NHL Similarity Scores

The similarity scores were derived using a method similar (no pun intended) to the one used by Doug Drinen over at Pro-Football-Reference.com.

It is important to note that this method does not attempt to find players who were similar in style of play. Rather, it attempts to find players whose careers were similar in terms of quality and shape. By shape, I mean things like: How many years did he play? How good were his best years compared to his worst years? Did he have a few great years and then several mediocre years, or did he have many good-but-not-great years?

Another important item to note is that players are only compared to other players who played a comparable position. In other words, forwards are compared to forwards; defensemen are compared to defensemen; and goaltenders are compared to goaltenders. This is not always perfect, but it works well enough in most cases.

Players with at least three years played and a career value greater than zero (see #2 below) will have two similarity tables on their player pages. The first displays the most similar players through a given year (i.e., through year n). Only the first n years of a player's career are used when computing these scores. The second displays the most similar players based on entire careers. In this case, all years are used for all players.

Here is an example using Sidney Crosby and Eric Lindros through six years:

  1. Get the adjusted Point Shares values for the first six years of each player's career and order them from greatest to least.
    SC 14.7 13.3 11.1 10.5 10.1  9.0
    EL 15.0 12.8 10.0 10.0  9.2  7.6
  2. Compute a career value for each player by multiplying his best season by 1, his second-best season by 0.95, his third-best season by 0.9, etc.
    SC = 61.08
    EL = 57.72
    If either player has a career value less than zero then we stop here. Otherwise…
  3. Calculate the absolute difference in the adusted Point Shares values.
    SC 14.7 13.3 11.1 10.5 10.1  9.0
    EL 15.0 12.8 10.0 10.0  9.2  7.6
        0.3  0.5  1.1  0.5  0.9  1.4
  4. Calculate the penalty by multiplying the first penalty by 1, the second penalty by 0.95, the third penalty by 0.9, etc.
    P = 3.96
  5. Compute the similarity score as follows:
    100 * (1 - (2 * 3.96 / (61.08 + 57.72))) = 93.3

Through six years, the similarity between Crosby and Lindro is 93.3 (with 100 being a perfect match). The career similarity score between these two players is actually much lower, as we use the same method but fill in all 13 years of Lindros' career and fill out the rest of Crosby's years with zeros.