The Secret Value of a Goal (And Why Some Don’t Matter)
We count goals as if they are integer numbers. To a certain extent, they are, though not all goals are created equally. For example, suppose in a hockey game, a team is up three goals with only a minute left in the third period. In this case, scoring again to increase the lead to four goals probably won’t impact winning probability (unless, of course, you are the Dallas Stars playing the Vancouver Canucks… Dallas managed to blow a three-goal lead in April 2025, with a minute left in the third period).
What I am proposing is a new way to count the value of each goal based on time and situation. The formula determining the value is as follows:
Change in Goal Differential State Winning Percentage (at that time interval) / Tied Game Winning Percentage (at that time interval)
For example, say a team’s win probability is 90% before the goal, and the goal bumps it up to 95%. The increase is 5 percentage points. If a tying goal at that moment is worth 50% win probability, then we calculate: (0.95 − 0.90) ÷ 0.50 = 0.10.
Here is why this is useful: if the score of a game is close, goals become more valuable. As the amount of time left in the game decreases, the value of a goal in a close game also increases. Think of it like currency. In a tied game, a goal is worth a full dollar. In a blowout, that same goal might only be worth a dime. The formula measures exactly how many ‘cents’ a goal is worth in the context of the game.
What if the score is not close? Generally speaking, as the score gap in any given game increases, teams (either ahead or behind) are more likely to go into “autopilot” mode. Considering the “blowout” factor weights relatively inconsequential goals less, and gives more weight to “important”, or close-score goals. This stat is easy to read, with anything scoring above 1 as more impactful, and anything below 1 as less impactful.
Here is the formula methodology: Using Python, I scraped Evolving-Hockey’s NHL play-by-play (PBP) data from every regular season and postseason game since 2007-08. For each game, I created a variable to indicate whether the game ended in a tie, a home win, or an away win. I then grouped every goal by time of game scored, and goal differential (close game or a blowout). From there, I was able to calculate winning percentage based on timing and situation. I chose 30 sample sizes among the rolling sum to ensure statistical significance.
Here is the output from my Python when I visualize the data:
As the chart illustrates, as a game progresses, the impact of goals which increase the lead increases. Somewhat surprising were instances of goals scored with less than five minutes left in the game. In those instances, the impact of the goal on winning probability was fairly limited.
So the next time someone brags about a hat trick, ask them: Yeah, but did it matter? Goals win games, but only the right ones at the right time.
