Can forwards affect SV%

[BarryBSB]

There'd be a 1 in 1024 chance of this happening randomly

As a further hypothetical, what if we were to identify 20 such 4 year streaks over a 10 season sample.

 

[Doctor No]

The math professor in me loves hypotheticals.

Supposing that save percentage is not affected by forwards, and that everything observed is just due to random chance, 1 in 16 skaters will finish better than their team average for four consecutive years.

Suppose that there are 450 players who have four consecutive seasons in the league (that'd be 15 per team).

Just by random chance alone, I would *expect* 28 players to be identifiable as exceeding their team average on-ice save percentage for four consecutive years.

Twenty would be a piece of cake - I bet you could find them right now if you wanted to. And that's over four years. You give me a ten-season sample? You'd be able to stumble over twenty accidentally.

 

[lomiller1]

IMO something people miss when thinking about shot quality is that bad defensive players don't just give up more high quality shots they give up more shots from all ranges. The increased number of longer shot attempts given up tends to cancel out the increased number of high quality shot attempts as far as save % goes.

You are much better off looking at either shots from high % areas or total shot attempts given up because it actually makes a pretty good proxy for scoring chances.

 

[tfong]

It's not what I believe, it's what is true Whatever ability a forward has to affect his goaltenders' save percentage is so tiny that it's buried by randomness. You're always going to find outliers -- players who seem to have an abnormally high or low on-ice sv% -- that doesn't mean those players are causing it.

Oh I see what you mean. So really its because the measurement of on-ice sav% is so un-precise that you can't use it to judge the defensive play of a forward currently.

 

[CDN24]

I would expect the stat to mean something, they key will be they why does that happen. I would expect some 4th liners to have the highest on ice save percentage as they will be matched up against some of their counterparts who should have low shooting percentages- they are on the 4th for a reason. Similarly I would expect lower numbers for some noted shutdown centers as they play a much larger percentage of their time against guys with better shooting percentages. Much like the old shut down 3rd liners of yesteryear were usually minus players. Success came in limiting Gretzky to 1 goal not getting his line to zero.

 

[Micklebot]

Short answer is absolutely, however in practice no.

I mean, if I stepped out on the ice instead of whoever, sv% would plummet, because my team would effectively be on the PK the entire time, but players at my skill level don't last in the NHL (AHL, ECHL, or even most beer leagues).

The problem is, top defensive players tend to play against top offensive players, and even then, they are just one person in a 5 man unit.

 

[Ole Gil]

It's just common sense. Forwards obviously can affect the sv%.

Other things that affect save%: Linemates, defensmen, goalies, opposition, luck, on the rush vs. in the zone, odd man chances, ice quality, shot distance, screens, etc..............

There are enough things going on that you'll be hard pressed to find a nice pile of dots with a perfect line through them, but you can easily imagine a situation where a forward taking a particular action leads to a lower % shot. So a forward who does those kinds of things more often...

 

[Bear of Bad News]

It's just common sense. Forwards obviously can affect the sv%.

Of course they *can*, in theory. Do they, in practice?

"Common sense" isn't always true, which is one of the points of analytics to begin with.

(By the way, no one in this forum ever expects to find a "nice pile of dots with a perfect line through them", despite what you appear to believe).

 

[Ole Gil]

Have you ever seen a forward break up a 3 on 2 with a backcheck? Ta da! They've done it in practice. Amazing!!

Trying to find the correlation between a single forward and save% usually doesn't work because there are just shy of a bajillion moving parts.

 

[Bear of Bad News]

Have you ever seen a forward break up a 3 on 2 with a backcheck? Ta da! They've done it in practice. Amazing!!

Trying to find the correlation between a single forward and save% usually doesn't work because there are just shy of a bajillion moving parts.

Not the question I was asking, so let me be more clear. Does it make a difference at the end of the season, IN PRACTICE?

Yes, some forwards can limit other teams' shots. Do they disproportionately limit high- (or low-)opportunity shots, or is the limiting impact spread across the spectrum of all shots?

Are you suggesting that some forwards choose to only limit high-opportunity shots, but would choose to not limit low-opportunity shots?

 

[Ole Gil]

Not the question I was asking, so let me be more clear.

Does it make a difference at the end of the season, IN PRACTICE?

In practice, if you break up one 3 on 2, and have no effect on sv% for the rest of the season, you've impacted sv% IN PRACTICE!!

 

[Bear of Bad News]

In practice, if you break up one 3 on 2, and have no effect on sv% for the rest of the season, you've impacted sv% IN PRACTICE!!

Thanks for the contribution to the thread.

 

[Ole Gil]

Yes, some forwards can limit other teams' shots. Do they disproportionately limit high- (or low-)opportunity shots, or is the limiting impact spread across the spectrum of all shots?

Are you suggesting that some forwards choose to only limit high-opportunity shots, but would choose to not limit low-opportunity shots?

Yes, there are certainly forwards who choose to disproportionately limit high-opportunity shots, and that would obviously lead to a higher sv%.

Although, I do see the point that determining how some forwards would affect the sv% is not so straight forward.

 

[eklunds source]

Yes, there are certainly forwards who choose to disproportionately limit high-opportunity shots, and that would obviously lead to a higher sv%.

Although, I do see the point that determining how some forwards would affect the sv% is not so straight forward.

The point isn't that "no forward can possibly have any long-term affect on his goaltenders' save percentage"... The point is "any affect that forwards have are so small compared to the fluctuations we see that we can safely ignore it, or consider it good fortune".

It's not about nailing down one ultimate formula that aggregates everything into one easy rating, it's about finding loose trends in imperfect data with not enough of a sample size.

 

[Bee Sheriff]

Why stop at your own team's forwards? The other team's forwards take shots, those affect save percentage too.

 

[Viktri]

I mentioned above that given a sample of 16 players, you would expect one to outperform the team average in four straight seasons solely due to random chance.

That person might be Stepan.

Yes, this issue is significantly more complicated than it may initially appear and there seems to be a mixing of 2 separate issues and too many unsubstantiated/unrecognized assumptions:

The question being asked is whether forwards affect SV% - the answer is probably yes but still some assumptions are made.

Thought experiment: if all forwards on the ice do nothing and it becomes a 5 skaters vs 2 defenders + goalie, and this gives skaters more room to shoot from advantageous (to the shooter) angles and make more passes, then we would answer yes - but the assumption is made that shots from different areas of the ice will lead to different SV%. The key is that the factor that leads to the affecting of SV% is the difference between where a shooter will shoot from and this could be affected by a forward, any impact on this would lead to an impact on SV%. I believe this assumption to be true, but I wouldn't go so far as to say it must be true 100% of the time - some forwards are crest hunters after all.

Now the question being implied is whether a good defensive forward will impact SV% positively, that is a higher level of forward defense = higher SV%.

I take several issues with this:

1) What is good defense? Back check so that the opposing forward can't get a shot away is not measured by save % and yet I would say that this should be categorized as ideal defense (and not attainable; 0 shots against during an entire game).

2) If back check is impossible, the forward will try to shoe-horn the shot into a low percentage area - but are all the goalies the same or do some have differing weaknesses and can we adjust for these sv% weaknesses; how useful is average data in this scenario about individual performance?

3) Defending a shooter VS passer and all the statistical data differences that would be collected I think this is pretty complicated

4) Can we adjust for the differences in defensemen and how they impact the way that goalies and forwards defend?

There are many more significant issues, and I believe that even the fundamentals of this question are in question (this despite the fact that I do believe forwards can impact SV% - just not nearly in the same way that has been implied - for the better).

I think that as statistical data collection becomes more granular, we'll be able to eventually be able to answer the question.

 

[theButton]

I think a conclusive answer to this question would be damn tough to prove statistically. But how about the opposite? Can anyone make a convincing argument that forwards do not affect SV%?

 

[eklunds source]

I think a conclusive answer to this question would be damn tough to prove statistically. But how about the opposite? Can anyone make a convincing argument that forwards do not affect SV%?

Forwards cannot affect save percentage to a degree that we can track.

Whatever abilities a forward has in being able to limit high percentage scoring chances, force shots from farther away from the net, prevent opponents from gathering rebounds, force opponents to hurry their shots so they aren't as accurate... All those factors, and anything else you can think of that a forward might do to make his goaltenders life easier, add up to a very small fraction of what random chance says it must be.

I'll do this 'experiment' step-by-step one more time to demonstrate how we know this.

There were 300 forwards who were on the ice, at 5v5, for at least 1,000 shots against, between the years of 2007 and 2011 (4 seasons). Between 2011 and 2015, that list contains 358 names. Doing a check between them, there are 195 names who show up on both lists.

By comparing how those 195 forwards performed in the first 4 seasons with their next 4 seasons, we can see how strongly they are correlated. If a forwards' play is a strong determining factor in how their goaltender performs, we will see that generally, forwards who are above average in years 1-4 are also above average in years 5-8, and vice versa. The degree of correlation tells us how strongly forwards can "drive" on-ice sv%.

Before we look at the results of those 195 forwards, here's some examples of what we might see...

Here are 2 variables that are highly correlated -- comparing how many defensive zone faceoffs a team gets versus their corsi percentage:

*The chart says offensive, it's definitely defensive zone faceoffs

You can tell immediately that they're closely related -- if you know what one value is, you probably have a pretty good idea of the possible ranges of the other. Most of the dots are on or very close to the line.

Here are 2 variables that are notably correlated -- comparing a teams' goals % to their corsi %.

Generally, teams with a better corsi % post a better goals %, but its not guaranteed. There's definitely a link there, but if I told you a teams' corsi% was 50%... you could guess their goals% wasn't below 40% or above 60%, but you probably wouldn't be able to guess whether they were in the top half or the bottom half of the league.

Here are 2 variables that are essentially, not correlated at all -- comparing a teams' shooting percentage with their save percentage. Clearly, the data points are essentially random, with very little relation between each other. Knowing a team was excellent in save percentage gives you essentially ZERO information about how they shot.

Finally: 2007-2011 vs 2012-2015, forwards, 1000+ shots faced in both groups:

Essentially, totally random. For every forward who plots right along the line, there are dozens that are nowhere near the line. A pearson coefficient that low 0.03 is meaningful in theory but useless in practice -- if you know a forward is among the very highest in the league in on-ice save percentage over four years, you don't know for sure he'll be among the highest again over the next 4 years -- just that he probably won't be one of the very worst. It's not useful information; it doesn't allow us to make predictions going forward; the correlation is too weak to be of any note.

There's a lot of ways this isn't perfect -- for example, you could normalize a players' on-ice sv% versus their expected on-ice sv% (given the history of the goaltender)... but that's about 10x the work and as you can see from that graph, there's very little promise to find meaningful information there.

Any "measure" that is directly related to "goal events" is almost assuredly destined to fail (at the single-season, or even multi-season level). There are not enough goal events for random chance to even out. That is the entire premise behind Corsi/Fenwick/Shots being valuable.

 

[theButton]

Very cool post, thank you.

If it's possible, who is the player at/near .940 in both samples on your last graph?

 

[eklunds source]

Very cool post, thank you.

If it's possible, who is the player at/near .940 in both samples on your last graph?

That player is Shawn Thornton (0.9413 between '07-'11, 0.9390 between '12-'15).

He is on of just three players more than 2 standard deviations above league average in both data groups, joined by Brandon Sutter and Radim Vrbata.

There are zero players more than 2 standard deviations under league average in both data groups (possibly because of survivorship bias - perhaps fewers players had their careers survive a 4-year plunge in on-ice sv%).

edit: i just noticed that with the trendline in that previous graph, you can see league average 5v5 save percentage rising between 2007 and 2015.

 

[CanadienShark]

As a center if he's defensively responsible I can see that being plausible.

After all, his job is to cover the dangerous area of the ice on defense. So if he's preventing those high % shots, or turning them into less dangerous ones it'll be reflected in the goalie's sv%

Whether there's enough events over the course of a season to make it something statistically significant I'll leave to someone who's better at numbers than me.

I obviously don't have a concrete answer, but I think it would be statistically significant (assuming MLR 1-6 of course). A small sample size doesn't actually affect the statistical significance test, it just means you should take it with a grain of salt. I may be wrong though, as I'm only one course into econometrics, which was an intro-type course (although a 3rd year one). Of course though, trying to produce an accurate regression model would be nearly impossible.

Of course they *can*, in theory. Do they, in practice?

"Common sense" isn't always true, which is one of the points of analytics to begin with.

(By the way, no one in this forum ever expects to find a "nice pile of dots with a perfect line through them", despite what you appear to believe).

I don't know why I laughed so much at this.

 

[Sysreq]

Forwards cannot affect save percentage to a degree that we can track.

(Words. Words. Words.)

All of this is well and good, but I feel like its missing the point. We shouldn't be asking what is the Sv% when this player is on the ice.

We need to ask instead how does this goaltender's Sv% change when this player is on the ice.

All this data needs to be normalized around expected Sv%.

 

[Micklebot]

All of this is well and good, but I feel like its missing the point. We shouldn't be asking what is the Sv% when this player is on the ice.

We need to ask instead how does this goaltender's Sv% change when this player is on the ice.

All this data needs to be normalized around expected Sv%.

Heres a similar graph using rel sv%, (on ice sv% - Off ice sv%). Does this address you're concern?

I think the issue is people want to believe (and it makes sense) that good defensive players will lower the number of quality shots against. The problem is, they also lower the number of non-quality shots, so the effect on sv% is minimal unless they do so disproportionately. Every time a forward prevents a bad angle shot from even getting to the net, or blocks an unscreened point shot, he in effect lowers a goalies sv% (assuming it would have been an easy save otherwise) because now a higher percentage of the total shots against are from higher quality area. The inverse applys when he breaks up a 2 on 1.

 

[charlie1]

I already posted this but it seems that it was missed.

https://public.tableau.com/profile/emmanuelperry#!/vizhome/SampleSizeAnalysis/Dashboard1

There is a small but detectable effect of individual forwards on Sv%.

 

[hatterson]

Forwards cannot affect save percentage to a degree that we can track.

When you said this the first thing that jumped into my mind was pitch/catcher framing in baseball. I read an article a while ago (wish I could find it again) that looked at how some stats evolved. It noted that, as of a decade or so ago, the common view was that although there may be a pitch framing skill that catchers have, it's effects lay below the detection threshold.

Now, there's still debate on the exact value of it, but many numbers have the impact of an elite pitch framing catcher at multiple wins per season, (in the 2-5 range). That's just for their framing ability, it doesn't include anything they do in regards to live play defense (fielding, throwing runners out, etc.) or hitting.

Basically in the course of a decade or so framing went from something below detection to something that might be the most important defensive contribution of a catcher.

I'm not meaning to imply that save percentage is definitely in the same boat, just saying that's it's worth noting that simply because it's below the detection threshold now doesn't mean it always will be.

As tracking methods improve (the new real time tracking seems to be a very promising step) it's entirely possible that we discover an interesting conclusion that is currently being hidden behind a lot of other random noise.

Here are 2 variables that are essentially, not correlated at all -- comparing a teams' shooting percentage with their save percentage. Clearly, the data points are essentially random, with very little relation between each other. Knowing a team was excellent in save percentage gives you essentially ZERO information about how they shot.

As an aside that doesn't really relate to this thread, this information essentially nullifies PDO as a combined stat. Not that anyone should be really surprised by that, but it's still worth noting.