victory, this roster was built to win now. So unsurprisingly the Packers have some big names with big contracts, which shows up here. Of all the 3GML teams, the Packers have the most players falling within the Fair or Good Value categories; only 9 of their currently rostered players fall outside of 1σ. Those that do are all big name players. The best value on the Packers Roster is clearly Juju Smith-Schuster. He boasts an 86 OVR in Madden, and an 87.2 CAM rating, due in large part to his late 2nd round rookie deal. On the flip side, his running mate Brandin Cooks is rated at 85 OVR, but caries a whopping $16.8m cap hit in 2020, resulting in a 73.4 CAM rating. The worst value is OT Jake Matthews, who has a 78 OVR in Madden and a 68.3 CAM. Take this with a grain of salt, however, as Jake Matthews is clearly better than his Madden rating, and has only a $10.7m cap hit this year. If this is the worst CAM rated value player on Green Bay’s roster, then the Pack is in a good place. Oh, and the Packers have the highest number of players (26) above league average CAM rating.
Houston Texans
Figure 3: CAM rating for 3GML Houston Texans.
The Houston Texans roster follows similar trends to the Packer roster. Over the last 2 seasons, the 3GML Texans GM has been very aggressive in Free Agency and in trades, bringing in a number of big name players on big contracts. However, what stands out the most about this roster isn’t those players, but rather the players drafted by the Texans. Genard Avery, Darius Slayton, Derrius Guice, Orlando Brown, Jr., and Lamar Jackson all fall within the Excellent Value category, each playing well above their draft value. This is particularly true with Jackson, who boasts the highest CAM rating on the Texans roster and at the most important position in football. On the other side, former draft pick LB Myles Jack sits as the player with the worst value, holding a $15.4m cap hit in 2020 and only a 79 OVR Madden Rating. This is in part due to a down year for Jack in 2019 (along with the entire Jaguars defense), but also in part due to a quirk of Madden. Because Jack is listed as a MLB, he immediately takes a 2 OVR rating decrease. If he is listed as an OLB, his rating increases to 81 OVR, and his CAM rating moves from “Very Poor” to “Poor” value. While still not ideal, a $15.4m cap hit is still very high for a position that is typically undervalued in both Madden and in real life. Luckily, the Texans GM has shown a propensity to make up for big spending through excellent drafting, as shown here.
San Francisco 49ers
Figure 5: CAM rating for 3GML San Francisco 49ers.
The 49ers are perhaps the team with the least consistency of all of the 3GML rosters. On one hand, they boast the highest number of players in the Excellent to Exceptional Value range, with 9 players total. On the other hand, the have the most players in the Very Poor Value range, with 3 players total. It should come as no surprise that 2 of the 3 players are QBs. Teddy Bridgewater signed a 3 year, $63m dollar deal to be the starter of the 49ers for the foreseeable future. With only 75 OVR Madden Rating, his $14m cap hit brings his CAM rating down to 63.1. The Niners also chose to resign Marcus Mariota on an incentive-laden contract, with a $7.5m cap hit that could escalate to over double that value if he wrests control of the starting position. His high cap hit at only 68 OVR brings his CAM rating down to 57.3. It seems the shadow of Andrew Luck’s surprise retirement is looming large over the 49ers’ roster. On a high note, the 49ers can rely on a strong running game with Alvin Kamara and Phillip Lindsay holding team-high CAM ratings at 90 and 89.2 respectively. Lindsay’s CAM rating at only 85 Madden OVR puts him in the Exceptional Value range.
Conclusions
The Cap-Adjusted Madden Rating highlights some interesting patterns for each team, reflective of the strategy each GM takes with respect to roster construction. While a good initial foray, the CAM rating shows an inherent bias towards players on rookie contracts that must be addressed in future iterations. As well, the CAM rating reflects Madden biases against non-skill position players, which will not be easy to address. Also, the current CAM version, while sensitive to cap hits, does not take into account the overall salary cap. I will work on accounting for the salary cap in the next version of the CAM rating, and hope to share that work in the next installment of this series.