How draft prospects are evaluated with data
Teams use age, efficiency, shooting signals, defensive event rates and measurements to judge draft prospects. This article explains each input, how models combine them, and where the numbers run out.
Every June, a front office has to turn a few hundred games of video, a stack of scouting reports and a table of statistics into one decision: which 19-year-old or 22-year-old do we trust with a roster spot and, in the first round, a guaranteed contract? The decision is made with incomplete information about a person who is still growing, who will change teams, coaches and roles, and whose current numbers were produced against opponents who will not be the ones he faces as a professional. It is a forecasting problem of the hardest kind.
Data does not solve that problem, and nobody serious claims it does. What data does is narrow it. A good statistical read of a prospect tells you where a player sits among thousands of earlier players who looked similar on paper, which of his numbers are likely to carry over and which are likely to fade, and where the scouts should spend their attention. This article walks through the main inputs teams and public analysts use: age, per-minute production, efficiency, shooting signals, defensive event rates, measurements, and the context that surrounds all of them. It also explains how those pieces are combined and why the combination still misses.
If you want the pick-by-pick business side of the draft, see The draft: how picks are valued and traded. This piece is about the evaluation underneath it.
What a draft model is trying to predict
Before any number is chosen, a team has to decide what it is predicting. "Will he be good" is too vague to model. Common targets include whether a player ever becomes a rotation regular, how many minutes he plays over his first several seasons, whether he becomes an All-Star, or a blended measure of career value built from the all-in-one statistics described in Win shares, BPM and VORP: the all-in-one stats. Each target gives a different answer. A player who is likely to be a solid ten-year role player and a player who has a small chance of being a superstar can have the same expected value and need completely different handling.
That distinction matters because the draft is lopsided. A very small number of players produce a very large share of the value, which is why the top of the draft is treated so differently from the middle. A model that predicts the average outcome well but cannot identify the rare outlier is of limited use to a team trying to find a franchise player. Conversely, a team picking in the middle of the first round is not hunting a superstar; it wants someone who will reliably contribute. The right statistical question depends on where the pick sits.
It also matters that the sample is small. The NBA draft has two rounds, so only a few dozen players are selected each year, and a model built on, say, twenty years of history is working with a few hundred first-round players at most. That is thin for a problem with this many moving parts, which is one reason careful analysts are modest about their models. The caution in Small samples in basketball: when a hot streak is noise applies to the history of the draft itself.
Age comes first
If one idea separates a thoughtful draft evaluation from a lazy one, it is age adjustment. A player who averages 17 points as a 19-year-old freshman and a player who averages 17 points as a 22-year-old senior have not done the same thing. The younger player has produced at that level while still developing, and the usual expectation is that he has more improvement ahead. The older player has produced it at or near his physical and skill peak, against younger opponents, and has less room to grow.
The aging pattern behind this is described in Aging curves: when basketball players peak and decline. Players generally keep improving into their mid-twenties, so a prospect's age at the time of his production is a rough measure of how much of that improvement remains. In practice, analysts compare a prospect to others of the same age: how does this 19-year-old's rebounding compare to other 19-year-olds who played major college basketball, and how did those players turn out?
Age is also why the one-year college stay has shaped the draft so much. When many of the best prospects spend a single season in college, the evaluator gets one season of data and a very young player, which makes every number carry heavy weight. The history of that arrangement is covered in The one-and-done era: how players spend a year in college. A prospect with three seasons of college data gives the model much more to work with, but the older version of him has to clear a higher bar to be considered a good bet.
One caution. Age data can be unreliable for international players, and birth dates can be imprecise for players whose records come from leagues with less documentation. Teams verify what they can, but a model that leans heavily on age leans on an input that is not always certain.
Production, adjusted for minutes and pace
Raw averages are misleading because minutes and pace differ. A player who scores 14 points in 20 minutes is producing at a different rate from one who scores 14 in 34, and a team that plays fast creates more shots for everyone than a team that grinds. Evaluators therefore convert counting stats to rates: per 40 minutes, per 100 possessions, or as a percentage of the team's possessions or rebounds while the player is on the floor.
Those conversions come with their own warning. Per-minute numbers for a player who rarely plays are noisy, and a bench player who stars in ten-minute bursts against tired opponents may not be able to sustain the same output over 30 minutes. A sensible read looks at both the rate and the playing time, and discounts the rate when the minutes are few. The measures involved are explained in Usage rate: how much of the offense a player uses and Rebounding rates: why rebound percentage is fairer than rebounds.
Usage deserves special attention. A prospect who produced efficiently while using a large share of his team's possessions has shown that he can handle a heavy load. One who was efficient only because he took a handful of easy shots has shown less. And a prospect who used a huge share of possessions on a weak team, with little help, may have been inflated in volume but deflated in efficiency by circumstances he could not control. The numbers alone do not say which situation you are looking at; film does.
Efficiency and the shooting signals
Scoring efficiency is among the sturdier indicators, but it needs interpretation. True shooting percentage, which folds in threes and free throws, is the standard starting point; see True shooting percentage and effective field goal percentage. A prospect with a high true shooting percentage on high usage is a strong sign. A high percentage built mostly on dunks and layups in a college game is less convincing, since the same player will face far more size and athleticism in the pros.
The most important thing a draft analyst asks about shooting is not "how well did he shoot" but "will he shoot as a pro." College shooting percentages from three are a noisy guide, partly because many college players attempt few threes and partly because the college line is not in the same place as the NBA line. Analysts look instead at a cluster of signals that tend to travel better:
- Free throw percentage. It is a cleaner view of shooting touch than three-point percentage because the distance and the circumstances do not change. Players who shoot well from the line, especially big men who rarely shoot threes, are more likely to develop a perimeter shot than their college three-point numbers suggest.
- Three-point volume. A player who takes a lot of threes in a college offense, and makes a decent share, has shown willingness and a repeatable motion. The sample is larger and therefore more trustworthy than a player who took 20 threes all season.
- Shot variety. A scorer who makes catch-and-shoot threes, pull-ups off the dribble and shots off screens has shown more than one way to score. A prospect who only finishes at the rim or only shoots when open has a narrower set of tools.
- Free throw rate. The ability to get to the line, covered in Free throw rate and the value of getting to the line, is a sign of attacking ability and often translates, because drawing contact is a habit.
In short, the question is whether a player's shooting is a skill he owns or an output he enjoyed. Regression toward the mean, explained in Regression to the mean: why extreme seasons rarely repeat, hits hot-shooting prospects hardest: a freshman who made 45 percent of a modest number of threes is probably not a 45 percent shooter.
Defense and the event rates
Defense is the hardest part of a player to measure, as Defensive metrics: why defense is hard to measure explains, and the problem is worse in college, where the tracking data is thinner and scheme and teammates matter enormously. Draft analysts lean on a handful of imperfect stand-ins.
Steal and block rates, usually expressed per possession or per minute, are the best-known. Evaluators have found that prospects who produce steals and blocks at high rates tend to be athletic and anticipatory, and that those traits carry into professional defense better than some other college numbers do. A guard with a high steal rate often has quick hands and active feet. A big man who blocks shots at a high rate often has timing and length. The two events together, sometimes nicknamed "stocks," are a decent signal that a player impacts the game on defense.
The signal has limits. A high steal rate can mean gambling, and gamblers can be punished by better offenses. A big who blocks shots but fouls often may be hurting his team. Rebounding rate, fouls per minute and the team's defensive performance with the player on and off the floor round out the picture. Teams also separate what a player did from what his role allowed. A lanky wing in a zone defense has fewer chances to prove he can stay in front of a quick guard than one who played man-to-man.
On the offensive side of the ledger, assist rate and turnover rate, covered in Assist percentage, turnover percentage and the passing numbers, tell you about decision-making. A big man or wing who passes well is more valuable than the scoring line suggests, because passing is a skill that makes a whole offense function. A guard with a high assist rate and an ugly turnover rate may be a creator who needs to cut down on mistakes, or a player whose college team gave him no other option.
Measurements and athletic testing
Basketball has a measuring ritual of its own. At the draft combine, prospects are measured for height without shoes, weight, wingspan, standing reach, hand size and body fat, and they run through athletic tests: a standing vertical jump, a maximum vertical jump with a running start, a lane agility drill, a shuttle run and a sprint over three-quarters of the court. They also take part in shooting drills and scrimmages. Teams hold private workouts and medical examinations in addition.
Wingspan and standing reach are the measurements with the clearest basketball meaning. Reach influences how far a player can contest a shot, how high he can finish, and how much room a defender must give him. A big man with a long wingspan can protect the rim from positions in which shorter-armed players cannot. A wing whose arms are long relative to his height can be a disruptive defender. Evaluators therefore look at a player's measurements in relation to his position, not just his height.
Athletic testing is useful but should be read carefully. Jumping and quickness numbers are measured in a gym on a given day with a given amount of rest, and they correlate loosely with game performance. A very poor athletic profile can limit a player's ceiling, particularly at the guard spots, but a great athletic profile does not guarantee that a player will use his gifts well. The best prospects tend to combine production, skill and tools. A tool-laden player who has not produced is a project; a producer with no tools is a risk; and both appear on every draft board.
Medical information belongs here too. A team's doctors review a prospect's history of injuries and his bone and joint health, and a flagged concern can move a player down a board or off it. That information is private and is not something public analysts can model, which is one reason outside draft rankings and team boards sometimes differ.
Context: the level of competition and the role
A number means different things depending on where it was produced. A player who dominated a smaller conference may have been up against players who would not be rotation members at a major program. Analysts therefore adjust for the strength of opposition, using team ratings of the kind discussed in Strength of schedule and how to adjust for it and in College efficiency metrics: KenPom, NET and the ratings used to select the field. Performance against strong opponents, and in the NCAA tournament, receives extra attention, though the sample is tiny.
Prospects from outside college are harder. International players come from leagues with different rules, different court dimensions and different styles, and the minutes they receive may be limited by the demands of a professional team. A teenager who plays 10 minutes a night for a top European club may be more impressive than a college freshman who plays 30 on a weaker team, but the numbers will not show it. Giannis Antetokounmpo, taken 15th in 2013 out of a Greek second-division club, is a famous example of how little a stat line can say about a player's trajectory when the competition and the minutes are unusual. Players coming from high school, from prep schools or from development leagues present the same difficulty: the competition varies, the sample is small and the statistics are less standardized. The pathways are described in How international players come to the NBA and the WNBA.
Role and teammates complete the context. A prospect who played alongside two other future pros had fewer shots and smaller numbers than the same player might have had on a thinner roster, and his efficiency might have been helped by the defensive attention the teammates drew. Analysts compare his production when different teammates were on the floor and ask whether the role he filled is one he will have in the pros.
How models combine the inputs
A modern draft model usually takes a prospect's age, per-minute production, efficiency, shooting signals, defensive event rates, height and wingspan and, where possible, competition level, and then estimates how players with similar profiles fared. Some approaches are statistical regressions that weight each input based on how well it predicted past outcomes. Others look for "comparable" players, a method that makes the output easy to explain: the prospect resembles these ten earlier players, and here is how they turned out.
The strength of such models is discipline. They apply the same lens to every player and are not swayed by a great highlight, a charismatic interview or the recency of the last tournament game. They are also useful at flagging cases where the consensus ranking looks out of line with the data, which is often where the most valuable discussions between analysts and scouts begin.
Their weakness is the one stated at the outset: the history is small and the world changes. A model trained on players from one era may mishandle a game in which the three-point shot is far more central than it once was, and it cannot see what a player will become after he is drafted. It also cannot see how a coach and a development staff will use him. Two prospects with identical profiles can end up in entirely different places because one landed with a team that gave him minutes, a clear role and good coaching, and the other did not. The final outcome is partly a result of the destination.
What the misses teach
The draft's history is full of choices that look bad in hindsight, and they are instructive because the warning signs were often more visible than people remembered. In 1984, Portland took Sam Bowie second, ahead of Michael Jordan, who went third; Bowie's career was undermined by leg injuries, a risk that no box score could have measured. In 2001, the Washington Wizards made Kwame Brown, a high school player, the first pick overall, and in 2003 Detroit took Darko Milicic second, ahead of Carmelo Anthony, Chris Bosh and Dwyane Wade. Those selections show how hard it is to project a teenager with no meaningful record against strong competition. The 2003 class is revisited in The greatest NBA drafts: 1984, 1996, 2003 and more.
The other direction is just as telling. Dirk Nowitzki went ninth in 1998, Stephen Curry went seventh in 2009, Manu Ginobili went 57th in 1999 and Nikola Jokic went 41st in 2014. None of those careers was predicted by the draft position, and each says something about what the numbers and the scouts under-weighted: skill and shooting touch in a player who did not look like his position, a shooter's trajectory in a small guard, the translation of a European player's feel for the game, and the passing and skill of a big man who did not test like an athlete. In a number of those cases the skills that mattered were visible in the data, such as free throw shooting, passing rates and age. They were underrated because the player did not fit a mental picture.
Taken together, the lesson is not that the draft is random, because it is not. A first-round pick is far more likely to be a useful player than a late second-rounder. The lesson is that the uncertainty around any single pick is large, and that a method which is honest about that uncertainty will do better over many picks than one that is confident about each.
What to take away
When you read a draft ranking or hear a prospect described by his numbers, a few questions will carry you a long way. How old is he, and how does his production compare with others his age? Are the numbers per-minute or per-game, and on how many minutes? Does his shooting look like a skill, given his free throw percentage and the volume of his threes, or like a hot year? What do his steal, block and rebound rates say about his defense, and what does his role say about how much to trust them? How was he measured, and does the body fit his position? And what was the competition?
Data will not tell you who will be a star. It will tell you who has the pattern that most stars share, who is being overrated by raw points per game, and who has a skill that the standard stat line hides. Used that way, it is a good companion to the scouts and a poor replacement for them. You can follow players as they move from college to the pros on the players page and see how college programs are rated on the college hub.