2017-18 stats and underlying metrics thread [Mod: updated season]

[lomiller1]

It's no secret that I have talked to *members* of the Jets organizations. I've had solid conversations with some of these, and have heard good things about other members of the organization (supposedly Larry Simmons is really smart and actively looking to improve things in many areas).
I know that Maurice is interested in things from a team level perspective.

Now, the real question is about HOW one uses statistics and, most importantly, how do they attack situations where their own notions are counter to what statistics may indicate.

Another question is how do they deal with things that may be more explanative than predictive.

A couple thoughts:
I don’t think it can be stressed enough that the real purpose of statistics is to separate meaningful information from noise. Even smart people are prone to seeing patterns where there is just noise, or allowing noise to obscure real patterns and then go out hunting for confirmation of what they think they see.

The problem with team’s internal analytics branches is they don’t have independent third parties looking at and challenging their results. This would make them prone to mistakes creeping in and messing with their conclusions.

I don’t know how heavily I want my coach involved in analytics. Many of the things coaches need to deal with need to be addressed before they rise above the level of noise. I don’t want my coach waiting 30 games for something to become a statistically significant pattern before acting. I also don’t want them dabbling and finding patterns that are really just noise. On the other hand, if it is established as a patter I want them to understand why they math shows that and be willing to act.

 

[garret9]

Sorry for the entry-level question here, when people say Corsi is more predictive than wins/GD, how predictive are we talkin? It strikes me that it can't be all that predictive because gamblers would be having a field day.

If a team has elite Corsi stats (i.e. they are number 1 in the league) what is the probability that they are in the top 16 teams at the end of the season? Is that question answerable?

I touch up on this in that Tweet thread I posted.

More is a relative term, and in this case we mean relative to goals and wins.
So, for example, if you had two columns of team rankings in performance in Corsi and goal differential at game 20 in the season, the former would closer correlate to goal differential for games 21-82.
The reason for this is that despite shot quantity not being the whole story (non-luck variation is about 66% of standings, and shot quantity is about 1/3-2/3's of that non-luck variation), it is the most "stable" of the non-luck variation factors.

Examples:
A good goaltender will still have streaks of bad games, but is more likely to be good than a bad goaltender.
A good shooter will still have cold streaks of no goals, but is more likely to be good than a bad shooter.
A good shot differential player will still get out shot at times, but is more likely to be good than a bad shot differential player.
All factors of skill have "susceptibility" to variation, randomness, and just being human, just the last one is a lot more stable.

Gamblers, and Vegas, already use shot quantity, and the other factors, to help themselves. That said, because hockey has such a high amount of luck variation in the standings (again about 33%), it's always going to be a low return gambling sport.
True randomness/coinflip (+ home team advantage) and a perfect model is a separation from being correct 54% of the time vs about 66%.

 

[Whileee]

One solution for the Jets is to play Wheeler and Scheifele a lot less and Hendricks and Tanev more, which will be correlated with future team success because Wheeler and Scheifele are really dragging down the Jets' CF% this season and Hendricks and Tanev have a positive relCF%.

 

[garret9]

One solution for the Jets is to play Wheeler and Scheifele a lot less and Hendricks and Tanev more, which will be correlated with future team success because Wheeler and Scheifele are really dragging down the Jets' CF% this season and Hendricks and Tanev have a positive relCF%.

Or better yet, fix the issues causing Wheeler and Scheifele to get pinned in...

 

[JetsFan815]

11-4-3 (the one that matters most of all)

+10GD (contributes a lot to above)

Cute but I think you know that is not what I meant. The post I was replying to said that NHL teams have moved beyond Corsi for evaluation, and I asked him to show a metric where the Jets looked good. I am assuming NHL teams are using something a little more complex than the Win-Loss record after moving on from Corsi

Good enough for what?

I think you know the answer to this question. It is a good enough baseline for future wins. I don't disagree that NHL teams might have more sophisticated means at their disposal, they might have better xG models and more data at their disposal in general.

What my point is that I very much doubt that any of those metrics you claim that teams have moved onto make the Jets look much better. Shot attempt quantity is going to be one of the biggest inputs in any decent evaluation model since the game is literally about outscoring the other team. Jets have been doing poorly in Corsi and the publicly avaialable xG model, I doubt that the xG model that teams use make the Jets look like a top team. Going by the eyetest, just look at the number of threads where people post "we stole 2 points tonight" or "helle won that for us"

 

[Jimby]

How about a comparative analysis? As someone pointed out in another thread not only are the Jets and Devils records identical but their corsi is pretty much the same as well

 

[mcpw]

This thread has jumped the shark...

 

[winnipegger]

I touch up on this in that Tweet thread I posted.

More is a relative term, and in this case we mean relative to goals and wins.
So, for example, if you had two columns of team rankings in performance in Corsi and goal differential at game 20 in the season, the former would closer correlate to goal differential for games 21-82.
The reason for this is that despite shot quantity not being the whole story (non-luck variation is about 66% of standings, and shot quantity is about 1/3-2/3's of that non-luck variation), it is the most "stable" of the non-luck variation factors.

Examples:
A good goaltender will still have streaks of bad games, but is more likely to be good than a bad goaltender.
A good shooter will still have cold streaks of no goals, but is more likely to be good than a bad shooter.
A good shot differential player will still get out shot at times, but is more likely to be good than a bad shot differential player.
All factors of skill have "susceptibility" to variation, randomness, and just being human, just the last one is a lot more stable.

Gamblers, and Vegas, already use shot quantity, and the other factors, to help themselves. That said, because hockey has such a high amount of luck variation in the standings (again about 33%), it's always going to be a low return gambling sport.
True randomness/coinflip (+ home team advantage) and a perfect model is a separation from being correct 54% of the time vs about 66%.

Ah, so even with a perfect model you would only be predicting a correct outcome about 66% of the time (accounting for 100% of non-luck variables) because 33% of the time it's just bouncing pucks and randomness. And Corsi is far from a perfect model and in fact only narrowly outperforms other simple metrics like GD. This separation process of luck vs non-luck strikes me as insanely complicated.

For example in the game vs Arizona #2 Laine's assist was him chopping the puck off of someone's stick which happen to land right on the stick of Lowry who was in a high danger scoring zone. Now does this fall into a category of randomness or what. There are so many pucks going off shinpads and skates that end up in prime scoring locations. But then the player (Lowry) chose to take a route that led him to that high danger zone, so it seems this goal was a result of both luck and non luck factors. Maybe Laine knew where that chopping motion was going to send the puck. These lines are blurry to me.

 

[garret9]

I think the easiest way to look at luck is this:
If every team was actually perfectly equal and it was a pure coin toss, we would still expect some teams to win more games than others.

 

[Whileee]

I think the easiest way to look at luck is this:
If every team was actually perfectly equal and it was a pure coin toss, we would still expect some teams to win more games than others.

...yes, but we wouldn't be very good at predicting which teams were which.

 

[RageQuit77]

I think the easiest way to look at luck is this:
If every team was actually perfectly equal and it was a pure coin toss, we would still expect some teams to win more games than others.

Yes. But only with limited sample sets.

If there would be infinitely long season (sample set), result of an individual game determined always by unweighted pure coin toss of 50/50 probability, average luck per team would become limitlessly more equal between teams. Farer to the infinity we would toss our coins, more close to the theoretic, statistical limit we would reach. In infinite NHL season the Lim probability would be in the infinitum: 0.5 win/lose ratio for each team.

 

[Jimby]

About the Jets high shooting percentage and PDO... Before yesterday's game I was looking at some stats to see the underlying numbers and who exactly has the high shooting percentage that is supposed to regress. I found that there are 13 different guys with a 5 on 5 SH% above 10. Eight of them have chipped in just one or two goals. For the team's SH% to regress I would think a whole lot of guys would have to stop capitalizing on chances like Perreault and Hendricks have done in the last couple of games. Are those fluke goals? Not in my book. Meanwhile, the guy scoring most of the goals - Liane - has a respectable 7.69 SH% which I am sure he will have no trouble maintaining.

Regarding corsi, it seems one big reason why the Jets team corsi is lower than some teams (I just checked the Oilers) but our shooting percentage is higher may have to do with the fact that their D have a whole lot more shot attempts. I suspect most of those are further out and not that hard to stop resulting in their high corsi and low shooting percentage. I doubt the Oilers corsi is predictive of better things to come considering who is actually taking all those shot attempts. So, the Jets could easily increase their corsi to the Oiler's level just by having the D flail away from a distance.

 

[PhilJets]

One solution for the Jets is to play Wheeler and Scheifele a lot less and Hendricks and Tanev more, which will be correlated with future team success because Wheeler and Scheifele are really dragging down the Jets' CF% this season and Hendricks and Tanev have a positive relCF%.

Wasnt this the same concern with ESL last year? But they were producing at a better clip.

Schiefele for a few days was leading the league in scoring.
Laine then took over and lead the Jets in scoring until he was injured. Top 10 in scoring and ahead of Matthews by a giod margin.

I kept saying ESL is a quick strike line. They defy the odds.


Edit:
2016 vs 2017
Main difference was the record of course but there were many factors last year.

No little and trouba to start the year

No kulikov and myers injury


Goaltending was the main factor. And it was year 2 of the youth movement.

And oh the f up sched

 

[Jimby]

Dom Luszczyszyn tackles the Jets poor corsi and comes the conclusion that it is all Wheeler's fault. I came to a similar conclusion a while back

By the numbers: Jets can be even better and it starts with...

Regarding shot quality, the HDCF% at 60% supports those charts posted earlier. As for the top line's play so far this year - I think what we are seeing is a classic Wheeler slow start only this time he is getting points during his slow start. His lack of hitting, low takeaway and high giveaway numbers support the slow start idea as does the eye test.

 

[Gm0ney]

Dom Luszczyszyn tackles the Jets poor corsi and comes the conclusion that it is all Wheeler's fault. I came to a similar conclusion a while back

By the numbers: Jets can be even better and it starts with...

I thought one of the big takeaways from the article is that the Jets are trending the wrong way in just about every metric compared to last season: 5v5 Corsi-For is down, 5v5 Corsi-Against is up. 5v5 High Danger Chances For are down, HDCA is up. PP HDCF is down. PK HDCA is up.

Everything good, they're doing less of. Everything bad, they're getting more of.

It boils down to this: "If they can keep it up, then who cares about Corsi, it's just very unlikely that they can."

 

[Whileee]

Did a quick adjustment of the Jets' CF% in relation to scores... just a reweighting of the Jets' CF% to the NHL average time distribution by score (down2+, down1, tied, up1, up2+) increases the Jets raw CF% by almost 1% (~47% to ~48%). In other words, if the Jets had the same time distribution by score as the entire NHL, with the same CF% in each score state, their CF% would be approx. 1% higher. That only covers some of the reasons for the Jets poor shot metrics, but does provide some quantification of the score effects. Corsica does some adjustments for score etc., but it's not clear what assumptions they use. I would note that the score distribution lumps together all of the 2+ differences, which probably misses some of the bigger differences that might occur when scores are more extreme.

 

[Whileee]

I thought one of the big takeaways from the article is that the Jets are trending the wrong way in just about every metric compared to last season: 5v5 Corsi-For is down, 5v5 Corsi-Against is up. 5v5 High Danger Chances For are down, HDCA is up. PP HDCF is down. PK HDCA is up.

Everything good, they're doing less of. Everything bad, they're getting more of.

It boils down to this: "If they can keep it up, then who cares about Corsi, it's just very unlikely that they can."

Who would have guessed this?

 

[Gm0ney]

Who would have guessed this?View attachment 84685

It looks like someone is focusing all his efforts on strategizing to get the most out of the worst players on the team to the detriment of the best players...

 

[Whileee]

It looks like someone is focusing all his efforts on strategizing to get the most out of the worst players on the team to the detriment of the best players...

Either that, or the Jets have a terrible top-6 and an outstanding bottom-6.

Or maybe the Jets have a different coach for the top-6 and the bottom-6...

 

[Aavco Cup]

Not sure what this is showing exactly but it looks good seeing the Jets where they are on this chart. Maybe someone stats can shed some light?

@nnstats
ran a Bradley-Terry model on the NHL season to date:
- implicitly accounts for schedule toughness
- included a home advantage component
- counted OT/SO games as ties
#Isles jump from 9th to 4th leaguewide, #Leafs drop from 8th to 16th

 

[WPGChief]

Not sure what this is showing exactly but it looks good seeing the Jets where they are on this chart. Maybe someone stats can shed some light?

@nnstats
ran a Bradley-Terry model on the NHL season to date:
- implicitly accounts for schedule toughness
- included a home advantage component
- counted OT/SO games as ties
#Isles jump from 9th to 4th leaguewide, #Leafs drop from 8th to 16th

Ever done those sites that puts one thing versus another, and you have to choose one, and then it puts that thing against another thing, forever and ever? As in, you're asked "which is better: A or B"; you choose B; "which is better: B or C"; you choose B again; "which is better: B or D"; you choose D; "okay, which is better: D or C"; and so on and so on. Comparisons that will eventually result in an aggregated ranked list.

This works very well in sports as it's always 1v1s against different teams. It looks like Namita did this with the past results of the 2017-18 season thus far, but changed OT/SO to ties, and then for this chart it is compared against the median NHL team. So, when the Jets face your middle of the pack team, they have a ~65% chance to beat them based on the results of games thus far.

Granted, I suck with trying to define models, so I would do some further googling.

 

[Whileee]

Ever done those sites that puts one thing versus another, and you have to choose one, and then it puts that thing against another thing, forever and ever? As in, you're asked "which is better: A or B"; you choose B; "which is better: B or C"; you choose B again; "which is better: B or D"; you choose D; "okay, which is better: D or C"; and so on and so on. Comparisons that will eventually result in an aggregated ranked list.

This works very well in sports as it's always 1v1s against different teams. It looks like Namita did this with the past results of the 2017-18 season thus far, but changed OT/SO to ties, and then for this chart it is compared against the median NHL team. So, when the Jets face your middle of the pack team, they have a ~65% chance to beat them based on the results of games thus far.

Granted, I suck with trying to define models, so I would do some further googling.

Good description.

Basically a more comprehensive way of adjusting results for quality of competition.

 

[Jimby]

About Scheifele waiting for better opportunities when taking shots....

 

[garret9]

About Scheifele waiting for better opportunities when taking shots....View attachment 84911

Is it better though?
3 seasons in a row his 5v5 Corsi and goal production per min has gone down at similar percentages.

 

[JetsFan815]

About Scheifele waiting for better opportunities when taking shots....View attachment 84911

Here is Scheifele's shot chart from the 15-16 season where he had a very good CF% and a much higher rate of individual corsi attempts. It doesn't really look all that different in terms of shot location than it does for this season that you posted above. This should further raise doubts about the "Scheifele is becoming an anti- Corsi player" narrative