How has regular season NHL goal scoring changed over time?
We often talk about how much the game of hockey has changed. It is faster, more skilled, less physical, has improved gear and new strategies, etc.. Here we will take a closer look at some data to see if and how scoring and other basic metrics have changed over time.
Gathering the data
For this article, we will use NHL.com/stats team summary data from the inaugural 1917–18 season to the current 2020–21 season. This contains basic information like goals, points, wins, etc., summed over each season. The dataset is not entirely complete as earlier seasons are missing information like shots on goal (SOG) and faceoff percentages (FOW%).
To automate the process of gathering the data, I scraped the website using RSelenium and rvest. This process is covered in detail here and the code is linked at the bottom. It is also worth mentioning that this data was scraped on May 2nd, 2021, meaning the entire 2020–21 shortened season is not yet finished. Additionally, we are also only looking at regular season data here. Playoffs to come!
Analyzing regular season NHL data
Before looking into goal scoring, saves, etc., let’s first just take a look at season length. Each point represents a team’s season total, and since teams typically play the same number of games, they are mostly overlapping here. The colors here correspond to a decade of seasons and are consistent with the rest of the plots.
An interesting piece of history is that the Montreal Wanderers only played 6 games in the inaugural season, thwarted by a fire burning down their home arena, whereas the other three teams played 22 games. The Toronto Arenas went on to win the first Stanley Cup that season. We also observe that a few points deviate from the trend in the last two seasons. This is because the 2019–20 regular season was cut short due to the pandemic and the 2020–21 regular season was not over when I downloaded the data. The other dips correspond to lockout seasons. Nonetheless, the overall fact remains that NHL seasons have increased in length.
Goals for over games played (1917-2021)
Now moving into the most interesting metric: goals. Each point in the plot below corresponds to the team average goals for (GF) / games played (GP) over the course of a season. The NHL started with just four teams and has grown to 32, hence the increase in density of points over time.
As we can see, NHL games were initially very high in scoring, with season-long averages of nearly 5 GF/GP (therefore about 10 goals per game). This quickly began to drop all the way to 2 GF/GP until the 1930s when the averages increased and steadied around 3 GF/GP. In the 70s, goals started to spike again reaching a max of nearly 4 GF/GP in the mid 80s. Afterwards, there was a steady decline to the 2000s where GF/GP eventually began to gently increase to today’s game where the average is just under 3 GF/GP.
The seasons from 1997–98 to 2003–04 are commonly referred to as the “dead puck era” because of the dip in production. During this period of time, more physical and slower play was the name of the game, whereas today revolves around skillful and fast skaters.
Nowadays, we often debate how McDavid’s place in history will stack up against players like Crosby and even Gretzky. The question always seems to come back to whether it is even possible to compare them because the game is so different. Is McDavid’s 2020–21 (24 years old for the Oilers)1.83 points-per-game (PPG) average akin to Gretzky’s 1985–86 (also 24 years old for the Oilers) 2.69 PPG average?
Firstly, note that we are now talking about PPG, which includes goals and assists. As a very rough test, we can first ask whether there is a significant difference between GF/GP between their respective seasons (unfortunately our data set does not contain season PPG numbers but GF/GP should still capture the idea). This does not compare the players directly but instead asks generally about scoring across the league at these time points. The league average was 3.97 GF/GP in 1985–86, which was significantly higher (pvalue = 1.195e-09) than the current 2.88 GF/GP of the 2020–2021 season. Naively, if we adjust for this ~27% decrease in scoring, then Gretzky’s PPG would sit around 2. Still higher than McDavid’s but closer than it might appear just looking at the stat lines!
However, it should be restated that this is not a thorough test by any stretch of the imagination. We can see from the black dots that the Oilers had an extremely high GF/GP during the Gretzky Era, largely driven by him. Likewise, McDavid boosts the current GF/GP average, but likely to a lesser extent. Simply adjusting for percent change does not capture all the complexities, so perhaps this something to be explored in a future article.
Shots on goal (1959–2021)
Next we take a look at SOG to see if volume of shots can in part explain the fluctuation in goals. Interestingly, when we compare SOG over time, the season averages are relatively constant. There were a few years during the late 90s and early 2000s where SOG dipped and correlated with a decrease in GF/GP, however, team SOG/GP averages hovered around 32 for the most part.
What this would suggest is that SOG, certainly when considering them in isolation, do not explain the variability in GF/GP. It is worth noting that here we are just talking about SOG, not shot attempts (SA) which includes off-target shots, hit posts and blocked shots. It would be interesting to see how SA and defensive stats, like shot blocks, have changed over time and have perhaps leveled out SOG.
Save percentage (1959–2021)
A major topic of debate is how goaltending has changed since the beginning of the NHL. Often times people will argue that today’s goalies are better, bigger and have improved tactics and gear. As a result, people claim it must be harder to score in today’s game.
Unfortunately, NHL.com/stats does have this data from as early as 1917, but we are still able to look as far back as 1959. Initially, Sv% was ~91%, then there was a steep decrease from the 1970’s to mid 1980s. In other words, a higher percentage of shots were finding the back of the net. Again this overlaps with the heart of Gretzky’s career, but we can also see that his team, the Oilers, held some of the lowest recorded Sv% in the history of the NHL. As the saying goes, a good offense is the best defense…
Since these are team averages over the course of a season, a single player should not have too much influence on the trend, however, Gretzky undoubtably contributed to the drop in Sv%! Over the next 10 years, Sv% climbed its way back up to ~92% where it still hovers around today, although the tail end of the graph indicates this is beginning to drop again.
Power play percentage (1977–2021)
Finally, here we just compare PP% over time, beginning in 1977. Aside from a slight decline during the 1990s from ~20% to ~16%, PP% has been been more or less constant. Interestingly, the decline in the 1990s also overlapped with a local decrease in GF/GP and SOG, but increase in Sv%.
Every season there are teams that really struggle with converting on power plays and others sitting around 30%. For example, in the 2020–21 season, the Ducks sit last in the league at an abysmal 9.2%, while the Oilers are again at the top with 27.2%. As we have seen before, PP% is a major predictor for game outcomes!
Conclusion
The NHL today is a very different league than it used to be. Looking across 100+ seasons of data, we observed that GF/GP has fluctuated over time, while other stats, like SOG and PP%, have been relatively constant. There exist many speculations and explanations as to why metrics might vary or stagnate. Some point to the bigger and better goalies, others point towards improved strategy and more defensive game plans, or to more penalty minutes, or even to the spreading of talent across the league as it continues to grow in size. The list goes on and on. Any thoughts are welcome below!
This dataset and analysis was solely focused on basic regular season data. Upcoming articles will take a deep dive into playoff numbers as well as other categories and advanced stats.
Code availability
Code for scraping the data can be found here. Code for cleaning the data and creating the visualizations can be found here.
