Every year during April, right around the NFL Draft, you hear a lot of chatter in the scouting community about finding the “sleepers”, or the “hidden gems” in the draft. Finding players that can over-perform their perceived draft stock is one of the biggest goals for every GM in the league. But just which players are the biggest steals in NFL history? Which players have over-performed their draft stock the most?
Let’s put this question into a more concrete form. We are looking for the players that have enjoyed a career much more successful than other players drafted in the same draft position. For example — George Kittle got drafted in the 5th round with the 146th pick in the 2017 NFL draft but is now a 2x All-Pro and one of the best Tight Ends in the league. That makes him much more successful than others drafted with the same 146th pick in other years. Other steals in recent years that come to mind include Richard Sherman(154th pick), Antonio Brown(195th pick), and Geno Atkins(120th pick).
It is hard to quantify success in the NFL (and in all sports for that matter) but I decided to use ProFootballReference’s Weighted Career Approximate Value (CAV) as a gauge for success. CAV is a weighted summation of a player’s every seasonal value and ranges from 0 (the vast majority of players) to 177(Peyton Manning). It is far from the perfect metric to measure a player’s career success but it is sufficient in a broad-case usage such as ours. I gathered the CAV of every single drafted player since 1970 (NFL merger). The number of drafted players differs each year as the older drafts had more rounds (in 1970 there were 17 rounds) but modern drafts have around 255 picks so I only worked with the first 255 drafted players in each draft.
Here is every single player’s CAV mapped out. As you can see, most do not make any impact in the league and crowds towards CAV = 0. Spoilers alert who do you think can possibly be that little lonely dot on the top right is? Anyway, our goal is to find which players are the biggest outliers in this data. Which players have accumulated significantly more CAV than the average player in their draft position? To find how successful an average player is in each draft pick, I got the averaged CAV of all players in each draft position. We can call this averaged value the expected CAV.
No surprises here. The higher draft positions tend to have better careers and thus higher expected CAV. So a player picked with the 1st overall pick can be expected to have around 65 CAV while 255th, the last pick can be expected a mere 5.8 CAV.
From here, we can compare the expected CAV of each pick to the CAV of every player drafted in that pick. By subtracting each player’s own CAV with their draft position's expected CAV, we get the player’s CAV over expected. This value tells us just how much a player played better than they were expected.
Now we can finally find the outliers. There are many ways to do this but I defined the outliers as those players with CAV over expected that was higher than the expected CAV by 4 Standard Deviation. Using this definition, we were left with 90 players.
These 90 players in orange represent the best value picks in NFL history. Some GOAT-level players such as Peyton Manning (the dot on the top left) did not make it despite having the highest CAV of all players at 177 because he was drafted first overall and first overall picks had an expected CAV of 65. Meanwhile, Marques Colston made it with 73 CAV because he was drafted with the 252nd pick, which had expected 5.7 CAV. Outperforming his draft position average by over tenfold.
So who are these players? If we simply look at the highest value over expected, the winner is (no surprises) Tom Brady with 165 CAV over expected. Brady had 176 CAV and the expected CAV for his draft position (199th) was 10.7. It is important to note that the expected CAV of the 199th pick was heavily biased(much higher than other picks nearby) due to Brady’s own high CAV.
Here are the top 5 players with the highest CAV over expected in NFL history
If we simply look at the CAV over expected, we will just get some of the best players of all time. Since even the highest expected CAV is in the 50s, Hall of Fame level players who racked up over 100 CAV will not be terribly impacted by this. CAV over expected over-values earlier picks. That’s why there are so many late first-rounders here, the expected CAV drops significantly after the first dozen picks, yet there are many HOF players drafted in the late first. If you are satisfied with this, you can stop reading the article now. But I personally was not satisfied with this. I wanted to find the true underdogs that got drafted late and had a very successful career. But how could we find these players?
There’s also another problem. When you look at this graph, do you see how some of the orange dots, the outliers, are lower than some blue dots that are right beside each other? That means there are players that got drafted near these outliers AND had better careers than them, yet did not get chosen as an outlier.
This is because we compared each player’s CAV with their draft position’s expected CAV. And since we had a sample of only 50 players for each draft position (1970–2020, one pick per year at each draft position), it meant that if there were even a few players that had great careers in the same draft position, it would bring up the expected CAV for that draft position and thus the threshold for the outlier for that draft position was raised high. Or put simply, the expected CAV was biased. Take pick 207 (late 6th round) for example. Coincidentally, there were multiple players drafted here that had great success in the NFL. The pick 207 club has 8 players with over 30 CAV, including 4x All-Pro, Superbowl champion Jessie Armstead (88 CAV) as well as 3x Pro Bowler, SuperBowl champion Antoine Bethea. You can actually see Armstead pretty clear in the graph as a lone blue dot in the sea of orange.
He is surrounded by players that have been selected as an outlier, but because of his own success along with few others, pick 207 has an expected CAV of 10.54. Much higher than other picks near 207. Meanwhile, the draft position 100 picks higher, pick 107, has an expected career value of 8.04.
So in order to fix this, we have to reduce the bias for expected CAV. To do this, I first removed the outliers and recalculated the expected CAV to reduce the outlier’s bias. But I also recalculated this expected CAV using an average of the previous, current and next pick’s expected CAV so that there would be less variation. So for example, the expected CAV for the 31st pick would actually be the average of the 30th, 31st, 32nd picks expected CAV. By doing this, the bias was significantly reduced. As shown in the graph below, the averaged expected CAV has much less variance between each pick, making it much more desirable.
*note: I realize I could just regress the expected CAV to make a smooth curve and kill any bias but there were some inherently interesting trends hiding within the data like the last pick of the rounds performing better than expected and I wanted to preserve that.
Using this improved expected CAV, I recalculated the outliers and this was the result. Notice how there are MUCH fewer blue dots that are in the sea of orange dots.
It’s still not perfect but it’s much better. Now let's take a look at the new outliers' top 5 highest CAV over expected.
You can see that the values for expected CAV and CAV over expected has changed. However, I found this answer to be boring. I came into this question wanting to find the true underdog stories. The guys that got drafted in the 5–7 round yet still went on to have a successful career. Guys like Jessie Armstead. They were marked as an outlier but how do we find them?
We could do this by finding the z-score. Z-score is just a metric that tells you how far away the value is from the mean in terms of standard deviation. In our case, it will tell us how much better our player was from the average player in his draft position. We pretty much used this to find our outliers. Now let’s use this to find the biggest sleepers out of these outliers. The z-score heavily punished players with higher expected CAV, thus putting more value into late-round players. Here were the top 5 players with the highest z-score.
*note that this z score was using averaged mean and SD previously mentioned
Now that’s more fun. Todd McClure was drafted in the 7th round and went on to start for the Falcons for the next 14 years. He went 9 years without missing a game. Is he the greatest underdog story of all time? Probably not but his draft position of 237 had an expected CAV of 2. He was the only player picked at 237 to have gathered more than 25 CAV. But meanwhile, out of the outliers, the player with the lowest magnitude was Jerry Freaking Rice. He had a CV of 159 but he was drafted 16th which had an expected CAV of 35 giving us a z score of 3.77. High expected CAV significantly impacts z-score and thus over-values late-round picks. Z-score does indeed find the players that outperform their draft position the most…but this is still unsatisfying. When we talk about the biggest “steals” in the draft, being a solid starter for a long time is great, but we want to see some greatness as well. Some gold-jacket level stuff.
So CAV over expected over-values earlier picks while z-score over-values late-round picks. Let’s just combine them to get the best of both worlds! I normalized CAV over expected, and z- values and simply added them. After combining these values, we finally have our biggest steals in the NFL draft.
Tom is still #1. By quite a large margin too. You don’t need any statistics to tell you that though, taking one look at the chart could’ve told you that. He is miles away from the closest player. Drew Brees, holder of almost all passing records in the NFL is #2 as he was drafted with the last pick of the first round. The 7x Pro Bowler Zach Thomas at #5 is also notable, being one of the two players drafted after the 5th round to have over 100 CAV over expected.
Note that due to the huge variation in CAV over expected (thanks Tom), when min-max scaled, the CAV over expected had some gap so I (arbitrarily) used a 1/2 multiplier on the z-score before adding them. I think this picked the biggest sleepers of all time pretty well, even if it was not perfect. Overall, the outliers highlighted the all-time great out-performers and then we filtered them to find the best among the greats.
There were some issues with my method, with some top 10 having lower CAV than other outliers nearby. But without putting too much effort into it, this satisfies my curiosity. No matter how you slice it, it's pretty simple. Tom Brady is the greatest (steal) of all time.
*final note: someone smarter than I come up with a better way to reduce bias than just averaging a few of them. Maybe rate of change? Maybe doing a regression fit and then finding an average between them? Let me know if you have better luck than me.