Game by Game: How To Use The KPI Scorecards

It’s early in the NHL season, which means small sample sizes abound. That makes it tougher to do any real analysis on what’s happened, so I took a lot of user feedback and decided to write up how to use the new “KPI Scorecard” that I’ve been posting on Twitter after each game.

In the business world, “KPI” stands for “key performance indicator.” In other words, metrics that best measure the success of a business objective or initiative over time. Crossing this idea over into hockey, I decided to aggregate a few different important metrics by displaying them on this scorecard. As you can imagine, I essentially consider them the “KPIs” of determining whether a player’s game was good or bad.

So, that’s really the genesis of what the scorecard aims to do. Now, let’s talk about what it tells us.

There are two sections of the scorecard: the individualized game stats (left) and the season averages (right). The season averages benchmark the player’s performance over the 2019-20 season against their outputs for the individual game the scorecard describes. Let’s break it down further.

There are four metrics the scorecard presents: time on ice, individual expected goals, on-ice expected goals, and game score. Additionally, each column is conditionally formatted (in Isles colors!) such that:

• Dark Blue is really good
• Dark Orange is really bad
• White is neutral (or average)
• There are gradients for each color to reflect the degree of “good” and “bad” based off of historical benchmarks (i.e. on-ice expected goals for percentage at 50%)

Now, for the metrics themselves:

Time On Ice (TOI)

Source: Natural Stat Trick

This is the most self explanatory metric, but it is important to show how often players are playing. It’s important context when reading across the table, as well as to get an understanding of the type of trust that a coach has in a certain guy. Additionally, a player who has all “blue” metrics playing for 20 minutes provides different context than one who has played for five minutes. It’s not the full story, but it is an important notation when looking at game performance.

The only real important caveat here is that time-on-ice incorporates all situations, so it’s more than just 5v5. Power play, penalty kill, overtime, etc. minutes are all displayed on the scorecard.

Individual xG (ixG)

Source: Natural Stat Trick

(Ed Note: this was incorrectly labeled on prior versions of the scorecard as 5v5 ixG. The scorecard displays *all situations* iXG.)

Individual xG measures how many goals a player is expected to score in a given game as determined from a few factors. These include the coordinates from which an attempt was taken, the type of attempt it is (wrist shot, rebound, one-timer, etc.), and other contextual factors to determine the likelihood that the attempt becomes a goal.

These numbers are typically very low, as most attempts have a small chance of actually becoming goals due to advanced defensive systems, goaltending ability, and all of the factors that we know contribute to low scores in hockey. However, it does give an indication of the quality of attempts a player is taking in a specific game.

For example, in the above chart Mathew Barzal had 0.28 ixG against St. Louis. What this tells us is that Barzal was expected to score 0.28 goals based on all of the attempts he took during that game. He was credited with the Isles’ second goal in that game, so he technically outpaced his performance. But this metric gives a better sense at the type of shot attempts each individual player took during the game.

On-Ice 5v5 SVA xGF%

Source: Natural Stat Trick

It’s good to know how much individual contribution a player has towards scoring, but that’s only one half of the game. Expected goals for percentage (xGF%) tells us the share of expected goals that a team has while that player is on the ice. Basically, it measures the total of all expected goals while a player is on the ice, agnostic of who is actually taking the shot attempt, and calculates the percentage of those goals the (in this case) Islanders have.

This metric is best used at 5-on-5 play, as special teams skew the totals in different directions. This is important because players who play only power play or penalty kill will see a drastic change in their overall totals. So to best normalize for this, using 5-on-5 play allows us to analyze performance in an equal way.

The “SVA” in this metric stands for score-venue adjusted, which accounts for where the game is being played and what the score is at the time of an attempt. This is relevant because we know due to “score effects” that teams play more aggressively when they are behind. That will inflate expected goal totals as the dynamics of the game shift according to that context. This adjustment normalizes that as well to give an appropriate amount of credit towards attempts depending on whether a team (in this case, the Islanders) is winning or losing.

To put this into action, from a 5-on-5 score-adjusted perspective Adam Pelech was on the ice for 0.43 expected Islander goals and 0.28 expected Blues goals during Monday’s game. The expected goal share formula would divide the 0.43 number into the total number of 0.71 to give him a total of 60.65% share of expected goals.

This gives context into whether a player is on-ice for more quality offensive opportunities and provides a more comprehensive light into in-game performance than any preceding shot-based metric.

(Note: There are other ways to gauge offensive and defensive effectiveness by using rate stats (such as expected goals for/against per 60 minutes), but those metrics are best used over time and not over an individual game.)

Game Score

Source: Hockey Stat Cards; Created by Dom Luszczyszyn

Finally, the game score column acts to measure the single game productivity of a player. This is one of my personal favorite metrics. The thinking is that production goes beyond just goals and assists, but includes other elements and events that occur over the course of a game. Such events may include expected goals (for & against), penalties drawn and taken, and shots on net.

You can read more about the updated formula for 2019-20 here.

In a lot of ways, Game Score acts as somewhat of a proxy for production itself. Players that produce on the scoresheet will see higher game scores. Players that are consistently in their own zone and getting scored on will see a worse score.

Production is inherently important; it’s how teams win games. But there is more within the process of scoring a goal that shows contribution, which is where this metric succeeds.

All of the aforementioned metrics show something different, but they are all (mostly) centered around expected goals, which is extremely important to me (especially as a “process guy”). Teams that are putting themselves in position to score more than the other team should theoretically win more games. Not only that, we can see which players are actively affecting quality chances through this scorecard. It also coincides pretty directly with Barry Trotz’s ideas behind “shot quality” while not losing its shot share metric roots.

On a personal note, I hope this piece was helpful. You can look forward to a new KPI Scorecard for every game, posted approximately 30-60 minutes after the final horn sounds.

--

Thanks to Natural Stat Trick, Hockey Stat Cards by Cole Palmer, and Dom Luszczyszyn for making all of this data public and usable.