Editing GameScore

To see the running list of GameScores as they are updated weekly, navigate to the PlayerCards page.

When I first released my rendition of GameScore, I did so based on subjective relative weightings of importance. Essentially, I broke the quantifiable measures of stats into three tiers based on their perceived effect on an individual game's outcome. The highest weighted measures were scoring stats, the next highest were stats that measured scoring chances, and the lowest weighting was given to measures of events that created scoring chances. I have since recalculated my weights based on more quantifiable, and therefore more sound rationality. 

Each non-scoring stat was changed to reflect their relationship to a scoring play, therefore weighting them on their probability to affect the score and outcome of the game. The changes are as follows:

Description of Changes

• Individual Goal Scored - No change
• Goal Differential - I decreased the weight to GD because plus/minus has been proven to be a poor indication of a player's performance/value. I've settled on the value of .25. 
  
• Individual Primary Assist - A primary assist is a scoring play, therefore there is a direct affect on the game's outcome and the measure should be weighted heavily. However, there is a degree of separation between a registering a goal and registering an assist. The player scoring the assist is not the individual affecting the tally on the scoreboard, but they are as close as possible. There is a slight decrease between the weighting of a Primary Assist and Goal Differential to account for the difference between goal scoring stats and assist stats. 
• Individual Secondary Assist - A secondary assist has a slight degree of separation from a primary assist because it relies on the contributions of teammates. Therefore, the stat was weighted marginally below the weighting of a primary assist.
• Scoring Chance Differential - The weighting for this measure was decreased to reflect the percentage of scoring chance events that produce a point. This measure was calculated by dividing the League Average P60/the League Average SCF60. This produces an outcome similar to Shooting Percentage, but instead of relating the number of goals scored to the number of shot attempts, the number of points produced is divided by the number of scoring chances produced by a player and their teammates. The league average is  16.7%, meaning nearly 17% of all scoring chances produce a point. Hence, the weighting for this measure was assigned a value of .17.
• Individual Shots For - The weighting for this measure was changed to reflect the League Average Shooting Percentage which has a value of 9.09%. This means that 9% of all shots produce goals which is reflected in the weighting of .09.
• Corsi Differential - This adjustment follows similar a rationale to what was used to weight Scoring Chance Differential. I found the league average Points per Corsi For (P60/CF60) value to essentially create a Corsi Scoring Percentage. Again, this stat is to be interpreted like Shooting Percentage. The league average Corsi Scoring Percentage is 2.7% which means that almost 3% of all Corsi-for events produce a point that effects the game's outcome.
• Individual Penalty Differential, Takeaway Differential, and Blocks - These metrics were all proven to have minuscule affects to scoring event frequencies as their degree of separation is far removed from a scoring play. However, they are worth noting in regards to a player's performance.
• Individual Hit Differential - This measure was removed as the results of a hit is either reflected in the Takeaway Differential if the player loses the puck or not worth mentioning at all if the puck is not turned over. Just keep your head up next time, kid.

Results

The statistical breakdown of the new GameScore stat is summarized below:

Here's what you should know from above:

• The highest score was a 5.53, which was Travis Zajac's hat trick performance on December 1st against the Blackhawks.
• The lowest score was a -2.87 outing by Ben Lovejoy on November 26th against the Penguins. Lovejoy had no points (shocker), a -2 Goal Differential, and was buried in advanced metrics.
• The average score is .25, which was a tad higher than the previous GameScore's metadata that had a .22 average.
• The Standard Deviation is smaller, meaning that the data is tighter to the average (1.76 compared to the previous value that was over 4 points). With the changed GameScore, there are less outlier performances and the measure is more consistent with non-analytical measures of performance.
• The quartiles indicate a reasonable measure of good/bad games. To clarify, if a player has a GameScore above the upper quartile measure of 1.24, one can say that the player scored better than 75% of all performances to date. On the contrary, if a player records a performance under the lower quartile of -.86, one can say that the player had a performance within the lowest 25% of all performances to date.

Final Formula

GS = ((GF-GA)*0.25) + (G*1.00) + (A1*0.90) + (A2*0.85) + ((CF-CA)*0.03) + ((SCF-SCA)*0.17) + (ISF*0.09) + ((ITKA-IGVA)*0.01) + (IBLK*0.01) + (IPENDIFF*0.01)

Replicate My Results

Follow these steps to replicate my results:

1. ​
2. Go to corsica.hockey.com/skaters
3. Click Custom Query in the top left
4. Indicate a specific Team or Select All in the Team field, select "All" in the Strength State field, select all reports in the order they are provided in the dropdown menu, and indicate the correct date (10/12/16-present for this season).
5. Uncheck Aggregate Games
6. Download as CSV
7. Open in Excel
8. Create a new column and use the formula below:
   1. =((S2-T2)*0.25)+(AM2*1)+(AN2*0.9)+(AO2*0.85)+((G2-H2)*0.03)+((AA2-AB2)*0.17)+(AR2*0.09)+((BB2-BA2)*0.01)+(BC2*0.01)+(DJ2*0.01)
9. Check the formula's cell alignment with the intended formula design:
   1. ​GS = ((GF-GA)*0.25) + (G*1.00) + (A1*0.90) + (A2*0.85) + ((CF-CA)*0.03) + ((SCF-SCA)*0.17) + (ISF*0.09) + ((ITKA-IGVA)*0.01) + (IBLK*0.01) + (IPENDIFF*0.01)
10. Done.