clock menu more-arrow no yes mobile

Filed under:

Universal Measures of Player Value: Delta

Analytics has tried to find a universal currency for player values. Delta was another attempt.

Dilip Vishwanat

Up next is Delta.  Tom Awad developed Delta in 2010.  On the surface, Delta sounds simpler than GVT.  “Simply weight each shot by its expected chance of resulting in a goal, do the difference of for and against as in Corsi, and voila! Delta, the love-child of Corsi and plus-minus, is born....  Using this approach, we can create raw Delta scores, which simply represent what a player's plus-minus would have been with average goaltending on both teams, average shooting ability on both teams, and neutral luck.”  Raw Delta scores then get corrected for Situation (primarily face-off location)(called S), Quality of Opponent (O), and Quality of Teammates (T) to create DeltaSOT.

Output:  Predicted goals.
Input: Shots For and Shots Against.
Zero-point: Weighted Shotss For = Weighted Shots Against.
Slope:  Not constant.  Each shot, whether for or against, has a different value.
Symmetry:  Delta is symmetric.
Comparability:  Goalies don't have a Delta value.
Complexity:  Although Delta seems simple, it isn't.  Shot value is difficult to compute.  The adjustments for SOT are also complicated.

Delta comes up with some weird results.  In 2009-10, the top 5 players by DeltaSOT were Nicklas Lidstrom, Brian Rafalski, Patric Hornqvist, Mark Giordano, and Dustin Penner.  Hart trophy Winner Henrik Sedin was 27th.  Norris Trophy winner Duncan Keith was the 8th defenseman and 25th overall.  Jack Johnson was dead last by a wide margin, so at least Delta got that right.

2010-11 was even weirder.  The top 5 were Chris Campoli, Joe Pavelski, Lubomir Visnovsky, Ryane Clowe, and Chris Higgins.   Hart trophy Winner Corey Perry was 197th.  Norris Trophy winner Nick Lidstrom was 280th.  Matt Martin was last.  Jack Johnson was number 855, 5th from the bottom.  Score another one for Delta!

The biggest problem with Delta is, despite being more complicated than Corsi, it is actually worse at predicting the future.  Eric T. showed that the correlation between a player's Delta one year and his Delta in the next year is 0.35. For Corsi the correlation is 0.56.

Delta was another attempt at a universal currency for the value of players.  Like GVT, Delta has proven to have its problems.  Delta is hard to calculate.  Despite the complexity, Delta is less useful than Corsi by itself.