Dr Statlove or: How I Learned to Stop Watching&Love the Bomb

[mikey287]

Just moving slappy and I's (and others are certainly welcome) long discussion regarding advanced statistics and their relevance to the game to a different thread so the discussion of the playoffs could continue uninhibited in "Other NHL Games" thread...

Last post of that thread:

Shot quality exists. No one denies that. Those links all say so. Its just as big of an influence as you believe it is. Over the long term, and over an entire season, and over an entire roster it largely evens out.

I knew you were going to cite roster turnover, but with 4 years worth of data and 30 teams, that argument is simply not plausible. You yourself cited to Boston or Phoenix as examplars of stability in terms of how they operate and what they do in the game; we could probably come up with a bunch of teams that played relatively the same way year to year in that time period (eg, the Pens). So to say, well, data points over 120 distinct 82 games seasons show no correlation, but I am going to discard them because a team changed 3 or 4 players per year (and often those with lesser ice time) or changed coaches in one of those years -- it just doesn't stand up. Either there is some carryover on a large scale basis from year to year, or there isn't -- you can't be logically inconsistent and pick and choose when it suits you. Moreover, the fact that the correlations are practically zero -- man. There's just nothing there. Trying to explain it away with a bunch of generalities isn't persuasive IMO.

See, this is what I mean...this is what turns me off from the movement. If there aren't numbers for it, if there isn't enough understanding to see how it would effect the game, it is just ignored. To think that changing a quarter or a fifth of all skaters has no impact just shows the disconnect between the game on the ice and the game on paper.

For instance, look at the impact of losing Willie Mitchell and Matt Greene in Los Angeles. Forced rookies into the lineup, forced Rob Scuderi to have to shoulder too much of the load and he struggled with it. Doughty had to focus more on playing defense and couldn't open it up as much as he normally would and had his worst production since his rookie year even though 23 years old is supposed to be in your statistical prime, right?

Jonathan Quick last year: 1.95 GAA, .929 save pct. 10 shutouts. Jon Quick this year: 2.45 GAA, .902 save pct., 1 shutout. Worst save percentage of his career.

Strong finish to Quick's season though, right after they acquired Robyn Regehr to shoulder some of the load defensively.

From the start of the year through March, Quick's save pct. was .895. Then Regehr was acquired and from April to present is .928.

And then dear Lord, to discount coaching changes too...this is worse than I thought...

The very first thing I said to you re: Reimer is this:

Perhaps your view is that this season is too small a sample size? But that's unlikely because you then write...

I said that because I'd have no issues with that. I personally like the guy from what I have seen, and believe that this season and his rookie season (which you keep ignoring for some odd reason in terms of information -- maybe because it doesn't help your conclusion that he "blows") and this year are more reflective of his abilities than his second year, when he got concussed early in the year, was out for a couple months, and came back to play below average. There's nothing at all wrong with you concluding that based on your own scouting report plus the small sample size, that you believe he'll revert to "normally blowing" or whatever it is exactly that you view his natural state. But, your initial position was SAVE% is useless.

Save pct. doesn't seem to have any great indication value today about the talent of the goalie in the net. That's why career journeymen have the three best save pct. marks in NHL history* - Tim Thomas, Brian Elliott, Craig Anderson. And all in the last three years. It just seems really unlikely that save pct. is a direct indication of talent, otherwise it would be used to predict NHL success, right? Like Nicola Riopel, who is on the brink of NHL stardom...or Mike Murphy...or a thousand others...it doesn't even have correlation in the NHL: Roman Cechmanek, Rollie Melanson in his Isles days, Wayne Stephenson in his Flyers days, etc.

Maybe Reimer becomes the best goalie of all time or whatever, but he doesn't look the part...and the rookie year is being largely ignored because it's not uncommon in today's game for rookie goalies with a different technical setup to have early success until there's a video review of his tendencies because there's no "book" on them...

A 20 foot wrist shot on a 1-on-2 is not the same as a 20 foot wrist shot on a 2-on-1.

This is the problem. Of course shot quality exists on an individual level of shot x versus shot y taken in the same game. Again, the difference is that over the long term, the "quality" equalizes. That's why data from multiple seasons doesn't show any meaningful change in shooting percentage or in save% when goalies move from Team X to Team Y. You're focusing on small events to make larger points, when the mountain of data that exists shows the opposite.

I don't think there is much productive conversation to be had here frankly, and like you said, we don't need to jam this thread up any more with this stuff.

So, again, Grant Fuhr faced the same quality of shots as Billy Smith? Marc-Andre Fleury the same as Henrik Lundqvist?

Taking data from multiple seasons further introduces even MORE variables to it...every season is not created equally. Why would one assume that all the fluidity that surrounds the game, both on the ice and off, is effectively equal? It literally doesn't make sense.

It's the little things that are impactful though...Marc-Andre Fleury's even strength save pct. this year was .927 which is tied for 17th in the league (somewhere between 17th and 20th, I won't convert them to an additional decimal). If he makes five more saves on those shots this season, it jumps to .934 and he's now in the top-10 in the league. If he makes two more saves, two, on those shots, he jumps to 14th - tied with a Vezina finalist. These instances don't matter?

There is no mountain of data...well...there is, but it's relevance is really quite questionable. Like I said, just because you have a lot of something doesn't mean that it's quality data or that it's significant to the discussion at hand. It seems to gloss over things that matter and write them off as "doesn't matter" because they can't assign a number to it and/or they don't understand the item's significance on the game....

[columbia]

DIGR: A Defense Independent Rating of NHL Goaltenders using Spatially Smoothed Save Percentage Maps
http://www.sloansportsconference.com/?p=648" onclick="window.open(this.href);return false;

For those who haven't sat through a couple semesters of advanced statistics, the results for 2010-11 were:
http://online.wsj.com/article_email/SB1 ... _email_bot" onclick="window.open(this.href);return false;

The results for 2009-10 are in his paper (and discussed in his presentation) here:
http://www.sloansportsconference.com/wp ... e-Maps.pdf" onclick="window.open(this.href);return false;

Discuss....

[ExPatriatePen]

Ohhhhhhhh myyyyyyyyyyy, a statistics thread with hockey as the data source.

Oh joy, oh joy, oh joy.

One of the most misunderstood fields of study by layman, statistics.

I work with stats every day and far and away no expert on statistical methodology, but this thread should be fun and VERY contentious.

I love it.

[Penspal]

Props for the title, its what brought me here.

Not poo-poo ing on this thread, I want to be clear on that. Obviously brighter and more educated posters than I discussing stats. And its certainly appears that the applications of stats in hockey is growing at leaps and bounds.

However, there are still many intangibles that I do not thinks stats can capture, but do play a role on how a player/team performs.

Example A players first shot in a game on his first shift.... Is that comparible to his 4th shot in the 3rd period (warmed up so to speak). Yeah maybe?
Example Recognition of "grey events" happening immediately following a penalty expiring, delayed call extra man on, PK effeciveness after killing 3 consecutive penalties
playing injured or with injuries, peformance after fighting (sticking up for teammate maybe), playing with or without consistent linemates, team systems, different goalie style (cover puck vs. clear/deflect out of play), west coast travel, rule changes, good/bad ref calls, ice conditions,2nd game back to back vs rested, divorce proceedings, cold/flu, contract year vs. 1st year long term, etc, etc, etc There are so many factors that are not tracked that play a part/role in performance.

Stats apparently were ONE of the convincing factors that lead the Pens to acquire James Neal. Hard to agrue with their use in that case. However, he did not perform in his first year, stats did not (and likely they should have) shown it would take time? Anyway, I guess what I'm saying is that they are an extremely useful tool in the toolbox, and becoming more and more important, but hockey, like life, cannot be viewed in isolation

[slappybrown]

Save pct. doesn't seem to have any great indication value today about the talent of the goalie in the net. That's why career journeymen have the three best save pct. marks in NHL history* - Tim Thomas, Brian Elliott, Craig Anderson. And all in the last three years. It just seems really unlikely that save pct. is a direct indication of talent, otherwise it would be used to predict NHL success, right? Like Nicola Riopel, who is on the brink of NHL stardom...or Mike Murphy...or a thousand others...it doesn't even have correlation in the NHL: Roman Cechmanek, Rollie Melanson in his Isles days, Wayne Stephenson in his Flyers days, etc.

Maybe Reimer becomes the best goalie of all time or whatever, but he doesn't look the part...and the rookie year is being largely ignored because it's not uncommon in today's game for rookie goalies with a different technical setup to have early success until there's a video review of his tendencies because there's no "book" on them...

Tim Thomas' lifetime save% is .920. His case is certainly an outlier as he didn't become an NHL regular until age 31, but there are other reasons for that (he also put up a .941 in the season before the strike in the A; maybe it clicked for him; maybe Boston's scouting didn't realize what they had; but if you want to minimize his performance with Boston, then go right ahead).

Brian Elliot's lifetime save % is .909. He had one exceptional season over 38 games. A single 38 game stretch in evaluating any player is laughably small to the point of statistical irrelevance, particularly where Elliot had played close to 200 games in the NHL before his .940 in 38 games, and his save % was sub-900. No one points to Brian Elliot as an indicator of the value of save % in evaluating goalies in a positive sense (in fact the opposite, his overall performance is generally subpar), and citing him is nothing more than you cherry-picking a small sample size to create a strawman.

Craig Anderson is an even worse example than Elliot. He's a lifetime .915 guy, which means he's at about the league adjusted save % average for goalies if you rough out his entire career/the NHL averages from 02 to present. He had one season, last year, that was a true outlier with a .941. He put up that .941 in 24 games!!!! He's played 299 games outside of that .941, and in most every season in which he saw meaningful time he puts somewhere around .910 to .920. Pointing to a 24-game stretch in support of the argument that "If Save % is so meaningful, why isn't Brian Elliot headed to the hall of fame" demonstrates a fundamental misunderstanding of statistics.

I have no idea who Nikola Riopel is, but this: http://en.wikipedia.org/wiki/Nicola_Riopel" onclick="window.open(this.href);return false; tells me that this is probably another situation where you pull one 59 game season where he puts up a .931, and ignore the other 120 or so games where his save % was ballpark .910 or so. Or that his cup of coffee in the AHL appears to have been a disaster. You of all people should know that the vagaries involved in junior hockey and the significantly more disparate talent spread in juniors makes projecting out NHL performance less than a direct one to one comparison (for a non G example, why isn't Yannick Riendeau in the NHL? There are probably many reasons, but no one -- at least not me -- ever made the claim that statistical performance in juniors is the lead indicator of NHL success). And you probably could have used a better example than Riopel if you wanted to go this route.

It is unclear what you are trying to say with the Rollie Melanson example. Is your view because he was better with the Isles as opposed to his post-NYI career, his performance with the Isles...is what? Wayne Stephenson doesn't seem to have any save% information -- http://www.hockeydb.com/ihdb/stats/pdis ... p?pid=5173" onclick="window.open(this.href);return false; -- so please feel free to share.

I suppose I have the same question with Cechmanek. http://www.hockey-reference.com/players ... mro01.html" onclick="window.open(this.href);return false; Nothing controversial far as I can tell. His lifetime number is .919. He put up 921, 921, 925, and 906 in 4 seasons in the NHL. What's surprising about that? Are you saying that his pre-NHL save% figures (which I don't have) indicated he should have performed better? Are you saying he didn't actually perform as well as those % figures indicate? I certainly don't have a wealth of Cechmanek-based memories beyond disliking him and the Flyers generally so feel free to share.

With respect to comparing goalies across eras -- the Fuhr example you used -- I have no idea where you are getting that from. No one said that 1986 hockey is akin to 2006 hockey, nor would anyone who watched 5 minutes of a game in either era. Making that comparison in absolutes is pointless and reductionist, and yet another entirely invented argument. The comparison is that the shots that Grant Fuhr faced in 1986 were, in all likelihood, of a similar quality to those faced by his peers in 1986. And by similar quality, I mean that analysis has shown that shot quality accounts for somewhere between 10% to 15% of shooting percentage. I doubt anyone has gone far as far back as the 80s to compare the shooting percentage variation over multiple seasons, but those done in the 2000s and since have shown that there's simply no correlation from year to year between shooting percentage performance. You can dismiss it all you want, but that's just reality.

If you, however, did want to compare save% across eras, what you could turn to is -- HORROR OF HORRORS -- MATH AND STATISTICS AND NUMBERS RUNNNNNNNNNNNNNNNNNN:

[C]onsider the 2012-13 NHL season. League-wide, there were 41,827 shots overall and 3,684 goals scored, for a goals against average of 8.81%. Vezina Trophy winner Sergei Bobrovsky had a goals against average of 6.87% -- 29% better than league average, giving him a SV%+ of 129. Taking that 29% better performance into account combined with the number of shots he faced, we say he saved 21.48 goals more than an average goalie would have, given equal playing time.

These stats are especially helpful for comparing goalies across eras: Patrick Roy led the league in save percentage in the 1991-92 season with a .914 SV%; in the 2012-2013 season that would have placed him as the 26th best goalie between Ben Scrivens and Jonas Hiller. However, we can look at his SV%+ of 130 for that year and see that he was actually 30% better than the league average at stopping goals -- better than Bobrovsky, even.

Re: Reimer, which is how this started when you said he "generally blows", he's another example of why looking at stats in a vacuum isn't useful or productive. His second season was marred by a concussion, so is he more likely to be the guy that put up a .923 in 70 games in years 1 and 3, or the guy who put up a .900 in year 2 when he was loopy in 34 games? I'd suggest he's more likely to tilt towards the player he was in year 1 and 3 (and to the extent we want to accept that shot quality exists beyond the 10-15% or so most of us nerds ballpark it at, he was playing behind a lousy defense and seeing a lot of rubber, I think the Leafs have been top 5 in SA every year he's been there, but that info was in the thread this all started in and I don't feel like looking it up).

Obviously, they went out and got Bernier, so maybe they don't think Reimer is the long-term solution there. Then again, Nonis is an idiot (see the Bozak deal, Grabovski debacle, the Bolland and Clarkson contracts, losing MacArthur to a division rival on a cheap deal, signing two useless facepunching meatbags in Orr and McLaren on one hand versus getting Bernier at a reasonable cost/contract on the other), so who knows how this will play out.

And guess what, people who read and digest all these hockey stats love watching hockey. Its not this binary choice between (1) DERP I WORCH DA GAMES or (2) HERP I LURVE EXCEL AND HOCKEY NUMBERS. The people in 2 are 100% in group 1 as well. That's what this all comes down to -- the false conclusion in your mind, as reflected by the thread title, that learning about the game and challenging conventional wisdom equals "you don't watch the games, nerd."

[mikey287]

The noteworthy examples of journeymen having "incredible" seasons was to say that Thomas had two of these in two decades since he's been drafted or so...and yet there's all this noise about him, but these other goalies are just one season away, maybe one and a half whatever, from being in that territory. I'm well aware of the sample size, and it's my point of contention with Thomas. You nailed it.

Re: Stephenson, Melanson, Cechmanek, others. Further illustrations about team aspects of impact on save pct. - not an independent measure of netminding. Many different factors involved, all must be considered. Goes beyond raw numbers.

Re: eras.

I don't believe I said anything about 1986 hockey being akin to 2006 hockey. I believe my comparison was shot quality of B.Smith vs. G.Fuhr. Then M.A. Fleury vs. H.Lundqvist. Perhaps I worded it poorly, my apologies.

The comparison is that the shots that Grant Fuhr faced in 1986 were, in all likelihood, of a similar quality to those faced by his peers in 1986.

Highly unlikely. A random sampling of any number of games vs. similar opponents (or all the games, if one has that kind of time) would not allow one to come to these kind of conclusions. Marked differences in personnel, style, coaching, utilization of different zones/rink geometry. It's that understanding and context that saw Fuhr garner many Vezina votes despite never finishing top-5 in save pct. in the save pct. era.

Not scared of numbers - big math fan - just scared of the hasty conclusions that some can come to with them. Hockey, unlike baseball, features limited isolated 1-on-1 matchups. Many options given the fluidity of the game. Difficult for numbers to compensate for. Good strides have been made, thought-provoking. But ultimately, there are too many variables to overcome.

Re: Reimer. Proof in the pudding. Talent evaluation vs. statistical evaluation. Reimer was not thought to be an adequate full-time solution to get the team to where it wanted to be - though numbers would have suggested otherwise. Talent evaluation properly identified a weakness of the club and addressed it. Reimer kept around to provide rest for Bernier, as I don't believe Bernier has had a 25+ game season in the NHL yet. Time for adjustment is necessary.

Other instances abound on that front.

Re: Watching vs. stats. Wouldn't necessarily challenge whether they watch or not...it's the understanding and conclusions drawn that are in question. The conclusions drawn - as with Reimer for instance - would likely not been the same conclusions drawn if the goalie was seen and evaluated in the context of the game.