Crystal Blue Regression: Leafs, Avalanche, Ducks, Among the Most Likely to Regress in 2014


Picture taken by Sarah Connors, posted to Flickr – via Wikimedia Commons

With the Winter Classic coming up, or should I say the Winter Classics since the NHL handles marketing success like the kid who found the cookie jar, we also ring in the rough middle of the season. It’s a time for reflection, maybe a chance to re-assess your decisions, lifestyles; and if you’re analyzing the NHL, it’s the perfect time to recognize trends that may or may not continue. Also known as “regression,” here I’m dealing with a concept everyone understands to a degree; you invoke it when you see a friend sink a half-court shot in basketball and say, “Yeah, bet you can’t do that again.” The trend, supported by a history of not making half-court shots, suggests that it is unlikely for your friend to sink the half-court shot, even if they recently made one. In the NHL, possession stats like Corsi are considered better predictors of future success than stats that can be influenced more greatly by luck, like goals (and, consequently, wins), shooting percentage, or save percentage. Much like your friend and their half-court shot, there are teams that are defying their odds (established by possession measures) to succeed, which can easily happen with less than a half-year of performance.

So how do we determine who could start coming up empty on their half-court shots? Well, the easiest way is to simply compare teams’ point percentages (or the percent of games they’ve received at least a point) to their rank in possession. To add necessary complexity, though, it is worthwhile to observe a team’s PDO, or their 5v5 shooting percentage plus save percentage (usually expressed as a whole number around 1000). PDO basically helps identify where the luck might be coming from, as very, very few teams in NHL history have been able to defy regression to league averages at even-strength (usually via save percentage, not shooting percentage). To include all of these measures, I put together a table thanks to the data compiled at Darryl Metcalf’s excellent site,


Note: The first “PDO Rk” column is actually the “P% Rk” column.

The column marked “Diff” refers to the difference in rank between the team’s points percentage and possession metric; the lower, the more they appear to performing better than expectations. This differential is important – regression works in both directions, just as a team can play better than expectations, they can also play worse than expectations. I also added a latter two columns of the breakdowns of rank by shooting percentage and save percentage to further determine where our PDO is coming from. It’s worth noting that ranks, like index numbers, are pretty rudimentary, arbitrary stats devices, but for this kind of exercise they’ll do just fine. So, using the table above, I can see 8 pretty strong candidates for regression; ranked in order from less to more likely to regress, they are:

8. Montreal Canadiens

The Canadiens might be a little surprising to only fall to 8th on this list, as they are near the top of the table above, but it’s pretty clear they are benefiting from some strong play by a goaltender who is, in fact, a solid player. Carey Price might not always be able to bail this team out, but with the shooting underperforming slightly (on a team that does have the talent to shoot well), the regression might not nearly be as severe for them. Prediction: Decline slightly

7. Washington Capitals

Washington is probably the most curious team on the list, a team not aided by any element of PDO that just seems to be getting the right bounces at the moment. Of course, you can’t rely on bounces, but we’re also talking about a team that boasts prime shooting talent that could help offset possession deficiencies. Shooting percentages are not normally the best thing to rely on, but when you have one of the best snipers of the generation taking a large number of your shots (as Alex Ovechkin does), you might be able to overcome some of those odds. Prediction: Decline slightly

6. Phoenix Coyotes

 If I had to pick one team on this list I have trouble feeling strongly about, it’d probably be the ‘Yotes, if only because how do you look at that lineup anymore and say that it looks definitively bad or good? Ditto, goaltending? What’s for sure, though, is they’re overperforming their possession, and I just have trouble saying they should be considered above-average in shooting or goaltending (which is where they are currently falling in PDO). All signs point to an albeit slight correction. Prediction: Decline slightly

5. Florida Panthers

The Panthers, going into the season, looked the part of a team that is trying to build itself up. Understandably, then, you’d think they would have their struggles this year. Surprisingly, they’ve played a pretty sound possession game, and have been undermined by some pretty poor shooting luck and goaltending. While I have my reservations about the goaltending improving much (Markstrom is just not moving in the right direction this year, and Thomas is done), I do think they should gravitate to the average for shooting percentage. Prediction: Improve slightly

4. Colorado Avalanche

A couple of years ago, there was a virulent battle between Avs fans and the stats crowd for this very reason; the end result was a victory for statistics, insofar as you can consider a bevy of Colorado faithful hating stats folk a “victory.” The Avalanche have the bad confluence of defying possession and PDO measures, and in a tough Western Conference you’ll eventually have to face the music. In fact, Colorado has already felt some of the sting, with tough games for Varlamov and Giguere exposing a weak underbelly. That being said, their forward shooting talent is real and could keep them in the playoff hunt. Prediction: Decline slightly

3. Toronto Maple Leafs

The stats-inclined Leafs fans have been dreading this moment for the last couple of months, but the fact of the matter is that this team is grabbing wins but not passing the smell test. Kessel and surprisingly-good goaltending out of tandem James Reimer and Jonathan Bernier have brought them this far, but I’m not about to assume either goalie is a top 10 goaltender yet and the forward depth does not impress me. It could get ugly for Nonis, Carlyle, & Co. Prediction: Decline substantially

2. Anaheim Ducks

The reality for the Ducks is not that they’re a bad team; it’s just that they aren’t a #1 team. I couldn’t think of any clearer message than their defeat on December 29th at the hands of the Sharks, a truly good team, where the Ducks simply did not look in-control of the game. What’s more, they hold the best shooting percentage in the league at 5v5, and I might buy a slightly-above average shooting team, but not more. Their possession figures also suggest a team that should be in playoff spots 5-8 rather than 1-4. Ducks fans should be nervous about that goaltending carousel as well. Prediction: Decline substantially

1. New Jersey Devils

Talk about personnel shakeups: the Devils lost one of the most complete players in their history two years ago, and then lost one of the league’s most prolific goal-scorers this century the following year. Maybe that’s why they didn’t want to turn their backs on Marty Brodeur so early in the season, but the truth is that Cory Schneider is the better goaltender, he was brought in to assume the starting role, and they need to make the smart choice if they want to get into the playoffs. Outside the net, the Devils are playing an incredible game, with Elias and Jagr playing like they just turned 30, and Zajac and Greene turning in monumental efforts. The stars haven’t aligned for their shooting, and if they make the commitment and give Schneider his due, look out. Prediction: Improve substantially


Regression is a funny thing, and even above I probably make it seem too simple. The reality is, I can only project with confidence about three teams to regress strongly, because the same odds that inform my regression predictions restrain those predictions from being too drastic. I can’t quote this tweet from Pierce Cunneen (@pcunneen19) often enough (it’s actually part of a series of tweets on regression):

“we still should expect/predict that the player or goalie or team should produce at their average talent level”

In other words, people look for dramatic corrections where shooting drops to 2%, or save percentage drops to 78.5% (a real possibility in Calgary, maybe), or a team loses 10 in a row, but the fact of the matter is that regression progresses in the same way the first half of the season did: gradually. (Does that make sense? I think it does…).

3 thoughts on “Crystal Blue Regression: Leafs, Avalanche, Ducks, Among the Most Likely to Regress in 2014

  1. Poor form to be the first comment on my own post, but it’s also a fun exercise to look at the teams on that table in the middle of the pack…meaning, they are basically reaping what they sow out there.

  2. Ben,

    When looking at PDO, shouldn’t we use all even strength minutes rather than just close minutes. That causes some big changes in PDO numbers. For example, Devils have a PDO of 99.3 in all even strength minutes.

    To me that would give a better idea of their even strength luck.

  3. Pierce,
    This is a really good question…I think I just settled with score close because those were the numbers I was dealing with for the possession figures (and, I figured if score close still comprises the bulk of the minutes, the it should do fine).

    That being said, would it be better to do all even-strength versus 5v5? Bill James once said don’t compromise your sample size except for good reason, and I think that’s a good axiom. I’m okay with the 5v5 Close figures because they still represent the largest amount of minutes, and they are the most all-things-being-equal minutes, so I still stand by what’s up there.

    But, if I were to switch to all even-strength, I’d hold pretty close to the above list, though I’d drop Washington and Phoenix in favor of bringing in the New York Rangers and Carolina as improving in 2014.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s