really so will lose just 4 more game without hemsky? I thought that will be wayy more. Lets see how this season unfolds.

The thing of it is that more than all of that differential occurred in 2006-07, when a) Hemsky missed a disporportionate number of games (18 = 45% of his 40) relative to the number of actual Oiler games (82 = 30% of their 272). The Oilers were by far worse in ‘06-07 than at any time since then (including the first third of ’09-10, depressing as it’s been). More specifically, Ales missed a stretch of 8 games during that Historically Bad collapse down the stretch. Oilers were outscored 28-8 in those games, and if you want to pin that on all on Hemsky’s absence, well … I’d put at least some of it on Bryan Young, Danny Syvret, Sebastien Bisaillon, and (lest we forget) Toby Petersen playing the blue, not to mention the rare invocation of the little-known Latitude 53 rule** which forced the Oilers to dress only 17 skaters for a number of games. It was disastrous before Hemsky went out and after he got back, so his own numbers would have taken a hit, but hardly commensurately.

(** Latitude 53 Rule: “If any team south of Latitude 53 has a shit streak, give them a break. If the team north of Latitude 53 has a shit streak, kick ‘em while they’re down.”)

So I say let’s throw out 2006-07 because it’s disproprotionate not to mention the most ancient history cited, and review just the last 2.3 seasons. Prepare for a surprise.

I had to convert Scott’s initial GPG data back to integers (so somewhere might be ±1 due to rounding), I get this for 2007-10:

With Hemsky: 168 GP, +462 / -507 = -45
" " " " " " " " " " " : +2.75 / -3.02 = -0.27 /60

Without Hemsky: 22 GP, +62 / -62 = even
" " " " " " " " " " " " : +2.82 / -2.82 = even

… and what do you know, over that span the Oilers are actually a quarter of a goal a game better without Ales than with him.

Before people start giving me the gears for unfairly criticizing Hemsky again, let me point out:

1) These are pure numbers over a well-defined period, no interpretation or opinion involved.

2) Yes it is a small sample size. But so is a sample of 40 games missed which are 6 times as significant as the 232 games played, so distribution bias will play a huge role.

3) I don’t for a second think the Oilers are better off without Hemmer. It is, after all, a small sample size.

4) It is such a small sample size, to draw conclusions like this study proves Ales has a value of $6.16 MM is about three significant digits too precise. :)

Ales would be healthy if he’d just WORK HARDER.

I guess if you want to limit it further that’s your choice but do keep in mind that Hemsky was also in for seven games of that stetch without Smyth in 2006-07 and the Oilers GD for those 7 games was 22-12 so, bad as they were, the team did perform better with Hemsky over that stetch than without him.

As I said, his own numbers would have taken a hit, but hardly commensurately. I’m pretty sure the time Ales missed was at the nadir of the slump, where the short roster BS was happening and every bad thing possible — including an injury to Hemsky — was happening. Bottom line: I wouldn’t put much stock in the 2006-07 data, especially when the same methodology for the more recent three years yields contrdictory results.

As for the team overall being worse in 2006-07, they had 71 points at the end of the year.

Certainly. What I referred to as Historically Bad was that last 20-game segment. Ales missed 40% of those 20 games, <12% of all other games under review. It’s certainly enough to corrupt the data, I’m sorry to say.

Now, I just don’t see it as "fair" for you to make a point about how poor the numbers look after taking out one of the segments in which he performs the best.

Say what? Let’s review the 20 game segment which I think should be thrown out entirely in any kind of critical analysis:
With Hemsky: 12 GP, +15/-41
Without Hemsky: 8 GP, +8/-28

If this is “one of the segments in which he performs the best”, I’d hate to see the other segments. That entire group of games was completely out of character with any other identified group of games for this team in the last dozen years at least. Fact is Hemsky missed 40% of them which contributed to the problem for sure, but he hardly solved the problem while he was in there, now did he? My concern is that those 8 games he missed make up 20% of all games he missed during the time of your study, whereas the 12 games he played represent barely 5% of that sample.

I can understand making some kind of adjustment to the data but throwing it out entirely seems like it’s done for a reason other than "fairness." It seems more like it’s done for "helping Bruce’s point."

My point is about math, not about Hemsky. If you’ve missed that, you need to reread what I wrote. Skewed sample sizes during a time of extremely misrepresentative data are going to corrupt any study, one way or another.

I like the idea of using one-game in to compare one-game out which I described in more detail above.

Except the data in question has Hemsky playing SIX games for every one he misses. So one-in-one-out is disproprotionate. The 40 games he missed have SIX times the weight per game than the 232 he played.

Anyway, go right ahead and do that. I think you’ll be surprised at what you find. My own method to throw out the 20 game segment is effectively “just” 1.5-games-in-to-1-game-out, but has much more to do with removing games from the study which are unrepresentative of the Oilers rather than of any one player.

I haven’t yet run the numbers so don’t know if it will make Hemsky look better or worse but I do know that it would give the "swoon" period the same amount of clout on both sides of the ledger (with and without) and would enable us to use all the data.

First of all, we’re not using “all the data”, we’re using 2006-10. You have argued against using 2005-10 because Hemsky missed disproportionately few games in 2005-06 and I have argued in favour of using 2007-10 because Hemsky missed disporportionately many games in 2006-07. The fact that each such study would yield vastly different results suggests very strongly that the method may be interesting but it’s unreliable. Too many other factors at play in a team game.