I have always liked statistics. I have always believed in analytical approaches to measure performance and I am a full-time supporter of the analytical thinking in sports. The goal of this statistic, then, is clear: have the most realistic and objective picture of what is happening in the field of play, avoiding (to a certain extent) any subjective opinions or conclusions based just on the eye-test.
Statistics have been recorded for decades, since a long time ago, for practically every sport on earth. The problem with those statistics — and numbers in general — is that they can be hard to understand for some and can turn into complex metrics that are not usually presented to the reader in the best possible way. That is the point where statistics (say, numbers) and visualizations (say, the way of presenting them) must meet each other in order to facilitate the understanding of the former.
On top of that, multiple metrics (much more in the case of hockey, or so it seems) can be hard to understand just because of the name they were once assigned. Corsi, Fenwick, PDO,… Those are just some of the most common and used ones in the field but even with that their names are almost not related at all with what they are meant to represent and can be found with different names depending on the source making use of them. This can turn into a problem for the casual fan that wants to delve into the analytical side of the sport, even more if his background is that created from watching game broadcasts in which terms such as “playmaker”, “iron man”, “defensive defenseman” or “possession player” are often used but have nothing to do with the numbers generated by those already mentioned metrics. This is why new ways of looking at players’ performance must be created and properly presented in order to make it easier to understand what is truly happening on the ice and how to measure each player’s abilities.
One of the most interesting approaches I’ve found around this idea is that of Nick Abe, from XtraHockeyStats, who developed a rating system in which he uses his own metric (eGF, expected goals for) to quantify the ability of players in multiple categories. By grouping statistics created around the idea of the eGF/eGA (always accounting only for even-strength 5v5 play) he came up with ten different ratings for each player, which can be found next (a detailed explanation of the calculation of the ratings can be found on Nick’s blog):
- Offensive Possession: A player’s ability to generate offensive chances for his team. Based on the player’s generation of eGF relative to line mates and relative to the league.
- Offensive Awareness: A player’s ability to convert chances into goals. The less chances a player needs to score, the highest this rating would be for him. Based on the player’s GF being higher (or lower) than his eGF relative to line mates and relative to the league.
- Offensive Overall: A player’s overall offensive ability, weighted both Off. Possession and Off. Awareness.
- Defensive Possession: A player’s ability to limit opponent’s chances. Based on the player’s ability to reduce the eGA relative to line mates and relative to the league.
- Defensive Awareness: A player’s ability to limit opponent’s chances from turning into goals. Based on the player’s ability to generate a higher (or lower) GA than his eGA relative to line mates and relative to the league.
- Defensive Overall: A player’s overall defensive ability, weighted both Def. Possession and Def. Awareness.
- Goal Scoring: A player’s ability to score goals. Based on goals per 20 minutes and IPP (individual point percentage).
- Passing: A player’s ability to generate offense with passing. Based on assists per 20 minutes and IPP.
- Durability: The number of minutes a player played over the past 3 seasons. It works as a quick way to determine the reliability of the player’s rankings.
- Overall: A player’s overall rating, a weighted (with an emphasis on offense) combination of Offense/Defense/Durability. It shouldn’t be understood as a catch-all statistic, as it measures the overall ability of players and therefore some with a tremendous offense but horrendous defense would be rated lower than a more above-average balanced played.
It should be clarified that players with less than three seasons of data available (i.e. rookies at the end of the past season, the 2016/17 one) have their ratings penalized in comparison with those of players having played the prior two (still penalized) or three seasons.
As introduced on the Durability rating explanation, ratings come from weighting the past three seasons of play of the player at hand. This comes with two advantages. First, while analysing a single snapshot of a player’s performance (say, that from 2014/15 to 2016/17) we look at how his game has been not only in a reduced period of time (a single season) but instead in a multi-season environment that helps us know the trends his game is following and what to expect from it in the future. Second, while analysing the evolution of a player’s abilities, comparing multiple snapshots of his performance over time it is easy to know if the player has improved or worsen his game a lot because big jumps in the value of any of the used ratings are reduced from the fact that each snapshot contains information from three seasons that are shared along the evolutionary line of the player’s performance. This means that for a rating to experience a big jump (either up or down), the player must have improved or worsened the abilities linked to that ratings a lot. All in all, these ratings are useful predictors of future performance.
The ratings have a mean of 70 and a standard deviation of 5. As a rule of the thumb, the following categories can be generated and kept in mind when looking at the ratings in order to quantify the abilities of each player and trying to come with a relative position where they should be allocated in any team’s lineup:
- 60: Poor quality. Fringe NHL level.
- 65: 4th-Line Forward / 4th-Pair Defenseman
- 70: 3rd-Line Forward / 3rd-Pair Defenseman
- 73: 2nd-Line Forward / 2nd-Pair Defenseman
- 76: 1st-Line Forward / 1st-Pair Defenseman
- 80: Elite level
* * *
With the ratings and their meaning understood it was the time to come up with a way to present them better than a table, which can be hard to comprehend and extract knowledge from. Even applying some simple visualisation techniques such as colorising the cells from red to green in order to highlight the best and worst abilities of each player, it is hard to reach conclusions from it. The following table includes every player from the 2016/17 Pittsburgh Penguins roster in alphabetical order. Cell colours go from red (rating of 60 or less) to white (rating of 70) to green (rating of 80 or more).
Even applying a little visual aid, it is practically impossible to get an idea of where each player stands or how two players compare. This is why I decided to create the Skater Rating Charts (SRC) and the Skater Rating Evolutions (SRE). By applying simple visualisation techniques and employing two common plots in the shape of radar and line charts it becomes much easier to comprehend how players abilities compares between them, in which areas a concrete player have the edge, and how the career of one or two players have developed over time in terms of their overall rating.
Expanding on the already presented table of data, we can generate a few charts in order to see which are the best and worst abilities of Pittsburgh’s first line of forwards (based on the last Stanley Cup Final), that comprised of Jake Guentzel, Sidney Crosby and Bryan Rust. As an important note, keep in mind that the scale of the chart varies from one to another in order to generate meaningful shapes, so don’t compare how big those shapes are (they all will reach the border for some rating), but rather the actual shape of the radar figure.
The first thing to clarify is how the chart is organised in terms of each rating position. At the top the overall rating for the player is presented. The closer to the top, the best the player is. The axis that follow the Overall rating to the left and to the right is related to the two other overall values of the model, the Defense Overall and the Offense Overall. The components of those overall ratings follow them both at their left and their right. This makes it easy to know if a player is good on defense (left side), offense (right side) or both defense and offense by just taking a quick look at the chart. Keeping this order in mind, the Goal Scoring and Passing Ability ratings follow those related with the offensive side of the game clock-wise. Finally, the Durability rating closes the chart.
Now looking at the players introduced earlier and looking at their SRCs it is not hard to infer that all of them are forwards given that their shapes are all right-sided. It is also get to the conclusion that both Guentzel and Crosby are the more complete offensive players, while Rust falls a little short in terms of Offensive Awareness, while he has some more defense-related abilities than the other two. Another interesting point to look at is the Durability rating. We all know Crosby is a seasoned veteran that has been Pittsburgh’s face for quite some time, thus his elite rating for that category. Guentzel and Rust, on the other hand, have played for the first and second time respectively during the past season. That puts them on a similar, albeit still low, Durability rating (Guentzel makes for the different by having a higher TOI accrued in one season than Rust in two).
In order to truly highlight the difference in ability between players of the same or different teams, SRCs can be tweaked to include two shapes instead of one. This makes it easier to compare the abilities of different skaters by using the same scale. If we compare Crosby’s shape with that of Rust we get the following, easy to read, chart.
It is hard not to know who is carrying who in this line. We already knew Rust was a limited player in terms of his abilities, but this clear any doubt about the difference between a perennial MVP-caliber player and a depth guy.
* * *
The next logical step once I created these SRCs was to work on some other type of chart helpful at the task of representing the evolution of the careers of different players in terms of their overall rating, trying to depict tendencies in order to discover regression or progression in the development of young players or recognise declining arcs during the final years of a player’s career. This lends to the creation of the Skater Rating Evolutions (SRE).
By using a simple line chart it is easy to compare the evolution of two players’ career in a quick to understand way. Let’s take a look at the careers of Patrick Marleau and Joe Thornton, life-long and longtime San Jose Sharks.
Both drafted in the late 90s, we are able to track their careers back to the 2008/09 season, the first from which we have data. During the first three three-season weighted snapshots there is not much difference between them, but things started to go Thornton’s way from the 2012/14 point on. The overall ability of Joe jumped from 77 to 81 while Marleau started a slow but steady regression that sunk completely by 2016 to later resurge thanks to his last-season production. Thornton, on the other hand, kept improving his game up to a maximum of an elite level 83 rating two seasons ago to regress hard during this past one, although still being ahead (between elite and 1st-line compared to 2nd-line by our rule of the thumb) of Marleau.
Although not shown in this introduction, another potential use of the SRE charts is related to the analysis of the evolution of different ratings for the same player. While the most obvious use is to compare careers in terms of Overall rating, the fact that those charts can highlight improvement or decline in different abilities can paint a clear picture of how a player is changing his playing style (i.g., an unidimensional shooter starting to approach the game from a more passing-driven perspective).
* * *
This concludes the introduction of both SRC and SRE, two new visualization tools aimed at improving the understanding of the sport of hockey and the performance of its players. Now that the model and the charts have been introduced, I’ll start making use of them in multiple articles here at Holyfield, where they will help improve the written content of the pieces and ease the understanding of certain angles and decisions made by NHL teams during the past and most of all the exciting future ahead.