Actually the title is a bit of a misnomer. Corsi, as we think about it, isn't exactly wrong it just doesn't tell very much of the story. One of the things that I've struggled with this season is hearing constantly about Carlyle, his crappy system, and how he's an idiot because if CF% could just go to 50% that we would be a great team. The reason I've struggled with that overall concept is not so much because I know if Carlyle's system is good or bad, but because my "eye test" of the Leafs gives a lot of players a failing grade. My personal view is that running any other system, we probably would have achieved exactly the same result.
The issue, of course, is that most people assume that CF % improvements represent a linear improvement in the team's performance. That is to say that if you moved from something like 40% CF to 45% CF that you could expect to move from 40% GF (i.e. scoring 40% of the goals) to 45% GF. The problem is that is almost sure to be false. You may receive some improvement in GF% for a given improvement in CF%, but it is very likely to be minimal.
Why is it likely to be minimal? Because you can't just generate scoring chances (and Corsi events) without taking risks and when you take risks bad things tend to happen, or put another way you tend to give up better scoring chances. The below table illustrates what happens in the 5 different 5v5 game states (via extraskater.com). -2 and +2 represents trailing and leading by 2 or more goals and all figures are time-weighted league wide averages.
As you can see from the table, when teams are trailing by 2 or more they dominate Corsi (55.6% CF versus the league average of, naturally, 50%), yet their GF% is only slightly over 50%. This is because, in order to generate those additional Corsi events, risks are taken and better opportunities to score are given to the other team. This fits with generally accepted wisdom on score effects, when a team has a lead they will turtle, giving up fewer quality chances but more chances overall. Correspondingly Both SV% and SH% are lower for the team that is down by 2 than up by 2. Now, obviously, you'd still rather have 51% GF than 50% GF, so teams are in fact better off with the massive 5.6% CF increase, but not nearly as well off as you would think. For reference, 50% GF at even strength will put you on the playoff bubble.
Now the obvious problem with this analysis is that certain teams (that are just bad) will tend to trail more, so a time weighted comparison of the numbers will have more Buffalo at -2 than Chicago at -2. Which is fair. So I took the top 5 and bottom 5 (I left out Florida in favour of Toronto, because who doesn't love looking at Leaf stats) teams and compared their game state stats.
As you can see even the mightiest teams in the league (by GF%) are subject to turtling with the lead and taking more risks when down. Except the Bruins, their improvement in CF% when down is minimal. Even the crappiest teams, like the Leafs enjoy a nice boost to CF% when trailing. Which leads into their GF% in different game states:
As you can see, for the most part, GF% seems to go up a bit (though generally less than CF%) when trailing and down when leading. The difference here compared to league averages may be due to team skill or team system. Which leads to the final table, the PDO of these teams in different game states:
|Blues||- 2.90||- 4.70||100.80||1.40||3.40|
|Blackhawks||- 4.30||1.90||100.20||1.50||- 2.50|
|Bruins||- 0.20||- 1.40||102.90||0.10||- 1.50|
|Leafs||0.90||- 2.90||101.60||0.80||- 0.80|
|Flames||5.30||- 2.00||98.50||0.10||- 1.80|
|Islanders||2.00||- 0.90||98.60||- 1.10||- 1.70|
|Oilers||- 4.30||1.00||99.80||- 1.70||2.70|
|Sabres||0.30||0.90||98.70||- 5.60||- 1.50|
As expected, the top teams generally have a much lower PDO when trailing and a higher one when leading. The Flames seem to be some kind of weird aberration with a massively higher PDO when trailing. I would say you can draw some inferences from some of these charts based on game state.
The first is that teams like St. Louis (who actually get outscored when trailing) are heavily reliant on their system and having the lead. When faced with adversity or the need to score a goal they have to take more risks and break out of their system, which leads to much worse than expected results. That is to say that Ken Hitchcock is probably doing a fantastic job in St. Louis and when they start taking risks and not using his system, they aren't a very good team.
The second is for teams like Buffalo who perform dramatically better when trailing than leading (much better than league average), they are probably not employing a system that is optimal for that team. If they used a more run and gun type approach they would improve from "godawful" to "terrible". The Leafs, on the other hand, are pretty inconsistent across game states and that's probably due to the fact that their players are also inconsistent. They lose a lot of 1v1 battles, which is the basis of consistent play, but they have incredible speed and can pounce on turnovers by the other team, so if the other team doesn't make turnovers they get shut down.
The final inference is that Corsi (or shots for/against or fenwick or whatever you want to use) may be a good approximation of "possession" but it doesn't do enough to describe quality of possession or quality of chances given (or taken). On the whole, there is strong evidence to indicate that for a given skill level you can only increase CF% by reducing PDO. While this may still lead to net gains in GF% the effect is much more moderate than you would think. In the case of the Leafs, if you average their -2 and -1 game states you get a CF of 48.3% and a GF of 48.85%, an improvement of 7.3% on their tied CF but just 1.15% on their tied GF. They would certainly be "better" but would still be on the outside looking in for the playoffs.