Do the Leafs start on time?

If you’ve watched Mike Babcock’s media availability, you will certainly be familiar with him bringing up the concept of ‘starting on time’. Often, he does so when the Leafs look flat to start a game, or go down a goal (or three) early, seemingly before the game has even begun. This has been a criticism levied at the team and at the coach. Putting aside whether ‘starting on time’ is on the coach or the players (or some combination of the two), this issue has been particularly notable this season. There have been 10 Leafs games in 2019/2020 where there was a first period goal. The Leafs have given up the first goal in seven of those 10 games (the only game without a first period goal was the Leafs’ home loss to the Blues — the Leafs gave up the first goal in this game too, but it came in the second period).

This has led to fans wondering how a team could so consistently fail to ‘start on time’, with some remarking that this has been a years-long issue under Babcock. The natural question is of course, do the Leafs actually do this?

The first problem we run into when trying to evaluate whether the Leafs fail to ‘start on time’ is that the phrase means different things to different people. Does it mean that the Leafs consistently give up the first goal? Does it mean that they consistently give up the first goal in the first period, or in the first five minutes, or in the first two minutes? Does it mean that their play is poor in the first five minutes of each period? Does it mean that their shot share is poor, or their expected goal share? You get the picture. There are many possible definitions, some reasonable, and some unreasonable.

In order to judge whether the Leafs start poorly, we have to actually define that concept. The most natural and simple way to look at this is who gives up the first goal. However, no one says a team ‘starts poorly’ if they give up a 1-0 goal in the second period. So let’s first restrict ourselves to looking at games in which there are first period goals.

Looking at Goals

As mentioned above, the Leafs have given up the first goal in such games 70% of the time this season (7/10). However, last season, this was not the case. The Leafs played 68 games where there was a first period goal. In 37 of those games, they scored first. This means that they gave up the first goal in such games 46% of the time last year, so they struck first more often than not. Their GF% was 55% last year, so they scored first as much as we would naively expect them to based on their overall team quality.

Can we restrict this further? What if we choose a more stringent cutoff for when the first goal is scored.

2018/2019 Maple Leafs First Goals

As it turns out, this shows that the Leafs didn’t just start on time, they hit the ground running. The more you restrict the definition of starting on time to be limited to the very start of the game, the better they look.

As you might expect, the picture is not nearly as rosy when it comes to this season. If we recreate the same table with the first 11 games of 2019/2020, there’s a different pattern.

2019/2020 Maple Leafs First Goals

This season, the Leafs have been playing from behind early. The next question is, what happened between last year and this year? The common stats answer is “well, things just happen sometimes”. There may be an underlying hockey reason why the Leafs give up more goals at the start of games this year than last year. However, this is very much a phenomenon that is particular to this season, and not some property of the Leafs (and Babcock) generally.

Looking at Shots

Now, let’s step back from goals. In hockey analysis, one reason we shy away from goal-based analysis is that goals don’t happen that often. As such, there is a huge degree of variance in any analysis that focuses exclusively on them. As alluded to earlier, another possible interpretation of ‘starting on time’ would be to control the flow of the game from puck drop. To this end, we can instead explore how the Leafs shot share evolves over the course of the first period (we maintain this restriction as the idea of ‘starting on time’ implies focus on the first period).

Rather than using counts and tables, I think it is clearer to show how the Leafs shot rates (for and against) evolve over the course of the first period, both for last season and this season. Unlike most hockey analysis, we will not restrict to a certain game state (5v5, 5v4, etc.). This is because we want to capture overall levels of team play - if there is a consistent pattern for the Leafs to draw or take penalties early in games (resulting in shot dominance or Carlyle-ing, respectively), that is a part of ‘starting on time’ and should be reflected in this analysis.

Let us first consider 2018/2019. If we plot the Leafs average (smoothed) shot rates (per 60 minutes) as a function of the time elapsed of the first period, we get the following.

The black line represents the Leafs shot rate for, and the red line represents their shot rate against. The dashed black line is the Leafs average shot rate for, and the dashed red line is the Leafs average shot rate against (once again, per 60 minutes).

As expected, the shot rate (both for and against) is low to start. As every game starts with a neutral zone face-off with rested players on both ends, this is logical—it takes some time for somebody to get out of the neutral zone and generate something. As the period progresses, the Leafs get outshot (slightly) in the few minutes of the game (on average), and then mostly outshoot their opposition until the final 2.5 minutes of the period, where their offense dries up and leaves them in the red. Shot rates for and against both increase towards the end of a period, as teams scramble to get any shots on net.

Last year, the Leafs generally outshot opponents slightly — nothing here indicates that their first periods were particularly different from that. It’s tempting to read a narrative into the pattern I just described, but in reality, shot rates are random variables and they are going to vary. This is not to say that there is no possible reason why this variance occurs, just that we have to be careful about retroactively applying a narrative once we see data.

In any case, it’s possible to look at the figure above and think “Aha! The Leafs get outshot early, as expected!” As I said off the top, a lot depends on your definition of ‘starting on time’. Unless we define that exclusively as “being outshot in the first 2.5 minutes”, the Leafs starting slowly is not an idea that is very well-supported by this analysis. As soon as we go beyond 2.5 minutes (which is obviously still very early in the game), the Leafs look much better. So the answer to the question of whether the Leafs were slow starters last year is “it depends”. Which is never an answer that will get you a lot of RTs on Twitter, but that doesn’t make it any less true.

Related

Narratives and Random Number Generators

Now, let’s turn our attention to this season. As covered above, the Leafs have conceded goals at the start of games. We can recreate the plot above using the 11 games that the Leafs have played to see if their poor early-game results go beyond goals and extend to shot differentials too.

There are a few obvious things to note. For one, this chart is far more volatile, a consequence of the smaller sample size (I kept the smoothing parameters constant across these charts, in part to illustrate this). We also see more extreme results (note the scale of the y-axis in comparison to the previous chart). Finally, the dashed red line has moved downwards, as the Leafs have better defensive results this year than last (at least, by shot rate).

Turning our attention to the actual point of the plot, we see that once again, the pace of play is slow at the very start of the game, after which the Leafs tend to outperform their opponents until midway through the period, after which it is more back and forth (notwithstanding the huge spike in shot rates at the very end of the period).  There is no real  sense of the Leafs underperforming (by this measure) in the early part of games — if anything, they’re not ending the first period on time!

Now, this is not a complete analysis. There are a few notable shortcomings, including:

1. This is only a shot-based analysis. It might look slightly different if we look at expected goals, or scoring chances, or some other measure of team performance. That said, those are typically highly correlated with shots, so I think this is a relatively minor concern. Due to the data issue with expected goals in the 2019/2020 season, I would be hesitant to glean anything from an analogous analysis using expected goal rates instead of shot rates.
2. We do not account for score in any way with the plots. The shot rates are not score adjusted, and we don’t delineate between periods where the Leafs lead, trail, or are tied. As a result, there might be an effect of the Leafs shots for being inflated and their shots against being deflated as a consequence of score effects. I did also plot the same charts with score-adjusted shot rates, and visually, they were very similar. Additionally, filtering out data based on score state would reduce a small sample of 2019/2020 data to a near-nonexistent sample.
3. This doesn’t account for brain-farts or egregious plays that lead to high quality chances against (such as Mitch Marner’s turnover on the power play in the October 21 game against Columbus).

These caveats are, in my opinion, relatively minor. As such, I am comfortable saying, based on this analysis that:

1. The Leafs did not start slowly in 2018/2019 when looking at goal-scoring.
2. Depending on your point of view, you could argue that they did fail to ‘start on time’ in 2018/2019 in terms of carrying play, as their shot share in the very early parts of games was lower than their overall shot share (though by a fairly small amount). However, this is a narrow definition. Expanding the horizon slightly, even to include the first five minutes (let alone the first 10) would change the conclusion.
3. This season, the Leafs have fallen behind early in a majority of their games, however, their shot share during the early portions of the game remains strong, and the shots-based analysis does not suggest there is a systemic cause of their poor goal results in the first period of games. Putting aside the frustration of watching your team go behind very early into a game, it’s not something I would expect to see going forward, outside of anomalous events such as Marner’s gaffe against Columbus.

* Unless the definition of starting slowly is exclusively limited to the first 2-3 minutes of a game

There is obviously a lot more that can be done here, but to keep this at a reasonable length, I’ll cut it off here. Let me know of your thoughts, and possible extensions to this work to more accurately determine if this claim holds water.

Acknowledgements

This work would not be possible without Evolving-Hockey, and their new feature allowing their Patreon supporters to download NHL play-by-play data into a CSV. If you’re interested in hockey stats and analysis, I highly recommend subscribing to them.

Thanks to Alan (@loserpoints) of Raw Charge for letting me know about Evolving-Hockey’s aforementioned features, and for his help in creating this article.