finding no real value-added for determining when a collegiate RB breakout age. RB breakout Age was not a factor in my initial machine-learning RB model posted in March.

But I am not one to quit so easily. I did some additional research, and have come up with a new, better way to identify RB breakouts. You may recall, from the initial model, that adjusted yards per team offensive play was a major factor. Using that as a basis for breakouts proved to be incredibly strong.

I went back into my database (261 RBs), and recorded the age at which a player recorded at least 2.0 adjusted yards per team offensive play. I then created a weighted average with that number and final age of the RB, much like I do with the WR position (if a RB never broke out, just final age is used). The results proved to be extraordinary.

x85BA	RBs	Hits	Hit Rate
Age 19	17	10	59%
Age 20	47	15	32%
Age 21	82	15	18%
Age 22	73	9	12%
Age 23	37	0	0%
Age 24	4	0	0%

As you can see, breakout age and the resulting adjusted age can tell us a lot about incoming RBs. An adjusted age of 19 led to a whopping 59% hit rate in the sample, and the hit rate decreases in a reasonable fashion as age increases. Interested in an older RB prospect? If they haven’t broken out, or didn’t break out until late, it is likely best to avoid them. No RB with an adjusted age of at least 23 was a hit in the sample.

RB Breakout Age: Model Impact

Of course, the final test for how much to care about breakout age is to see how statistically significant it is in models. Fortunately, adjusted age proved to be very significant when run through the machine-learning process. It led to the development of a new, six-model ensemble. Here were the inputs listed in order of significance:

1. Final season adjusted yards per team offensive play
2. ESPN Grade
3. ESPN overall rank
4. Adjusted age
5. Speedscore
6. Binary return touchdown check

To give an idea of just how powerful adjusted age is, it came in with the same level of significance as the overall rank, which is essentially our proxy for draft position. Outstanding. Here’s how the new model rates the 2020 RB class. I have included the age metrics in the table so you can see who sees a bump as a result.

Player	School	fAge	bAge	x85BA	Prob
Jonathan Taylor	Wisconsin	20.96	18.96	19.26	82.2%
J.K. Dobbins	Ohio State	21.05	21.05	21.05	54.0%
Cam Akers	Florida State	20.54	20.54	20.54	36.9%
D'Andre Swift	Georgia	20.98	-	20.98	33.1%
AJ Dillon	Boston College	21.68	21.68	21.68	30.5%
Clyde Edwards-Helaire	LSU	20.74	20.74	20.74	28.2%
Zack Moss	Utah	22.06	22.06	22.06	24.2%
Antonio Gibson	Memphis	21.54	-	21.54	19.6%
Eno Benjamin	Arizona State	20.73	19.73	19.88	17.2%
DeeJay Dallas	Miami	21.3	-	21.3	5.7%
LaMical Perine	Florida	21.93	-	21.93	5.7%
Anthony McFarland	Maryland	21.84	-	21.84	3.7%
Darrynton Evans	Appalachian State	21.49	-	21.49	3.5%
Ke'Shawn Vaughn	Vanderbilt	22.67	21.67	21.82	3.4%
Salvon Ahmed	Washington	21.02	-	21.02	3.1%
Javon Leake	Maryland	21.43	-	21.43	1.6%
J.J. Taylor	Arizona	22	-	22	1.4%
Levante Bellamy	Western Michigan	23.1	-	23.1	1.3%
Raymond Calais	Louisiana	21.76	-	21.76	1.3%
Patrick Taylor Jr.	Memphis	21.69	-	21.69	1.2%
Joshua Kelley	UCLA	22.13	21.13	21.28	1.1%
Rico Dowdle	South Carolina	22	-	22	0.6%
JaMycal Hasty	Baylor	23.32	-	23.32	0.5%
Mike Warren	Cincinnati	21.15	20.15	20.3	0.5%
Scottie Phillips	Ole Miss	22.25	-	22.25	0.4%
Brian Herrien	Georgia	21.91	-	21.91	0.4%
Sewo Olonilua	TCU	22.11	-	22.11	0.3%
Reggie Corbin	Illinois	23.83	-	23.83	0.2%
Tony Jones	Notre Dame	22.12	-	22.12	0.2%
Darius Anderson	TCU	22.32	-	22.32	0.1%

Our top RB in the previous model sees an even bigger bump when factoring in age. Jonathan Taylor broke out at the tail-end of his age-18 season, and it has his success probability all the way up at 82%. The only other prospect to break out before 20 was Eno Benjamin, who saw his success probability nearly double since the first exercise. On the other end of the spectrum, Zack Moss saw his probability decline due to advanced age and some declining ESPN grades.