Pelton: How my NBA draft projections work

Three primary factors go into my projections for the NBA draft:

• A player's projected performance as a rookie

• His age

• His rank in the latest top 100 from my ESPN colleagues Jonathan Givony and Mike Schmitz

All three of those factors have helped predict how well college and international players will perform in the NBA. Let's take a closer look at them.

Projected performance

For college and international players alike, my projections start with translating what players have done before with what they will do in the NBA for 14 core statistics: 2-point percentage; 3-point percentage; free throw percentage; offensive and defensive rebound percentage; assists; steals and fouls per 100 team plays; blocks per 100 2-point attempts; percentage of individual plays devoted to 2-point, 3-point and free throw attempts; turnovers; and usage rate.

Some stats tend to decline once players enter the league more than others. For example, free throw percentage rarely changes much, but players shoot much worse percentages from the field. For college players, this process also adjusts for strength of schedule to put players who faced different levels of competition on a level playing field.

Utilizing multiple seasons of data yields better projections than looking solely at the most recent season. In fact, for college players I've found that earlier seasons tend to project NBA performance better than more recent seasons because older prospects no longer have the same experience advantage in the NBA. So the player's most recent season is weighted two times, the season before that (if available) three times, and two seasons before that five times. The weights are opposite for international prospects, who haven't demonstrated the same pattern because they are playing against older opponents.

The last factor in the projections is regressing performance to positional averages for NBA rookies (for college prospects) and replacement-level performance (for international players). This helps account for outliers, particularly for stats that tend to fluctuate, such as 3-point percentage. As a sophomore at Arizona, Derrick Williams shot 56.8 percent on 74 3-point attempts. Williams has made 29.9 percent of his 3s during his NBA career.

It doesn't make sense to regress a point guard to the same assist rate or block rate as a center, so this is the one place where positions affect my projections. This tends to have more impact for players who saw limited action in college or overseas than experienced players. It also can be problematic for versatile players, such as Ben Simmons, whose stat lines don't look like any one particular position.

WARP projections

The statistics-only version of my projections estimates the wins above replacement player (WARP) that a prospect will average during his first five seasons in the league, adjusted to favor immediate projection by discounting performance in future years.

To calculate this, I've used past players in my database -- which is largely complete back through the 2006 draft and includes a handful of players back through 2003 for whom per-play college stats are available -- to run a model projecting WARP based on the player's projected win percentage (the per-minute version of WARP, akin to PER) as a rookie and his age.

Because players tend to develop through around age 27, their age makes a large impact on their projections. All other things being equal, each additional year of age tends to reduce a player's projected NBA value by about 0.5 WARP per season.

Since 2018, the WARP projections have included a higher replacement level for centers and power forwards than perimeter players -- particularly wings. Note that this change has not been applied retroactively to past projections before 2018.

Consensus projections

Inspired by draft analyst Layne Vashro's Humble model (Vashro is now working for Kroenke Sports & Entertainment, the parent company of the Denver Nuggets), in 2015 I introduced consensus projections that also include the league's assessment of a prospect's talent. For past prospects, I measure this with where they were actually drafted. For current prospects, I estimate this using their rank in our top 100.

Using the actual draft slot plays a factor in subjective scouting analysis and improves projections, so I tend to use it as my primary ranking of players, while also showing the stats-only projections when possible.

Past projections

To give an idea of how my projections have worked -- and when they haven't -- here are the rankings for the 2006 through 2019 drafts, based on my current projection model: