By: Albert Carreno
A hotly debated question in hockey is how can the defense around a goaltender impact his performance? The answer to this question may seem to be fairly intuitive. If a goaltender is surrounded by a stronger defense, they will perform better according to goaltending metrics like save percentage and goals against average since a stronger defense is better at preventing quality offensive opportunities for the opposition from developing, right? Well, the answer might not be as simple as anticipated. The reality is that goaltending and defense in hockey are incredibly hard to quantify from a performance standpoint. There are many different metrics in hockey, but most of them are dedicated to offensive performance and puck possession. As such, it can be difficult to gauge the correlation between a strong defense and a goaltender’s performance. However, by taking a few different approaches in analyzing the available data, there is a way to at least provide evidence for different sides of the argument about the impact defense on goaltending.
One opinion may be that defense has no impact on goaltending. People who align themselves with this mode of thinking likely take a firm stand in what the goaltender’s role is: protect the puck and stop shots. They may believe that the goaltender is the last line of defense and that it is his responsibility and no one else’s to prevent goals; that he is the only one in control of saving every shot directed towards the net. This is a more traditional philosophy surrounding the question of goaltending. Regardless, it does seem to have some merit according to regression analysis comparing high danger chances allowed per game and shots against per game to save percentage at even strength. Generally, high danger chances represent shots or scoring opportunities from places in the offensive zone where the odds of scoring a goal are high. Here are two scatterplots to better visualize the information being presented. The data shown was taken from a sample size of the last five NHL seasons.
Firstly, as we can see, there was little to no correlation between high danger chances allowed per game and team save percentage. The data points indicate no visible trend in either direction. One could argue there is a very slight negative correlation, but even so, the results are not statistically significant in any way because of how small the correlation is. The points are scattered all over the graph which typically represents no correlation. More of the same is what is going on in the second scatterplot where save percentage is plotted against shots on goal against per game. In fact, there is even less correlation here: the line of best is nearly completely horizontal, a true indicator of zero correlation. These variables are so uncorrelated that there is hardly even much of a difference in save percentage for teams giving up 20 shots at even strength versus teams giving up close to 30 per game. These results were incredibly surprising to find out firsthand since they oppose much of what I believed to be true. Nonetheless, these results can help give validity to the idea that quality of defense is irrelevant when compared to goaltending.
Another viewpoint, likely the one most people take, is that quality of defense has some impact on how a goaltender performs, but it’s not a critical factor. This is also a reasonable stance to take as basic logic says that quality defenders can lower the amount of difficult saves a goaltender has to make, but that the goaltender still needs to have a lot of skill and ability and that that is just as important. When analyzed statistically through data analysis, we can also see that there is some evidence to back up this perspective. This can be seen through comparing the corsi% for the top defensive pairing of a team and high danger shota allowed over the whole season with the goals against average for a team over the last five seasons. Corsi% provides a rough snapshot of how often a team possesses the puck in a game by comparing their total shot attempts(including those that are blocked and missing the net) with those of the opposition. In this case, we are taking the Corsi% of the top defensive pairing of a team. Each team has three defensive pairs; the top or first pair is usually considered the most effective pair defensively on each team and as a result spends the most time on the ice per game. Thus, a team’s defense is highly dependent on this first pairing. If they are largely in control of the puck and are driving the play without allowing the other team to create much opportunity offensively, it should theoretically make it easier on their goaltender. The results from these regression analyses are shown in the scatterplots below.
Looking at these graphs, we see a different story from the first two graphs presented. The top plot comparing high danger shots allowed in total with goals against average shows a moderately positive correlation with goals against average. In a large number of cases, it seems that the more high danger shots a team gives up in the regular season, the higher the goals against average of a team becomes. That said, the correlation here is not high enough that we can safely generalize and call this a rule of thumb. This correlation coefficient r for this plot is .56 and the r squared is .314. What this means is that for every unit change in the independent variable(high danger shots allowed) there is a 56% positive change in the dependent variable(goals against average), and that 31.4% of this change in the goals against average can be explained by the amount of high danger shots allowed.
For the second plot that depicts the correlation between the Corsi% of a team’s top defensive pair and goals against average, there is a correlation coefficient r of -0.43 and an r squared of 0.183. Similarly, what this means is that for every unit change in the Corsi% of the top defensive pairing for a team, there is about a 43% negative change in goals against average. Also, 18.3% of this change in the goals against average can be explained by the Corsi% of a top defensive pairing. Compared with the first scatterplot, the correlation shown in this plot is weaker, but it is still high enough to be considered at least close to moderate. This means that both of these variables, which are considered to be among the better ways to quantify defensive performance with the data that is readily available online, do have an actual effect on goaltending performance. As explained, it is far from a major effect, but it is still a visible effect that validates the popular opinion that goaltending is only somewhat affected by defense.
Lastly, there are those that believe defense is vital when it comes to goaltending performance. Without good defenders, people in this camp believe that goaltenders are going to have their work cut out for them from having to face so many difficult shots and that the fatigue from being under siege for most of the game will prove insurmountable, causing them to perform worse than if they were on a team that protected the puck better and gave up less. In analyzing this argument statistically with empirical data, I decided to take a different approach. Instead of using linear regression analysis, which is of course reliable and highly effective, I decided instead to compare case studies of different goaltenders and compare their numbers on highly defensive teams that minimized risk as much as possible versus when they were on teams that struggled to establish a good defensive structure in games and gave up a lot of goals in general. The dot plots below show the save percentages and goals against averages of six goalies on defensively talented and defensively limited teams.
The first dot plot deals with the goals against average of these six goalies. The first group of observations represents the goals against average of these goalies when they were playing for defensively limited teams. The second group is the goals against average for goalies when they were on defensively talented squads. As we can see, every single goalie did significantly better in terms of goals against average when they were on good, defensive minded teams compared when they were not. In particular, every single goalie allowed at least half a goal less on the better teams, with three allowing at least three quarters of a goal less. This is a remarkable improvement.
The second dot plot is a similar story. The only difference here is that the first group of observations represents the performances of goaltenders on bad teams and the second represents the performances on good teams since unlike goals against average, the higher the save percentage, the better. This plot has very similar results to the first plot. Every single goaltender showed improvement in their save percentage when they were on a good defensive team. A couple of goalies even improved their save percentage by .025 which is no easy feat. All told, there is definitely evidence to back up the assertion that defense is paramount and has major effects on the performance of a goaltender when taking a close look at some case studies.
After reading all of this, you may find yourself a bit confused. Evidence for every side of the argument was presented, but no clear stance was taken. That is because it is truly impossible to narrow down how impactful defense really is on goaltending without the use of more advanced metrics that have either not been created yet or information that is unfortunately in the hands of corporations and is not publicly accessible. The point I want to make here is that whatever side of the spectrum you fall on with this debate, there is valid data to support your argument. However, I will say that not all data and analysis is created equal. I think it is fair to say that the evidence presented for defense having a very large impact on goaltending performance is the weakest. It was taken from a small sample size and no advanced statistical methods were used to carry out the analysis of the data. It still provides irrefutable evidence, but not enough to truly prove one’s point when compared with the evidence for the first two sides of the argument. As such, I personally believe we can simply combine the first two sides of the argument and safely say that defense has a small to moderate impact on goaltending performance, but not a major one. Whether more sophisticated metrics and analysis will arise one day to prove or disprove this claim is still uncertain, but as far as what tools are readily available right now, I think this is a fair conclusion to come to.
All statistics and data were taken from MoneyPuck, Natural Stat Trick, ESPN, or Hockey-Reference.