Expected goal and shot value can be calculated by using the success-rate of that specific play, as well as the success-rates of all succeeding plays to determine how likely that play has been to either produce a shot-on-net or a goal.
For example, we know that for a pass to the slot to result in a shot, the player passing the puck to the slot has to be able to succeed with his pass. Also, the player receiving the pass has to be able to get his shot through to the net.
Using 100 attempts (to simplify the explanation), if David Desharnais is successful with 39% (tops among forwards) of his attempted passes to the slot, we know that the player receiving the pass will have 39 shot attempts as a result of those 100 passes (100*0.39=39).
If the player taking the shot is Max Pacioretty, and he has been able to get 53% of his attempted shots through to the net, we can then calculate the expected shot value of that initial pass by calculating how many of those 38 attempted shots would result in a shot-on-net (0.53*39=20.67). This tells us that a pass to the slot from Desharnias to the Pacioretty has an expected shot value of 20.67; as it has succeeded in the past at a rate of 20.67%.
The expected goal value would be calculated using Desharnais pass to the slot success-rate combined with Pacioretty's true-shooting-percentage (percentage of attempted shots that result in a goal). For example, Pacioretty's TSP was 6.6% (0.066*39=2.57). So a pass to the slot from Desharnais to Pacioretty has an EGV of 2.57, as that play was successful in the past at a rate of 2.57%.
This system will allow us to potentially determine the goal or shot value of any play on the ice. The key is to include every potential succeeding play as well.
For the sake of simplicity, I used the average success-rates of Montreal Canadiens forwards to calculate all succeeding plays. In other words, the average success-rate of Habs forwards within each succeeding play was used in the calculation. That said, the success-rate of any combination of players can be calculated using this model. The success-rates used in this post where determined using the data from every Montreal Canadiens 2013-14 regular season and 2014 playoff game. A list of all the events I track can be found here.
As with any model based on success-rates, sample size has a direct impact on the results.
EXPECTED GOAL VALUE OF AN ATTEMPTED PASS TO THE SLOT
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Pass is made to the slot by the player listed, shot is attempted by an average Canadiens forward
EXPECTED SHOT VALUE OF AN ATTEMPTED PASS TO THE SLOT
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Pass is made to the slot by the player listed, shot is attempted by an average Canadiens forward
EXPECTED GOAL VALUE OF AN ATTEMPTED CYCLE PASS
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Cycle pass is made by the player listed; pass to the slot, and shot are attempted by an average Canadiens forward
EXPECTED SHOT VALUE OF AN ATTEMPTED CYCLE PASS
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Cycle pass is made by the player listed; pass to the slot, and shot are attempted by an average Canadiens forward
EXPECTED GOAL VALUE OF AN ATTEMPTED OPEN-ICE DEKE IN THE O-ZONE
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Open-ice deke and shot are attempted by the player listed (play does not involve a teammate)
EXPECTED SHOT VALUE OF AN ATTEMPTED OPEN ICE-DEKE IN THE O-ZONE
(ORIGINATING FROM THE PLAYER LISTED BELOW)
-Open-ice deke and shot are attempted by the player listed (play does not involve a teammate)
The next step in this process is to integrate tendencies into the expect goal and shot values. This will be done by using player tendencies to determine how likely each player was to attempt each play while they were in possession of the puck in the offensive-zone. Some of my work with tendencies can be found here and here.
For example, since Desharnais chose to make a pass to the slot 21.7% of the time he made a puck-possession play in the offensive-zone, we simply multiply Desharnais' ESV or EGV for that play by (0.217). This will allow us to value all of a player's play options based both on success-rate and likelihood.
I'll post my results integrating tendencies in an upcoming post.
