Shot Differential Part 2: Data Mining the Stretch Run

(Author's note: This post took me way too long to construct, contains zero cartoons, isn't very well written, and doesn't come to any real conclusions.  You may waste hours trying to decipher all the data tables and may fully hate me for it, but at least we'll all be a few hours closer to puck drop.  Ah, I kid -- you'll probably love it.)

A month ago, I took a statsy look at the Ducks' regular season in terms of shot differential.  One of the striking findings was that while the Ducks were outshot by a whopping 314 shots last season, they were somehow able to generate a +14 shot differential over their last 30 games.  Nice turnaround, right?

So I decided to dig deeper into the shot data from those last 30 games of the regular season, using NHL.com's Play By Play logs to make a huge spreadsheet (here's the one from the season finale, for example).  It's an incredibly effort-heavy process because while it contains useful data about who is on the ice for every event in a game, it pastes rather horribly into Excel and has to be fixed one game at a time, always noting whether Anaheim is the home or road team.  But I persevered through 30 games because I was curious: How did the Ducks generate that slightly-positive shot differential, and do those results feel repeatable?  Before I go into any findings, though, I've got several caveats to go through about this whole exercise:

• The 30 games-to-go mark is a kind of important line in the sand; it's worth mentioning that if I did this exercise over the final 31 games, the Ducks would have been outshot over the sample.  Even so, keeping shots close is an achievement for last year's Ducks -- in no other 30- or 31-game segment did the Ducks come anywhere near their opponents in shots.
  
• The Ducks were obviously a very charmed in this stretch, as it's the 30 games that saw them climb all the way up to 4th in the west without much use from their #1 netminder.  7 of these 30 games went into overtime, and none of them went to a shootout.  Ducks won 6 -- they were crazily good in 4-on-4 down the stretch.  So the Ducks were lucky or clutch, take your pick, but close games were going Anaheim's way.
  
• I'm not really a strong proponent of shot-counting analysis, even though that's exactly what I'm about to do.  Reason one: Shots are not goals; they do not win games.  And while over time counting shots does more or less approximate goals, it does so by diluting the incentives -- hockey isn't about taking the most shots on net to win a game, it's about taking the right shots on net to win a game, and I still don't think we're capturing the right shot data to measure that (How many shots-on-goal were shot through screens last season?  How many came on odd-man rushes?  Somebody upstairs should start recording those goal-enhancing factors on every shot attempt.).  Raw shot counting certainly has value, no doubt, but in a game theory sense, it's not an expression of the true intent of the game.
  
  Also, I'm a little wary of attaching things like Corsi metrics to individuals -- at any point in time, one player doesn't affect the flow of play so much as his four on-ice teammates do; he will obviously make some impact but he's also a bit bound by where he plays.  So it's not always a "talent" stat that emerges (that guy sure is good at Corsi!) so much as a situational stat (that guy plays in a good Corsi environment, plus he might be good at Corsi), and I'm not sure we always treat individual rankings in the right way.  So don't read this post too damningly; some findings will appear striking, but I really don't mind shrugging them off as "well, shot analysis is sometimes bullshit".  Because yeah, it sometimes is.
  
• I'm probably including too many columns in my tables, for whose benefit, I'm not sure.  This has made the tables tougher to read, but in my defense, SBN's table formatter kind of sucks.  For each table I'll point out what takeaway observations I note, but generally the columns to focus on are the 3rd and 6th -- Shots on Goal and Attempted Shots.  I've included goal differential and some breakdown of shot attempts (shots missed vs. shots blocked) for context and reference, but this post's main focus is how the Ducks got their slightly positive shot differential.
• This exercise would have been way, way more useful had this spreadsheet contained all 82 games -- how do these 30-game results compare against the 52 games that came before it?  But that kind of effort isn't imminent -- this spreadsheet is a real bitch to make.  A fuckload of copying, a fuckload of pasting, a fuckload of formatting, a fuckload of parsing.  But it works now!  Fuckload yeah!
• Also, because I don't know every goalie's number on every opposing team, empty net goals are mixed in with the regular goals.  5-on-5 really just means there were six players on the ice per each side -- most likely one of them was a goalie, but not always.

Caveats noted?  Onwards!

Okay, the following tables are dense with data, and there's quite a few of them.  But let's go over the basics, looking at the second column, "Goals", as our walk-through example -- if you get the idea of this column, the rest should hopefully make sense.  At the bottom of the Goals column we can see that over the 30 game sample, the Ducks scored 99 goals and allowed 89, for a +10 goal differential.  The rows, then, are splits about how that happened.  Anaheim had a +2 goal differential at even strength, a +22 goal differential on the power play, a -16 differential on the penalty kill, and a +2 differential in penalty shots.  The subsequent columns show total shots on goal, shots fired wide, shots that were blocked, and total shot attempts.

Last 30 Games -- Goals, Shots, and Shot Attempts by Manpower Situation

Takeaway #1: The Ducks did have a +14 shot differential over the final 30 games, but it did not come as a result of taking more shot attempts than their opponents, in fact that margin wasn't close at all.  Opponents took 193 more shot attempts than Anaheim did, but because so many went wide or were blocked, the Ducks did manage to come away outshooters. 

Now a smart blogger would take a moment here and talk about team styles, or coaching philosophies, or how the Ducks play in such a way that encourages opponents to take wild and hurried shot attempts or something.  But I'm not really playing the role of explainer today; I have no idea really why the Ducks were forcing so many shots wide or weren't making more shot attempts themselves.  Maybe they were lucky, maybe Ray Emery has a death stare, maybe it is systematic -- tough to say.

Takeaway #2: The Ducks did not outshoot their opponents at 5-on-5 over this stretch, but rather kept shots close enough that they could make up for it by excelling on special teams and 4-on-4.  The special teams discrepancy is pretty significant, and at first I had to go look to make sure that power play time wasn't severely weighted in the sample.  Power play time did favor the Ducks, but not to the same degree as shots produced -- 172 power play minutes vs. 160 shorthanded minutes.  The Ducks' power play did generate shots better than the opponent's.

So the initial answer(s) to how the Ducks got a positive shot differential over the final 30 games?  Forcing opponent shot attempts wide, blocking shots, and working the power play and 4-on-4 situations to their full advantage.  Also, they hit the net on all three of their penalty shot attempts, unlike Olli Jokinen.  :)

Now here's another fun way to slice the data -- how the Ducks' shot metrics split based on the scoreboard:

Last 30 Games -- Goals, Shots, and Shot Attempts by Scoreboard Situation

Takeaway #3: Man, the Ducks were a very, very good team down the stretch when down by a single goal.  Now I should note, I think the general pattern of this table holds true for most teams: when a team is trailing in a game, its shot ratios look better than when it is leading -- it reflects not only shifted priorities (trying to tie vs. trying to hold a lead), but also there's some make-up refereeing and double-shifting and desperation and whatnot as well.

But wow -- that "Down 1 goal" row -- pretty exceptional stuff; the Ducks were "clutch" as fuck.  Now I'm not really sold on the idea of long-term repeatability of this, but it is good to know that the Ducks' desperation mode appears to be a very good weapon -- it was a key reason for Anaheim's late-season success and part of how they got the positive shot differential.

* * *

Now that we've had a taste of "how" the Ducks outshot their opponents, let's dig deeper into one of the components -- keeping things reasonably close shot-wise at 5-on-5.  Even though the Ducks were outshot by 43 shots 5-on-5 over the 30 games, that did mark an improvement from earlier stretches in the season.  What individuals enjoyed that relative success?

Here's a list of the forwards, ranked by descending shot differential.  Don't pay too much attention to the specific order, as many of these guys end up pretty close and I probably should have used some sort of shot ratio to rank, but do notice who's near the top and bottom of the list.

Takeaway #4: Wow, Bacon McMuffin did a lot of shot-bleeding down the stretch!  Now here's where I want to be cautionary -- McMuffin was probably the most important bottom-six forward for Anaheim down the stretch, but he does get murdered in these numbers almost because of that.  He had to play with Corsi-bleeders pretty exclusively once Getzlaf returned from injury, and he couldn't fix the problem himself.  But the lesson here isn't so much that McMuffin is bad, it's that even as the Ducks were winning close games, there was still lots of room for improvement particularly from Anaheim's third line.

But hey, Jar Jar Ruutu did pretty well on these metrics during his short stint.  As time passes, he'll probably get lumped in with Brad Winchester as "deadline deals that didn't pan out", but they were much different players in Anaheim in terms of on-ice reliability.  I would have no problem if the Ducks had retained Ruutu.

Similarly, here's results for individual defensemen:

Takeaway #5: Wait a goddamn second -- Andy Sutton was useful at 5-on-5 down the stretch?!  So says the data, at least.  Obviously this was cushy minutes, but maybe it did take the guy most of the season to find a groove he could play in.  I was as happy as anybody that the Oilers traded for Sutton, and still sort of am -- he was a healthy scratch too often for that money -- but in the right situations, the guy can be effective.

Takeaway #6: It's sort of interesting to look at Cam Fowler vs. Beauchemin and Sbisa.  Fowler is the Ducks' d-man who suffers the most in plus-minus, but his 5-on-5 shot differential wasn't bad at all, the same as Toni Lydman.  Then again, his shot-attempt differential is the worst, and goals-against are going in -- there is some problem outside of shot totals, I'd think.  Meanwhile Sbisa and Beauchemin have pretty reasonable plus-minus numbers, but their shot-differential numbers are McMuffin bad.  Is that a function of actually playing too much with McMuffin, or are they the cause of it?  Again, I'm not explaining today, but they too appear to be lightning rods for 5-on-5 shot bleeding.  But it didn't really kill them on the scoreboard.

Anyways, here's another fun feature about the spreadsheet I've put together -- I can even look at how defensemen fared together!

Here are the pairings that had some 50+ shots for/against at 5-on-5 over the stretch, again sorted by shot differential.  Again, Sutton comes out strong, and Beauchemin-Sbisa are solidly behind the pack shot-wise; incredibly pretty much all the 5-on-5 discrepancy can be put on them.  And yet most of the goal-bleeding is being done by Fowler-Lilja for some reason.  Luck?

Takeaway #7: Now that the effort to construct this spreadsheet is completed, playing around with it is kind of cool!

Again, none of this is meant to be a slight against anybody's play -- this is just a particularly-successful 30-game sample that is being parsed to death in an imperfect manner, and while it does reveal quite a bit about how this sequence of results transpired, there's always some bullshit mixed in.

* * *

So now back to the main question: is the manner Ducks' success on these 30-games repeatable or reliable going into the coming season?  Tough to say, obviously -- the Ducks had many things go right over that stretch, with many dramatic comebacks and late wins; that probably doesn't carry over very well.  And it's obviously going to be important that whatever the Ducks were doing on their special teams carry over.  Todd Marchant has retired, and it's possible Teemu does as well -- they won't be doing it with exactly the same personnel.

But at the same time, there's definitely areas that can improve, especially on the 3rd line.  And that's really where Andrew Cogliano fits in, hopefully.  His game has flaws for sure, but the bar for improving the Ducks appears to be pretty damn low -- if he and some of the kids can forge a semi-competent third line, that's better than what the Ducks had over that 30-game stretch.  Plus there are some players where we could expect some improvement anyways -- McMuffin, Fowler, and Sbisa are all going to be a year wiser, and Beauchemin is no longer settling in. 

Bottom line: I gleefully don't know.  I was talking to my buddy the other day and telling him that this season feels pretty unique from a blogging sense -- I can see many reasons why the Ducks will be good, and I can see just as many reasons why the Ducks will be bad; the spectrum of potential outcomes seems much wider than any other season I can remember.  So many questions to be answered, and I'm getting pretty excited to see how it pans out.

Go Ducks.