Expected points (biggest surprises), plus 2010-11 projections

[Mayor Bee]

In the thread on worst Cup-winning teams, I touched briefly on expected win percentage. It's in a Bill James book, so I'll give him credit for it, although it may have been devised by someone else (and been referenced as such).

Basically, it looks like this:
(Last year's wins times 4 + two years ago's wins times 1 + three year's ago's wins times 1 + a .500 record times 2) / (the same thing as above, only with "total games" substituted for "wins")

In the case of hockey, we have to use points instead of wins. And we also have to adjust what a .500 record is for the era of the OT point, so I set it as what would be the league average in points for the particular year whose expected points we're trying to figure out.

So I ran it through for every team in NHL history, and here's what I came up with.

The biggest disappointments were...
1) 1989-90 Quebec - Expected to have 68 points, actually had 31
2) 2006-07 Philly - Expected to have 93 points, actually had 56
3) 1997-98 Tampa - Expected to have 77 points, actually had 44
4) 2005-06 St Louis - Expected to have 89 points, actually had 57
5) 1970-71 Detroit - Expected to have 87 points, actually had 56

The biggest surprises were...
1) 1992-93 Quebec - Expected to have 59 points, actually had 104
2) 2006-07 Pittsburgh - Expected to have 63 points, actually had 105
3) 1974-75 Buffalo - Expected to have 77 points, actually had 113
4) 2001-02 NY Islanders - Expected to have 60 points, actually had 96
5) 2005-06 Carolina - Expected to have 77 points, actually had 112

Most teams fall within a very close range of their expectations. There's 1,247 team seasons that I have logged, and only 19 of them fall outside of a 30-point range. Most of the huge surprises or disappointments are a direct result of a lot of fortuitous circumstances bouncing one way or the other, which skews that season's numbers as well as the next.

But, in any case, I also ran the projections for 2010-11 as well. I operated off the idea that the average point total for a team next year will be 91.6 (the average of the last three years). So here's how it looks.

Obviously, there will be swings based on retirements, player developments, player decline, and so on. But here is what we see for next year...(really going out on a limb, I might add)

Eastern Conference
1) Washington - 100
2) New Jersey - 92
3) Pittsburgh - 90
4) Buffalo - 88
5) Boston - 87
6) Ottawa - 84
7) Montreal - 84
8) Philly - 84
9) NY Rangers - 83
10) Carolina - 79
11) Florida - 77
12) Atlanta - 77
13) Tampa - 74
14) Toronto - 74
15) NY Islanders - 73

Western Conference
1) San Jose - 99
2) Chicago - 94
3) Detroit - 94
4) Vancouver - 90
5) Phoenix - 89
6) Nashville - 87
7) Los Angeles - 85
8) Anaheim - 84
9) Calgary - 84
10) Colorado - 83
11) Dallas - 82
12) St Louis - 82
13) Minnesota - 81
14) Columbus - 77
15) Edmonton - 69

Click to read more...

 

[John Flyers Fan]

I'll take the "over" on Wash, Pitt, Philly, Chicago and LA.

 

[Weztex]

Most teams fall within a very close range of their expectations. There's 1,247 team seasons that I have logged, and only 19 of them fall outside of a 30-point range. Most of the huge surprises or disappointments are a direct result of a lot of fortuitous circumstances bouncing one way or the other, which skews that season's numbers as well as the next.

You should sort disparities using point percentage. The higher number of games played today give recent teams bigger variations. Teams playing in the 60's are missing 24 possible points. The 1970-71 Wings would then be a bigger disappointment than the 2005-06 St Louis.

 

[seventieslord]

I don't like this model at all.

"Within 30 points" is close?

 

[Canadiens1958]

The Model

I don't like this model at all.

"Within 30 points" is close?

The model accurately reflects the O6 era where the standing were fairly compact with teams rarely showing great changes positive or negative from season to season. Even the arrival of a generational talent like Bobby Orr had little impact on the Bruins final position.

Post expansion - in terms of disappointments it is very revealing It clearly shows the teams that suffered from bad coaching hires - Detroit with Ned Harkness 1970-71, 2005-06 Mike Kitchen in St.Louis or those with coaches that the game had passed by, Michael Bergeron in Quebec and and Jacques Demers in Tampa Bay.

In terms of surprises it does a good job of illustrating how quickly young teams can mature - Buffalo, Quebec, Carolina, Pittsburgh, especially interesting as it pertains to the salary cap era. Making a good model not only from the standpoint of explaing but predicting as well.

Within 30 poinst being close. More or less represents the diff between the first and 20th teams in today's NHL.

Feel free to propose an alternative. All ears.

 

[seventieslord]

The fact that I can't do better, and don't care to, doesn't make the model any less inefficient...

 

[BM67]

A model that predicts that only 4 teams will have totals over the expected average, and by no more than 8.4 points at that, doesn't look very good. The projected points total up to some 200+ points less than the 09-10 season.

 

[Mayor Bee]

Understand that it was about 1:30 AM, and I had just spent the last 20 minutes going, "Why the HELL won't this thing sort properly?"

I'm out of town for the next two days, so I'll take a closer look on Thursday or Friday and see what adjustments need to be made or can be made.

 

[seventieslord]

If a team is trending downwards or upwards, this won't account for that. For example, they go from 100 to 90 to 80. Most likely occurrence is that they get 70. This model says they will get 86... I think.

 

[goalsversusthreshold*]

If a team is trending downwards or upwards, this won't account for that. For example, they go from 100 to 90 to 80. Most likely occurrence is that they get 70. This model says they will get 86... I think.

While you're right that that's an important distinction, a majority of teams trending consistently in either direction would decrease their velocity in that direction. The further you get to the end of a range, the less likely you are to continue in that direction as fast. I'd definitely bet on a team that was 80 last year and 90 the year before that finishing above 70.5 points, all other information being equal.

What's that formula based on? Is it just someone's guess? It seems to lower the range unrealistically. I find it hard to believe there will only be one 100 point team next year considering there were 11 this year and at least 5 every year since the lockout.

 

[overpass]

While you're right that that's an important distinction, a majority of teams trending consistently in either direction would decrease their velocity in that direction. The further you get to the end of a range, the less likely you are to continue in that direction as fast. I'd definitely bet on a team that was 80 last year and 90 the year before that finishing above 70.5 points, all other information being equal.

What's that formula based on? Is it just someone's guess? It seems to lower the range unrealistically. I find it hard to believe there will only be one 100 point team next year considering there were 11 this year and at least 5 every year since the lockout.

The formula was developed back in the 1980s by baseball stats guru Bill James. While it was designed for baseball, it should work well for hockey also. One possible difference is that the coefficient for the 0.500 seasons might be different for hockey, but I don't think it would be too different.

Models of this type have been proven to be more accurate than simple trend extrapolation, at least in baseball. I expect the same holds in hockey.

While the point totals for the league seem compressed, keep in mind that it is forecasting individual teams, not leagues. It's likely that at least one team will have over 110 points, but it's not likely that any individual team will have over 110 points.

 

[Canadiens1958]

Correct

The formula was developed back in the 1980s by baseball stats guru Bill James. While it was designed for baseball, it should work well for hockey also. One possible difference is that the coefficient for the 0.500 seasons might be different for hockey, but I don't think it would be too different.

Models of this type have been proven to be more accurate than simple trend extrapolation, at least in baseball. I expect the same holds in hockey.

While the point totals for the league seem compressed, keep in mind that it is forecasting individual teams, not leagues. It's likely that at least one team will have over 110 points, but it's not likely that any individual team will have over 110 points.

Quite correct in your explanation. However the baseball model does not have to account for the "loser point".

How does one account for the "loser point" in hockey since it is not guaranteed is another issue.Theoretically possible to have a season without any "loser points" or with a "loser point" every game.

This past season there were app 300 "loser points" awarded, ranging between 5 -15 with the average being about 10 per team. An appropriate adjustment would not be difficult and would change little in the overall relative positioning.

 

[Bear of Bad News]

This past season there were app 300 loser points awarded, ranging between 5 -15 with the average being about 10 per team. An appropriate adjustment would not be difficult and would change little in the overall relative positioning.

The original poster does this (or claims to - he didn't post an Excel spreadsheet or anything).

 

[ContrarianGoaltender]

The original poster does this (or claims to - he didn't post an Excel spreadsheet or anything).

I don't think he does it correctly, which is why the values all seem low.

Take Washington. They had 121 points this season. Since this year's point totals account for 50% of a team's projection in 2010-11, there is no way the Caps should project to only 100 points.

The calculation should be (121 x 4 + 108 + 94 + 91.6 x 2) / 8, or 109 points. I think the OP was using the league average points in the denominator, which results in understating all results in the shootout era.

 

[Bear of Bad News]

Good point - I match your calculations (although I get 92.0 as the "league average" points for 2009-10).

 

[goalsversusthreshold*]

The formula was developed back in the 1980s by baseball stats guru Bill James. While it was designed for baseball, it should work well for hockey also. One possible difference is that the coefficient for the 0.500 seasons might be different for hockey, but I don't think it would be too different.

Models of this type have been proven to be more accurate than simple trend extrapolation, at least in baseball. I expect the same holds in hockey.

While the point totals for the league seem compressed, keep in mind that it is forecasting individual teams, not leagues. It's likely that at least one team will have over 110 points, but it's not likely that any individual team will have over 110 points.

That makes sense in principle. If you roll 30 dice, it's very unlikely that none will land on six, but any individual die is less likely to roll on six than not. But even accounting for that, individually I'd expect teams like the Sharks, Blackhawks, Red Wings, and Capitals would finish with more than it says.

The Sharks, for instance, have finished with at least 100 for four straight years and have finished with at least 95 all but one of their last eight seasons. I'd bet at even odds on them finishing with at least 100 this year.

As for loser points, I know that someone on Pucks Prospectus treats the average record as .550. I don't know if he's treating OTLs as half a win or just using the wins:regulationloss ratio (the second would give more volatile results) to derive that number though. Is there a way you could substitute the average points per game for the .500 record part?

 

[Mayor Bee]

I found the error! Turns out I'd misplaced two different characters in the formula and let an Excel self-correct "fix" a problem, and it was giving me the original numbers from the first post. So I adjusted it to what it's supposed to be, ran it back through, and got the correct ones.

And if you needed further proof of my recent sleep deprivation, it took me an hour of staring at it with my mouth hanging open. So here are the CORRECTED 2010-11 projections.

Eastern Conference
1) Washington - 109
2) New Jersey - 100
3) Pittsburgh - 99
4) Buffalo - 96
5) Boston - 95
6) Ottawa - 92
7) Montreal - 92
8) Philly - 91
9) NY Rangers - 90
10) Carolina - 87
11) Florida - 84
12) Atlanta - 83
13) Tampa - 81
14) Toronto - 80
15) NY Islanders - 80

Western Conference
1) San Jose - 108
2) Chicago - 103
3) Detroit - 102
4) Vancouver - 98
5) Phoenix - 97
6) Nashville - 95
7) Los Angeles - 92
8) Anaheim - 92
9) Calgary - 92
10) Colorado - 91
11) Dallas - 89
12) St Louis - 89
13) Minnesota - 88
14) Columbus - 84
15) Edmonton - 76

 

[goalsversusthreshold*]

I found the error! Turns out I'd misplaced two different characters in the formula and let an Excel self-correct "fix" a problem, and it was giving me the original numbers from the first post. So I adjusted it to what it's supposed to be, ran it back through, and got the correct ones.

And if you needed further proof of my recent sleep deprivation, it took me an hour of staring at it with my mouth hanging open. So here are the CORRECTED 2010-11 projections.

Eastern Conference
1) Washington - 109
2) New Jersey - 100
3) Pittsburgh - 99
4) Buffalo - 96
5) Boston - 95
6) Ottawa - 92
7) Montreal - 92
8) Philly - 91
9) NY Rangers - 90
10) Carolina - 87
11) Florida - 84
12) Atlanta - 83
13) Tampa - 81
14) Toronto - 80
15) NY Islanders - 80

Western Conference
1) San Jose - 108
2) Chicago - 103
3) Detroit - 102
4) Vancouver - 98
5) Phoenix - 97
6) Nashville - 95
7) Los Angeles - 92
8) Anaheim - 92
9) Calgary - 92
10) Colorado - 91
11) Dallas - 89
12) St Louis - 89
13) Minnesota - 88
14) Columbus - 84
15) Edmonton - 76

Note: for a projection of the actual standings, not just points, Buffalo would be above Pittsburgh and Vancouver would be above Detroit.

 

[Mayor Bee]

If a team is trending downwards or upwards, this won't account for that. For example, they go from 100 to 90 to 80. Most likely occurrence is that they get 70. This model says they will get 86... I think.

It's unlikely for a team to continue to freefall past a certain point, just as a team on the rise is unlikely to climb past a certain point. A team like the 1995-96 Red Wings is a massive aberration.

Teams tend to gravitate toward average, and in a capped league, it's much more likely to happen. The fact that some teams tend to excel or tend to bring up the rear is built on other factors that conspire to undermine this effect. To use the Detroit example, they were able to sign several future HOFers during the heart of their run for a salary that no other team in the league could have signed those players for.

Quite correct in your explanation. However the baseball model does not have to account for the "loser point".

How does one account for the "loser point" in hockey since it is not guaranteed is another issue.Theoretically possible to have a season without any "loser points" or with a "loser point" every game.

This past season there were app 300 "loser points" awarded, ranging between 5 -15 with the average being about 10 per team. An appropriate adjustment would not be difficult and would change little in the overall relative positioning.

The league average in points since the loser point has looked like this:
1999-00 - 86.07143
2000-01 - 86.06667
2001-02 - 86.03333
2002-03 - 87.16667
2003-04 - 86.83333
(Implementation of shootouts)
2005-06 - 91.40000
2006-07 - 91.43333
2007-08 - 91.26667
2008-09 - 91.40000
2009-10 - 92.03333
2010-11 - (projected) 91.60000

Good point - I match your calculations (although I get 92.0 as the "league average" points for 2009-10).

92.03333, to be exact.

That makes sense in principle. If you roll 30 dice, it's very unlikely that none will land on six, but any individual die is less likely to roll on six than not. But even accounting for that, individually I'd expect teams like the Sharks, Blackhawks, Red Wings, and Capitals would finish with more than it says.

The Sharks, for instance, have finished with at least 100 for four straight years and have finished with at least 95 all but one of their last eight seasons. I'd bet at even odds on them finishing with at least 100 this year.

As for loser points, I know that someone on Pucks Prospectus treats the average record as .550. I don't know if he's treating OTLs as half a win or just using the wins:regulationloss ratio (the second would give more volatile results) to derive that number though. Is there a way you could substitute the average points per game for the .500 record part?

That's what I was feebly attempting to do before catching the error, and it remained after catching the error(s).

To the first point, just a week or so ago, I managed to incorrectly call a coinflip on six consecutive tosses. Not only that, I called tails every single time.