Today, I want to talk about a concept I call Schedule Pace or maybe more accurately, Relative Schedule Pace.

NHL teams each have a schedule that starts and finishes at approximately the same time. What I’m after is a definition of “pace” that addresses the fact that over the course of a season, some teams inevitably run well ahead or well behind other teams in terms of the number of games played when they face each other.

To what extent does it happen, and when it happens, does it matter?

I first became interested in this concept a couple of seasons ago, when the Oilers (for reasons I can no longer remember) started the season playing a ridiculous number of games in a very short period of time. As a result, the majority of their matchups that season (almost 60 games) were against opponents who had played fewer games than the Oilers had.

Those teams sometimes had played as many as four or five fewer games than the Mighty Oil!

Was that the reason the Oilers sometimes looked like the more tired team even in “guaranteed win*” situations? Was that the reason the Oilers suffered an unusually high (even for the Oilers) injury count that season?

Maybe.

It turns out, this might have played a part, because schedule pace can make a difference.

I compared the records of home teams (data source: NHL.com) over the past four seasons between those that were disadvantaged (had played more games) vs those that were even up vs those that were significantly disadvantaged (had played >3 more games than their opponent).

Notice that I ignore OTL/SOL in this analysis. Many argue ignoring the Bettman point is the ‘right’ way to assess a team’s record. I wish I had such noble intentions … the truth is that the nature of the NHL data I use makes it easy to figure out who won a game, but it’s a massive PITA to figure out if the game went to OT or SO. Feel free to pretend I’m doing it this way because it’s right and not because I’m lazy. 🙂

In any case, if you decompose those numbers, it’s clear that almost all of the observed difference is because of significant disadvantage situations. It really doesn’t make much difference to your win chances if you’ve played a game or two more than your opponents (unsurprisingly). But if you’ve played significantly more, it’s a different story.

While the traditional way of looking at easy/hard schedules is to look at either rested vs played-the-night-before, or perhaps the difficulty of teams faced, schedule pace is one of those things behind the scene that does seem to matter at the extremes.

Being four or more games more fatigued than your competition appears to negate most of a team’s home ice advantage.

(At the end of this article, for the power geeks among you, I describe the quickie Bayesian analysis that I used to confirm that the ‘significantly disadvantaged’ numbers are likely meaningful, despite the limited sample size)

Here’s a chart that shows the relative pace numbers for all the teams for 2018-19. The brown bars are games where the team has played fewer games (has an advantage). The green ones show games where the team has played more games (is at a disadvantage). Games, where the game count are equal, are not shown:

The chart is sorted by the average disadvantage – in other words, not just the raw number of disadvantaged games, but the extent of that disadvantage. That’s a bit of a relief for Oilers fans – like last year, the Oilers have a pretty balanced pace, neither advantaged nor disadvantaged.

This year, the Panthers have the advantage of a slow-paced schedule for most of the season, while Pittsburgh faces a disadvantage.

As I noted earlier, it’s not just how many, it’s how deep.

The histograms for the three teams that give specific counts for game disadvantages are shown below. A shift to the right shows a disadvantage, while a shift to the left indicates an advantage. It’s the extremes that matter.

Notice that the Oilers’ chart is nicely balanced. Yay us!

Florida, on the other hand, has a strong bias to the left, showing that they play fewer games than most of their opponents. Almost a fifth of their season finds them with a ‘4 or more’ game advantage. In two cases, they will go into games having played six and seven fewer games than their opponent!

And then at the other end of the spectrum, our poor dethroned Cup champions play a significant proportion of games at a schedule disadvantage – five games with the dreaded ‘4 or more’. Hey, at least they have some Cups to show for it. As we’ll see in the next section, the Oilers had to deal with this scenario despite being a lottery team. Again.

And here’s what the same chart looks like back in 2015, plus the detailed histogram for the Oilers. I assume you can see what I mean about the Oilers’ brutal pace that season.

So there you have it … a look at what I call “schedule pace”.

While the traditional way of looking at easy/hard schedules is to look at either previous game rest levels or difficulty of teams faced, pace is an idea few spend time on.

Understandably because for the most part, it is not a big deal. But it seems to matter at the extremes. Being four or more games more fatigued than your competition appears to negate most of a team’s home ice advantage. Likely it works mostly the same way – probably more so – on the road.

And, thankfully, the Oilers are not at the extreme downside this year. Although one of these years, it would be cool to see us advantaged.

A bit of background on whether the observed differences we see are meaningful.

The raw data over four seasons indicates a relationship between being pace disadvantaged and having a weaker record, dropping from a win% of  54.3% when pace is equal, to 54.0% for any (1 or more games) pace disadvantage, all the way to 51.5% when looking only at significant (4 or more games) disadvantage.

To suss out whether the observed differences were meaningful, I did a quickie Bayesian analysis, where I modelled the win-loss records as a simple proportion (i.e. used a Beta distribution), and assigned a modest prior distribution of a completely breakeven season i.e. Beta(41,41). For no other reason than this prior gives a reasonably broad anchor to neutrality – I’m a believer that such priors are better than both uninformed and unjustly informed priors.

The result of running the different records through the Bayesian update process produces posterior distributions that look like this:

The home-ice advantage is clearly visible, as all of the distributions are generally shifted well north of 50%.

It’s also pretty clear that, in aggregate, there is really no meaningful difference between being disadvantaged and being neutral. There might be if we had lots more data, but not with just four years worth.

However, some of that is because the effect of the extremes is drowned. When we compare the situation between significant disadvantage and neutral, there is a noticeable difference, despite the large uncertainty caused by the low sample size.

By using the standard Bayesian technique of drawing and comparing a large number of samples (1,000,000) from each distribution, these results indicate that there is an ~80% chance that the observed difference is meaningful.

At some point, it might be worth calculating a regression to estimate the relationship and the error bars directly. For now, though … good enough!