With October officially in the rearview mirror, it’s time to begin our evaluation of the Philadelphia Flyers and their play so far in 2016-17. Ten games is around the point of the year when it’s possible to get a read on whether a team is a potential contender or in need of dramatic improvements to even have a shot at the playoffs, and the Flyers are now right at that mark.
Their win-loss record doesn’t provide a ton of clarity. At 4-5-1, Philadelphia is a sub-0.500 hockey team, but that’s not “bad” enough to qualify as a truly disastrous start. They’re not in the basement of the Metropolitan Division or the Eastern Conference, but they’re certainly not close to pressuring the true front-runners either.
So let’s break down each element of the Flyers’ performance thus far. Most would agree that the team’s power play has not been a problem — they currently rank third in the NHL in goal-based efficiency and are in the top-half of the league in all of the shot generation metrics. By the same token, few would argue that the penalty kill has been especially impressive, as it ranks 24th in the league and is right around middle-of-the-road in shot prevention.
Then, there’s the goaltending, which has clearly not delivered strong results as of yet. All that can be said about Steve Mason and Michal Neuvirth right now is that their true talent levels are almost certainly higher than they’ve showed so far. Whether one or both of the Philadelphia netminders actually does bounce back remains to be seen, but considering their track records, the most likely scenario is a return back to the norms of the past few seasons.
That leaves the Flyers’ performance at even strength, specifically during 5-on-5 situations. Since the majority of each game is played with teams on equal footing, evaluating 5v5 play tends to be the most complete way to measure the effectiveness of a roster. But not all evaluations will lead to the same conclusions.
There are two legitimate schools of thought regarding the Flyers as a 5v5 team this season. Some believe that Philadelphia has largely carried play against their opponents so far, and has been let down by bad luck and subpar goaltending. Others hold that while the team may be controlling the games for large stretches, repeated defensive breakdowns have resulted in Philadelphia regularly losing the quality chance battle on a nightly basis, causing poor results.
As it turns out, both sides make strong points. The Flyers have driven play at 5v5 through ten games, when looking at pure volume, but they also are struggling to beat their opponents in quality shot creation. That leads to legitimate questions when trying to project Philadelphia’s future performance at 5-on-5. Will the disparity between their shot volume and shot quality continue through the final 72 games of the season? Are their volume numbers likely to move down to match their poor performance in chances? Or can we expect the Flyers to begin winning the quality battle over the remainder of the season and move in line with their tendency to outshoot the opposition on the whole?
Understanding Corsi and Expected Goals
The driving force behind the bulk of “advanced” metrics in hockey is the belief that shot creation and shot prevention is a better way to evaluate teams and players than on-ice goals. While winning the ultimate battle on the scoreboard is obviously is the primary focus of every team, goals are relatively infrequent events. A six-goal performance in one game does not mean that a team is now an offensive powerhouse — there are a number of unsustainable reasons for such a performance. Maybe a role player scored on all six of his shots on goal. Maybe the opposing goaltender had the worst game of his career. That doesn’t erase the game, but those causes do not make it likely that the team will repeat the performance.
This is why the analytics community has leaned toward overall shots as a way to measure team and player quality. Shots occur far more often than goals, so counting the total number created and allowed by a team results in a far larger sample than simply using the instances when a shot actually makes it into the back of the net. It makes sense intuitively, as well — teams that outshoot their opponents tend to win more games than those who are constantly in the defensive zone.
But the biggest reason why the community gravitated toward shots as their main evaluation tool is because they have proven to be a better predictor of future outcomes than goals. Interestingly enough, past goal differential itself is not a particularly good predictor of future goal differential. If you want to guess if a team will outscore opponents in the future, look at whether they’ve outshot their opponents so far that season, not if they’ve outscored them. The team with a strong shot differential but a poor goal differential is more likely to post more goals than their opponents than another squad high in the plus column in goals but getting way outshot.
That’s the concept behind Corsi, which adds up all shot attempts (shots on goal, unblocked attempts and blocked attempts) and turns it into a differential. For example, if a team generates six attempts in a period and their opponent only four, the team would be said to have a Corsi For percentage of 60%. There have been further advances in improving Corsi — such as adjusting the metric based on team performance in different score scenarios — but that’s really the gist of the stat.
And by Corsi so far this year, the Flyers look like a pretty solid team. During 5v5 situations, they’ve generated 51.32% of the total shot attempts, after adjusting for score. That ranks them 10th in the NHL, and 6th in the Eastern Conference. We’re not talking about a world-beating squad, but at least at 5-on-5, that’s the performance of a playoff-bound team.
But Corsi certainly has its detractors. The most common criticism of the stat is that not all shot attempts are created equal — some are inherently more dangerous than others. Corsi assumes all attempts, from the one-timer in the slot to the flip shot from the point, to be of equal value. As a result, some have created “weighted” versions of Corsi, which adjust each attempt for perceived quality.
That’s the thought process behind Expected Goals. Using the shot type and location data tracked by the NHL, the analytics community has created a number of models that approximate the goal likelihood of each shot occurring on a team’s watch. For the purposes of this analysis, we’ll use the most widely-cited public version of the stat — the xG available on Corsica.Hockey, currently the premier analytics site on the internet.
By this metric, the Flyers are far less impressive. Again using the score-adjusted version, Philadelphia comes in at a 47.27% Expected Goals For percentage at 5v5, far below their solid Corsi ranking. In fact, they currently sit 24th in the NHL in this stat, with only the Columbus Blue Jackets ranking below them in the East.
Philadelphia’s strong Corsi backs up the argument that the team has carried play this season, spending lots of time in the offensive zone and blasting away at opposing netminders. But their poor xG speaks to the legitimate criticism — that the Flyers may be generating more sheer volume than their opponents, but other teams are creating enough high-quality scoring chances (and preventing enough Philadelphia chances as well) to overcome the volume deficit.
But which is more important for the future?
Both Corsi and Expected Goals tell us what happened at 5v5 over the past ten games for the Philadelphia Flyers. But the real question is what the discrepancy between the two stats means for the rest of the season. After all, 4.05 percentage points is a pretty large gap — only nine NHL teams between 2011-2015 had a greater disparity between a strong 5v5 Corsi performance in October versus a poor one in Expected Goals.
Some would argue that winning the overall shot attempts battle (Corsi) bodes well for the Flyers moving forward. After all, it means they’ve carried play, right? Once they clean up the defensive breakdowns and the quality of shots evens itself out, Philadelphia can continue with their formula of outshooting the opposition and ride that to more wins.
Not so fast, says the other side. Who’s to say that the Flyers will ever clean up the issues that have led to their early scoring chance deficit? Winning the raw shots battle is meaningless if opponents’ higher quality chances provide more overall value. And what if past xG is a better predictor of future shot volume outcomes than the past shot volume outcomes themselves? Maybe their Corsi will trend down over the rest of the season to match their poor xG.
This is when it helps to look at how these situations have played out in the past. Luckily, Corsica.Hockey has full-season data going all the way back to the 2007-2008 season, and we can use that to see which of October Corsi and October xG has proven to be a better predictor of future outcomes. That should tell us which of the two statistics we should take more seriously at this stage of the season.
What is a better predictor of future xG — past xG or past Corsi?
So far this year, the Flyers have a solid Corsi, and a poor xG. Obviously, if they want to take pressure off their goalies, they want to improve their performance in terms of Expected Goals, which in turn should lead to a better 5v5 goal differential and then more wins. But does their poor October xG make it unlikely that their rest-of-season xG will improve? Or does their relatively strong October Corsi imply that better xG outcomes are right around the corner?
To answer this, I looked at the performance of every team since 2007-08 in October from a 5v5 Corsi and xG standpoint (removing the lockout year of 2013-14 since no October games were played), and compared it to their results over the rest of the season. By measuring the strength of the relationship between past performance and future outcomes, we can see which metric is more “predictive” of what will happen moving forward. We do that by placing two sets of data in a scatterplot graph, and measuring the relationship (or R-squared) between the two. The stronger the relationship, the more predictive it is.
We’ll start with comparing the October 5v5 Expected Goals performance of all teams since 07/08 with those same teams’ performance in the metric through the remainder of the year. Our result is an R-squared value of 0.257 — not insignificant, but not ridiculously high.
Now, let’s substitute October Corsi For percentage for October xG%, and measure its relationship to future xG%. If the r-squared is higher here, then October Corsi For has done a better job of predicting rest-of-season xG than actual October xG.
As you can see, the relationship is much stronger here. This is a big point in the Flyers’ favor — in the past, October Corsi For% (where the Flyers have performed well) has predicted rest-of-season xG% better than October xG% (where the Flyers performed poorly). This bodes well for the possibility that Philadelphia’s xG performance will creep up over the remainder of the season to fall more in line with their solid October CF%.
But what about the other theory — that their CF% will sag over the remainder of the year and fall closer in line with their poor xG%? To answer that question, we’ll run the same test, except looking at rest-of-season Corsi For performance.
What predicts future CF% best?
Just like we did with xG%, let’s start by measuring the relationship between October Corsi For percentage and rest-of-season Corsi For percentage for the teams in our dataset.
An r-squared of 0.4732 is pretty strong, especially in comparison to the relationships we found while studying xG% predictability. But is it a stronger relationship than the one between October Expected Goals For% and rest-of-season Corsi For percentage?
Obviously, the answer is no. October Corsi For percentage has historically done a far better job of predicting rest-of-season CF% than October xG percentage has. This again bodes well for the Flyers — we shouldn’t expect their strong 51.32% Corsi For to regress just because their xG hasn’t been anything special.
Essentially, at the ten-game mark, if the Flyers had to choose one of Corsi or Expected Goals to be their strong suit, they chose the right one in 2016-17. Corsi is simply the more meaningful metric in terms of projecting future play-driving performance at this stage of the season.
Conclusion
The issue with evaluating the October performance of the Philadelphia Flyers during 5v5 situations is that different statistics tell different stories. By raw shot attempt measurements, like Corsi, the Flyers look like a team on the cusp of contention. But when checking weighted statistics like Expected Goals, they seem more like a basement-dweller.
So which measurement holds more weight at this point of the season? It turns out that October Corsi For performance is a better predictor than xG For% of both rest-of-season Corsi For percentage and rest-of-season Expected Goals For percentage. As a result, it’s more likely that Philadelphia’s poor xG will rise up to meet their currently solid CF% than the other way around, a reassuring thought for Flyers fans.
It’s fair to note that Corsica’s version of the xG stat is not the only one in existence. Hockey-Graphs contributor DTMAboutHeart has developed his own xG metric, and reports superior predictiveness than the one in use on Corsica. However, even that version of the stat (which is private) does not reach maximum usefulness until Game 30. In the early stages of the NHL season, Corsi For percentage is still superior.
The only remaining question is simple — why is Corsi a more useful stat than xG early in the season? My guess is that coaching comes into play here. It seems like it would be far easier for a coaching staff to clean up early shot quality issues in the offensive and defensive zones than to fix issues with driving play on the whole, which would likely necessitate a full-fledged system overhaul. Cutting down on scoring chances allowed might just be a matter of adjusting some coverage strategies on the cycle, or finding the right distribution of roles. These are relatively minor tweaks, not massive adjustments.
In any case, the Flyers find themselves on the right side of the 5v5 advanced statistical ledger in the early season. It’s now up to Dave Hakstol, his assistants, and the players themselves to extract more value out of that solid Corsi For percentage and prove the historical data correct.
