In preparation for the 2016 draft, ESPN’s Analytics Team dusted off its NBA Draft Projection model, which debuted last year on FiveThirtyEight.com. This model is designed to project the NBA success of college prospects early in their careers, or, more precisely, it is projecting a player’s statistical plus/minus (SPM) in years two through five in the league.

That time frame was chosen because it reflects the number of years a first-round pick is under team control, and does not penalize players for poor rookie season, which are often outliers for a variety of reasons (drafted to a bad team, needs time to develop, etc.).

The model’s main inputs are college statistics (adjusted for pace and level of competition faced), Chad Ford’s Top 100 Prospect Rankings and player information such as age, height, weight and position. For 2016, it used information from the 2001 to 2011 draft classes to predict SPM for players in later classes, with steps taken to adjust for the players that barely saw the court or never played in the NBA.

Like many models, a prospect’s scouts rank is by far the most important variable when predicting NBA success, but by adding other variables the model reduces the uncertainty in projecting SPM by about 10 percent compared to using Chad Ford’s Scout’s rankings alone. Aside from scouts’ rankings, demographic factors such as age, height and BMI had some of the biggest impacts.

Not surprisingly, younger prospects generally achieved higher SPM in their first five seasons, but that is largely a function of the draft; players usually enter the draft when they believe they will be drafted, so the top players will generally enter at a young age.

Among the 14 opponent-adjusted college statistics included in the model, offensive rebounding percentage has the biggest impact on the projections for big guys and steal percent is important for guards. It’s likely these stats are capturing a level of athleticism transferrable to the NBA and may be able to isolate a player’s skill in that area of the game. Three-point rate (percentage of shots taken behind the arc) also proved valuable, which could be a sign of the changing nature of today’s game.

Bringing all of these variables together into a random forest survival model produces two main outputs. The first is a player’s draft grade, which inputs a player’s average SPM projection on a 0-to-100 scale. The players with the highest draft grades are most likely to be successful in the NBA but may not necessarily have the highest ceiling.

To understand the risk and reward of each prospect, a player’s SPM projection is also broken out into a player’s chance to play at the level of an All-Star, starter, bench player or bust early in his career. Based on this methodology, depending on the number of college prospects in the top 100 each year, there are expected to be about 2.5 All-Stars, seven starters, 30 bench players and 40 busts per college entry class.

It’s important to note that this is measuring the percentage chance each player reaches these levels in his first five seasons. For example, Kyle Lowry made the All-Star team in 2015 and 2016, but did not reach that level until his ninth season.

The next and most natural question is how the model has done in predicting past classes. No model will be ever be 100 percent correct, but this one has proven to be effective and well calibrated. Since draft classes from 2001 to 2011 were used to train the model, we can look at the 2012 to 2015 classes to see how it played out.

Among the top 10 most-likely players to play at an All-Star level in the 2012-15 draft classes, there were certainly hits and misses. Some of the misses were a product of injuries and others were a result of the overall uncertainty surrounding the draft. Nonetheless, among this group, about three players would be expected to play at an All-Star level in their first five seasons, and given the experience of each player, that projection may not be too far off.

Looking only at the 2015 draft class, five of the top eight projected college players made the First- or Second-Team All-Rookie teams (seven of 10 spots were college players), with the top projected prospect, Karl-Anthony Towns, winning NBA Rookie of the Year.

Now, no algorithm or general manager will ever be able to perfectly forecast the NBA Draft. Every year there will be examples of players that slipped through the cracks and other seemingly safe picks not working out. One thing that the model does not explicitly measure (though it is accounted for in the scouts' rankings) is a player’s drive, leadership and intangibles.

Injuries are also another factor that cannot be forecast beyond accounting for a player’s BMI. Based on what can be measured, however, the model is a valuable and accurate tool to help sort out the players at the top of the draft and identify sleepers in the 2016 draft class.