The Quick and Dirty Take data from the selected player's 2020-21 NBA Season Calculate the selected player's percentile for each stat against

Introducing the Crafted NBA Similarity Feature

August 13, 2021

As we launch our brand new player similarity finder, we thought we might as well go ahead and explain the thing.

The Quick and Dirty

  1. Take data from the selected player's 2020-21 NBA Season
  2. Calculate the selected player's percentile for each stat against the filtered set of player seasons
  3. Find the most similar offensive and defensive seasons
  4. Combine offense and defense to find the most similar player seasons since 1978
  5. Profit, intellectually

The Nitty Gritty

A quick and easy way to compare current players to players from the past using data for every player season where the player played greater than 500 minutes. Initially, this was created to be a way to compare young players against players from the past with the goal of providing further context into how good a player might be expected to become in the future, but eventually we decided to expand it out to all players. The original thinking was, if we know how these players from the past turned out, there's a reasonable chance that this player will turn out to be of similar quality to these players they were most similar to at their current age. Take Immanuel Quickley, for example:

The first thing you will see is the player's "scouting report", i.e. how they rate at each stat against the other players in the set (the filters default to the players current age +- 1 and their position). Note that any time you change a filter, the players in the dataset as well as the scouting report and similar players will also change. In Quickley's case, we are looking at all Guards who were 20, 21, or 22 years old during the season.


Now, based upon that set of 608 players, we will measure every other player in the set against the 607 other players to get their percentiles for each stat. We then take the difference between each player's percentile for each stat and find the difference between their percentile and Quickley's. The least amount of difference between the percentile's, the more similar the player is. Now, the most similar players:



This Quickley guy...pretty...pretty...pretty good.

Stats Used

We tried to cover the main areas of play with our stats, but due to a lack of defensive box score stats (particularly in the pre player-tracking era), we had to settle for what was available. The defensive stats are pretty self explanatory, so we will skip those and get straight to the offensive ones. We think we have struck a nice balance between measuring how much a player does and how effective they are at doing it. Any stat with an 'r' in front of it means that it is relative to the league average from that season. This was necessary to close the gap in the rise in efficiency (three point shot!) and style (less offensive rebounding) over the years. 

Similarity Score Guide, presented as human relationships

50 and under - never seen this person before, lives in a different hemisphere

50-60 - same country, different region

60-70 - grew up in the same state

70-80 - same city, dad went to high school with aunt Becky

80-90 - high school classmate, probably have a secret handshake

90+ - conjoined twins

*Shout out to the incomparable FBREF for the inspiration.

Back to Blog

Sign up for our newsletter

Be the first to know about new features, upcoming releases and exclusive insights.

We care about the protection of your data.
We will never sell your information.