The Importance of Context in Statistics: A Case Study on Goaltenders Analytics

Statistics are an essential component of sports analysis, and in the world of hockey, they are constantly used to evaluate player performance. But how useful are they if not contextualized? Analyzing numbers in isolation can lead to misleading conclusions. Let’s see why.

Analytics image

The Problem with Isolated Statistics

Let’s consider an example outside of hockey. In a restaurant X, out of one hundred dishes served, fifteen are pizzas, and seventy-five are fish-based. If we relied solely on these numbers, we might conclude that fish is much more popular than pizza. But is that really the case? This interpretation ignores fundamental variables such as pricing, the restaurant’s location, the type of clientele, or even how the waiter presents the menu. In other words, numbers provide an indication, but without context, they can be misleading.

Goaltender Statistics: Goals Against and Save Percentage

In hockey, the two main statistics used to evaluate goaltenders are:

• Goals Against (GA)

• Save Percentage (SVG)

  
  

At first glance, these metrics seem reliable: fewer goals allowed and a higher save percentage should indicate better performance. But is that really the case?

A Real Comparison

Imagine two goaltenders who played the same game:

• Goaltender A: allowed 2 goals on 10 shots, with an 80% SVG.

• Goaltender B: allowed 1 goal on 15 shots, with a 92.5% SVG.

• 
  

If we only consider the numbers, Goaltender B appears to have played better. But can we truly reach this conclusion without additional information?

The Importance of Context

To accurately evaluate a goaltender’s performance, other key variables must be considered:

1. Shot distance – A shot from the central slot is much more dangerous than one from the blue line.

2. Game situation – Did the shot come from even strength, a penalty kill, or a breakaway?

3. Shot type – Was it a Goal Chance or a simple shot? (For a classification of Goal Chances, refer to the dedicated blog article.)

4. Defense quality – Did the team allow many rebounds or screens in front of the net?

   
   

A goaltender with a lower SVG might have faced tougher shots and more complex situations, while another with a high SVG might have stopped mostly harmless shots.

Goalie AI

Toward a More Comprehensive Statistic

There are statistical methods that take context into account, weighting the actual value of shots and their probability of generating a goal. The most relevant among these is Expected Goals Against (xGA), which measures the number of goals a goaltender should allow based on the quality of the shots faced.

xGA can be calculated for a single play, a game, a period, or an entire season and depends on various factors, including:

• Shot distance and angle

• Shot type (slap shot, wrist shot, etc.)

• Shot or Goal Chance situation (deflections, screens, lateral passes)

• Game situation (Power Play, Even Strength)

• Rebounds allowed

  
  

For even greater accuracy, additional variables such as shot release time, defensive pressure on the shooter, and rebound placement could be included.

Once Expected Goals are calculated, they can be compared with Expected Shots (which assign value to each shot based on the same variables) and then used to calculate the Expected Save Average. This allows for more normalized data and enables goaltenders to be compared using Goals Saved Above Expected (GSAx), which represents the difference between Expected Goals and actual goals conceded.

Are We Sure?

One might think that with these advanced methods, goaltender evaluation is definitive. However, there are still issues to consider:

1. Statistical accuracy – Human error and artificial intelligence imprecision can lead to significant discrepancies over a season. A goaltender who plays the puck frequently might face 3-4 more shots per game, significantly affecting the analysis.

2. Team quality – A strong team might allow fewer high-danger chances while conceding the same number of shots. Expected Goals might increase without real scoring threats, altering the goaltender’s evaluation.

3. Seasonal GSAx – Often, GSAx is presented as a total sum over a season, penalizing goaltenders who play fewer games. For a fair comparison, it is necessary to calculate the average GSAx per 60 minutes, normalizing the data for all goaltenders.

   
   

My Opinion

GSAx is a great method for normalizing league-wide performance and gaining a broad perspective, but it cannot be the sole evaluation criterion. From my point of view, three key variables are crucial for assessing a goaltender:

1. Number of slot shots faced – Indicates the level of pressure endured.

2. Rebounds allowed – Measures the goaltender’s ability to control rebounds.

3. Number of Goal Chances faced – Based on my observations from nearly 5,000 games, a goaltender concedes an average of one goal every 3.5 Goal Chances.

   
   

These three data points, combined with further observations, allow for a much more accurate evaluation of a goaltender’s performance.

Conclusion

Hockey is a team sport that goes beyond statistics. A strong goaltender is not necessarily the one with a 95% SVG over a season but rather the one who makes the decisive save at key moments, changing the outcome of a game.

Statistics are essential for providing a general overview, but they cannot and should not be the sole measure of judgment. We must use them to understand how to improve the goaltender, the playing system, and the entire team, without losing sight of the human and qualitative aspects of the sport.