Prior to his dismissal, consistency was a theme that now ex-coach John Hynes had been citing to explain the underperformance of his unit through November. Using statistical estimates of variance, we can provide a quantitative perspective to this claim with focus on to two primary questions:
The acquired dataset spans from the 2018-19 season through November 30th, 2019 and contains game by game GameScore and independent expected goals (ixG) data. The data has larger samples for players rostered in both seasons relative to those acquired this offseason.
Statistical estimates of variance allow us to quantify the variability or dispersion of any numeric value in a dataset. In other words, we are able to assign a numeric value describing the degree to which these values are tightly clustered around the mean or median. Take for example the below two charts showing GameScore over time. Player A, aside from 1 great game and 2 poor outings showed performances consistently around the 0-2 GameScore point range. Player B on the other hand sees much more variability and a larger range of values. One could argue that Player A is more consistent than Player B. However, one could just as easily argue that Player B reaches higher peaks of GameScore and therefore their variability may be preferred. This analyses will quantify the consistency of each player first and weigh magnitude and consistency second. I'll reveal these mystery players in just a bit.*
Who Have Been the Most (In)Consistent Devils This Year?
To answer our first question about consistency, the Median Absolute Deviation (M.A.D.) was calculated to provide a quantitative estimate of variability for both the GameScore and ixG for a player across his sample (More on M.A.D. here). The general idea is that the lower value, the less variability in performance and therefore greater estimate of consistency. This analysis could be performed with any “catch-all” metric, but I decided to use Dom’s GameScore metric since it takes into account other elements of the box score other than point production and ixG which provides an estimate of how much of a scoring threat each player was on a given night. I decided to use the Median Absolute Deviation instead of other measures like the Standard Deviation and Mean Absolute Deviation because it is considered a more “robust” estimate of variability. “Robust” is a term used to describe metrics that are more resistant to the skewing effects of outlier values. All it takes is one 4 point night from Blake Coleman on his Birthday/Thanksgiving to bump his season average sky-high and although he is not undeserving, this is an analysis on consistency - we should be making every effort to honestly portray a standard performance for each player.
The charts below show the relationship between the consistency (x axis) and magnitude (y axis) of a player's performance by GameScore (left chart) and ixG (right chart). A player's consistency is a measure of their M.A.D. value with a lower M.A.D. value indicating lower variability and therefore higher consistency.
Looking at the GameScore chart first, we can see that Kurtis Gabriel is the most consistent performer in the dataset (right-most point). It's a shame his median GameScore is actually negative meaning he was consistently a poor performer by this measure. You'll see Hughes and Simmonds in the top right quartile signifying the highest balance of GameScore consistency and magnitude - both of these players have low variablity in their performances and their median performance is roughly .4 GameScore points. Hall and Palmieri have the highest median performances but have GameScore consistency values below the median line of this dataset.
Looking at the ixG chart next, we are able to interpret one macro-level claim before diving into individual player performances. Not one Devil is in the top right quartile meaning no player is both above the median level of ixG and ixG consistency. The Devils have not had one player standout in their ability to generate dangerous scoring chances night in and night out. Perhaps John Hynes had a point. Hall, Hischier, and Simmonds had the highest median ixG scores with below median ixG consistency levels while a slew of players gathered in the bottom right quartile signifying consistently poor performances by ixG.
Weighing Both Production Consistency and Magnitude, Which Devils are Having the Best and Worst Seasons?
The second question raises an interesting perspective. Players sometimes have excellent nights where they pepper the stat sheet and achieve high GameScores and ixG values. These performances are what makes a player a star but it is unreasonable to ask even a star to consistently put up career nights on a nightly basis. That said, the players who are capable of reaching the higher GameScore and ixG values on any given night should not be penalized in this analysis for nights where they are not able to sustain above average play. To explore this perspective, we will multiply each player’s consistency value percentile by their median ixG or GS value percentile. This method will account for consistency and magnitude and theoretically tell us which Devils were able to produce at the highest level with the most consistency.
Among the best Devils by this analysis from a GameScore perspective are Hughes, Simmonds, Hall, Palmieri, and Hischier while the worst active performers are Mueller, Severson, Subban, Greene, and Rooney. From an ixG perspective, the best Devils have been Hall, Hughes, Palmieri, Wood, and Hischier while the worst active performers have been Mueller, Greene, Zajac, Rooney, and Bratt.
Special thanks to Cole Palmer (@cepvi0) who generously pulled the data used in this analysis. Cole runs hockeystatcards.com which provides Single Game Stat Cards, Nightly Skater Leaderboards, Skater Cards and so much more. I highly recommend you add HST to your advanced stats sites and toss him a follow.
*Player A is Jack Hughes and Player B is Taylor Hall