Following the weekend’s games, only the Tampa Bay Lightning ranks ahead of the Montreal Canadiens in total goals scored among Eastern Conference teams. And yet, Montreal ranks 13th in the NHL in expected goals, a statistic that assigns a value to every unblocked shot attempt based on various characteristics, such as shot location or whether it is a rebound attempt.
The Canadiens also sit seventh in the overall standings, despite a negative expected goals differential, a negative shot attempt differential and a negative shot differential. Add to that the fact that the Canadiens have a negative goal differential at five-on-five.
In short, the Canadiens are beating the models, but will it last?
The case for: puck movement
Public models do not have access to puck movement prior to a shot. They can sometimes infer that movement through publicly available events, but it is impossible for public models to know whether a shot taken from a specific area of the ice was preceded by a pass coming from the opposite side of the offensive zone. And yet, this puck movement is exactly where some of the Canadiens’ best players excel. Whether it’s Slafkovsky, Demidov or Hutson, a core part of the Habs’ offensive creation comes from puck movement and passes that sometimes travel from one side of the offensive zone to the other.
Former New York Rangers goaltender Stephen Valiquette set out to quantify the impact of puck movement prior to a shot on the probability of scoring a goal. In his research, he found that if the puck crosses an imaginary line connecting the two nets (dividing the ice into two equal halves) before a shot, that shot is significantly more likely to result in a goal.
The line featuring Slafkovsky, Kapanen and Demidov has a 61.9% goals share, but only 46% expected goals share, according to Natural Stat Trick. Is it possible that this gap is explained by the passing abilities of the two wingers on that line? It is worth noting, however, that Slafkovsky has been on the ice for more goals against with Kapanen and Demidov than with Suzuki and Caufield, even though he logged 100 more minutes with his former linemates. In short, this style of play may also come with certain risks.
The case against: the lack of above-average finishers
Expected goals models are agnostic to shooter quality. These models assume that shots are taken by average shooters. As a result, elite shooters generally outperform their expected goals. For example, Leon Draisaitl has recorded 50-goal seasons while posting 31, 32 and 36 individual expected goals, according to Natural Stat Trick. That said, a sufficiently large sample is required to ensure that this is a real talent rather than luck.
At the moment, 12 of the Canadiens’ 15 leading goal scorers are outperforming their individual expected goals. Outside of Cole Caufield, which of these shooters truly possesses an above-average shooting talent?
The case for: above-average passers
According to research by Chris Boyle, when a shot occurs directly after a pass that crosses the imaginary line dividing the ice into two equal halves, we mentioned earlier, shooters convert on average 30% of their shots, compared to 8.5% otherwise. It is highly unlikely that Kapanen is a true-talent 18% shooter, but when flanked by Demidov and Slafkovsky, he becomes one. As such, it is possible that the Canadiens’ shooters’ overperformance is not due to luck, even if it does not reflect their true shooting talent.
The case against: one-goal games are random
This is a well-known reality in the professional sports world: games decided by a one-score margin involve a fair amount of luck. Year after year, professional sports teams that win a large share of close games try to convince themselves that it is a skill: “our bullpen is excellent,” “our players are clutch,” “we know how to protect a lead.” We hear these refrains often. And yet, these results have little predictive value from one season to the next. A team that wins many one-goal games in a given season is a prime regression candidate the following year.
The Canadiens have managed to win several of these games since the start of the season, which is why their negative five-on-five goal differential and their plus-two goal differential in all situations have not prevented them from sitting just one point off the top of the division.
The case for: overtime is not random
Many of these one-goal wins have come in overtime. The Canadiens have only 14 regulation wins. In the Eastern Conference, only the Rangers and the Blue Jackets have fewer regulation wins than Montreal. But here, are we really dealing with luck? These one-goal games involve a real skill that is unique to these situations: three-on-three overtime. Habs fans will not need to be convinced that their team is among the best in this aspect of the game. Perhaps this is, therefore, a true talent, rather than the randomness of professional sports? Or perhaps we are simply in the process of generating a new narrative similar to those mentioned earlier?
The case against: the limits of the human brain
It is always possible to convince ourselves; we do it all the time. The human brain is fallible, and the heuristics it uses to understand the world are efficient but limited. We are wired to pay attention to information that confirms what we already believe. This is why publicly available statistical models do not always attract the attention they should. We hope they will confirm our opinions, rather than using them to learn something new about the sport we are passionate about.
We are also allergic to randomness and constantly in search of explanatory patterns. This text may well be guilty of that. Sometimes, luck plays a larger role than our brain is willing to accept.
The reality is actually quite simple: statistics such as expected goals and shot attempts have greater predictive power than the traditional statistics we are naturally drawn to, such as actual goals. The good news for Montreal fans? The rolling 10-game average from Natural Stat Trick estimates that the Canadiens have finally crossed the 50% shot attempt share and 50% expected goals share thresholds since mid-December.