Filed under:

# How Rolling Average Charts Help to See High Level Trends

Tough week for the Islanders, but using rolling average and probability, there are positive takeaways to explore.

Three-game losing streaks are not fun, as this week proved. The Islanders dropped two games on their Florida loop, and now head back to Brooklyn for a Tuesday game against Vancouver. But, there’s actually been some interesting data which shows the Islanders performing at a somewhat higher level than they were at the start of the season.

Yesterday, fellow contributor Travis Flynn touched on this idea by looking at expected goal rates among Isles’ centers in recent games. Within, he used a five-game rolling average to provide context to the data. This is a really good way to look at trends, using sample sizes that are not small enough to be too noisy, but not large enough where they define an entire season.

The morning after each game, I post three 10-game rolling average charts on Twitter, each of which set to provide further context and insight regarding the Islanders’ recent play. I want to use this article to go into the specific meaning of each of those charts.

Before we do that, we have to talk about what rolling averages are. It’s official definition, per Wikipedia, is a calculation to analyze data points by creating a series of averages of different subsets of the full data set. Let’s look at this somewhat visually.

So, as we can see, by taking the last ten games in repetition, we can create a moving average of how a team has trended in any facet of the game over a period of time. For the following charts, I use ten games, though there’s no real community standard. Again, the key idea is that the sample is large enough to reduce noise while being small enough to show a moving trend through time.

Chart One: Defensive Improvement, aka #TrotzWatch

The first chart looks to see how the Islanders are faring in different types of attempts against per 60 minutes of 5v5 play. It’s broken out by total attempts, unblocked attempts, shots on net, scoring chances, and high danger chances. The idea behind this chart is to show how the Islanders are faring at limiting opponent’s offensive opportunities over time, given that is the hallmark of Barry Trotz’s reputation.

So far, there has not been much improvement from the ten game increments of Game 10 through Game 16, but we can start to see the line graphs tilt downwards for overall and unblocked attempts. That’s a good thing, and shows that the team is trending - slightly - towards allowing less attempts per game.

Chart Two: Shooting Percentage

If it seemed like the Islanders were scoring at will earlier in the year, it’s because they were. At one point, their high danger shooting percentage was literally double the 2017-18 league median, so it was only fair to assume that things would start to level off. And… they have.

When people describe the Islanders as “lucky,” citing a high PDO, it’s a little bit of a misnomer. I’m guilty of it too. But in reality, It’s not that teams are lucky or unlucky by the definition of the word. Hockey, by nature, is a random sport. It’s a fast game with a limited amount of opportunities to score, which is deterred constantly by goaltending. Through all of that, teams do go through periods of hot streaks and cold streaks, and a lot of it can be pretty inexplicable. So that’s where the idea of “luck” comes in.

Basically, the Islanders got off to a supremely hot start through the first 13 games of the season, but have since expectedly regressed closer to league averages in shooting percentage, especially for higher quality opportunities. This coincides pretty much in line with the end of the Isles’ five game winning streak. Again, not a surprise considering two of the biggest correlations to winning are shooting percentage and goals.

Regression to the mean, or the idea that a sample will eventually move towards its average, is a worthwhile concept here. That doesn’t mean the Islanders will go ice cold in the coming weeks, but it’s worth monitoring as the team works through ebbs and flows of the season.

So it’s not really a matter of luck, more that water will simply find its level over time. But because it’s not predictive in terms of when that will happen, it’s been simplified down to the idea of luck.

What counteracts that? Shot attempt share.

Chart Three: Shot Attempt Share

Here’s where we can combine some learnings from the first two charts and blend them with the idea of shot attempt share and shot quality share. Looking at how the Islanders are trending, we can see clear upticks over time in all attempt types as the year has progressed. This is good, because the more shot attempt share a team has, the more opportunity they are giving themselves to score. Simply put, you have more chance to score a goal on fifteen shots than ten shots.

Since we can establish the idea of opportunity, this is also where some caution needs to be taken about the idea of all things being equal through time. For example, the Islanders were at 39.74% High Danger Chances For after 10 games. If we were to assume they would stay at that level over the course of the season, they’d basically need a goaltending and special teams miracle to make the playoffs.

Their rolling 10-game average after Game 16 is now 45.56%. Granted, this is still a really bad number - and if they stay here all season, they’d still need that goaltending & special teams miracle. But, we can see in just six games the team has improved by a quite a few percentage points. What that at least says is that the Islanders have had more opportunities to get shots on net, get scoring chances, score goals in recent games.

So why hasn’t that happened? Well, as we look above, we can blend the idea of shooting percentage with shot share and see that as the Islanders have gotten more attempts, they’ve scored less. The good news is these are not mutually exclusive ideas - the Islanders could continue to get more of the shot attempt share as they hit another hot run with regards to shooting percentage. Keeping that in mind, we can tell a fairly comprehensive high level story off the top with just these two charts.

Additionally, as the team continues to adapt to Barry Trotz’s system, there’s a chance they could continue such an upward trajectory as it pertains to shot attempt share. Of course, that’s no guarantee, but it also shows that the assumption of what a team is in a ten game sample may not be indicative of what they are in another ten game sample. So, it’s important to keep monitoring these charts because the dynamics switch constantly.

Conclusion

Rolling average charts through time help to bridge the gap between a final judgment and dynamic reassessment. They help to see improvements and setbacks, not just at a high level but specifically when in time these high level performance changes occur. And because we’re dealing with ten game samples, the lack of noise creates a smoother line making trends easier to identify from the outset. Ultimately, there are a lot of different ways to use such a chart. They can be used at a team level, as seen above, or at an individual level.

But no matter how they are used, if the sample is big enough, they can be a very effective tool in truly understanding how a team is performing at certain aspects of the game over time.

All data within the charts is pulled from Natural Stat Trick