Blocked Shots, Hits, Giveaways/Takeaways Discussion

[Randall Flagg]

Yesterday Pi and Liger were debating various statistics in the Botterill thread, and I had said I'd plot some numbers on the rest of them, and I figured I'd stick them in a new thread since that thread isn't about numbers.

Pi and Liger were arguing over whether faceoff percentage, hits, and blocked shots were strongly/weakly correlated/uncorrelated with on ice success. So I had plotted the hit rankings of every team to finish 1st -> 31st in the standings, every season since 2013-2014.

A stat that would do a good job describing the standings is, uncontroversially, goal differential. Generally, teams with a large positive goal differential have a good record, as they acquired it by winning a lot of games. Same goes for negative differentials and losing. So, this is what an effective descriptive stat would look like (an ideal one is obviously a perfectly straight line of y = x on this particular plot)

I haven't done a lot of statistical analysis, but from my understanding the R^2 value tells you how much of the linear (ideal) variation in the y axis (goal differential here) is due to variation in the x-axis (standings position). This R^2 value is quite high for a hockey stat, which makes sense - most hockey advanced stats are used for predictive purposes, and nothing can describe wins versus losses better than the thing that determines wins and losses (goal differential in games).

So if we skim the standings and sort by goal differential, we're probably also doing a pretty good job of sorting teams from best to worst.

Pi suggested doing the same thing with the hit statistic.

This is 1/60th as descriptive as goal differential. It's more or less a scatter plot. Fun things to point out - only ONE of the last 35 teams to finish in the top 5 of the NHL standings were also top 5 in hits. Sorting the standings by hits any given year will give you, more or less, a random distribution of the teams, you won't see the good ones at the top and bad ones at the bottom, or vice versa, you'll see a mix. Remember, this is at a team level. You could try to do different analyses for players, I'm not prepared to do so and it's not what we're here to discuss.

He also suggested doing this with faceoff percentage:

There is a little more correlation here, though it's still quite weak. In particular, there is a cluster of points at the top, while everything else is more or less randomly distributed. What this tells me is, as long as you aren't bottom ~7 in faceoffs, it doesn't really matter whether you're 1st or 15th, as far as your likelihood of being high in the NHL standings. It's not a particularly compelling stat.

Here's the plot for blocked shots:

I included the line's forumla here, because the slope is NEGATIVE. This implies that it's actually slightly better to have a smaller number of blocked shots as a team than a higher number, as we talked about in the other thread, because of the implication that you don't have the puck as often if you're racking up that stat. At a team level, it's worse than useless to use as a barometer of success.

Takeaways are slightly stronger predictors than faceoffs, but it still looks like more or less randomly generated data. I have the impression that they aren't recorded very reliably.

Giveaways are, like blocked shots, essentially random data with a slight negative correlation, implying that teams with more giveaways are sliiiiiiiiightly more likely to be good than teams with less giveaways (which is consistent with the "having the puck more" argument)

So I decided to do two stats that proxy how much you "have the puck," expected goals percentage and score-adjusted corsi. Keep in mind: These stats are used for predictions, it's where they perform the best (predicting team and individual level success over the NEXT batch of games, not describing where teams/players CURRENTLY rank). They blow goal-based counting stats out at doing that, so I'm not even using them in their proper regime by running this analysis, but I did it anyway.

Both of these predictive stats are better team-level descriptors than ANY of hits, blocked shots, giveaways, takeaways, and faceoff percentage. This is why stat dweebs use them instead. It was also interesting to take note of the outliers when doing these last two. Every team was an outlier in the other stats but in these ones, you'd see Carolina underperforming their high xGF/Corsi a couple years before exploding on the scene in the actual standings. You would see other teams like Montreal finish really high in the standings despite bad numbers, and the following season they'd be 20 spots lower in the standings, and you'd remember that the first year was Price's Vezina year, the next was his injury year, stuff like that. Washington was a consistent outlier in having bad underlying stats with high standings finishes - they have had the best team-wide shooting talent since 2007 in the entire league, allowing them to consistently get away with this. These stats showed you crashes of Colorado, Calgary, the year before they happened.

I like to critique advanced stats a lot because I don't think that they are used correctly by very many people, but this was a fun reminder of why they're more interesting than the numbers that were all NHL.com had for a few years, that get talked about like dinosaur numbers nowadays.

Like I said in the other thread, the mechanisms that drive players to hit and block shots are emblematic of a commitment and focus for the current game, and are things we all want this team to have, so I think that's what pi means when he says he wants to see hits and blocked shots. But using the numbers as a comparison tool is statistically useless. You'd be two to three times as much better off using advanced stats in the wrong domain for the same discussion. I know, I know pi won't ever do that. (I encourage nobody to do that.)



 

[triumph_communes]

Oh cool Rsq

If you want to make an ‘analysis’ you’re also going to need a p value

Cheers

[Randall Flagg]

4 hours ago, triumph_communes said:

Oh cool Rsq

If you want to make an ‘analysis’ you’re also going to need a p value

Cheers

Yeah, for some reason excel didn't give the easy option for that like it did R^2, and quite frankly I woulda been fine just squinting at the general shape of the data since it's just hockey. I've never rigorously learned statistics so if excel won't pop it out I'm not going to bother right now

[skaught]

Is standings position inverted on the graphs?

Edit: nevermind, the y axis is rank

Edited February 8, 2020 by skaught

[triumph_communes]

7 hours ago, Randall Flagg said:

Yeah, for some reason excel didn't give the easy option for that like it did R^2, and quite frankly I woulda been fine just squinting at the general shape of the data since it's just hockey. I've never rigorously learned statistics so if excel won't pop it out I'm not going to bother right now

https://www.excel-easy.com/examples/regression.html

[LTS]

For blocked shots I think it would make more sense to track the number of blocked shots / shots allowed (or possibly even opponent shot attempts, but that could be really tough).  

Naturally, as you said, teams that give up a lot of shots (ie. don't have the puck) are likely to show higher shot block totals.  Of course, it might still end up looking the same when you plot it out.  

Cool stuff though... thanks for doing the leg work.

[Randall Flagg]

31 minutes ago, LTS said:

For blocked shots I think it would make more sense to track the number of blocked shots / shots allowed (or possibly even opponent shot attempts, but that could be really tough).  

Naturally, as you said, teams that give up a lot of shots (ie. don't have the puck) are likely to show higher shot block totals.  Of course, it might still end up looking the same when you plot it out.  

Cool stuff though... thanks for doing the leg work.

Definitely gonna do that when I get a second. good call!

Edited February 8, 2020 by Randall Flagg

[SwampD]

This was fantastic. Awesome job, Flagg.

The only thing I would take issue with is when you say it is “statistically useless”. Hockey is a “game of inches” and other than blocked shots, those plots are not horizontal. Better teams are better at hockey stuff and it all makes a difference, as you’ve shown.

Again. This was awesome.

[pi2000]

Very well done Flagg.

Buffalo is DFL in FOW% and blocked shots/60, 30th in takeaways/60, and 24th in hits/60.

What I take from your analysis is that as long as the team is at some minimum ranking, they have just as much chance at success as any other team at the same minimum?

If I'm trying to fix the Sabres and Im in the hunt for a #2 center, I'm going to look closely at FOW%.  And I'm also looking at bottom 6 guys that can play physical, are skilled defensively and willing to block shots... which is a skill believe it or not.

Guys like Larsson, Girgensons, Okposo, etc.. are not the bottom 6 guys I'm looking for.   They were miscast early in the developmental stages of their careers as top 6 players who were relied upon for offense.  They're simply not good enough at the details of defending to fill the role this team needs.  They need a few more Sobtoka's IMO.

[LGR4GM]

1 hour ago, SwampD said:

This was fantastic. Awesome job, Flagg.

The only thing I would take issue with is when you say it is “statistically useless”. Hockey is a “game of inches” and other than blocked shots, those plots are not horizontal. Better teams are better at hockey stuff and it all makes a difference, as you’ve shown.

Again. This was awesome.

 

17 minutes ago, pi2000 said:

Very well done Flagg.

Buffalo is DFL in FOW% and blocked shots/60, 30th in takeaways/60, and 24th in hits/60.

What I take from your analysis is that as long as the team is at some minimum ranking, they have just as much chance at success as any other team at the same minimum?

If I'm trying to fix the Sabres and Im in the hunt for a #2 center, I'm going to look closely at FOW%.  And I'm also looking at bottom 6 guys that can play physical, are skilled defensively and willing to block shots... which is a skill believe it or not.

Guys like Larsson, Girgensons, Okposo, etc.. are not the bottom 6 guys I'm looking for.   They were miscast early in the developmental stages of their careers as top 6 players who were relied upon for offense.  They're simply not good enough at the details of defending to fill the role this team needs.  They need a few more Sobtoka's IMO.

Right, so neither of you understand statistics or a regression line. Got it,  thanks. 

[LGR4GM]

There's literally an entire graph showing block shots and hits mean nothing and you're still talking about it like it is a major consideration. Pi, do you work for the Sabres? It should explain their odd player moves. 

[Curt]

39 minutes ago, pi2000 said:

Very well done Flagg.

Buffalo is DFL in FOW% and blocked shots/60, 30th in takeaways/60, and 24th in hits/60.

What I take from your analysis is that as long as the team is at some minimum ranking, they have just as much chance at success as any other team at the same minimum?

If I'm trying to fix the Sabres and Im in the hunt for a #2 center, I'm going to look closely at FOW%.  And I'm also looking at bottom 6 guys that can play physical, are skilled defensively and willing to block shots... which is a skill believe it or not.

Guys like Larsson, Girgensons, Okposo, etc.. are not the bottom 6 guys I'm looking for.   They were miscast early in the developmental stages of their careers as top 6 players who were relied upon for offense.  They're simply not good enough at the details of defending to fill the role this team needs.  They need a few more Sobtoka's IMO.

For FO% and Takeaways, basically yes.

However, for Hits and Blocked Shots they are basically useless as a team statistic.  They tell you nothing about whether the team is good or not.

For Giveaways, a team that has more is actually slightly more likely to be good.

Edited February 8, 2020 by Curt

[inkman]

2 hours ago, Curt said:

For FO% and Takeaways, basically yes.

However, for Hits and Blocked Shots they are basically useless as a team statistic.  They tell you nothing about whether the team is good or not.

For Giveaways, a team that has more is actually slightly more likely to be good.

Right. Because in order to giveth something away, thou need to hasth. /Mike Tyson voice

[bob_sauve28]

Is there some sort of official NHL definition of what a "Hit" is? A lot of these terms as so subjective. Like face offs are always scored a win or loss, but that's absurd. Many face offs just end up in a scramble or something. 

[Curt]

17 minutes ago, bob_sauve28 said:

Is there some sort of official NHL definition of what a "Hit" is? A lot of these terms as so subjective. Like face offs are always scored a win or loss, but that's absurd. Many face offs just end up in a scramble or something. 

There is not, and that’s part off the problem.  They are recorded at the discretion of the scorekeepers at each NHL rink.  Even the number of shots recorded is much less accurate than one would assume.

[SwampD]

5 hours ago, LGR4GM said:

 

Right, so neither of you understand statistics or a regression line. Got it,  thanks. 

I wonder why stats folks are often thought of as know-it-all *****. Oh, right. Got it, thanks