≡ Menu

Age and the NFL Draft

In yesterday’s Wall Street Journal, Kevin Clark noted that the Eagles targeted college graduates in the 2014 NFL draft. Six of the seven players selected by Philadelphia are on track to get their degrees before entering the NFL, which is important to Chip Kelly.

Kelly said a degree is more than proof of intelligence. “It’s also, what is their commitment?” he said. “They set goals out for themselves and can they follow through for it? A lot of people can tell you they want to do this, this and this. But look at their accomplishments.”

Kelly’s quote has a certain air of truth to it, but is it verifiable? Do players with college degrees turn out to be better pros than players who don’t obtain their degrees? Unfortunately, I don’t have historical data on whether players graduated college before entering the pros. So this post can’t and won’t answer that question.

But we do have player age for all NFL players, subject to a big caveat [1]Unfortunately, we do not have such data on players who were drafted but did not make it to the NFL. This is a potentially serious issue with trying to analyze Kelly’s claim: if a non-graduate … Continue reading So here’s what I did:

1) Record the top 250 players selected in each draft from 1990 to 2009.

2) For each of those 5,000 players, calculate their Career AV in their first five years.

3) Create an expected AV curve for players by draft slot, which mirrors the myriad of other draft curves I’ve created.

av thru 5 years 1990 2009

In that chart, the average AV produced by each draft slot is in blue; the red line represents a smoothed curve. For the remainder of this post, the expected AV based on draft position reflects the red line.

4) Calculate the age as of September 1st of each class year for each prospect.

5) Run a regression using age and expected AV based on draft position to predict actual AV.

The result?

Actual AV = 34.7 + 0.91*DraftAV – 1.4*Age

The key part is at the end of the formula: for every year of age, a player’s expected AV in his first five seasons …. decreases. [2]Note: The R^2 is 0.34, which just means you can’t guarantee how good a player will be based solely on his age and draft position. The p-values associated with both age and draft value, however, … Continue reading In other words, to the extent age serves as a proxy for college graduates, Kelly’s hypothesis isn’t just not true, it’s counterproductive. Youth appears to be undervalued in the NFL draft. [3]And arguably is even undervalued by this method, as — all else being equal — we would expect a 23-year-old player to produce more value in his first five years than a 21-year-old player. … Continue reading

After eliminating the players who never made it to the NFL, that left 4,234 players in my sample.  The average age of those players on 9/1 of the year in which they were drafted was 23.1 years. By far, the player who most overachieved relative to draft position was Terrell Davis, who was 22.8 years old on September 1st of 1995, the year he was drafted. Davis actually graduated from Georgia, so he’s an example of age not necessarily being a great proxy for earning a degree. Next on the list of overachievers was Ray Rice (21.6), Emmitt Smith (21.3), Zach Thomas (23.1), and Patrick Willis (22.6). Thomas and Willis were four-year players, while Rice and Smith left early.

Of course, you can’t get too much from isolated examples. And while the regression results are conclusive, they’re also a bit abstract. So here’s what I did next.

The number 4,324 is evenly divisible by 23, which means we can group our draftees into 188-player groups. I sorted that list of players from biggest overachivers (Davis, Rice, Smith… at 1, 2, 3…) to biggest underachievers (Charles Rogers at 4,322, Ryan Leaf at 4,323, and Akili Smith at 4,324). So Group 1 has the biggest overachievers, Group 2 has the next 188 best overachivers, Group 3 has the 188 best overachievers after them, and so on, while Group 23 has the 188 worst underachievers (and Group 22 has the 2nd worst set of 188 underachievers). Then, I calculated the average September 1st age of each group:

age draft 1990 2009

As you can see, the average age of the biggest overachievers was 22.8; that’s the youngest average age of any of our 23 groups.  The next three youngest groups are groups 2, 3, and 4; in other words, the biggest overachievers are the youngest players, and the relationship is pretty clear. Using age as a proxy for the binary “did he earn a degree” category is tricky, but the analysis here is clear that young players are undervalued.

One last interesting note. If you look at the far right of the graph, you see that the curve begins to slope downwards. This means the biggest bust are pretty young, too (although still older than Groups 1, 2, 3, or 4).  Evaluating younger players is probably trickier: not only is less tape likely available, but the projection of a player’s physical projection is trickier, too.

Players within 3 months of 21.5 years of age on 9/1/XX represent the most undervalued group in the study. There were 133 of those players, and on average, they overachieved by 5.3 points of AV.  Players within three months on either side of 21.0, 22.0, 22.5, and 23.0 produced about 1.5 more points of AV than we would expect based on their draft position. As you would expect, the standard deviations were higher there, too.  But it seems the risk is certainly worth it when it comes to drafting young players. Although that doesn’t appear to be the case when it comes to quarterbacks. [4]I used this method to test whether quarterbacks were undervalued by AV. The regression results were inconclusive; the coefficient on age was -1.6, but the p-value was only 0.12 (sample size of 204). … Continue reading

References

References
1 Unfortunately, we do not have such data on players who were drafted but did not make it to the NFL. This is a potentially serious issue with trying to analyze Kelly’s claim: if a non-graduate was selected in the draft but because of his “lack of follow through” he fails to even make a roster, he would be a shining example of Kelly’s claim but would be ignored in this study.. That’s a problem, but there’s no way around it.
2 Note: The R^2 is 0.34, which just means you can’t guarantee how good a player will be based solely on his age and draft position. The p-values associated with both age and draft value, however, are statistically significant at every level.
3 And arguably is even undervalued by this method, as — all else being equal — we would expect a 23-year-old player to produce more value in his first five years than a 21-year-old player. Of course, the issue regarding players who can’t even make it to the NFL remains.
4 I used this method to test whether quarterbacks were undervalued by AV. The regression results were inconclusive; the coefficient on age was -1.6, but the p-value was only 0.12 (sample size of 204). The biggest reason for the differing results, I suppose, is that AV is not the ideal way to measure quarterbacks.
{ 28 comments }