Shot danger evaluation

[morehockeystats]

Based on the work of Micah Blake McCurdy presented at SEAHAC 2019, I was able to create a model that evaluates the danger of the shots based on static recorded parameters, as well as the skater's ability to score and goalie, correspondingly, to save above and below expectations.

Puts a whole new meaning into SVP...

• The main part
• The addendum

To summarize, the average probability of an unblocked shot attempt going in is about 0.06. However if we dissect these attempts by their available properties, we may conclude that the actual probability would wildly vary, with some even going above 0.5 . Naturally, for a goalie letting in 3 goals on 40 shot attempts that together are expected to produce 2.040 goals is not the same as letting in 3 goals on 30 attempts that carry an expectation of 3.512, but a mere SVP would tell an entirely reversed story.

Same for the skater. A player that consistently manages to send more pucks in that the sum of the probabilities of the shots he takes should be regarded as a more skilled shooter, whereas a player that misses rebounds in front of the goals, well, umm...

Questions and discussion are welcome!

Click to read more...

 

[Filthy Dangles]

Appreciate the work. My gripe with shot quality stats are that two big factors are ignored imo.

Whether the shot was clear sighted or not and if a pass was made forcing the goalie to move.

Btw I see one of the factors is whether the shot came off the rush, how does that work? Afaik, shots are not recorded or distinguished as being off the rush or not in the official scoring sheet for games.

 

[morehockeystats]

Appreciate the work. My gripe with shot quality stats are that two big factors are ignored imo.

Whether the shot was clear sighted or not and if a pass was made forcing the goalie to move.

Btw I see one of the factors is whether the shot came off the rush, how does that work? Afaik, shots are not recorded or distinguished as being off the rush or not in the official scoring sheet for games.

Of course there are more factors to be added, the two mentioned above are definitely among them.
There is a widespread definition of a rush, slightly varying from source to source, but remaining similar. I use the following formulation:

A shot attempt from offensive zone which is preceded by a non-faceoff event in non-offensive zone within five seconds.

 

[Henkka]

Great work. Sounds like the best model in theory I've seen so far.

Are the royal road pass, behind the net pass and goalie screens added? If not, these formulas tent to overrates selfish players, and underrate best team-first players.

The Most Important Line on the Ice You've Never Heard Of

If the passing factor is still missing, it inflates linemates scoring, who are playing with greatest playmakers. Before those effects are not added, I think we are still just half of the whole "truth".

In a nutshell, Crosby linemates turn to be regular 30-goal scorers and regress under 15 goals without him (or Malkin).

It would interesting to create somekind of key-assist formula, and then implement it together via somekind of line scoring formula and dig the effectivity out for individual scoring.

I don't have the recources to create that, but I think could work as an idea.

 

[morehockeystats]

Great work. Sounds like the best model in theory I've seen so far.

Are the royal road pass, behind the net pass and goalie screens added? If not, these formulas tent to overrates selfish players, and underrate best team-first players.

The Most Important Line on the Ice You've Never Heard Of

If the passing factor is still missing, it inflates linemates scoring, who are playing with greatest playmakers. Before those effects are not added, I think we are still just half of the whole "truth".

In a nutshell, Crosby linemates turn to be regular 30-goal scorers and regress under 15 goals without him (or Malkin).

It would interesting to create somekind of key-assist formula, and then implement it together via somekind of line scoring formula and dig the effectivity out for individual scoring.

I don't have the recources to create that, but I think could work as an idea.

Of course these three are not explicitly added, because there is no information about the shot assists, let alone miss assists. The first and especially the second one are sort of added implicitly, through the ice quadrant and the shot type.

A question is, what is the truth that we're looking for?

I am aware that even static factors like home/away are not incorporated, and the quadrants should in fact be replaced with differently shaped sectors. But... The thing is that each of the factors I did get to use, plays a role, because the probability of a shot attempt going in is different for each value of the factor. I'll work on it more - for example the mere on-ice presence of players may add or subtract from the probability.

Thanks for the feedback.

 

[Midnight Judges]

In a nutshell, Crosby linemates turn to be regular 30-goal scorers and regress under 15 goals without him (or Malkin).

In 14 seasons, Crosby has had a guy on his line score 30 goals 2 times (Kunitz and Guentzal), and Kunitz was a 25 goal scorer prior to becoming a Penguin.

 

[gumgum]

In 14 seasons, Crosby has had a guy on his line score 30 goals 2 times (Kunitz and Guentzal), and Kunitz was a 25 goal scorer prior to becoming a Penguin.

Woah.