Bayesian Analysis, Preseason, and Ryan Strome and Anders Lee.

Jonathan Willis wrote a post for Oilers Nation about how we should use Bayesian Analysis when it comes to preseason and training camp.  It’s a great post and I recommend you read it, but I want to expound upon it here quickly as it comes to the Isles.

“Bayesian Analysis” is a fancy name for a pretty simple and common sense concept.  If we know something about a player, a team, or anything really and then we acquire new knowledge about the player/team/thing, we shouldn’t just throw out our prior knowledge – even if the new knowledge contradicts our old understanding.  Instead we should evaluate how strong our prior understanding (our “prior”) was and how strong the new evidence is and combine the two to try and get a more complete picture.  If our prior is very strong, it should take very strong new evidence to change our mind to make us believe that our prior was wrong.  If our prior is weak, of course, it takes less strong evidence to change our minds.   You can do this with math (Willis demonstrates it in the post above), but it’s a really basic concept.

This is a pretty important thing to remember for preseason.  NHL Preseasons are VERY short.  We have only 7 games (two of which are split squad, so players can at most play in 6 of them) – of which players won’t play in every one of them.  We have a month of camp – which is not the same as game time obviously – for teams to look at players.  This is a very short time to evaluate over 40 players, 25-30 of which have a shot at making the roster.

In other words, the new information a team can obtain through preseason is never going to be very strong.  Even the best possible camp a player can have shouldn’t be considered that strong of evidence, as this is pretty much the definition of a “Small Sample Size.”  And so, if you have a strong prior at ALL about a player or group of players, training camp/preseason pretty much should NEVER change your mind.

——-

Let’s look at two potential NHL forwards to see an example of what I mean.

Anders Lee:  Anders Lee is a guy who has never been a pretty highly rated prospect.  He was drafted late in a draft as an overager, wasnt even ranked top 100 CSS pre-draft, and some experts (Corey Pronman) didn’t even have him in the Isles top 5 prospects – hell Pronman didn’t even have him top 10.  So our prior going into last year was basically not to have high expectations.  Then he had a good AHL season and an incredibly good 22 NHL games – both in traditional and fancystats.

So going into camp, what’s our projection of Lee as an NHL player?  We should think he probably is ready to be one, and we should think there’s a chance, although not necessarily the highest chance, that he’s a top 6-er.  So our prior here, in an atmosphere where there are a ton of bodies competing for 12 jobs, isn’t super strong that he deserves one of them (especially given his waiver status).  So Training camp and preseason performance should influence us, and the team, a little bit.  It shouldn’t matter more than 22 NHL Games of course – seriously, think about that for a second – but it may influence us a little.

Ryan Strome:  By contrast, Strome has been the Isles top or one of their top 2 prospects since his draft day.  He was drafted 5th overall, and had strong juniors #s.  He debuted in the AHL last year and managed to LEAD THE LEAGUE in POINTS in mid-December, forcing a call up – all at age 20 (by contrast, Lee was 23).  Strome finished the AHL with 1.32 points per game in 37 games.  He then played 37 NHL games, managed 18 points despite frequently being with lousy linemates, and had good overall fancystats with a +3% relative corsi while not being sheltered.

So going into camp, what’s our projection of Strome as an NHL player?  Well we should be pretty damn sure he’s already capable of being one and we should be pretty damn convinced he’s a top 6-er.  He could have the absolute worst camp ever, absent injuries, and our opinion on Strome should still remain pretty much the same: He should make the team.  This doesn’t mean he should necessarily play Center over other guys, but there’s no chance you should be persuaded that he doesn’t deserve one of 8 wing spots over any of the other guys.  Our prior is that strong.

——

In short, everything we know about Ryan Strome, makes the idea that he has anything to prove in training camp absolutely absurd.  He’s got the pedigree, 37 amazing AHL games, and 37 strong NHL games behind him.  What exactly in 6 (less actually) preseason games and a month of camp should be able to change our minds about him?  How exactly can you believe, from anything in camp, that the team would be stronger for Strome off the team?

Only by forgetting what we know about Strome – about any players – prior to camp, can you make that argument.  And that’s obviously stupid (hey if Tavares had a bad camp, leave him off!).  Shouldn’t it be for the team as well?

Context is Important – But its just as Important not to OVEREMPHASIZE it.

One of the most frequent complaints “anti-analytics” people make about so-called advanced stats is that these stats fail to adjust for context – some players have harder minutes than others.  This complaint is of course bunk, and a complete straw man – analytics has come up with numerous methods on calculating how difficult a player’s minutes are: From measures of competition, to measures of teammates, to measures that contextualize where players start their shifts most often (zone starts) – we have multiple metrics to explain EACH of these things.  Sure you can use fancystats incorrectly and completely ignore context, but good analytics work doesn’t do that, and it’s pretty easy to see where someone is and isn’t attempting to take into account context.

That said, with the new season coming up, it’s important to remember not to go TOO FAR when doing this.  That is, context is important, but context doesn’t explain EVERYTHING.  And people often tend to forget this.

Continue reading

The Process of Drafting Goalies: – A look at Isles’ goalie drafting since 2008

I wrote a piece on Hockey-Graphs this week about how we should and should not draft goalies in the NHL and whether things will change.  I wanted to apply this to the Islanders’ drafts from the Garth Snow era.  For those who are unfamiliar with how analytics treats goalie drafting, and I went over how we deal with goalies in general in my last livecast:

But in sum, there are a few points to remember about goalies:

1.  Goalies are HIGHLY variable over single seasons due to small sample sizes and the small differences between NHL goalies.  A goalie can be great one year and then be lousy the next year -> in fact when a goalie does have a great year, we do expect him to take a clear step back the next year due to regression.  With relatively few exceptions (Rask), you should never sign goalies to long term or expensive contracts, because the odds of failure are really high.
2.  Goalies peak REALLY early (age 24 or thereabouts) and then decline throughout the rest of their career.  This is on average, of course – some individual goalies may peak later and some will not suffer the effects of aging for a while longer than we’d expect, but for the most part, goalies decline every year and expecting a goalie over 24 to get better is making a sucker’s bet.

The above two points would seem to emphasize the importance of drafting goalie talent – after all, drafted players are the cheapest (not expensive) and will be on your team at the youngest age, near a goalie’s peak, so as to avoid the issues of goalie decline you get on the free agent market (most goalies don’t hit UFA till age 28 or even mid 30s).  But this brings us to point 3:

3.  Goalie prospects are far more variable than skater prospects and have a much higher rate of failure, due to the difficulty of scouting goalies and the extreme small sample size of pre-draft goalie prospects.  As such, you generally should NOT draft goalies early, and probably never draft a goalie before the third round.

So yeah, this doesn’t make it easy – to a large extent, you simply have to get LUCKY to draft a good goalie.  But it helps to have a good process, such that bad goalie drafting luck doesn’t result in you missing out on better skating prospects than you have to.  So let’s look at Isles drafted goalies under Garth Snow:

 
Continue reading

Josh Bailey vs Colin McDonald – Comparable?

In the Lighthouse Hockey comments on my last post (Crossposted at this blog as well), one commenter objected, arguing that Colin McDonald is a superior player to Bailey who simply hasn’t gotten the opportunities Bailey has.  I’d like to go into this comparison a bit more in depth than I did in the comments, where I misremembered and made an incorrect reference to how Bailey played with Reasoner (I was thinking of Grabner).

What do we know about Colin McDonald?

CMac

Above are Colin McDonald’s statistics as gathered through behindthenet.ca, stats.hockeyanalysis.com and my own neutral zone tracking.  What you see is a player who probably doesn’t drive possession, or makes his teammates slightly worse when he’s on the ice at pushing the puck forward – you can’t see it in these statistics, but if you use his two-year WOWYs at Hockey Analysis (See HERE), you find teammates have been on average 0.94 percentage points worse with him on the ice than off the ice.   McDonald has absolutely played with brutal linemates last year – Matt Martin and Casey Cizikas are unequivocally not very good at all and has at best played with decent linemates (Grabner, Aucoin).  That said, the WOWYs suggest that while he’s not as bad as his relative corsi would suggest last year (-7), he’s not good at driving play forward.

This shouldn’t be a surprise really – as I’ve noted before int he neutral zone #s (see the right side of the above #s), McDonald is a guy who makes a LOT of his line’s zone entries, whether he’s with the 3rd or 4th line (those 30.4 and 30.6% rates are amongst the highest on the entire TEAM), but he dumps the puck like crazy.  Some of that is his line position – while on the 4th line this year his dump rate was an abysmal 28% – but even on a more skill oriented line on the lockout year, his carry in % was a pretty bad for a forward 40%.  A guy who takes so many entries and dumps so damn often will have a hard time driving play unless he’s particularly good in other areas, and it doesn’t seem like CMac is one of those guys.  He also doesn’t deny opponents the ability to carry-in, even when playing with better players at that skill as he did during the lockout year.

As for scoring, CMac doesn’t really do that much either.  Again some of that may be linemates (Matt Martin is the worst shooter on the team, and that’s CMac’s most common linemate), and it’s not like the point totals are horrible, but his #s the last two years come in at a point per minute rate (min 1250 minutes) that puts him 168th out of 221 qualifying forwards.

One thing Cmac does well is he draws penalties and doesn’t take that many.  That’s a useful skill – last year he basically netted the Isles 2 goals through a +12 penalty differential (1 goal last year).  This is an underrated skill yes, but 2 goals isn’t exactly huge/

So what we have here is a guy who slightly hurts possession, who doesn’t score very much and can draw penalties.  That’s not a great player, although it’s not the worst guy to have on 4th line – course you don’t want him to be on a line where he’s the best player, which was the case on last year’s 4th line!

Okay so what about Bailey?

Josh Bailey isn’t a huge possession driver either, although he’s more positive than CMac.  Some of this may be linemates, but again if you use a two-year WOWY (see HERE) and average it out you wind up Bailey’s teammates being 2.84% better with bailey on the ice than without in corsi.  This is probably being generous to Bailey’s effects – adding a third year will put him a little worse, but you clearly see a small positive effect over the years.  Bailey’s zone starts are similar to Cmac’s so that isn’t the difference either.

Bailey is also a better scorer, being 110th in EV points per minute over the last 2 years.  He also is okay in the neutral zone at carrying it in, and at denying opponents from doing the same (although Nielsen deserves much of that credit).  He’s not a GOOD scorer, but he’s okay, and he’s probably going to shoot at a higher % next year, since this year’s shooting % was almost certainly a fluke.  That said, he doesn’t shoot enough – the lockout year was the first time he had a decent shooting rate, and he stepped back to his career average this year.  The one thing Bailey isn’t is a drawer of penalties, he’s -1 over the last two years (net effect is basically null).  

So again, these #s don’t suggest Josh Bailey is a great player.  But he’s almost certainly a decent one who you would be thrilled with at third line wing, which is now what he is if we consider Frans 3C due to Grabovski.  And the chemistry with Frans has worked the last two years.  

Conclusion:

In short, CMac and Bailey aren’t really comparable at all, except as wings on this team.  CMac’s certainly proven himself to be an okay 4th liner, but this team has SO many forwards that he shouldn’t be guaranteed a spot – except over Matt Martin and Casey Cizikas.  Alas, those guys will probably get the spot over him.

[Cross Post] Don’t Trade Skill: PLAY Skill – Why should the Isles trade Bailey or Grabner?

[This Post was also posted at Lighthouse Hockey.  Feel free to go over there to comment, the comment section should be more lively there.  You can comment here of course, but the post is cross-posted here for posterity's purposes]

One thing that’s been taken for granted since the first two days of free agency, was that the Islanders were going to have to trade a forward.  After all, there are currently 13 one-way contract forwards on the roster (including Eric Boulton) as well as 3 highly talented prospects on two-way deals who you’d think the team would like to make the opening night roster (Anders Lee might in theory be sent down for roster reasons, but it seems highly unlikely that Ryan Strome or Brock Nelson are going anywhere).  Somebody has to go then – the team can’t really carry 15 forwards on the opening night roster (14 is very doable – 15 means the team basically can’t carry an extra D Man on the roster plus it means you’re scratching 3 healthy forwards in a night, which doesn’t seem like a good use of resources).  I’m telling you all nothing new, obviously.

The other thing that’s been taken for granted is that the team is most likely to accomplish this by trading Josh Bailey or possibly Michael Grabner.  After all, Bailey seemed to underperform for his contract last year – the overall point #s aren’t bad, but he simply didn’t score goals as he did the year before (He went from 1.27 G/60 in 13-14 to 0.36 at even strength), his worst rate in the last 4 years.  And then there was the insane streak where he basically had no point production whatsoever – nearly half of his points came after the olympic break too.  Add in a not cheap (although not expensive) contract with 4 more years on it, and it’s not an unreasonable thought to think the Isles should deal him and they obviously have put him out there.  Michael Grabner is a slightly different case – Grabner is obviously valuable in ways Bailey isn’t outside of scoring (elite penalty killing, greater goal scoring potential, strong defensive play, ability to play with other centers besides Frans Nielsen, etc.) but like Bailey he had an insane cold streak which (stupidly) resulted in a one game healthy scratch and unlike Bailey, his backloaded deal gets a bit pricey for the Isles starting next year.

That said, if the Isles can afford to carry Bailey and Grabner’s salaries (and no report from Staple says otherwise, but rather that the Isles are hoping to swap one for a top 4 D Man), this basically makes no sense for multiple reasons:

First, you’d be dealing either guy from a position of weakness – both guys had down years last year and are extremely likely to bounce back stronger – Bailey’s shot rate dropped from 12-13’s increase, but his shooting percentage also dropped which is a likely fluke (Shooting % fluctuations tend to be random, and Bailey has always been an above average shooter, but was below average last year)- he should be good for at least a few more goals next year.  Grabner is similar – his drop in goals was nearly entirely shooting % driven – at his career rate of shooting (12.3%) he’d have averaged 22 goals per 82 games instead of his rate of 15 last year, and Grabner has twice shot above that mark.   Grabner may also have better linemates for scoring next year too.

In short, you’d be selling low on either guy, when both guys are likely to have a bounce back season next year simply due to regression.  That’s generally a nono for any sports negotiating – instead of selling high and buying low, you’d be selling low and hoping to buy low.  You’re less likely to get a great (presumably Defensive) piece back given the seasons each guy is coming off – you might be better off trading midseason.

And you really should only trade either guy for a clear 2nd pair caliber D man for a simple reason:  Both guys are pretty good players who are amongst the Islanders’ current top 12 forwards.  Yes, the Isles have 15 non-goon forwards, but why on earth would you want to play Matt Martin, Casey Cizikas, or Colin McDonald over either guy (arguably Cal Clutterbuck too)?  Even WITH their bad years both guys were better scorers than those 3 grinders, and they were strong possession players as well.  Bailey’s possession #s won’t likely be that great again unless he plays with Frans, but they’ve always been better than Martin’s.  Seriously, look at this:

2014islesfs_medium

Let’s take a quick tangent here:  Why do teams play GRINDERS on the 4th or third line?

The answer isn’t that teams are better off with grinders rather than skill players in those spots – there’s a reason why olympic teams stack their rosters full of skill players instead of grinders, skill players are simply better at playing hockey in most ways. No, the reason is simple: With 30 teams and a salary cap, there simply aren’t enough skill players for teams to roll 12 skill forwards, so teams rely on grinding lines to eat up ice time (getting their skill players time to rest) and to hopefully prevent opposing teams from scoring in the mean time until the skill players can get back on the ice.

But some teams – like Chicago or the Rangers last year – do have enough skill players to fit four lines, and it tends to lead to being a pretty successful team.  After all, this means that no matter when your opponents have out a line that basically can’t score and is limited to trying to playing defense, you’re icing a line that can defend AND score, giving you a clear advantage.

This doesn’t mean teams can’t be cup contenders with 3 skill lines and a grinding line – see Boston, the East’s best team for a few years now for example – but in those cases, the team tends to have skill forwards and D Men who are elite enough that the grinders’ issues aren’t that costly (and in Boston’s case, a top 2 goalie).  Still, teams like Boston and Pittsburgh would also rather have skating talent on the 4th line and don’t simply because their other players have them RIGHT AGAINST THE CAP.  The Isles don’t have that situation – so why should they suffer the same limitation?

Again, let’s bring up the #s again:

2014islesfs_medium

The highlighted stat is relative corsi – the difference in the corsi (Shot differential including blocked and missed shots) of the team while each forward is on the ice from the corsi of the team with them on the bench.  In short, with those 3 forwards on the ice (especially Cizikas and Martin, since CMac was better when he played with other centers), the dropoff on the Isles was dramatic – they were pretty damn bad.  But you can practically use any other metric and you get the same thing – raw corsi only has Nikolai Kulemin as worse and well that was largely the impact of playing on the Leafs.  Points per 60 at EV again has Martin and Cizikas clearly at the bottom (and not the first time in Martin’s case) with again Kulemin (again, they’re hoping this is a Toronto thing and he had tougher minutes than any Islander) and now Clutterbuck (who has a better track record) the only other guys below CMac.  And no, context doesn’t explain it either – while the grinders had slightly defensive minutes in terms of faceoffs, they weren’t extreme defensive minutes enough to account for this, and nor were they minutes harder really than any other option for the same minutes other than maybe Cory Conacher (The only guys who got really easier minutes than the Isles 4th line in terms of faceoffs were the Tavares line, for obvious reasons).

So it’s clear that Martin and Cizikas are clearly inferior to the other 13 forwards, and CMac is pretty likely inferior to the other 12 guys as well.  So you DO NOT HAVE A JAM OF FORWARDS THAT MAKE IT SO YOU NEED TO TRADE A SKILL FORWARD – you have enough skill forwards (11 or 12 depending on what you call Clutterbuck) to make a full team and then you have a bunch of not needed grinders.  Trade those guys!  Don’t trade skill forwards at their lowest value unless you can get a clearly very positive thing in return.

I recently posted on my IslanderAnalytics blog on how the way analytics helps teams the most isn’t by helping them outsmart opposing teams but by preventing them from making stupid errors that those other teams might make instead.  This is one such case -> there’s nothing brilliant about playing a team of 11-12 skaters and without a designated “checking line.”  But there is something dumb over playing a checking line when you’re not forced to by cap or personnel issues – it’s rare to have the opportunity not to, but it’s right there for the Isles to grasp.  There’s no reason to trade skill players just to play inferior players at the same spots – and the Isles shouldn’t do so.

How Analytics Help an NHL Team the most – Preventing Stupid Mistakes

In the last few weeks, four prominent analytics-minded individuals have been hired by NHL teams.  First Sunny Mehta of the classic Irreverent Oilers Blog (and poker player) was hired to head New Jersey’s brand new analytics department.  Second, Eric Tulsky (@BSH_EricT of SB Nation’s OutNumbered blog, amongst others) announced he was hired by an anonymous NHL team.  Then Kyle Dubas, GM of the CHL’s Soo Greyhounds, who was known to be using analytics to try and succeed despite being a small CHL market team, was hired to be the assistant GM of the Toronto Maple Leafs.  Finally, just yesterday, Tyler Dellow (@mc79hockey) was hired by his favorite team, the Edmonton Oilers.  These very public hirings have given rise to people wondering how much the signings will improve the team.

However, while all the sexy talk about analytics revolves around these guys changing how teams play, the biggest way an analytics guy, or just analytics in general, can improve these teams is much much simpler:  By preventing teams from making stupid mistakes.

Phil Birnbaum wrote a post on this, referencing baseball, which applies very much to hockey here.  Analytics only helps a team outsmart other teams to the extent other teams don’t have people looking into the same things.  By contrast, Analytics will always help you avoid stupid mistakes that could set you back quite a bit.

Take a look at this past free agent market.  The Isles were looking for a top 4 D Man.  Let’s say they followed the analytics and came to believe Anton Stralman was their man, and offered him 20 Million over 5 years.  That would’ve probably been fair value for what Stralman offered – quite a bit due to his great possession play – yet it would’ve been more than what conventional analysis would’ve said Stralman was worth.

And here’s the thing:  THEY STILL WOULDNT HAVE WON THE BIDDING FOR STRALMAN.  Correctly following the analytics #s only helps when others aren’t doing the same thing – the Isles STILL can’t outbid richer teams who are on to the same targets, really.  In this case, being smart analytically wouldn’t have helped the Isles at all.

Now again, let’s look at another free agent D-Man: Brooks Orpik.  Orpik was considered a top 4 D Man in Pittsburgh by mainstream media.  He made Team USA for goodness sakes.  But the possession #s suggest Orpik is long past his prime and is no longer any good at all – rather he’s a freaking boat anchor.  By paying attention to the possession #s, the Isles would know not to bid much at all for Orpik, and thus the analytics serves a big benefit.  (You could argue the Isles couldn’t afford Orpik even if they thought he was good, but you can use any other bad D man, such as Derek Engelland instead).

This applies to more than free agency of course.  

Last year, there were no analytics publicly available that would suggest that Calvin de Haan was either ready for the NHL or that he’d be any good in it.  (It’s possible the team had such of course, but we didn’t, and there will be situations where the team doesn’t have enough data to make an analytics-based decision.)  Despite this, we DID have analytics available about the alternatives – Radek Martinek, Matt Carkner, and Brian Strait – all being bad options.  Analytics would’ve told you that, and would’ve possibly resulted in a call up of what turned out to be one of our better D.  Was there any guaranty that would be the case?  No.  But the unknown is more likely to be a better outcome than a known bad.

We’ll go into what are the known bads in an upcoming post – there are still a few on the Isles current roster.  Analytics won’t necessarily make the other guys any better, but avoiding the known bad options would already be a significant win.