Advanced Hockey Stats 101: Zone Starts (Part 4 of 4)
Here we take a closer look at the last of the main advanced stats, zone starts. This also serves as a primer for future articles delving into the validity and usefulness of hockey stats to make machine learning-based predictions.
Agenda
- Zone starts definition & theory
- Season-average association of player zone starts with points
- Season-average association of team zone starts with points
- Game-by-game association of team zone starts with wins
Data
2019–2020 season total player data was downloaded from hockey-reference.com. 2018–2019 home game data was scraped from nhl.com/stats. A tutorial on how do to this in R can be found here. Afterwards, I applied some basic cleaning and filtering, for example, keeping players with > 20 games played.
Zone Starts
As the name suggests, zone starts simply refers to the number of faceoffs in each end of the ice and not in the neutral zone. Note, these stats consider the number of faceoffs, not the number of wins. There are two main categories:
oZS% = offensive zone faceoffs / (offensive zone faceoffs + defensive zone faceoffs)
dZS% = defensive zone faceoffs / (offensive zone faceoffs + defensive zone faceoffs)
A team has an offensive zone faceoff if the opposing goalie covers the puck, opposing team ices the puck, gives up a penalty, etc.. Therefore, zone starts are theoretically a measure of offensive pressure, however, they are imperfect because a puck starting in one end of the ice doesn’t mean it stays there.
Zone starts can serve as a measure of player deployment since a team is often allowed to make a line change prior to the faceoff. For example, if one team ices the puck, then the other team is awarded an offensive faceoff and can make a change, while the defensive team cannot. Sometimes a specific line is better suited for a given situation so the coaches act accordingly. As a result, a player can tally an offensive, or defensive, faceoff even if he played no part in getting the puck to a given zone. Unfortunately, this does confound using zone starts as a measure of offensive pressure at the individual skater level.
Average player zone starts over the 2019–2020 season
The distribution of player zone starts from the 2019–2020 season is left-skewed with a mean just below 50. The grey lines demarcate +- 2 standard deviations from the mean, indicating the large majority of skaters had an oZS% between 35-63%.
Below, we see the top and bottom 10 skaters by oZS%. Torey Krug tops this list with a staggering 68.4 oZS%, meaning when he was on the ice (or put on the ice), far more faceoffs were in his opponent’s end than his own. Jay Beagle and Tyler Motte of the Canucks are at the bottom of the list, indicating most of their faceoffs were in their own end. However, this is not necessarily indicative of their play and/or lack of offensive pressure. Perhaps they were frequently deployed in the case of a defensive zone faceoff.
How do player zone starts relate to individual points?
Below, we visualize points per game versus oZS%. Labeled are the top 10 forwards by oZS% + Draisaitl, McDavid and Panarin (top 3 by points per game).
There is a lot of variation, however, forwards with a higher oZS% generally have higher points per game averages. This makes sense under the notion that more offensive faceoffs = more pressure = more scoring opportunities. Draisaitl is again an interesting case because he had a relatively low oZS% for his league-leading point per game average.
One confounding factor is that this analysis used points per game which is not limited to even-strength play. Although, when we use EV points per game, we see a similar overall trend:
Finally, this relationship is much weaker when considering defensemen. Scoring is not their primary role and thus points per game is not the best measure of their play. Some defensemen, like Andreas Englund, had 3 points over 24 games despite having a 65.3 oZS%. Other defensemen with a large offensive upside, like Cale Makar and Torey Krug, had both high oZS% and points per game averages.
Season total team points versus zone starts
While there is a positive correlation between oZS% and individual points, to determine if zone starts has much use for predicting game outcomes or standings, we have to consider team points. Below, we see each team’s point total after the 82-game 2018–2019 season versus oZS%.
For the most part, teams with higher oZS% had higher point totals. One standout was the Lightning who ended the season with a staggering 128 points but an oZS% < 50. They won a record-tying 62 games in the 2018–2019 season, which goes to show that oZS% does not paint a full picture of team performance. Finally, it is also worth noting that the range of oZS% by the end of the season is quite narrow as the majority of teams are within ~5% of each other.
Game-by-game association of zone starts and wins
Now, to evaluate how zone starts relates to individual game outcomes, we will consider each home game of the 2018–2019 regular season.
Surprisingly, home teams that lost in regulation (0 points), tended to have an oZS% > 50. Therefore, despite having a majority of offensive faceoffs, more goals were allowed than scored, and vice versa for the away team. Similarly, a team that lost in overtime (1 point) tended to have an oZS% > 50, while a team that won (2 points) tended to have an oZS% < 50. Perhaps the most interesting finding is that there are several instances in which a team lost despite having an oZS% close to 80. The opposite cases also exist where teams won despite having an oZS% of 20.
Overall, these distributions are fairly wide and share considerable overlap. The results are also counter-intuitive and contradictory to the season-total outcomes, so this game-by-game analysis may be capturing a spurious relationship. One explanation for this inconsistent relationship is that the former analysis considered each team individually while the latter considered all teams combined. Teams are not created equally so averaged effects may be heavily skewed. Another major consideration is that zone starts only incorporates even-strength faceoffs whereas a lot of point production occurs on the power play. Additionally, the sample sizes of faceoffs during individual games are small, so these associations are more influenced by puck luck and game-to-game variation.
oZS% on win percentage
Based on the violin plots above, I next asked how oZS% was related to win percentage. To do this, I first grouped game oZS% into 15 equally sized bins and averaged the number of wins. Interestingly, as the proportion of offensive faceoffs increases, the win percentage decreases. The first bin contains games in which the home team had an oZS% ~32. Despite having more defensive faceoffs, the home team actually won ~80% of those games. At the other end, teams with an oZS% of ~70 had a game win percentage < 35%.
This figure alone suggests that an average team is more likely to win if they have more defensive zone faceoffs than offensive zone faceoffs. However, a closer look revealed that teams were not evenly distributed across the bins. For example, the Calgary Flames occupied bin 15 (highest oZS%) 9 times and lost all 9 times, whereas the Vegas Golden won 5/6 games within this bin. On the flip side, teams like the New Jersey Devils occupied bin 1 (lowest oZS%) and won 5/6 times. This led me to believe that the correlation between zone faceoffs and wins was driven by a subset of teams.
To study this, I again binned games by oZS%, now into 5 groups to accommodate smaller sample sizes. Below, the size of circle corresponds to the number of games within a given bin. Each of these subsets only contained the 41 home games for each team since I used the same data from the previous two figures.
Generally, the same pattern emerges in which home teams won the majority of games when they had a smaller fraction of oZS%, and lost the majority of games when they had a larger fraction of oZS%. Similarly to what was captured by the violin plots, the team-based line plots also show a large amount of fluctuation in the middle bins where the oZS% was close to 50.
A few exceptions to the overall trend include the Jets, Panthers, Lightning and Golden Knights. Interestingly, the Devils had very few games in which they had a high oZS% (small circle in bin 5) and showed no pattern with win percentage.
Conclusion
Zone starts is another theoretical measures of offensive pressure and player deployment, however, as we have discussed, it is imperfect because a puck starting in one end of the ice doesn’t guarantee possession and production. Across an entire season, we observed that a higher oZS% correlated with more player and team points. However, when evaluating oZS% on a game-by-game basis, we found that winning teams, more often than not, had a low oZS%. This was influenced by small sample sizes and team-effects, so oZS% should be used very cautiously when making game predictions.
On its own, zone starts don’t appear to be overly informative. Perhaps when incorporating additional variables, like faceoff wins, it would become more useful since there are several other factors that influence whether an offensive faceoff actually leads to a goal.
