Assessing the Western Conference Playoff Landscape by @Oilersnerdalert

Who’s the favourite? We know the Oilers are going to play San Jose first, but in the big picture, who should the Oilers want to play most/least? Enquiring minds want to know!

There is no magical way to ‘know’ what’s going to happen in the various series, but we can look at how the season has shaped up and make some decent guesses about who to favour.  I’m going to walk through a few different methods and we’ll see what we see.

Don’t Use Standings

First, one thing that I’ll start with is to stay you can generally throw the regular season standings (or points) out the window.  There is a fairly weak relationship between points and post-season success (though it may surprise you to learn that the PresCup winner, far from being ‘cursed’, actually tends to do quite well in the post-season).

Head to Head Score Adjusted Corsi

Our first data stop will look at how the various West teams fared head to head in 5v5 score adjusted Corsi. Since the refs tend to put away their whistles in the post season, the importance of 5v5 is magnified, so this gives us a sense of how the ‘playoff’ part of their games went.

http://i.imgur.com/diHK3SO.png – source @OilersNerdAlert

From an Oilers perspective, MIN and ANA look like good matchups, while the others, especially CGY, CHI, and S.J look like tougher matches. But wait! Didn’t we go undefeated against CGY?? Not to mention STL? How is this possible?

Actually, this is the weakness of any statistical look – small sample size. In the case of CGY vs EDM, the Oilers blew out the Flames early in several games. For example, the first game of the season, the gameflow looks like this:

http://i.imgur.com/rbbg7Cn.png – source @OilersNerdAlert

The Oilers basically ran up the score early, dominated for most of the game, and then went into a defensive shell – which allowed the Flames to run up the fancystat counters, even as their loss was already written. Same story in the last game of the season, where the Oilers ran up a 5-0 edge and then coasted.

Now you might ask, isn’t score adjustment supposed to take care of that? Well – yes, sort of. It takes care of that at a statistical level. The score adjustment is done based on league wide averages.  So it works really well when you get at least 10 to 20 games of data, enough where the averages start to apply in a meaningful way.

But a score adjustment at a game level, or even at a season series of 4 or 5 games, while it will almost always push the fancystats in the right direction, won’t necessarily be reflective of that game or games. Especially if there were blowouts, as there were twice in this series.

So I wouldn’t take these numbers too seriously. That’s why I put this look first – it’s actually not that reliable IMO. It’s more for interest.

And as we’ll see later, if the Oilers face the Flames, take the Oilers all the way!

Head to Head Records

I don’t actually know if anyone has tested to see if head to head records have any predictive power for the post-season (my gut says probably not), BUT I sure do like this!

Goal Differential

I mentioned earlier that points are not that great a way to assess post-season chances. A much better a predictor is goal differential. (Read the detailed analysis here: https://www.stats.com/insights/nhl/debunking-myth-playoff-vs-regular-season-hockey/)

When we look at the teams sorted by goal differential, it gets pretty interesting:

http://i.imgur.com/KPObkSA.png – source NHL.com

Now Edmonton is starting to look more like a powerhouse than a weak sister, yes?  If Talbot Talbots and McDavid McDavids, the Oilers can beat anyone.

And of course, the weak sister in the West is in fact … Da Flames.

By the way, you might be wondering – isn’t this basically PLUS MINUS, and isn’t PLUS MINUS the pariah of the fancystats world?

Indeed, it is – at the player level. That’s for two reasons – the assignation of plus minus at the player level is extremely noisy, and because goals are such rare events, it takes multiple seasons to generate enough player sample size to overcome that noise – and by that time, your player has usually changed (situation, or even age!)

We don’t have as much noise, or as much of a sample size problem at the team level though, which is why goal differential works pretty well.

Score and Venue Adjusted [Corsi, Fenwick, Expected Goals]

Now let’s get on to some actual fancystats. I’m using Corsi, Fenwick, and corsica’s xGF.  Corsi has historically been something of a gold standard for predicting the future.  What’s interesting though is there’s an argument to be made that this relationship may be weakening as more teams pay attention to shots/possession and the resulting ‘market efficiency in action’ takes away some of the advantage historically measured by shot metrics.

We’re going to take the full seasons 5v5 measures and rank teams that way. So we’ll roll all three together to get a sense of where the teams fit:

http://i.imgur.com/OZPNHdY.png – source corsica.hockey

Ooh, that’s a bit ugly, isn’t it?  The Oilers are much weaker by this measure – ranking 7th, 2nd, and 6th out of 8.  So why such a big difference from goal differential?

Well, the easiest way to way outperform (or underperform) your underlying shot metrics is through the quality of goalering (you can also do it through special teams but I’d say that’s ‘harder’ in some sense).  So I think this really reflects the fact that Cam Talbot this season has been incredible – arguably a Top 4 or Top 5 goalie league-wide.

If the Oilers are to have success in the post-season, he’s going to have to continue his strong play. No surprise there.

San Jose looks a lot like the powerhouse that made it to the Cup Finals last year. Not going to be an easy series!

I guess the saving grace is that the Flames are still weak at 5, 8, and 8.

Tweaking the Fancystats

There’s an interesting tweak we can make to these numbers to increase their assessment/prediction capability. One of the things we know is that shot metrics in-season have their peak predictivity around 20 to 25 games, after which there is a slow decline in predictivity. Some of that is due to increasing randomness as games predicted declines.

But I think a significant part of it is also that teams change over the course of the season. Key players get hurt (or come back from injury).  Sometimes coaches change.  Teams get into a groove or fall out of one.

So we have this balance to find – we want the maximum amount of data possible, but if we use data that’s too old, it isn’t actually reflective of the team right now.

As it turns out, using the last 25 games gives adequate data volume and yet doesn’t get overloaded with games from early in the season that aren’t really indicative of a team now, producing a fairly high level of predictivity. (see for example Micah Blake McCurdy’s work on his Oscar prediction model).

So let’s look at two things – how a team did over the last 25 games of the season, and also the trend of that data, as a bit of a projection as to the direction of the teams level of play.  (Note: out of laziness, I’ve taken data for the Oilers from Feb 15th, which equates to 25 games. Other teams may be a bit more or less – ha ha, too bad for them! More seriously, it shouldn’t change the results much, if at all)

Let’s take a look.

Here’s the West teams from best to worst in SACF% over the last 25 games:

http://i.imgur.com/7ocNV8l.png – source corsica.hockey

Oooh.  Still sucks to be the Oilers on that basis though, doesn’t it?  But we also know that those 25 games started with a fairly poor stretch for the Oilers, but they’ve been coming on strong of late. And the opposite is true for the Flames. So let’s look at the trend over those games too.

http://i.imgur.com/2HcmCVp.png – source corsica.hockey (chart by @OilersNerdAlert)

Hmm, that’s encouraging, right? Despite the rather soft numbers the Oilers put up in the second half, in fact (as the eyes would suggest), the Oilers appear to be improving in a big way as they head towards the post-season. Yeah!

Cowtown on the other hand – again, as the eyes would suggest – are sliding back to Earth after the unsustainable hot streak that pulled them into the playoffs.

San Jose has solid numbers, but is neither hot nor cold.

I’ll leave you to mull over the rest of the trends.

Putting it All Together

We’ve taken a few different looks at how teams did in the regular season to get a sense of how they might fare in the playoffs.  Now, has this work given you the definitive guide to who’s going to win the West?

Ha, of course not! Statistics give you a sense of which way the probabilities lean, they are most certainly not fait accompli.

Rather, what we’ve got is some sturdy data to suggest which teams are leaning positive and which are leaning negative.

You still have to understand context though. Statistically, Anaheim is looking pretty good – but if Lindholm, Vatanen, and Fowler are out or not 100%, that’s a huge hit. Ditto San Hoser and Jumbo Joe. (In fact, one of the defining characteristics of Cup champions is that they are good and healthy when they hit the playoffs, and are still mostly healthy, or at least healthier than their opponents, by the time they get to the finals).

The Oilers meanwhile actually look pretty good, my friend!  Probably not to win it all, but I’d say we’ll be a tough out even for a legit contender.

And with McDavid and Dadbot on our side, anything can happen.

Bring it on!

Click the image above and pick up this sweet t-shirt!
Beer League Hero Written by:

I’m the Beer League Hero! I am from Camrose, Alberta but I make my home in Taipei City, Taiwan. I’ve been through the ups and downs and the highs and the Lowes, the Bonsignores and the McDavids, the Sathers and the Eakins but I’ll never leave my Oilers, no matter what!

They’re with me until the end and then some. GO OILERS GO!