The Puck Stops Here

Pnep (Roman Nepomnyaschev) has recently updated his Hall of Fame monitor. Previous incarnations have been mentioned in previous posts in this blog (here are forwards, defenders and goalies. Although these systems do a good job of ranking players based on what they accomplished in their careers, they can run into problems as they rely on awards won by players (which may carry poor voting from awards voters into rankings) and because some aspects of players skill are easier to measure than others (offence is easier to measure than defence).

Here are the updated formulas:

Forwards
Adjusted PTS/2 +
Adjusted PO PTS +
Captainship - 15 pts per season +
Top 10 in Goals after Season - 10 pts +
# 1 in Goals after Season - 100 pts +
# 2 in Goals after Season - 70 pts +
Top 10 in PTS after Season - 10 pts +
# 1 in PTS after Season - 150 pts +
# 2 in PTS after Season - 100 pts +
Cup - 50 pts (minimum 5 Adj PO GMS) +
Final - 25 pts (minimum 5 Adj PO GMS) +
All Star Game - 5 pts +
HART - 150 pts +
HART Runner Up - 100 pts +
BYNG - 75 pts +
1 ALL STAR TEAM - 75 pts +
2 ALL STAR TEAM - 50 pts +
CALDER - 50 pts +
CALDER Runner Up - 30 pts +
SELKE - 50 pts +
SELKE Runner Up - 35 pts +
CONN SMYTHE - 100 pts

Defenders
Adjusted Games/5 +
Adjusted Pts/10 +
Adjusted PO Games/2 +
Adjusted PO Pts/4 +
Captainship - 15 pts per season +
Top 10 in Goals after Season - 10 pts +
# 1 in Goals after Season - 100 pts +
# 2 in Goals after Season - 70 pts +
Top 10 in PTS after Season - 10 pts +
# 1 in PTS after Season - 150 pts +
# 2 in PTS after Season - 100 pts +
Cup - 50 pts (minimum 5 Adj PO GMS) +
Final - 25 pts (minimum 5 Adj PO GMS) +
All Star Game - 5 pts +
HART - 150 pts +
HART Runner Up - 100 pts +
1 ALL STAR TEAM - 75 pts +
2 ALL STAR TEAM - 50 pts +
CALDER - 50 pts +
CALDER Runner Up - 30 pts +
NORRIS - 150 pts +
NORRIS Runner UP - 125 pts +
CONN SMYTHE - 100 pts

Goaltenders
Adjusted Wins/2 +
Adjusted PO Wins*4 +
Cup - 50 pts (minimum 4 Adj PO Wins) +
Final - 25 pts (minimum 4 Adj PO Wins) +
All Star Games - 5 pts +
Calder - 50 pts +
Calder Runner Up - 30 pts +
1 ALL STAR TEAM - 150 pts (seasons 1917-81); 75 pts (seasons 1981-06) +
2 ALL STAR TEAM - 100 pts (seasons 1917-81); 50 pts (seasons 1981-06) +
Vezina - 100 pts (seasons 1917-81); 150 pts (seasons 1981-06) +
Vezina Runner Up - 75 pts (seasons 1917-81); 125 pts (seasons 1981-06) +
Conn Smythe - 100 pts +
Hart - 150 pts +
Hart Runner Up - 100 pts

In the years before awards existed, he uses projected award winners that can be downloaded here. Statistics are adjusted as described here.

This is a lot of work and it is very good work, but it still has problems. There is a non-symmetry across positions required to calibrate between different positions. For example, defenceman get points for games played and no other positions do. Does this make sense?

The biggest changes are adding points for being a team captain and for runs in the Stanley Cup playoffs. These are generally things that are done by good players - but not always. There are bad players who play on good teams and have Stanley Cup runs and there are good players on weak teams who do not. A team captain is a political position usually given to a top player on a team, but it isn't the best way to determine who is a top player. Its benefit is that it is easy to determine who was captain, so its not too difficult to add this to the HOF monitor.

It is argued that the Hall of Fame monitor is intended to determine who will be inducted into the Hall of Fame given the preferences of voters and not who should be inducted and they tend to look at things like captaincy and Stanley Cup playoff runs, so they should be included.

I disagree. I disagree because we have seen recent Hall of Fame inductions. They select unworthy players because they are nice guys who are borderline candidates who are friendly with the Hall of Fame committee (Dick Duff). If you really want to capture the biases of the Hall of Fame committee you should be given points for friendship with committee members. This is a rather subjective thing to do. How do I quantify if player X is a friend of Hall of Fame committee member Y. Did they appear on the same team at one time? Were they both in the same celebrity golf tournament? It's not something that should be analyzed.

I think pnep has done well using statistics that are available to rank players. He adds things that might be seen as important to Hall of Fame voters that are easy to quantify. It is a good list, but it should be used not to try to mimic the often irrational decisions of the Hall of Fame committee, it should be used to rank the best players of all time.

The problem with ranking the best players of all time is that often the contribution of a player does not fully show up in his statistics. This is the primary problem with sabermetrics and hockey. In order to have a truly definitive player ranking system, we must be able to quantitatively show a player's value toward making his team win games. The results of this ranking system should be used to predict award winners. It shouldn't use award wins as an input.

Is such a system possible? I don't think it is with the hockey statistics that exist right now. Too much of the game is undocumented statistically. Even if future games were better statistically analyzed, the old games will not have the "modern statistics". Thus ranking players of the past will be a guesstimate at best on how they would have fared. I don't have any such system myself. I doubt it is possible to have a system as successful as those that exist in baseball. I am very interested in any such attempts and will post and critique them here. This pnep system is the best that exists (that I have run across). I hope it can lead to bigger and better results in the future.

Here is an hfboards discussion of the newest hall of fame monitor.