Editor’s note: I asked loserpoints, a Raw Charge writer, to take a look at the Leafs and the Caps for us using a new analysis tool. This is his guest post for us, so give him a warm welcome please. Katya Knappe
Tonight, the Leafs return to the playoffs a year earlier than many people expected. That alone is worth celebrating. But moving beyond the first round is going to require them to once again defy expectations. As the last second wild card team, they face the Washington Capitals who have been dominant for most of the season.
By now, you’ve probably seen the gap between the two teams quantified in many different ways. Just about every model has the Caps favored, with their probability of winning the series ranging anywhere from 55% all the way up to 81%. But just in case you haven’t seen enough analysis of why the Leafs are the underdog in this series, let’s check out one more!
Last week, Ryan Stimson, published an excellent new article at Hockey Graphs. In that article, Ryan introduces the idea of using clustering to define specific player types. I highly recommend you read the article if you haven’t already. But to summarize, Ryan looked at a series of statistics and identified how player performance tends to cluster across those statistics.
He included shot metrics in his analysis but went further by also incorporating the zone entry and exit tracking from Corey Sznaijder as well as passing data tracked by volunteers as well as Corey and Ryan themselves. The analysis identified four forward player types and five defender player types. Because of the metrics included in the analysis, these categories are best thought of as offensive. No defensive measures were included in the analysis.
After identifying the player types, Ryan continued the analysis to identify which player types tend to be most successful together. By doing this, we can begin to look at how to optimize lines and pairings. As the Leafs head into their series tonight, let’s see how the two teams compare in terms of player types and what that might suggest about how well each coach optimizes the players on his roster.
Since this analysis is most relevant for forwards, we’ll start there. The forward types identified in Ryan’s work are:
Playmakers - Forwards who excel in all areas.
Balanced - Forwards who perform well in all areas but not quite to the level of the playmakers and who don’t excel in one specific metric.
Shooters - Forwards who are primarily shooters and aren’t as effective in passing or contributing to zone entries and exits.
Dependent - Forwards who aren’t particularly strong in any area and rely on their linemates to drive play.
The graph below shows the recent lines for both the Leafs and the Caps on the top half of the chart. The bottom half of the chart shows how the lines would look if we used Ryan’s work to optimize the combinations.
I’ve made some assumptions on the Leafs’ rookies. Mitch Marner graded out as a shooter. However, I attribute that to the lack of entry/exit and passing data. I’ve classified him as a playmaker because I feel that’s a more accurate representation. Kasperi Kapanen does not appear in the data set. I’ve classified him as a dependent player, which I think is fair at this stage of his career. Auston Matthews also graded as a shooter. I suspect that he might evolve into a playmaker at some point but given his relatively low number of assists compared to his goal scoring, I’ve left him as a shooter for now.
In comparing the two teams, the first big difference is depth. The Leafs ice six dependent forwards every night, while the Caps have only three. And those three are spread between the bottom two lines while the Leafs have their entire fourth line comprised of dependent as well as two thirds of their second line.
The second big difference appears to be player usage. Zach Hyman doesn’t seem to be the type of player who offers much to compliment Matthews and Nylander. The second line also appears to be less than ideal with more dynamic players getting less ice time on the third line. While the two lines are fairly close in ice time, the Kadri line does consistently get more.
By contrast, the Capitals are close to a perfect lineup. The only obviously questionable decision is having TJ Oshie on the top line. Swapping him for a playmaker like Justin Williams would be an improvement according to the methodology we’re using here but given Oshie’s results in that spot this season, it’s hard to argue too strenuously in favor of that. Brett Connolly is the another player who looks a little out of place but his goal scoring this year justifies that usage.
Using Ryan’s work, Petbugs created an excel tool that allows the user to optimize each team’s lineup. With that tool, we can see how much of an improvement could be gained by optimization. For both of these teams, there isn’t too much to be gained. The forward group for the Leafs above would be expected to earn a 49% share of the expected goals. The changes reflected in the optimal scenario would bump that up to about 50.3%.
However, if Auston Matthews is considered playmaker instead of a shooter, both the current and the optimal lineup would be expected to earn a 50.4% share of the expected goals. So whether or not there is anything to be gained by these changes really depends on whether Matthews is truly a shooter or if he is a more well-rounded playmaker.
The Caps lineup is already so well optimized that making a small change like swapping Williams for Oshie doesn’t do much to improve the expected results. They are already at 53% expected goal share using their current lineup and could only gain a couple tenths of a percentage point at best.
For two teams with good coaches, we wouldn’t expect to find much room for optimization. In theory, part of the reason that both of these teams are in playoffs is because the coaches do a reasonable job of finding the best combinations to use. If they didn’t, they likely wouldn’t win may games and wouldn’t be in this position. But even acknowledging that, the Leafs do appear to at least have some opportunity to try to improve the results they see from their current lineup.
Having looked at the forwards, let’s go through the same process with the defensive pairs. The defender player types are:
All-Around - Defenders who are strong in all areas
Volume Shooter - Defenders who are primarily shooters when they get into the offensive zone
Puck Mover - Defenders who shoot less and mostly contribute to getting the puck out of the defensive zone to start breakouts
Defense First - Defenders who don’t contribute much offensively
Neither of these teams has a defender who qualifies as all-around according to the clustering method used for the analysis. Both teams have mainly volume shooters and defense-first defenders. The Leafs lineup as shown above is already optimized as well as it could be according to Ryan’s work and would be expected to earn a 50.6% share of the expected goals based on the player types. Zaitsev is included above but Marincin is also a defense-first defender so he wouldn’t change the expectations much, if at all.
The Caps defense is a different scenario. Here, we do see some room for adjustments. Two volume shooters are preferred to a shooter and a defensive oriented defender according to this analysis. Because of that, moving Kevin Shattenkirk up to the top pair would seem to be advantageous. Taking that change into consideration, the model suggests that the Caps have an opportunity to gain about 1.5% of the expected goal share by making these changes, which a significant number. Especially for such a good team.
Player types are an exciting new way to think about lineups and have the potential to help guide roster decisions. The next phase of the analysis will likely introduce defensive metrics, which will hopefully make the model better and suggesting more well-rounded lines and pairs. As the Leafs head into this series, they are clearly the underdog and looking at the player types on each roster confirms that. The Leafs have some excellent top end talent but not nearly the depth of the Caps. And in a seven game series, that difference in depth will likely prove to be the difference.
