Expected % of Points - A different approach to measuring draft results

One of the things I have been trying to do for some time is come up with a way of measuring the results of a team's drafts while accounting for pick quality and player quality.

This is a challenge for a variety of reasons. The first challenge is that while getting a player of any NHL quality is generally considered a "success" there is nevertheless a difference in quality among players and any analysis that ignores this is going to be lacking something. For example, prior attempts to measure draft quality simply count all players reaching some threshold (e.g. 100 NHL games played) as "hits" and while this works on some level, it is also obvious that not all players reaching 100 games are of the same quality.

Further complication is the simple fact that drafts vary in quality, and should not be considered equal. A team should not be punished for the simple fact that there were poorer options available during the draft year.

I have developed a new approach to evaluation which attempts to address both of these issues. The premise is fairly simple - each selection is evaluated based upon the percentage of points produced by the player in relation to the rest of the draft.

Since we are using points, obviously there are going to be variances between players playing different positions. For the purposes of this post, we will using only forwards. Thus, we will be looking at how each team performs only when picking forwards.


The above chart plots what percentage of NHL points are produced by forwards taken at each position from 1970-2015.

In other words, forwards who are selected first overall typically score 8% of the points in the NHL draft. By using this data, we can derive an expected performance whenever a team selects a forward, based on the pick selection. This helps control for the quality of the draft as well as giving higher rewards to higher value players who produce a lot of points.

I have smoothed out the above graph to account for the outliers (bearing in mind that even using a "large" data set going all the way back to 1970, still means we are only looking at 45 drafts,) and worked out a calculated expected percentage for every draft position. This will give us a basis for evaluating each draft.

To illustrate, let us take a look at Washington's 2012 draft:

Selection xPct Pct +/-
11. Filip Forsberg 3.03% 9.95% 6.92
16. Tom Wilson 2.36% 4.06% 1.70
77. Chandler Stephenson 0.57% 0.70% 0.13
100. Thomas Di Pauli 0.45% 0.00% -0.45
107. Austin Wuthrich 0.42% 0.00% -0.42
167. Riley Barber 0.22% 0.00% -0.22

Washington selected 6 forwards in this draft. With the 11th overall selection, one would expect them to select a forward who gets 3% of the points in the draft. By selecting Filip Forsberg, they received (but later traded away) a player who to-date has scored nearly 10% of the points in the draft. We can this give them a "credit" of +6.92 for this selection.

Rounding out the rest of the draft, they get a smaller credit for selecting Tom Wilson at 16, more or less received the expected value from Chandler Stephenson at 77, and get very minor debits for not receiving any points from their selections at picks 100+.

Thus, we can calculate a "draft score" by summing the deltas, giving Washington a score of 7.65 for this draft.

This establishes the approach we can use for evaluating NHL drafts. In my next posts, we will look at defenders and goalies, as well as run some evaluations of all 30 NHL teams.

comments powered by Disqus