What's in Your Wallet? Third Line Checking with Rewards from LaddSteegBoll One Bank

Some say... I never dump the puck, and that our line does a miserable job of shutting down Jumbo Joe.

Some say this, indeed.  That would be Derek Zona over at The Copper & Blue, who's been doing some wonderful work breaking down the scoring chances in every playoff game.  But, after this line's mind-infesting work against the Sedins in Round 2, implying that Dave Bolland (and by extension, Kris Versteeg and Andrew Ladd -- the lightning in a bottle captured by Coach Q's line generator) was more lucky than good against Joe Thornton & co. was, well, a little agitating to us Blackhawks fans.

Derek put up a head-to-head breakdown of these scoring chances in a more recent post.  Interestingly, his results showed much of the same; that our belovedly pesky third line wasn't doing much to "shut down" that top line of the Sharks -- the one they call HTML.  If anything, according to Mr. Zona, all it did was open up opportunities for Sharpie's line (not so much Tazer's) in limited head-to-head TOI.

Head-to-Head Scoring Chances for Games 1 and 2 (click for chances per 15 mins) Reproduced, admittedly without permission (I hope this is okay)

I trust Derek's judgment on the scoring chances.  This is a relatively objective form of eye test.  But not only does this lead us to question our own eyes, but also what the best way to measure the effectiveness of a checking line is.

If you really want to "question" your eyes, call me a coward.  Go ahead, try it.

Though I never did a post on Head-to-Head +/- (my new name for Cross-Mutual +/-; I'll explain in my next report) for the Vancouver series, those numbers don't jump out at anybody either.  Bolland was +2 facing Henrik but even against Daniel (who he got to totally lose it).  The line was negative against Burrows, who got bumped up to the top line for most of the second half of the series (there's no chart here, so don't look for it).

People generally acknowledge the effectiveness of our very own Davey B-Rat against the Sedins, though, so let's look at some numbers from the Sharks series to see if actually they kept things going, or if our third line success was a figment of our homerism-twinged minds (and media).

	With triple the rewards, me and Ladd are having trouble deciding which category to choose.

There are a lot of stats that we can use to try to understand how Weapon's line handled the triple threat of Marleau-Heatley-Thornton.  I'll select a few to shed some light on how they match up to what we see on the ice.  Since these were the primary matchups, I'll limit the data to the Hawks' 1st and 3rd lines against the Sharks' top six.

Head-to-Head +/- and Corsi

Let's start with Head-to-Head +/-, since I've used it before and it gives you what really counts at even strength -- who scored, against who (remember, everything in this post is from a 'Hawks point of view, so pluses for us are minuses for Sharks).

Bolland and Versteeg's +4s on Heatley and Thornton were the highest among forwards.  Since Patrick Marleau was doing all the scoring -- and I mean ALL the scoring -- against us, it's amazing that he couldn't crack positive territory against a line that didn't have Toews or Hossa on it.

	Oh, I'm sorry, did THIS just happen?

But, just looking at the results doesn't tell the whole story, because goals often result from broken plays and lucky breakaways, rather than whatever style of play a line is predominantly imposing on another.  So, with the aid of my brand new program for parsing NHL play-by-play data, I found a way to do the same thing but with Corsi instead of +/-.

That's a pretty muddy picture, huh?  Against Joe Pavelski's line, it's obvious which matchup works there, so it would make sense that Coach Q wanted to keep Buff+DDN out there against them.  Of course, this begs the question of whether it was worth doing this, if VerBolLadd aren't exactly shutting down HTML -- but that's a question I'll answer in the last section.

	Does your leg brain itch?  Let me dig into it with the blade of my stick and see if that feels better.

In the meantime, we can clear this up by splitting the first two games from the last two.

Telling, isn't it?  Looks like at least from a shots-directed perspective (as per the definition of Corsi), something was working, and Todd McLellan didn't adjust until the series came back to Chicago.  Bolland's line stuck with Jumbo Joe, who got time with Devin Setoguchi and Randy Logan Couture.  H and ML got to spend time with Manny Malhotra.  What this did was:

• Isolate Thornton when he was bad; he was actually very good at times, but freeing up Marleau paid dividends.
• Get more ice time for guys beyond top 6; Malhotra is a FO machine and I can't fathom why McLellan didn't play Couture more in Games 1 and 2, given how well he'd played in the first two rounds.
• Allow Versteeg to outplay himself Pavelski, somehow.
• Reduce our checking line's advantage to drawing dumb penalties.  And scoring opportunistically.

	At our bank, you can forget about scoring chances... and still score.

But, even if these changes helped the Sharks, it didn't help them win Games 3 and 4.  Are they (or we) doing something wrong, here?

Weighted Shots-Directed Rating

After noting the limitations of Corsi in describing checking lines (this is fairly well known, and I've cited Behind the Net and its adjustments to Corsi before), I went back to my first super-long stats post and discovered a shocking fact that I neglected to highlight back then:  Shots Directed on Net Differential (basically a team Corsi) and Goal Differential are 23% negatively correlated.

That's right -- at least for our Men of Four Feathers, a positive Corsi rating means we score less.   Okay, so why are we tracking this statistic again?  Maybe we should shoot less?

	Hah, Steeger doesn't know how to dump the puck, I'll show him.  Oh wait, this is a shootout?  fffffffff!

Before all you young stats geeks out there start panicking, don't forget that Corsi is an illustrative statistic -- it does reflect the direction that pucks are being shot, even if it doesn't account for the quality of said shots.  But, if it won't tell us how our team is doing, what will?

To answer this question, you might think about trying to regress multiple variables against goal differential, something I might have alluded to in that post I mentioned, since those were all single-variable regressions.  But since that's tough to use for illustrative purposes here, instead I tried to devise my own rating system.

Basically, I took the components of Corsi: shots on goal (SOG), attempts blocked (AB), and missed shots (MS).  Each of these gets multiplied by some weight to come up with this rating, which I then regress against goal differential; how well my rating can predict goal differential comes out of the R-squared value.  Using Excel's optimizer, I came up with this formula as for the rating that best predicts goal differential in the regular season:

VerRATING = 0.66 x SOG − 2.08 x AB − 1.58 x MS

	Half a million for the stones, Taking trips from here to Rome, So if you ain't got no money take your broke ass home...

This is nothing like Corsi; in fact, it's bad if you don't hit the net.  Oh, that actually makes sense!  After all, I've already extolled the virtues of On-Net%.  It's more important to hit the net a lot than to just shoot more.  But, given how crazy things get in the playoffs, is this rating still relevant?

Here is the regression that came out of what I did above:

The pink squares in this graph represent playoff games.  So far, they seem to fit the pattern, so it ought to be safe to apply this to playoff games, like those against the Sharks.

This is still highly experimental, and I probably should have done this in a separate post.  Even though I used all game situations to derive the formula, wacky playoff numbers made me limit them to even-strength here.  Obviously there's a lot more work to be done with this rating of mine (and it needs a better name!).

Still, there's some interesting stuff to be taken away here.  Our top line does horribly in this statistic, because they generate so much offense that may be of the not-so-accurate variety.  Accordingly, the Bolland line does the same against HTML, for largely the same reasons.  However, overall, we do come out on top (Niemi has a +17 on Nabby), and we won the series.

What's happening here is that blocked and missed shots are being heavily penalized -- but remember that this is calibrated to determine success over the course of a game.  Toews's block on the PK leading to the Rat's shortie is a good example of how a block can be so bad (and that's also why many stats gurus prefer Fenwick over Corsi), even though it didn't happen at even strength.

	Boys gotta go... (that kid has an awesome playoff beard)

Well that was long.  But while it says that given the results, you can't say that Bolland's rattiness didn't have an effect on the Sharks' top line, it also doesn't prove it.  There's one thing left to test, however -- even if they only barely held HTML in check (or were beat at times), did they create enough additional opportunities for Daydream McNuggets?

Theoretical Line Matching: Corsi and WSD Rates

One way to check whether freeing our top line from HTML (who Toews et al. seemed to match up well against) was worth it is to pretend that they get an extra minute of ice time against HTML and that Bolland's line gets a minute less.

This is a pretty statistically naive methodology here, but it's basically saying that by shifting a minute of ice time from one matchup to the other, the lines' Corsi (or Weighted Shots-Directed Rating) would change proportionally with the rates the lines are producing these numbers.

I weighted each line's Corsi by each player's TOI relative to his line and the line's TOI relative to total even-strength TOI of the team.  Then, every time I added or subtracted a minute (+1 or -1 dTOI), I ended up with an equation:

totCORSI = 3.91 − 0.13 x dTOI

And doing the same for WSD rating:

totWSD = 1.36 − 0.02 x dTOI

Since both measures go down when we take away a minute from Bolland's line and give it to Toews's -- we can say that putting Bolland's line up against HTML resulted in more production overall.  So, it was worth it... even if barely.

Recap

So, what can we say?  None of these statistics look at the plays like you could with your eyes, but even if the Bolland-Thornton matchup was overstated by the media, the numbers do show that we held the better end of that matchup, in the end.

	When the ref drops the puck, you're supposed to go for the puck and he went for the wrist.  I think he got mixed up.  Ah, memories...

What you could gather from the above is that, while HTML outchanced our 3rd line, their pattern of play was far from dominant; by the time they got their act together in the latter half of the series, our guys happened to find ways to score, too (rewards!  I love rewards).

Basically, LaddSteegBoll did a couple of things right -- 1) they created a better matchup for the Captain's line, on a comparative basis, and 2) they held their own just enough to be worth it.  But, don't take my word for it; the statistics don't give us any clear indications here, and the fact that this line ended up inexplicably scoring as much as they did might be what made them look so good.

Besides, these are the only stats that actually matter:

2-1, 4-2, 3-2, 4-2

And most importantly, 4-0!

(sorry about the length; this post started as an attempt to shine some statistical light on this matchup and I got carried away -- for those of you complaining about the wait, though, at least this gave you something to do before the SCF)

This head-asplosion-inducing stats-post has been brought to you by:

	Vhat's in your vallet?