Skip to main content
The Most Promising Players In The NBA Draft According To My Computer

EDITOR’S NOTE (July 6, 2018, 2 p.m.): On June 20, about a week before the annual update of our CARMELO NBA player projections, we published this article using CARMELO’s preliminary data for players who were set to be selected in the NBA draft. In our haste, we failed to notice a problem with our file of NCAA player statistics. The file was compiled from multiple data providers, and the providers used different conventions to list certain statistics such as a player’s true shooting percentage. In particular, for some years, the statistics were listed as percentages (e.g., 51.2 percent), while for others, they were listed as decimals (e.g., 0.512).

This made the rookie projections that we published in the June 20 article a little weird. Because some categories were affected and others were not, CARMELO tended to emphasize unaffected categories such as height, weight, age and minutes played when selecting comparable players. (We discovered this issue and corrected it before we published the final version of CARMELO.) For the most part, the erroneous projections were fairly similar to the corrected ones — both versions like 76ers draft pick Zhaire Smith, for example. But there are some exceptions. Kentucky’s Kevin Knox, the Knicks’ first-round selection, graded out reasonably well in the buggy version of the projections, but the new, corrected version views him as a high-risk, medium-reward prospect — not hopeless, but not a good value with the No. 9 pick.

The original article has been rewritten to reflect the correct data and projections. The projections in this article differ very slightly from the ones on the CARMELO player pages because they use scouting rankings rather than a player’s draft position, as our official CARMELO projections do.

We usually don’t release our CARMELO NBA projections until after the NBA draft. But this year, in an effort to procrastinate from other modeling-related tasks,1 I finished them a little early. We’ll publish the complete set of CARMELO projections later this month, but with the draft scheduled for Thursday night, I wanted to share the system’s take on the best NCAA prospects.

Our methodology for CARMELO is pretty much the same as last year, with only minor tweaks. It works by identifying statistically comparable players — for instance, John Wall is currently similar to Detroit Pistons Hall of Famer Isiah Thomas and to Deron Williams. For NBA veterans, we use a database of player statistics since the ABA-NBA merger in 1976, and for rookies, we use a database of NCAA statistics since 2002, adjusted for pace and opponent strength, as provided to us by ESPN Stats & Information Group. The rookie projections also account for — indeed, heavily emphasize — where in the draft each player was selected. Because the 2018 draft hasn’t taken place yet, we can’t use that variable to evaluate this year’s prospects, so for now, I’ve used scouting rankings for both current and historical players.2

As I said, the changes from last year’s model are pretty minor, but one of them is potentially relevant in the context of this year’s draft, which is heavy on big men, including traditional centers such as Arizona’s Deandre Ayton. As ESPN’s Kevin Pelton has found, it’s become easier in recent seasons for teams to find once-desirable big men on the waiver wire or available for the minimum salary; the former All-Star center Roy Hibbert, who didn’t play at all in the NBA last year, is one perfect example. After evaluating the performance of players on minimum salaries over the past four years, we now use position-based replacement levels,3 which reflect that it takes a little bit more for big men to generate surplus value in the NBA than it does for guards and wings.

One last important warning: This list does not include projections for European players (so no Luka Doncic) or for other players who did not play NCAA basketball for some reason. Also, since Michael Porter Jr. played in only three NCAA games as a result of injury, we don’t project him on the basis of his NCAA statistics.4

At any rate, here goes: The top prospects as projected by CARMELO, non-Doncic, non-Porter edition. Players are ranked by their upside wins above replacement, a version of WAR that treats below-replacement-level seasons as zero instead.5

‘Stats + Scouts’ CARMELO projections for 2018 NBA draft

Not including European players or Michael Porter Jr.

Although CARMELO is reasonably deferential to the scout rankings, it has its share of disagreements. Although both CARMELO and the scouts are generally high on this year’s big men, CARMELO (in contrast to the scouts) likes Texas’s Mohamed Bamba and Michigan State’s Jaren Jackson Jr. slightly more than Ayton, for instance — and much more than Duke’s Marvin Bagley III. CARMELO likes Oklahoma point guard Trae Young, but it likes Kentucky’s Shai Gilgeous-Alexander slightly more. It’s high on Texas Tech’s Zhaire Smith but down on Kentucky’s Kevin Knox.

As both an empirical and a philosophical matter, we think it’s hard to beat the consensus rankings of NBA scouts and franchises. NBA teams are smart these days: Many of them have projection systems that are at least as sophisticated as CARMELO, plus they have lots of other information that we can’t possibly account for. So if CARMELO disagrees with the consensus of NBA teams, we don’t necessarily want to take CARMELO’s side of the bet. With that said, CARMELO is probably doing a few things right. It puts a lot of emphasis on a player’s age, for instance — that’s one reason it likes Jackson, who is the youngest player on our list. And it tends not to like players who score but don’t have good secondary skills, such as Knox (and to a lesser extent Bagley).

We can get a better sense for where CARMELO differs from the scouts by taking the scouting rankings out of the system and running “pure stats” projections instead. (Note that these projections still account for a player’s height, weight, position and age, in addition to his NCAA statistics.) We would definitely not recommend that NBA teams draft players on the basis of the list — it’s pretty wacky — but it helps to reveal how CARMELO “thinks”:

‘Pure stats’ CARMELO projections for 2018 NBA draft

Not including European players or Michael Porter Jr.

This list goes to show that ranking players purely on the basis of their college stats is probably an exercise in futility — or at least would lead to some very unconventional picks. On a pure stats basis, the top pick is … Kentucky’s Jarred Vanderbilt (?!?), who ranks just 56th in ESPN’s scouting rankings. Vanderbilt is a weird case, as injury limited him to just 238 minutes played for Kentucky. He was extremely productive in those minutes, averaging 19.7 points and 26.5 rebounds per 100 possessions, but his lack of shooting ability undoubtedly limits his upside, at least relative to where CARMELO pegs it.

Smith is more typical of the types of players CARMELO likes. At Texas Tech, he wasn’t a high-usage player, but usage rate tends not to translate very well from college ball into the NBA. On the other hand, scoring efficiency does, and Smith was highly efficient, shooting 55.6 percent from the field and showing 3-point range. His other statistics, such as blocks, steals and assist-to-turnover ratio, were also mostly pretty good, and he’s one of the younger players in the draft.

In other cases, CARMELO isn’t quite so contrarian. Jackson, Young, Bamba, Gilgeous-Alexander and Wendell Carter Jr. all rank in the top 10 according to both the pure stats and the scouting rankings. Overall, there’s a reasonably strong correlation (.53) between a player’s scouting ranking and his pure stats CARMELO upside score.


  1. Sorry, Micah.

  2. We’ll switch back to draft position once we publish the full set of CARMELOs.

  3. Centers have the highest replacement level, at just -0.4 points per 100 possessions, while shooting guards have the lowest, at -1.9 points per 100 possessions. Point guards (-1.7 points per 100 possessions) and small forwards (-1.6 points per 100 possessions) also have lower replacement levels than power forwards (-1.0 points per 100 possessions).

  4. We do have methods of projecting these sorts of players, but they’re quite primitive, so I’m leaving them off the list for now.

  5. The original version of this article used regular rather than upside WAR, because few players had negative WARs over the full seven-year window. In the corrected version, quite a few players are at or below zero, so it makes sense to use upside WAR instead, to more clearly distinguish them.

Nate Silver is the founder and editor in chief of FiveThirtyEight.