Welcome to Sunday Notes, where we try to rehash important developments occurring on Hockey Graphs and elsewhere in the CORSI twitter league in less than 500 words. I’m sorry if we forgot about your post, or misconstrued what you said. We don’t care. Don’t @ us. Just do better next time. – asmean
On wins and expectations
In a five-part series for Hockey Graphs, Arik Parnass determined whether NHL teams that use the best lead-protecting players consistently exceed expectations. The idea behind this was borrowed from baseball’s Pythagorean Expectations, which posits that a team whose winning percentage systematically exceeds expectations is an indication of skill rather than just variance. And that’s exactly what Arik found in a statistically significant but still pretty modest relationship: teams who allotted the most ice time to their good lead-protecting players won a higher percentage of their one-goal games. Precisely, scoring chances against was the measure that most strongly correlated with one-goal wins. This is somewhat of an intuitive finding, as one expects that players who are demonstrably good at protecting leads would be good at preventing scoring chances against.
Traditionally, hockey scouts relied on their eye-test acumen along with pretty rudimentary statistics to identify potential NHL-caliber players. This traditional approach, while widely adopted by most teams, has been largely inefficient. This is what makes Prospect Cohort Success (PCS), a project created by Josh Weissbock and “money puck”, a huge improvement over traditional resources afforded to scouts.
For a given prospect, PCS uses publicly available data from most hockey leagues to:
- generate a cohort group of historically comparable players according to age, height and points per game that played more than 200 NHL games
- determine the percentage likelihood that the given prospect will also go on to play more than 200 NHL games based on his cohort
The results they presented to Hockey-Graphs are telling: for most leagues, PCS significantly outperformed a regression of age, height and points per game in terms of predicting NHL games played. For forwards, PCS had a stronger correlation with both NHL games played and points than for defensemen. This finding can potentially inform drafting strategy, particularly in assessing the risk associated with drafting a defenseman early in the draft.
Check out all the results here.