Melvin Gordon’s Good and Bad Neighbors

Scott Spratt examines Gordon's closest recent comparable players based on a KNN algorithm.

| 2 years ago
Melvin Gordon

Melvin Gordon’s Good and Bad Neighbors

Melvin GordonI’ve read and heard a lot about Melvin Gordon in the weeks leading up to the NFL draft and the week since, and much of it has seemed to discourage fantasy owners. “Gordon played for a university in Wisconsin that has had previous backs become busts in the NFL, most recently Montee Ball.” “Gordon is not as talented as fellow first-round back Todd Gurley.” “San Diego overpaid to move up two spots in the draft to take Gordon.”

The thing about most of the criticisms of Gordon is they do not speak to what he actually is as a player. So rather than rely on those indirect comparisons, I wanted to use the data we do have on Gordon to draw some direct comparisons to him.

Player Year Drafted Height Weight 40Time
Melvin Gordon 2015 1.15 73 215 4.52


With Gordon’s draft status and basic measurables, we can make a fair estimate of his talent and style of play. To create the comparisons, I used a method called K-Nearest Neighbors (KNN). The idea behind KNN is simple. It uses a group of characteristics to rank how similar or different each item in a data set is to the item in question and then selects the K most similar items to average into a projection. Specific to the task at hand, I relied on a data set of rookie running backs drafted between 2008 and 2014. I used four characteristics of those players to create my model: draft slot, height, weight, and 40-yard dash time. In total, I had complete data on 108 backs, which is my sample.

When I ran Gordon through the model, I found that five of his 10 nearest neighbors gave me optimism for his transition to the NFL.

Melvin Gordon’s Good Neighbors, 2008-2014
Player Year Drafted Height Weight 40Time
Eddie Lacy 2013 2.29 73 220 4.55
Matt Forte 2008 2.13 74 217 4.44
Ryan Mathews 2010 1.12 73 218 4.37
Le’Veon Bell 2013 2.16 73 230 4.60
LeSean McCoy 2009 2.21 71 204 4.50


You can’t do much better than a best-fit comparison to the presumed No. 1 pick in fantasy in 2015, Eddie Lacy. However, Gordon’s specific tendencies as a player seem to make Lacy the least like Gordon of the five. Specifically, Gordon frequently bounced his runs to the outside rather than trust the holes provided by his interior linemen when he was in school.

Independent of the algorithm, I feel like Gordon’s best comparable of the five might be his predecessor in San Diego, Ryan Mathews. Of the group, he was the only player taken in the first round. He is very similar in size to Gordon, and I think their two speeds play more similarly on the field than they tested. Finally, I think both players have showed the potential to be productive pass catchers but have limitations there that create a low floor. Mathews, for instance, had a 12 percent drop rate and a poor 86.3 pass block efficiency as a rookie, but he still did record a 50-catch sophomore season when the team lacked alternatives to help on third downs. With some of the specialized players on the current Chargers roster, I think it is more likely that Gordon will cede that work to Danny Woodhead and Branden Oliver, but Gordon is capable of 30 catches even on early-down work.

Melvin Gordon’s Bad Neighbors, 2008-2014
Player Year Drafted Height Weight 40Time
Charles Sims 2014 3.05 73 214 4.48
Knowshon Moreno 2009 1.12 71 217 4.50
Mikel Leshoure 2011 2.25 73 227 4.56
Bernard Pierce 2012 3.21 73 218 4.45
Toby Gerhart 2010 2.19 73 231 4.50


The other five of Gordon’s 10 nearest neighbors were less positive. Both Bernard Pierce and Toby Gerhart were drafted into clear backup roles, but neither player has excelled in their opportunities to play since then. Charles Sims is in a similar situation in Tampa Bay, and he has not yet had the clear opportunity to prove himself one way or another.

The other two runners entered more similar situations as rookies to what Gordon has now, but injuries had major impacts on both players’ careers. Mikel Leshoure tore his Achilles tendon before he could even play his first game as a rookie, and while he provided some meager production once he returned, he never came close to his early promise. Knowshon Moreno has actually enjoyed two very productive periods in his career, but he had lost his starting job to veteran Willis McGahee in his third season even before a torn ACL and fumbling problems rendered him irrelevant for the next year and a half.

Leshoure and Moreno demonstrate Gordon’s potential downside, but their illustrated risk is one that applies to all backs. Just because they were similar players to Gordon does not convince me that they make a meaningful predictor of Gordon’s career health more so than the general injury risk that I’d apply to all rookie backs. As such, it’s hard not to look at Gordon’s comparable players and, when accounting for situation, become optimistic. The healthy players most like Gordon have pretty much all become productive players.

The composite projection the KNN algorithm creates for Gordon based on his 10 comparables rounds to 166 attempts for 658 rushing yards and five rushing touchdowns, and then 28 catches for 220 receiving yards and one receiving touchdown. Based on the unrounded totals, that is 120 fantasy points, which would have placed him as the No. 24 back in standard scoring in 2014 just behind Jonathan Stewart and Tre Mason and just ahead of Ahmad Bradshaw, Fred Jackson, and Isaiah Crowell. If anything, I think that’s somewhat pessimistic because of his favorable situation. In fact, I think it’s totally reasonable to draft Gordon as a top-20 back this season. Dynasty owners can reasonably be concerned about his career workload and other factors, but his forecasted short-term success is good enough to justify his selection in the first three or four rounds of a startup draft.

Scott Spratt is a Sloan Analytics Conference Research Paper Competition and Fantasy Sports Writers Association award winner. He also writes for RotoGraphs and contributes to ESPN Insider as a research analyst for Baseball Info Solutions. Feel free to ask him questions on Twitter – @PFF_ScottSpratt

  • Ben Fitzgerald

    The comparables you used don’t provide “a fair estimate of talent”.

  • Edward Misulis

    My only concern is that you used the common technique of player comparison forecasting, and I’m not even sure how well that actually predicts new players’ future production compared to other models. Would be great to see some articles that show how to do be accurate (or how to think about) injury, forecasting performance, etc.

  • Julien Bélair

    what was that? size and 40 yard dash to compare players…. I mean… wtf

  • Bust?

    Is Montee Ball really a “bust’? He’s a 2nd round pick who had decent production in limited playing time so far but was clearly the #1 Broncos back starting last year before getting hurt.

  • David Huie

    I don’t think draft position is a good metric to use with k-NN because it very subjective. If a player is taken at the n-th position in the first round, that is essentially saying that one team is willing to slot the player that high, not that the player’s ranking in relation to their peers is that high. A draft metric that probably holds slightly more weight is draft position among position class.

  • micah

    People keep saying gordons a reach and that gurley is way more talented. I think that’s lazy scouting because they see gordons from wisconsin and other backs from there haven’t been very successful, then they see gurley returning kickoffs and being big and fast. There is more to being a running back than being big and fast. In most of the games I’ve watched of gurley he’s running through huge holes straight up the gut, and his power is overrated.