At any rate, Tyler is a guy who not only asks good questions about our sport, but then has the gall to answer them with statistical research. I am very fascinated with this sort of game, although when it comes to actually doing anything, I am a hack.
(Just like my HTML skills. Good thing the team color is turning orange; it would be easier than me trying to figure out how to make this page purple.)
However, I do have a keen eye for criticizing others' work, and now that I have a forum; let's see what happens.
Tyler can confirm this (I can't find a good link but if you give me one I'll put it here), but I believe that his position regarding the Oilers' 8th-seeded success is that the Oilers' seed is understated; really they should have a higher (possibly 3rd?) seed, thus their postseason success would be less 'shocking'.
His argument is based on goal differential, calculating an expected goals for and goals against for a team and an expected standings based on those factors. Generally, this GD is sufficient enough; my problem is not with this method.
However, you see, there is this Roloson Factor (RF) to deal with. Tyler (and others) argue that the Oilers have suffered from abysmally poor goaltending all season, and the RF has corrected this problem. Thus, using their season's GA isn't fair; we should instead use only the games Roloson has played in. In the RF games, the GA indeed did drop noticeably. But troublingly enough, so did the GF.
I am paraphrasing here, in that I haven't bothered to look at exactly which games Roloson played in, nor am I pulling empty-net goals or anything, but here's how the Oiler season looks segmented by Roloson Day in Edmonton:
(Note the first segment covers 62 games of .589 hockey, and the second segment involves 20 games of .550 hockey. Edmonton by no means roared to the finish line.)
What can one make of this drop in Edmonton's goal scoring? Tyler cannot see a reason why that this 20-game anomaly should matter. He didn't see the Oil necessarily collapsing, and thus he decides to ignore it and use the season scoring average instead.
So loosely he grabs a GF from a season and a GA from a subset of games involving Roloson. This is where I see a problem.
I know where he's coming from, in that Roloson plays no part in the offense, so his effect really should only evidence itself in the defensive turnaround (which is substantial, particularly on the GA and PK numbers).
However, I think there is a connection between GF and GA that cannot be ignored. We observe in games that teams (and refs) push more for a goal in a 2-3 game than a 2-1 game. We observe how many games go to overtime or end as one-goal games. A trailing team more often than not 'gets the breaks'.
A better goalie implies that there is less time spent trailing, and probably more time holding a lead. I would expect a team that holds the lead more often to (a) score less, and (b) have less power plays. The latter comes both from spending less time on the attack and on biased refereeing.
Edmonton, in fact, (a) did score less, and (b) did have less power plays. Their PP% and PK% both improved, but they went from a 392-372 man-advantage count in the first 62 games to a 93-106 deficit in the last 20.
Now I don't want to knock the Oilers' playoff accomplishment by any means. That they earn on the ice, screw goal differential and us back-room pencil-pushers. And I don't really want to suggest that my numbers are right or anything; as always, the truth is somewhere in between. Hell, I'm not even sure that I'm saying that they don't deserve the 3rd seed, just by a lesser margin.
I'm just saying that discounting the fact that GF dropped along with GA is simplistic and convenient. The dropoff was probably too dramatic, but I would expect that there is a negative offensive effect to having a better goalie. When you lead more of the time, the balance of the game changes.
Note that I haven't really discussed goalie tendencies, like freezing or playing the puck. Both of these should influence a team's scoring rate as well. I also didn't mention Samsonov, who probably matters the most.
Stats lie to us all the time, but if we compile enough data they will eventually tell us the truth. Great work, Tyler. Now tell me everything I got wrong.