Thursday, October 31, 2013

The Hot Hand vs. The Large Sample Size

Tyler Dellow had an interesting post on how a goalie's save percentage has very little predictability from game to game, and how chasing hot streaks is, in his words, a "fool's game".  This is the kind of post that you want to point to when you hear some broadcaster talking about a goalie's "mental strength" in recovering from a poor outing and playing well the next night.  The goalie might be mentally tough, but Occam's razor suggests a more likely explanation is simple regression to the mean.

I agree with Dellow's general sentiment that goaltending evaluation is generally all about the large sample size and evaluating goalies based on single games is a poor strategy, but at the same time I think there are some legitimate merits to "riding the hot hand" as long as it is done responsibly and within reason.  I think there is a clear and understandable difference in mindset between the evaluation processes of a GM and a coach.  A GM wants to measure talent, a coach wants to measure talent and motivate his players to succeed through the options at his disposal.  I'm not sure a coach can be the ultra rational type because they do have to take into account the fact that they are leading a group of personalities on the team and want to make sure that the incentives are always properly aligned.

A hockey team is a group of competitive individuals that all want to play, and it is pretty much always in a team's best interests to encourage some level of internal competition to keep everybody sharp.  It is tough to have a legitimate internal competition when you are going to just keep giving the same guy playing time whether they have earned it or not.  When there is a situation where there is an elite goalie playing with a replacement level backup then it doesn't really matter what the coach does, even if the top guy is struggling you give him the minutes to play into shape and nobody is going to argue.  But if there is no clear #1, or an inexperienced guy who was supposed to be the starter is struggling (as Devan Dubnyk is in Edmonton, the occasion that prompted Dellow's post), then it's difficult to handle it politically if you want to keep riding a guy who is returning mediocre results, both externally to the media/fans and most likely internally in the room.  It can be an easy narrative to say that a team trusts or does not trust their goalie, and I'm not convinced that the effect is all that large, but there is at least something there, and coaches tend to be the perfectionist types that want to account for every single factor (the kind of thinking that leads to putting two RHS centres on the ice for right-side defensive zone draws late in the game, for example).

There's a saying in football that the most popular guy in the city is the backup quarterback.  Hockey fans can often latch onto the same kind of thinking in that the unknown abilities of the up-and-coming prospect or the journeyman AHLer might be preferable to the overly scrutinized flaws of the incumbent #1.  The difference between the sports is that a backup QB often never gets to take the field, while a backup goalie will always play at least some games.  In the grand scheme of things, if a team chooses to slightly overplay a backup or give a journeyman goalie a few not-necessarily-deserved starts because he is the "hot hand" and the main guy is struggling, that has very little impact on the team's overall chances across 82 games.  Even in a scenario where the starter is elite (say, .920) and  the backup is replacement level (.say, .905), a quick estimate indicates that an extra 5 games for the backup instead of the starter will cost the team about .001 in overall expected save percentage, which is a little less than one full point in the standings for a team with a typical rate of shots against.

That appears suboptimal, but if it has even the slightest effect in terms of motivating the top guy to improve his focus or work on his game then it might even be a positive move, despite the fact that the depth chart rankings make it overwhelming likely that the replacement goalie is in fact less talented.  Who wouldn't take a tradeoff of -.010 for 3 games in exchange for +.001 over 60, for example?  There's no guarantee that you get those results, of course, and I think the reactive nature of goaltending means that effort and desire is less correlated with success than at other positions, but I'm not willing to completely write off the benefits of competition either.

There can be cases where a coach gets legitimately carried away by the "hot hand" approach.  Probably the most dangerous one with the highest potential for serious consequences is when a goalie starts his career in the league or with a new team hot, and then coasts on that reputation for a long time before his team figures out that he isn't all that good.  In that case a reactionary coach would actually be preferable since they would be more likely to play other options rather than end up with an entrenched starter possessing overrated talent.  I also particularly hate the setup where a goalie who wins gets to stay in net for the next game regardless of how well he played the previous night, simply because it is based on such a high degree of luck, but that usually only comes into play in tandem situations where there probably isn't that much between the netminders to begin with and there wouldn't be much cost to even the most random of strategies.  On the whole it seems to me that it should be largely self-correcting.  If there is a large talent discrepancy, then it's probably going to reveal itself pretty quickly anyway, and if it doesn't reveal itself because the inferior guy is running hot then that's not really a problem, is it?

I guess in short, what I'm saying is that the high-level, super objective rational approach might be slightly superior from a purely odds-based approach, but I completely understand some coaching decisions because they essentially save themselves a lot of explanation at the cost of very little in terms of win expectation, and because they want to create at least a decent approximation of a merit-based environment that rewards success because of the motivational benefits.  If a coach starts the wrong guy come playoff time then there is good reason to be incensed, but for the most part I can't find myself getting too worked up about a little bit of suboptimal regular season goalie handling over the course of a long season.

Wednesday, October 30, 2013

Who Had the Worst Goaltending Offseason?

Heading into this season, I figured there were two teams that were in the running for the candidate of the worst offseason performance in the area of goaltending:  Philadelphia and Chicago.  The Flyers earned that nomination for their continued pursuit of the cheap goaltending tandem that rarely seems to work out for them, betting on Steve Mason and Ray Emery.  The Blackhawks would appear to be a less likely candidate for such a dubious honour, given that they are the defending Stanley Cup champions, but they gave Corey Crawford a surprisingly expensive long-term extension (6 years, $36 million) and elected to sign the 40-year Nikolai Khabibulin as their backup goalie.

(Note:  In a complete vacuum the Calgary Flames would probably win in this category because of their decision to ride Kerri Ramo and Joey MacDonald this year, but I'm giving them a pass because I figure that management was pretty blatantly looking to tank the season.  I'm also not ready to write off Devan Dubnyk just yet, which is why the Oilers aren't discussed further here.  The other team that seems to be taking on some risk in net is the New York Islanders and their continued reliance on Evgeni Nabokov, but he had a pretty strong season last year so at the moment they seem to be fine).

Obviously it's very early, and it's not a good idea to make too much of early returns because there is plenty of season remaining, but at the same time an evaluation of decision-making should not be entirely based on hindsight either.  While I'm far from a full believer in Steve Mason just yet, he has looked good enough to raise the possibility that he may have turned a corner in his career, which is enough for me to vault Chicago into the lead in the category.

Khabibulin was shredded last night by the Ottawa Senators and is off to an atrocious start (.818 in 3 GP).  The risk of signing 40-year olds is that they are prone to suddenly losing a step and falling off a cliff, and while it's far too early to make that pronouncement with a high degree of confidence, he simply looks terrible in the net right now.  Even if he was playing at a typical level for someone his age, it's very likely that he would be at replacement level or worse.  All in all, even just a one year contract at $2 million for Khabi seems to be a very poor bet by the Blackhawks.

I'm not a big believer in Corey Crawford, I think it is very possible he had a career year last year on a great team.  To my eyes, his playoff run was more impressive statistically than visually, and was buoyed by an unsustainably high PK save percentage.  I'm still unconvinced Crawford is anything better than league average, but even if you think he is worth $6 million a year right now, the problem is that he is already 29.  It might seem odd to say it, given that he only has 163 NHL games under his belt, but Crawford is very possibly already past his prime.  History generally shows that you usually need to be an elite talent to maintain an above-average level of performance through your mid-thirties.  Crawford is certainly not an elite talent, and as such the smart money would seem to be that he probably will not be able to maintain above-average play through the age of 35.

Extending Crawford at that pay scale was a curious move for the Blackhawks, a team that was once one of the trailblazers in pursuing the cheap goalie strategy.  They dumped Antti Niemi right after a Stanley Cup win (which given Niemi's recent performance looks worse in hindsight than it did at the time), and since then have been relying on cheaper tandem options.  It's possible that the team's strategy back then was largely influenced by the very poor results they reaped from big-money deals to Khabibulin and Cristobal Huet shortly after the 2005 lockout, and perhaps those memories have now faded enough in the rear view mirror that Chicago is ready to get on board with the long-term goalie contract trend currently sweeping the league.

The Blackhawks did make at least one positive move this summer by inking Finnish league standout Antti Raanta to a pro contract.  After missing some a few games because of injury, Raanta is off to a strong start at the AHL level.  At the age of 24, Raanta has already dominated a lower league, winning nearly all the awards he could win last season in Finland.  Finland is a goalie factory and even though many have already migrated over here to ply their trade in the NHL it is no small task to be named the best netminder there.  There are plenty of recent examples of European goalies who have been able to step right into the North American game and excel.  Chicago was likely assuming it would be better for development purposes to have Raanta prove himself at the pro level in North America for a season first before relying on him with the big club, but I would bet on a 24-year old Euro league standout any day over a 40-year old NHL journeyman.

Even with average goaltending Chicago is a Stanley Cup contender.  However, a big reason they won the President's Trophy last year going away was that they were 17-1-0 with their backup goalie in net.  If Crawford comes back down to earth and Khabibulin continues to struggle, the Blackhawks will be remain a top contender but will probably not be able to separate themselves from the pack as thoroughly as they did last season in the hyper-competitive Western Conference.

Monday, October 21, 2013

Why One Great Playoff Run Means Nothing

The vast majority of early season goaltending articles or blogs should be about the variability of small sample sizes, because now is the time of year where fanbases start to freak out and demand immediate action to address their team's perceived lack of goaltending, or conversely, compose detailed narratives about their young team gaining the confidence to come together and believe in one another when they should just be pointing at a .956 team save percentage.

In other words, sell short on Semyon Varlamov and Marc-Andre Fleury and buy stock in Henrik Lundqvist and Devan Dubnyk.

Fortunately, however, a goalie's early season struggles or successes usually fade pretty quickly into the rear view mirror.  Go back to 2012-13 and think about the seasons of Carey Price and Sergei Bobrovsky.  Are you thinking of the first seven games where Price was at .938 and Bobrovsky just .896?  A goalie's early season stats will impact his playing time and they can affect his awards recognition since a hot start can mean a goalie will spend most of the season among the league leaders, but often it just resolves itself as the goalie's play normalizes and the what-have-you-done-for-me-lately media finds another story.

The real small sample size misjudgments usually come in the playoffs.  Postseason samples are equally small, but they don't get large enough to normalize over any single playoff season, and people love talking about playoff runs and using that to impact how they view goalies.  The result of that is that a small sample at the right or wrong time can have a vastly disproportionate influence.

I could come up with dozens of examples, but maybe the best one is Steve Penney.  Penney was a terrible NHL goaltender, but because he managed to be starting for a large market team as a 22-year old rookie out of nowhere and got hot for one spring, he has carved out a spot in the collective hockey memory that is far greater than the anonymity he actually deserves.

Here is his career regular season save percentage record:

1983-84 (MTL): 115 SA, .835
1984-85 (MTL): 1344 SA, .876
1985-86 (MTL): 447 SA, .839
1986-87 (WIN): 134 SA, .813
1987-88 (WIN): 186 SA, .839
Career (TOT):  2226 SA, .859

From 1983-84 to 1987-88, Penney ranked 46th out of 47 goalies in save percentage with at least 2000 shots faced.  And he did that even though 60% of his sample came on a Jacques Lemaire-coached team with Chris Chelios and Larry Robinson playing in front of him.  If you look at little used goalies over the same time frame (e.g. all goalie seasons with 15 or fewer games played), the combined average is .859, which indicates that Penney was essentially a replacement level goalie, probably not appreciably different than some of his minor league playing partners like Mark Holden or Rick Knickle.  But you help get the Montreal Canadiens to the Conference Finals and do it with the media friendly narrative of the unknown who makes it big and that stays on your resume forever.

Percentage-wise, an .859 career goalie posting a .910 on 354 SA is the same level of outperformance as a .905 talent goalie getting up to .940.  And nobody is even three-quarters of the way to that sample size yet in this young season.  Crazy numbers, both good and bad, are essentially guaranteed, and they probably have very little to do with equipment changes or coaching changes or anything else, it's just variance at work.

One final comment on the subject of regression to the mean:  If you haven't read Brian Macdonald's Bayesian approach to analyzing goalies, I highly recommend it.

Thursday, October 17, 2013

The Maple Leafs and Shot Quality Against

With Toronto off to a 6-1 start despite once again being consistently outshot in games, it looks the Maple Leafs are going to draw an lot of the focus from both the hockey stats crowd and the more traditional hockey watching communities alike this season.

The Leafs have been winning games despite usually trailing in terms of possession, making up for it through their percentages.  A big factor in that is the sky-high 13.1% team shooting percentage, which may or may not be somewhat accounted for by talent and/or style of play, but here I'm going to focus more on the defensive/goaltending side of the equation.

The narrative from fans and talking heads is that the Leafs are willing to trade quantity for quality on the defensive side of the puck, keeping shots to the outside while taking away the dangerous areas of the ice.  I think that the overall numbers still suggest that shot quality effects in today's league are fairly minimal, but we're dealing with a single team here and it's at least possible they could be an outlier.  Yet an even bigger problem with the conventional wisdom is that it is completely contradicted by the situational goaltending stats.  Toronto goalies aren't putting up great numbers because their teammates are taking away the key areas of the ice at 5 on 5.  Their outstanding save percentages are on the penalty kill, a game situation where every team around the league tries to take away the key areas of the ice because that's basically the only strategic option a shorthanded unit has at its disposal.

Breaking down Toronto goaltending over the last two seasons by game situation, we get:

Even strength:  1383/1497, .924
Penalty kill:  222/244, .910
Power play:  42/44, .955

League averages from 2012-13 were .920, .865 and .918 respectively.  That means the Leafs were 5.1 goals above average at even strength and 12.5 goals above average on special teams.  There were an additional 1.6 goals saved because the team faced fewer than average shots against while shorthanded, which means that adding it all together, special teams accounted for just 16% of shots against but 73% of the team's goals saved above league average.

The penalty kill numbers from 2012-13 also buck the narrative that high save percentages were related to a team strategy of letting opponents fire away at will from the harmless areas of the defensive zone.  The Leafs were an elite PK unit in terms of shot prevention, allowing just 41.8 SA/60, good enough for 5th best in the league.  Compare that to even strength where nobody allowed more shots against per minute than Toronto's 33.7.

So far the 2013-14 numbers are pretty consistent at evens (33.4, 7th worst),  but the PK is also getting lit up on the shot clock with 63.6 SA/60, the 6th most league-wide.  I'd expect that number to drop as the season wears on, but if it continues to remain high then any save percentage regression is going to hit Leaf special teams awfully hard.

It could be argued that Toronto's shorthanded unit is so good that it reduces shot quality against, but there's no way it could account for the entire effect:  Any given power play shot has been 49% more likely to go in against a league average goalie than against a Toronto netminder since the start of last season.  It should also be noted that Randy Carlyle does not have a history of seeing his teams record amazing save percentages on special teams, despite some pretty good goaltending overall.  The highest number any of his Anaheim teams recorded on the PK was .889, and from 2005-06 to 2010-11 Duck goalies had a very ordinary special teams split (.926 at even strength compared to .877 on the PK, along with .910 on the power play).

I think Jonathan Bernier and James Reimer are a solid duo and have the talent to make an above-average tandem, but I don't think that Toronto's shorthanded save percentage is in any way sustainable.  The jury may still be out on how much Toronto's scoring performance is based on shooting talent and/or counterattacking skill, but the defensive shot quality arguments look pretty weak to me.  Evidence suggests that the Maple Leafs are being carried by some good goalies running unusually hot on special teams, and if that's true then the smart bet is that their save percentage numbers are going to regress pretty strongly over the rest of the season.