Why 5v5 Sv% is KING ...

[mounD]

I wanted to post this because, in the past couple days, I've been having a lot of discussions with people about how to accurately rate goaltenders in the NHL. Some people will just look at "raw" numbers like GAA, Wins, Sv%, etc. to value a goaltender's individual performance, but these statistics are HIGHLY team-based, and in the case of Sv% subject to wild fluctuations given the amount of 4v5, 3v5 opportunities a goaltender's team allows. If a team allows lots of 4v5/3v5 opportunities, and by extension more "quality" scoring opportunities -- which have a much higher rate of success than EV strength opportunities -- a goalie's raw Sv% can really suffer. Is that really the goaltender's fault?

I get the argument that a truly elite goaltender (in the Brodeur, Hasek, Roy mold) will find a way to make saves (including rebound saves) on the PK, most goaltenders are highly subject to allowing more goals-per-shot attempt when their teams are down 1 or 2 skaters. For these reasons, the best indicator of a goalie's true skill (and indeed future performance), therefore, is the metric of 5v5 Sv% (or EV Sv%). Because there are so many pitfalls associated with 4v5 opportunities, and the effect those opportunities have on a goalie's raw Sv%, 5v5 is the most accurate indicator.

Essentially, when predicting the future performance for a particular goaltender, it is best to look at two statistics: 5v5 Sv% and his team's PK% rate.

Think about how we do a "quick" evaluation for skaters when predicting future fantasy-relevant performance. While there are a number of ways to evaluate players (G, A production), including advanced statistics like Corsi, Fenwick, PDO, etc., we generally look at two statistics: a players TOI (including PP TOI), and their SOG production. Generally speaking, if a player is generating a lot of adoring opportunities (SOG) and receiving a lot of ice-time (+PPTOI) we can expect that he will produce sooner or later. The same type of outlook should be used for predicting goalies' future fantasy production, via 5v5 Sv% and his team's PK% rate.

If we look at these two statistics, we are likely to find goaltenders who perform well in 5v5 situations, and those whose teams can back them up with high PL% rates when the scoring opportunities become more lethal. If we overlap the group of goalies with very high 5v5 Sv% rates with the teams who have a very good PK% rate, we are VERY LIKELY to find goaltenders who will perform very well and post high raw Sv% numbers, and by extension lower GAA, more Wins, SO, etc.

Here are a few articles explaining these metrics, and why they are probably the best statistics to use when evaluating and predicting a goalie's production:

Evaluating / Predicting Goalies: http://www.snsanalytics.com/iZqyy8

Shot Quality / 5v5 Sv%: http://vhockey.blogspot.ca/2009/07/s...ntasy.html?m=1

James Reimer and Rebounds: http://theleafsnation.com/2013/7/15/...r-and-rebounds

[Souldjat]

Great stuff.

So is there a where can we actually find a goalies 5vs5 SV%? Is there some website out there that lists these stats in just like a normal stats page instead of bits of pieces of data being thrown into the middle of blog posts?

[mounD]

Great stuff.

So is there a where can we actually find a goalies 5vs5 SV%? Is there some website out there that lists these stats in just like a normal stats page instead of bits of pieces of data being thrown into the middle of blog posts?

Check out Stats.hockeyanalysis.com

You can set up requirements for the Stat reports; for instance, "2010-13 (3yr)" // determine stats by player position // on-ice goal stats, on-ice shot stats, on-ice Corsi, on-ice Fenwick stats, etc.

It's a FANTASTIC resource. It's my favorite website for hockey statistics.

[Pengwin7]

You know my thoughts.
Two comments.


1. I absolutely, 100% agree, that a goalie's SV% is the absolute best way to measure how quality that particular goalie is relative to other goalies.

2. In fantasy hockey, a HUGE contributor to statistical performance is the TEAM that the goalie plays for.
Chris Osgood had a career .905 SV%... which is probably good for about 800th among all-time NHL goalies.
Oh... and Osgood also had 401 wins... which places him 10th among all-time NHL goalies.

SV% is just one category.
We should not forget that.

[mister_mcgoo]

Great resource, thanks for the link, I've been using 5v5 numbers like this for evaluating goalies for awhile now but I hadn't seen that hockeyanalysis.com site before, that's really cool!

The only problem with it, as Pengwin alludes to, is that fantasy leagues count the whole S%, not the 5v5 number so you have to be careful how you choose to evaluate it.

For example, a goalie with a lower combined S% but strong 5v5 S% who is playing on a shit team that is steadily improving is a great candidate to buy low on. But if his team is not improving, or worse yet is in decline, it's not going to matter for the bottom line much and they're probably best avoided.

Question...how do you correlate these 5v5 numbers for goalie S% against how strong the team is 5v5? Because some teams play great 5v5 hockey but suck on the PK for example.

[mounD]

Great resource, thanks for the link, I've been using 5v5 numbers like this for evaluating goalies for awhile now but I hadn't seen that hockeyanalysis.com site before, that's really cool!

The only problem with it, as Pengwin alludes to, is that fantasy leagues count the whole S%, not the 5v5 number so you have to be careful how you choose to evaluate it.

For example, a goalie with a lower combined S% but strong 5v5 S% who is playing on a shit team that is steadily improving is a great candidate to buy low on. But if his team is not improving, or worse yet is in decline, it's not going to matter for the bottom line much and they're probably best avoided.

Question...how do you correlate these 5v5 numbers for goalie S% against how strong the team is 5v5? Because some teams play great 5v5 hockey but suck on the PK for example.

Well, I'm still pretty green about the whole thing, but from whay I understand team quality doesn't really have a great effect on a goalie's 5v5 Sv%: case in point, Florida Panthers' goalie's Roberto Luongo / Tomas Vokoun. Each of those goalies posted very quality Sv% numbers on really bad Florida teams. The thrust of the whole argument is that 5v5 Sv% tends to stay relatively constant from year to year, with the greatest (but not the only) fluctuations coming due to 4v5/3v5 (vice versa) opportunities affecting the numbers.

Clearly a team's PK% rates aren't stand alone: they are also affected by the goaltender, not just the team. I'm sure you've all heard the saying that "a team's best penalty killer is the goalie." And this is very true, as PK% is highly influenced by the goaltender as well as the team. That being the case, the stat isn't something we passively apply to goaltenders from the outside, but is rather a factor of goaltender play AND team quality in shorthanded situations. I didn't mean to say that goalies can't affect their team's PK% (and greatly so), just that when looking for "value" the two most important statistics to look out for are 5v5 Sv% and PK%.

For the record, this isn't meant to pigeon-hole is into only paying attention to Even strength. These numbers are intended to be an accurate predictor of Total Save%, and by extension GAA, Wins, etc. For instance, the James Reimer example from the posted article. Toronto's 5v5 Sv% in the samples was really good, but due to the Leafs' abysmal PK during those samples their team Sv% really suffered. This is to say: regardless of a goalie's skill at 5v5, a poor PK% rate can severely affect the overall numbers to the point where we may not pay attention to value that is, in fact, there.

Just like Mcgoo said, this is a great way to pick out goalies on the rise, and those on the decline. For instance, say Goalie A has a .925 5v5Sv%, but his team's PK% has been absolutely awful for the past three years, helping destroy the goalie's overall totals. If there is legitimate evidence that the team's PK is on the rise, then it's reasonable to believe that Goalie A will begin to post better Sv% numbers overall.

Bringing everything back to fantasy ... these statistics should ALWAYS be weighed against the quality of the team that a goalie plays for. It's been shown that GAA, Sv%, Wins, etc. are more determined by the team than by the goalie himself. It's important to remember this. 5v5 Sv% is just an indicator of a goalie's own skill, while PK% is the most important indicator for a goalie's performance outside 5v5 situations. It's only meant as a tool to recognize value that may not obviously be there.

Hope this helps.

[NewGarbageGoal]

I think it's pretty inarguable that save percentage is the most telling goalie statistic and the best indicator of ability, but I think calling even strength save percentage king is kind of overrating it a bit. Although I do agree even strength save percentage, overall, is a better indicator of talent then overall save percentage.

Like others have said, this is fantasy hockey so even strength save percentage doesn't really matter in the now in the sense that usually only overall save percentages matter. Plus goalie is the most fickle, unpredictable position in hockey really and they typically also take the longest time to fully develop as a player. Some would argue that, once a certain point is reached in terms of talent, there's little difference between most goalies in that a team's performance in front of the goalie matters more then the actual goalie. Obviously you don't want Michael Leighton or Justin Peters in net but a lot of people think the major difference between guys like Hiller, Halak, Price, Vokoun, Emery, etc. is the team they play for.

Goalies also sometimes get overrated or underrated because of the team they play for in both fantasy and real-life hockey. MAF is a perfect example of someone who is or used to be overrated in terms of regular season performance. People say he's a great regular season goalie and that he only falls apart in the playoffs yet if you look back at his career numbers they're really extremely lackluster even in the regular season despite having only played for an elite team most of his career.

The biggest point I'm trying to get at here is that there's lots of ways to look at goalies in real life, but as far as fantasy hockey goes overall save percentage is what matters and the biggest contributors to that are individual goalie talent and team quality.

If your point is to say that ES save percentages are key to evaluating goalie talent and that, as far as fantasy hockey goes, you need to evaluate rising and falling teams as well as goalies then, no offense meant here, but I thought that was kind of obvious.

[Axeman33]

Thanks for the link.

Great stuff there!

[leaftodd]

While I concur with the belief that sv% is the most important stat to evaluate goalies from a fantasy perspective I actually feel its not a stand alone tool. And I disagree even more with 5-5 being king.

Sv % has to be combined with an opinion of a goalies overall talent level along with what team he plays for. Who coaches and with what philosophies, both in general and in specific around game allocation between starter and backup, pp and pk, and overall type of play from a penalty generating perspective. All of this matters. Not any point in my opinion can be considered king.

[mounD]

FYI, if you don't want to take my word for it listen to Jeff Angus in his "Ramblings" from earlier today:

A very interesting read on evaluating and projecting goaltenders in fantasy hockey – from Michael Clifford.

http://sportsjerks.net/2013/07/25/fa...g-goaltenders/

Angus seems to think this argument (re: 5v5 Sv% + PK% being an accurate evaluation/ prediction of goaltender value) has some credibility, as he linked it in his daily ramblings. Just some food for thought, though I'd love to hear his take on this thread.

Cheers.

[Pengwin7]

Two examples, just to test you:

1) Brian Elliott [5-on-5 SV%]
2012-2013 (STL): .9244 (28th among 50+min goalies)
2011-2012 (STL): .9411 (3rd among 50+min goalies)
2010-2011 (OTT): .9003 (67th among 50+min goalies)
2009-2010 (OTT): .9028 (66th among 50+min goalies)

2) Jaroslav Halak [5-on-5 SV%]
2012-2013 (STL): .9091 (53rd among 50+min goalies)
2011-2012 (STL): .9374 (4th among 50+min goalies)
2010-2011 (STL): .9184 (45th among 50+min goalies)
2009-2010 (MON): .9348 (8th among 50+min goalies)

These numbers are all over the map.
Why?
Discuss.

[mounD]

Two examples, just to test you:

1) Brian Elliott [5-on-5 SV%]
2012-2013 (STL): .9244 (28th among 50+min goalies)
2011-2012 (STL): .9411 (3rd among 50+min goalies)
2010-2011 (OTT): .9003 (67th among 50+min goalies)
2009-2010 (OTT): .9028 (66th among 50+min goalies)

2) Jaroslav Halak [5-on-5 SV%]
2012-2013 (STL): .9091 (53rd among 50+min goalies)
2011-2012 (STL): .9374 (4th among 50+min goalies)
2010-2011 (STL): .9184 (45th among 50+min goalies)
2009-2010 (MON): .9348 (8th among 50+min goalies)

These numbers are all over the map.
Why?
Discuss.

To be fair, you raise a good point here; however, 50+ minutes is an INCREDIBLY SMALL requirement sample size. Statistical trends bear themselves out over very large (at least appropriately large) sample sizes, and to use a requirement as small as > 50 minutes played is the equivalent of "cherry picking" statistics to fit an argument.

If you had included large samples, and the numbers still supported your viewpoint -- that 5v5 Sv% fluctuates wildly from year to year -- I would be much more inclined to give credence to these statistics and your argument as a whole.

[Pengwin7]

To be fair, you raise a good point here; however, 50+ minutes is an INCREDIBLY SMALL requirement sample size. Statistical trends bear themselves out over very large (at least appropriately large) sample sizes, and to use a requirement as small as > 50 minutes played is the equivalent of "cherry picking" statistics to fit an argument.

If you had included large samples, and the numbers still supported your viewpoint -- that 5v5 Sv% fluctuates wildly from year to year -- I would be much more inclined to give credence to these statistics and your argument as a whole.

Well... if you are going to be a numbers guy:

i) Disregard the 50 minute thing. All that would do is change the "rank" of each goalie. The actual SV% value does not change. Increasing the minutes just weeds out the goalies that played seldom. I'm sure each of Halak & Elliott have played 500 min per season, probably closer to 2000 min per season.

ii) I don't understand these "large samples". The other day you were comparing the last 3 years of Mike Smith to the last 3 years of Hiller. I'm using the last 4 years.

So... my example is too small, but yours is fine???

Over the last 4 years:
Elliott has started 155 games.
Halak has started 161 games.
M.Smith has started 157 games.

These numbers seem similar to me.
But my comparisons are not big enough? Not valid for analysis?

Hmm... I don' t like this game.

[mounD]

Well... if you are going to be a numbers guy:

i) Disregard the 50 minute thing. All that would do is change the "rank" of each goalie. The actual SV% value does not change. Increasing the minutes just weeds out the goalies that played seldom. I'm sure each of Halak & Elliott have played 500 min per season, probably closer to 2000 min per season.

ii) I don't understand these "large samples". The other day you were comparing the last 3 years of Mike Smith to the last 3 years of Hiller. I'm using the last 4 years.

So... my example is too small, but yours is fine???

Over the last 4 years:
Elliott has started 155 games.
Halak has started 161 games.
M.Smith has started 157 games.

These numbers seem similar to me.
But my comparisons are not big enough? Not valid for analysis?

Hmm... I don' t like this game.

Well, the large vs. small sample size argument isn't to discount any of the statistics you included, but rather to weigh them together. For example, instead of looking at the goaltender's 5v5 Sv% for 10-11, 11-12, 12-13, etc. it's best to look at the goaltender's 5v5 Sv% from 2010-13 as ONE large aggregate sample. When this isn't possible, we should view the largest possible relevant sample we are able to attain.

For the purposes of the statistics you included it would be more appropriate to exclude small sample sizes. You state that the "50 minute thing" would only change the ranks of each particular goalie, that's not quite true. If a goalie has "only" played, say, 500 minutes that's less than 10 total games played (600 minutes). Over a sample size that small, it's very difficult to accurately judge a goalie's performance -- and even more difficult to project future performance -- given the fact that one blowout loss, or an impressive shutout, can wildly skew the results of that inquiry/sample. For this reason, we want to use the largest sample size possible.

When I was comparing the past three seasons for Mike Smith vs. Jonas Hiller, I wasn't using each season as a standalone statistic, but rather the entire aggregate of 2010-13 for each goaltender -- for the reasons described above. To apply this line of thinking to your examples regarding Halak and Elliott, lets take a look at those sample seasons: (I will post analysis post EDIT; I'm on my phone and if I move between windows I risk losing this response -- please wait to reply)

2009-10 Halak (2052 minutes) -- .9348 5v5 Sv%
2010-11 Halak (2585 minutes) -- .9184 5v5 Sv%
2011-12 Halak (2123 minute) -- .9374 5v5 Sv%
2012-13 Halak (614 minutes) -- .9091 5v5 Sv%

2009-10 Elliott (2332 minutes) -- .9028 5v5 sv%
2010-11 Elliott (2353 minutes) -- .9003 5v5 Sv%
2011-12 Elliott (1720 minutes) -- .9411 5v5 Sv%
2012-13 Elliott (997 minutes) -- .9244 5v5 Sv%

Disregarding the 2012-13 season (because the sample size is rather small), we are left with very consistent numbers from Jaroslav Halak over the 2000+ 5v5 minute samples. The 2010-11 season provides the greatey outlier season, at around 20 points difference from the 09-10/11-12 bookend seasons. This variance can be partially explained by sample size, as ~2000 minutes isn't an awful lot of 5v5 time for a starting goaltender to be rated on -- equating to roughly 30+ games worth of EV strength time. We wouldn't want to cast judgment on ANY goaltender based on only 30 games of experience, and if someone on these boards were to make sweeping statements about a goaltender with limited experience (Steve Mason post-09, Braden Holtby right now, etc.) I would be very cautious following that judgment. This isn't to say that the analysis is definitely incorrect, but merely that because of the sample size the judgment is inconclusive.

As for Elliott, if we exclude his #s from the 2012-13 season (again for the small relative sample size), we are left with the outlier season of 2011-12 in which he recorded a .941 at 5v5. Similarly to the above arguments regarding sample size, 1720 5v5 minutes isn't exactly a large enough sample size to cast judgment/predict based on a goaltender's performance -- equating to less than 30 games worth of 5v5 ice time.

I don't mean to say that these statistical outlier seasons aren't at least somewhat predictive of a goaltender's performance during those seasons; however, it is problematic to do so because the statistical sample is so small when compared against a goaltender's entire career at 5v5. I don't mean to completely discount these statistics. In fact, I wish to include them in our general assessment of Halak/Elliot, but instead of viewing them by themselves as individual statistics per season, we get a much clearer picture when we view them as part of larger samples (generally a 3-year window can be indicative, since over time we can expect the effect of outlying satistics to be much lesser than in a smaller sample).

I don't claim to know a lot about statistical analysis. I'm pretty much a newbie in that regard; however, the one thing I do know -- and wish to stress in this thread -- is that the larger the sample size, the more predictive and accurate the statistics themselves are. If we view statistics from relatively small samples, and make assessments and predictions based on those statistics, we are more prone to giving uninformed advice that may not accurately reflect the bigger picture. For the sake of argument, here is the "bigger picture" I'm talking about:

2009-13 5v5 Sv% for Halak/Elliott:

Halak (7400 minutes) -- .9281
Elliott (7400 minutes) -- .9131

These statistics are more likely an accurate picture of the skill level of each goaltender, IMO. Elliott's two most recent seasons (11-12/12-13) are relatively small sample sizes, and therefore we should avoid giving them increased statistical weight and instead look at the aggregate numbers to predict/project his actual skill going forward. The same could (and should) be said for Halak's 2012-13 season, as ~600 minutes isn't a relatively large sample size. I don't know about you, but I would much rather give increased statistical weight to a sample size of ~7500 minutes than a sample size of less than 2000 minutes. In the former we can expect a greater regression to the mean (read: a goaltender's actual ability/value), whereas in the latter we risk the pitfalls of particularly great/bad games affecting the accuracy of the sample's statistics.

Cheers.

[Pengwin7]

Disregarding the 2012-13 season (because the sample size is rather small), we are left with very consistent numbers from Jaroslav Halak over the 2000+ 5v5 minute samples. The 2010-11 season provides the greatey outlier season, at around 20 points difference from the 09-10/11-12 bookend seasons. This variance can be partially explained by sample size, as ~2000 minutes isn't an awful lot of 5v5 time for a starting goaltender to be rated on -- equating to roughly 30+ games worth of EV strength time. We wouldn't want to cast judgment on ANY goaltender based on only 30 games of experience

...
...

For the sake of argument, here is the "bigger picture" I'm talking about:

2009-13 5v5 Sv% for Halak/Elliott:
Halak (7400 minutes) -- .9281
Elliott (7400 minutes) -- .9131

These statistics are more likely an accurate picture of the skill level of each goaltender, IMO.

Long post... and good summary of statistics.
But now, for Part II:

1. It appears (from your statistics) that there should be NO reason to worry about Jaroslav Halak's performance in this coming season: 2013-2014.
His .9281 5-on-5 SV% is phenomenal.
Based on the information you collected, Halak should be the absolute bonafide #1 goalie for St.Louis this year and perform suchly.

2. It appears (from your statistics) that Brian Elliott is not even CLOSE to being as good a goaltender as Jaroslav Halak.
We should assume that Elliott has no chance to win the #1 goaltending job in St.Louis.


Are these your opinions on the St.Louis goaltending situation???