Recent homestand and advanced stats

[united]

Bingo, you should account for score effects in your analysis. Teams trailing by one goal take ~54% of attempted shots and teams trailing by two goals take ~56% of attempted shots. That becomes more extreme the closer to the end of the game you get.

[driveway]

It's pretty good compared to other stats. Including wins, weirdly enough.

http://blogs.thescore.com/nhl/2013/0...-told-the-cbc/

Fenwick Close on the road, an obscure statistic found off on some random corner of the Internet, is actually the most predictive of a team’s home record—more predictive than even road wins, road goal differential, or points earned on the road.

[gvitaly]

Originally Posted by Bingo

28 CHI Oct 15
27 EDM Oct 09
26 NSH Oct 31
25 ARI Dec 02
24 S.J Nov 26

Nobody would argue with the game in Chicago. The game in Edmonton could be an example of a team shooting from everywhere, as I never felt in danger in that one. The next two are the Arizona game just played and the win in San Jose, both felt like solid performances from the Flames (Arizona sloppy)

In all these games the Flames scored first. Moreover, the Flames spent most of these games either tied or preserving a one goal lead. When a team starts playing on the defensive/counter-attack it shows on the advanced stats.

Here are these games in terms of game time(MM:SS)-

Vs........Tied..........Leading(1 goal lead)........Trailing....Tied+1G lead
CHI.....42:15............22:20(22:20)............. ...................64:35
EDM....16:14............43:46(31:29).............. ..................47:43
NSH....31:49............15:43(11:33).............. .11:27.........43:02
ARI.....16:34............43:26(32:12)............. ...................48:46
SJS.....37:16............23:44(21:56)............. ...................59:10

As you can see in all those games the Flames played a good chunk of the game while they are tied, or have a 1 goal lead. It fit their strategy perfectly, by holding on and allowing them to catch the other teams with their speed. Also, don't forget that in the games against Chicago and Nashville the Flames were still considering themselves as an underdog... They came to make an upset, and not to play their own game.

EDIT: Just for comparison(probably not related because the Flames trailed a lot this year), times with a 1 goal lead in their best games-

Vs......1 goal lead
NSH....10:01
S.J......3:01
OTT.....8:14(In total 17:31 with a lead)
MTL.....10:39
CAR......2:14(In total 41:42 with a lead)

[Bingo]

Originally Posted by gvitaly

In all these games the Flames scored first. Moreover, the Flames spent most of these games either tied or preserving a one goal lead. When a team starts playing on the defensive/counter-attack it shows on the advanced stats.

Here are these games in terms of game time(MM:SS)-

Vs........Tied..........Leading(1 goal lead)........Trailing....Tied+1G lead
CHI.....42:15............22:20(22:20)............. ...................64:35
EDM....16:14............43:46(31:29).............. ..................47:43
NSH....31:49............15:43(11:33).............. .11:27.........43:02
ARI.....16:34............43:26(32:12)............. ...................48:46
SJS.....37:16............23:44(21:56)............. ...................59:10

As you can see in all those games the Flames played a good chunk of the game while they are tied, or have a 1 goal lead. It fit their strategy perfectly, by holding on and allowing them to catch the other teams with their speed. Also, don't forget that in the games against Chicago and Nashville the Flames were still considering themselves as an underdog... They came to make an upset, and not to play their own game.

Good point.

Like that three of the worst five on that list came in October, that's telling.

[Philly06Cup]

Corsi includes blocked shots; Fenwick does not.

Corsi is supposed to be correlated to puck possession. Fenwick is supposed to be correlated to scoring chances.

There's a lot of enthusiasm for these metric by fancy statisticians. There is some skepticism though.

See: http://blog.philbirnbaum.com/

He's a fancy sabermetric guy who is very skeptical of Corsi and Fenwick. A new metric being derived is 'TANGO' which does not weigh all shots equally, but gives more weight to goals.

"A metric that EQUALLY weights all shots is not a good metric (I’m looking right at you Corsi and Fenwick). It first ignores our prior that says that shots-that-are-goals contains more information than non-goal-shots. Secondly, it’s not supported by empirical research."


He also has a 7 part series on CORSI and shooting percentage: Talks about how PDO doesn't regress to 100, how quality of shots/chances matter. You know, the handwavey stuff people bring up all the time.

http://blog.philbirnbaum.com/2014/11...nto-maple.html

[EldrickOnIce]

Here is a derived regression equation using the last 5 NHL seasons as data:
Projected Team Points = (249.1 x 5-on-5 Close Fenwick) – 32.6

R^2 = 0.3332 (essentially meaning that 33% of the variation in points in explained by 5-on-5 fenwick close)

Source here:
http://www.sportingcharts.com/articl...tatistics.aspx

Predictions
(using current data from http://www.puckalytics.com/teamstats.html)

Pacific Division
Canucks 98
Ducks 98
Sharks 97
Oilers 91
Kings 87
Flames 86
Coyotes 82

Your president's trophy winners?
The Minnesota Wild with 109 points



I think the point being is that not one single variable will ever be an accurate predictor of... well... anything.
And I expect you could use a 5 variable equation that might explain 66% of the variation in this case.

But people use FF%Close because it has the 'best' single variable relationship to points. And it's a quick and easy lookup, FWIW...

[Flames Fan, Ph.D.]

Originally Posted by edslunch

Perfect correlation is r-squared = 1.0. The various flavours of Corsi and Fenwick are between 0.25 and 0.3 - correlated but not what I would term strongly correlated.

To put it more strongly, within the scale we typically refer to the 0.25-0.3 range as no or very weak correlation. And we're not even given a confidence interval, which I suspect is large. So it's fine if people want to use these numbers, as long as they realize that the basis is flimsy.

And more importantly:

[FBI]

What about a stat that takes your league wide rank in all of these categories:
Corsi
PP
PK
Goal Differential

and takes the average of them all? Then maybe we can see if all Stanley Cup Winners are in the top 5-10 of that stat, and if all playoff teams are top 16-20 or something and then see what we are this year so far.

[Street Pharmacist]

Originally Posted by FBI

What about a stat that takes your league wide rank in all of these categories:
Corsi
PP
PK
Goal Differential

and takes the average of them all? Then maybe we can see if all Stanley Cup Winners are in the top 5-10 of that stat, and if all playoff teams are top 16-20 or something and then see what we are this year so far.

Easy enough to do, but how to do the weighting? Equal?

[FBI]

Originally Posted by Street Pharmacist

Easy enough to do, but how to do the weighting? Equal?

Sure, or whatever ends up favoring us to make the playofs

[EldrickOnIce]

Or you could just, like the anti stats crowd does, claim the underlying numbers don't matter because the Flames look better visually than the numbers show.

Wait..
What?

[EldrickOnIce]

Originally Posted by FBI

What about a stat that takes your league wide rank in all of these categories:
Corsi
PP
PK
Goal Differential

and takes the average of them all? Then maybe we can see if all Stanley Cup Winners are in the top 5-10 of that stat, and if all playoff teams are top 16-20 or something and then see what we are this year so far.

Originally Posted by FBI

Sure, or whatever ends up favoring us to make the playofs

OK... How about this:
Regression equation based on only 2013/14 data.
Only 2 independent variables: FF%CLOSE and PDO
Of course R^2 will be high, as only one season data is captured, but is 89%

Equation: Points= -931 + 919.6*PDO + 2.07*FF%CLOSE

Projected final point totals, using latest statistics

TEAM PROJ PTS
Tampa Bay Lightning 122.933338
Pittsburgh Penguins 119.780441
Chicago Blackhawks 115.2540075
Toronto Maple Leafs 111.3845919
Nashville Predators 110.0110554
Detroit Red Wings 108.4704149
Calgary Flames 107.2005628
Los Angeles Kings 101.7983445
New York Islanders 100.3571109
Anaheim Ducks 100.2854949
San Jose Sharks 100.1700538
Washington Capitals 98.7687264
Minnesota Wild 97.439015
St. Louis Blues 96.4438702
Montreal Canadiens 96.3523011
Vancouver Canucks 95.6874454
Winnipeg Jets 94.6762664
Boston Bruins 92.192144
New York Rangers 91.9135178
Ottawa Senators 88.919887
New Jersey Devils 87.1761552
Florida Panthers 85.2254132
Dallas Stars 80.8105019
Philadelphia Flyers 76.690251
Colorado Avalanche 74.667893
Carolina Hurricanes 68.7401321
Columbus Blue Jackets 67.494152
Phoenix Coyotes 63.425564
Edmonton Oilers 55.4515731
Buffalo Sabres 53.1865762

Is this sort of what you mean?

[Bingo]

Updated best and worst list in both Corsi and Fenwick

Three of the 5 games in this losing streak appear in the top 5 corsi list

Corsi
1 BUF Dec 11
2 TOR Dec 09
3 NSH Oct 14
4 S.J Dec 06
5 OTT Nov 15

32 CHI Oct 15
31 EDM Oct 09
30 CHI Dec 14
29 NSH Oct 31
28 ARI Dec 02

Fenwick
1 BUF Dec 11
2 MTL Oct 28
3 OTT Nov 15
4 CAR Oct 23
5 MTL Nov 02

32 CHI Oct 15
31 EDM Oct 09
30 CHI Dec 14
29 CHI Nov 20
28 T.B Nov 06

[HUSKER4FLAMES]

There are three types of Lies

1. Lie
2. Damn Lie
3. Statistics.

A good statistian can make anything look statistical.

[Rick M.]

Originally Posted by Bingo

Updated best and worst list in both Corsi and Fenwick

Three of the 5 games in this losing streak appear in the top 5 corsi list

Corsi
1 BUF Dec 11
2 TOR Dec 09
3 NSH Oct 14
4 S.J Dec 06
5 OTT Nov 15

32 CHI Oct 15
31 EDM Oct 09
30 CHI Dec 14
29 NSH Oct 31
28 ARI Dec 02

Fenwick
1 BUF Dec 11
2 MTL Oct 28
3 OTT Nov 15
4 CAR Oct 23
5 MTL Nov 02

32 CHI Oct 15
31 EDM Oct 09
30 CHI Dec 14
29 CHI Nov 20
28 T.B Nov 06

Fascinating. Whatever predictive value these stats might hold, it's clearly not in predicting individual game results.

[driveway]

Our recent losses coincide with a significant drop in our shooting percentage.

We've shot 4.9% over the course of our five losses, while having a 5v5 Corsi (Within 1) percentage of 52.3%. Our season average for shooting percentage is
10.48% (including the 5 losses) and our 5v5 Corsi (Within 1) is 45.8%.

These stats seem to jive with the eye test. We're playing well enough to win, but not getting the bounces. Earlier in the season, we were getting the bounces.

What's troubling: our shooting percentage is likely to continue to trend downwards. Since 2010 only one team has finished an 82 game season with a shooting percentage above 10.4%

What's positive: A sustained stretch of 4.9% shooting is unlikely and if that Corsi stays above 50%, a long-term drop in shooting percentage should be balanced out by the increase in shot-attempts.

[Caged Great]

One thing that I have noticed is that the team that is losing the game tends to take more shots in order to try and come back. Usually teams sit on leads, especially when they are up by a goal or two as they don't want to risk things.

That would lead to great corsi results, despite not being very good.

There is no context to reflect this though.

[Bingo]

Originally Posted by Caged Great

One thing that I have noticed is that the team that is losing the game tends to take more shots in order to try and come back. Usually teams sit on leads, especially when they are up by a goal or two as they don't want to risk things.

That would lead to great corsi results, despite not being very good.

There is no context to reflect this though.

that and teams that are losing tend to "push" it a bit, we're certainly seeing the Flames trying to force things more now that they are tight. Giordano blasting a slap shot wide on a powerplay that has the puck end up behind the Flames net.

We didn't see that earlier.

So they are less patient, shooting more, probably seeing the quality of shot plummeting and with it the shooting percentage.

[driveway]

Originally Posted by Caged Great

One thing that I have noticed is that the team that is losing the game tends to take more shots in order to try and come back. Usually teams sit on leads, especially when they are up by a goal or two as they don't want to risk things.

That would lead to great corsi results, despite not being very good.

There is no context to reflect this though.

There absolutely is context though. Head to war-on-ice.com and you can alter the parameters for the stats you're looking at to see how different teams perform in different score situations as well as different strength-situations.

You can break it down by 5v5, Powerplay, shorthanded, 4v4, opposition goalie pulled, and team goalie pulled.

As for score-situations you can see All, Within 1, Leading, Leading by 2+, Leading by 1, Tied, Trailing, Trailing by 2+, Trailing by 1.

For example, over the losing streak, at 5v5, the Flames have had the following Corsi numbers:

All: 51.6%
Within 1: 52.3%
Leading: 41.7%
Leading by 2+: N/A
Leading by 1: 41.7%
Tied: 48.3%
Trailing by 1: 59.4%
Trailing by 2+: 45.8%
Trailing: 56.3%

So, you can see we're generating a lot of shots when we're behind, as is expected. The Trailing by 2 percentage is a little misleading, as the actual difference in Corsi events is only -4. That's the same difference in Corsi events as when we're up by one.

You can go deeper into the numbers if you want, and if you want to disregard them altogether that's fine, but don't say there is no context to reflect score-effects, that's simply wrong.

[Jay Random]

Originally Posted by driveway

You can go deeper into the numbers if you want, and if you want to disregard them altogether that's fine, but don't say there is no context to reflect score-effects, that's simply wrong.

When someone quotes Corsi or Fenwick close (i.e. within 1), however, those numbers leave out the most important score effect. A team leading by 1 is not at all in the same position as a team trailing by 1. In fact, a team trailing by 1 is more like a team trailing by 2 or 3 than it is like a team leading by 1.

Your own numbers, for instance, show the Flames’ Corsi when leading by 1 as 41.7%, but when trailing by one, 59.4%. Including both of those numbers in the ‘close’ number is a terrific way of obscuring the information that the numbers convey separately.

Remember the old joke about the statistician who had one foot in a bucket of boiling water, and the other foot in a block of ice. On the average, he was comfortable.