NHL 2018/19 Team Rankings

[willmma]

WE ARE DONE FOR 2018/19

Thanks for following the Elo rankings for the year. Looking forward to the next one.

Here is the summary for the year

	Team	Score
Highest High	Tampa Bay Lighning	2822
Highest Average	Tampa Bay Lighning	2685
Lowest Average	Ottawa Senators	2402
Lowest Low	Buffalo Sabres	2173

[TBODY]

[/TBODY]

And trend chart for the year (hard to see, sorry)
Score for each team by game

These team rankings are based on the ELO rating system.

This is purely mathematical based on wins and losses. Each team has a score. That score increases or decreases based on wins and losses. The amount of change is based on the score of the opponent. Eg. If a strong team beats a weak team then they don't gain a lot of points because it is expected. If a weak team beats a strong team then they gain a lot of points and the strong team loses a bunch. It balances out. Streaks can swing a team's rankings by a lot.

For the purposes of my calculations, each team started with 2500 points. K is 100 and doesn't change.

NOTE: I changed the scoring system so that OT/SO wins are lesser than regular wins (half a win). And a an OT/SO loss is half a loss.

After 1271 games played this season the rankings for Apr 08 2019 are:

#	Team	Lowest Score	Average Score	Highest Score	Elo Score
1	Tampa Bay Lightning	2464	2685	2822	2723
2	Washington Capitals	2341	2560	2761	2648
3	Nashville Predators	2270	2536	2741	2647
4	Carolina Hurricanes	2344	2540	2727	2622
5	Anaheim Ducks	2223	2424	2656	2613
6	Columbus Blue Jackets	2371	2527	2659	2605
7	New York Islanders	2400	2548	2722	2595
8	St. Louis Blues	2364	2537	2758	2594
9	Montreal Canadiens	2378	2512	2631	2589
10	Boston Bruins	2421	2582	2803	2579
11	Pittsburgh Penguins	2339	2545	2753	2560
12	Dallas Stars	2368	2512	2650	2547
13	Colorado Avalanche	2261	2484	2703	2523
14	Florida Panthers	2308	2468	2618	2518
15	Chicago Blackhawks	2227	2446	2624	2496
16	Los Angeles Kings	2285	2404	2578	2487
17	Winnipeg Jets	2355	2538	2666	2483
18	Detroit Red Wings	2247	2417	2618	2475
19	Calgary Flames	2447	2577	2723	2472
20	San Jose Sharks	2322	2552	2774	2455
21	Vancouver Canucks	2251	2444	2606	2447
22	New Jersey Devils	2295	2430	2666	2439
23	Arizona Coyotes	2302	2478	2627	2436
24	Minnesota Wild	2292	2483	2673	2404
25	Toronto Maple Leafs	2393	2546	2692	2403
26	New York Rangers	2243	2441	2662	2388
27	Edmonton Oilers	2296	2446	2635	2377
28	Vegas Golden Knights	2344	2516	2719	2376
29	Ottawa Senators	2220	2402	2579	2350
30	Buffalo Sabres	2173	2437	2674	2342
31	Philadelphia Flyers	2254	2475	2659	2307

[TBODY] [/TBODY]

Division	Avg Score
Central	2527
Metropolitan	2520
Atlantic	2497
Pacific	2457

[TBODY] [/TBODY]

Conference	Avg Score
Eastern	2508
Western	2490

[TBODY] [/TBODY]

Click to read more...

 

[JIMVINNY]

I like it. Doesn't mean much this early, but it'll be interesting to compare this to the actual standings later on down the road.

 

[willmma]

Updated the rankings for Oct 10 2018.

 

[willmma]

Updated rankings for Oct 11 2018

 

[willmma]

Updated rankings for Oct 15 2018 are

 

[Siignal]

This is some solid, interesting OC and will be interesting to follow and compare to subjective power rankings. You da man OP

 

[willmma]

Updated rankings for Oct 16 2018

 

[NyQuil]

We're 6th here and in the Power Rankings we are 20th.

Beating the Leafs probably accounts for a lot of that.

Split the difference? 13th?

 

[willmma]

We're 6th here and in the Power Rankings we are 20th.

Split the difference? 13th?

Yeah, I think a lot of the power rankings are based on preconceptions. After losing Karlsson the Sens were supposed to be a bottom feeder in the league... last place for sure, so the Sens have to overcome that mental hurdle in rankings. You are the only team to beat the leafs. Just beat a decent team in Dallas. 9th in the league in points. 7 points in 5 games, how can you be 20th!?

But ultimately, this is purely mathematical. The wins the senators got at the time they got them gained them enough points to put them in 6th position.

Note I consider an OT/SO loss to be half a loss, and OT/SO win to be half a win

This following shows the Senators trend. Every team started at 2500.

[2018-10-04,Chicago Blackhawks,4:2500 to 2525, Ottawa Senators,3: 2500 to 2475, OTSO] LOST 25 points
[2018-10-06,Ottawa Senators,5:2475 to 2532,Toronto Maple Leafs,3:2525 to 2468] GAINED 57 points
[2018-10-08,Ottawa Senators,3:2532 to 2478,Boston Bruins,6:2507 to 2561] LOST 54 points
[2018-10-10,Philadelphia Flyers,7:2440 to 2495,Ottawa Senators,4:2478 to 2423] LOST 55 points
[2018-10-13,Los Angeles Kings,1:2526 to 2462,Ottawa Senators,5:2423 to 2487] GAINED 64 points
[2018-10-15,Dallas Stars,1:2587 to 2523,Ottawa Senators,4:2487 to 2551] GAINED 64 points

The last two wins were big gains.

 

[NyQuil]

The last two wins were big gains.

Ottawa is in the midst of their season longest homestand of 5 games.

Things will be interesting on the road when less matching is possible.

 

[willmma]

Updated rankings for Oct 17 2018.
Devils are on fire.

 

[willmma]

Updated rankings for Oct 18 2018.

Latest changes:
[2018-10-17, New York Islanders (1): 2487 -> 2444, Anaheim Ducks (4): 2538 -> 2581]
[2018-10-17, Boston Bruins (2): 2620 -> 2559, Calgary Flames (5): 2543 -> 2604]
[2018-10-17, St. Louis Blues (2): 2427 -> 2392, Montreal Canadiens (3): 2535 -> 2570]
[2018-10-17, New York Rangers (3): 2410 -> 2393, Washington Capitals (4): 2467 -> 2484, OTSO]

 

[MoreOrr]

The Devils at 2"666"... sounds about right.

 

[joe dirte]

Cheers to this guy for doing this work. It's an interesting perspective.

 

[willmma]

Updated rankings for Oct 19 2018

Latest changes:
[2018-10-18, Philadelphia Flyers (3): 2446 -> 2405, Columbus Blue Jackets (6): 2511 -> 2552]
[2018-10-18, Arizona Coyotes (4): 2368 -> 2436, Chicago Blackhawks (1): 2498 -> 2430]
[2018-10-18, Boston Bruins (2): 2559 -> 2526, Edmonton Oilers (3): 2503 -> 2536, OTSO]
[2018-10-18, New York Islanders (7): 2444 -> 2493, Los Angeles Kings (2): 2435 -> 2386]
[2018-10-18, Colorado Avalanche (5): 2497 -> 2570, New Jersey Devils (3): 2666 -> 2593]
[2018-10-18, Buffalo Sabres (1): 2470 -> 2413, San Jose Sharks (5): 2420 -> 2477]
[2018-10-18, Detroit Red Wings (1): 2336 -> 2316, Tampa Bay Lightning (3): 2582 -> 2602]
[2018-10-18, Pittsburgh Penguins (3): 2466 -> 2541, Toronto Maple Leafs (0): 2661 -> 2586]
[2018-10-18, Vancouver Canucks (1): 2563 -> 2508, Winnipeg Jets (4): 2526 -> 2581]

 

[willmma]

Updated rankings for Oct 20 2018

Latest changes:
[2018-10-19, Nashville Predators (5): 2656 -> 2699, Calgary Flames (3): 2604 -> 2561]
[2018-10-19, Minnesota Wild (3): 2476 -> 2528, Dallas Stars (1): 2488 -> 2436]
[2018-10-19, Florida Panthers (6): 2374 -> 2414, Washington Capitals (5): 2484 -> 2444, OTSO]

 

[willmma]

Updated rankings for Oct 22 2018

Latest changes:
[2018-10-20, Colorado Avalanche (3): 2570 -> 2609, Carolina Hurricanes (1): 2492 -> 2453]
[2018-10-20, Chicago Blackhawks (4): 2430 -> 2497, Columbus Blue Jackets (1): 2552 -> 2485]
[2018-10-20, Nashville Predators (3): 2699 -> 2727, Edmonton Oilers (0): 2536 -> 2508]
[2018-10-20, Detroit Red Wings (4): 2316 -> 2355, Florida Panthers (3): 2414 -> 2375, OTSO]
[2018-10-20, Buffalo Sabres (5): 2413 -> 2459, Los Angeles Kings (1): 2386 -> 2340]
[2018-10-20, Tampa Bay Lightning (4): 2602 -> 2567, Minnesota Wild (5): 2528 -> 2563, OTSO]
[2018-10-20, Montreal Canadiens (3): 2570 -> 2542, Ottawa Senators (4): 2551 -> 2579, OTSO]
[2018-10-20, New Jersey Devils (2): 2593 -> 2518, Philadelphia Flyers (5): 2405 -> 2480]
[2018-10-20, New York Islanders (1): 2493 -> 2441, San Jose Sharks (4): 2477 -> 2529]
[2018-10-20, St. Louis Blues (4): 2392 -> 2467, Toronto Maple Leafs (1): 2586 -> 2511]
[2018-10-20, Boston Bruins (1): 2526 -> 2498, Vancouver Canucks (2): 2508 -> 2536, OTSO]
[2018-10-20, Anaheim Ducks (1): 2581 -> 2518, Vegas Golden Knights (3): 2488 -> 2551]
[2018-10-20, Arizona Coyotes (3): 2436 -> 2406, Winnipeg Jets (5): 2581 -> 2611]
[2018-10-21, Buffalo Sabres (4): 2459 -> 2517, Anaheim Ducks (2): 2518 -> 2460]
[2018-10-21, Tampa Bay Lightning (6): 2567 -> 2607, Chicago Blackhawks (3): 2497 -> 2457]
[2018-10-21, Calgary Flames (4): 2561 -> 2589, New York Rangers (1): 2393 -> 2365]

 

[willmma]

Updated rankings for Oct 23 2018

Latest changes:
[2018-10-22, Carolina Hurricanes (3): 2453 -> 2489, Detroit Red Wings (1): 2355 -> 2319]
[2018-10-22, Colorado Avalanche (4): 2609 -> 2641, Philadelphia Flyers (1): 2480 -> 2448]
[2018-10-22, Washington Capitals (5): 2444 -> 2507, Vancouver Canucks (2): 2536 -> 2473]
[2018-10-22, St. Louis Blues (4): 2467 -> 2462, Winnipeg Jets (5): 2611 -> 2616, OTSO]

 

[MNNumbers]

Just exactly how does the calculation work?

 

[tucker3434]

#2 huh? This is clearly the best ranking system in existence.

 

[willmma]

Just exactly how does the calculation work?

This site explains it well.

How To Calculate the Elo-Rating (including Examples)
In my ranking system every team starts with 2500 and I use a K constant of 100

Basically, a higher ranked team is expected to win against a lower ranked team.

The numbers in this example are not accurate at all but just to demenstrate

Say a team has a score of 2600 goes plays a team with score 2400
An expected win score is calculated (see site above). Expected score could be something like 70% or 0.7

A win is scored as 1 and a loss is scored as 0

If the team wins then they score 1, exceeded the expected score by o.3. Multiply that by 100 (K), that team earns 30 points to become 2630
The other team's expected score was 0.3, they scored 0, so they lose 30 points and become 2370.

If the lower team won then they exceeded expectations by 0.7 and earn 70 points: 2470
loser team loses 70 points: 2530

Beating a higher ranking team earns you a lot of points. Beating a lower ranked team earns you fewer points
Similarly losing to a higher ranked team costs you fewer points and losing to a lower ranked team costs you a lot of points

 

[willmma]

#2 huh? This is clearly the best ranking system in existence.

Yup, streaks can swing teams wildly. The Avalanche beat the red hot Devils and then went on a 3 game win streak. Their score jumped from 2497 to 2641.

The question is can they stay there? Losing to a lower ranked team (all but one) costs a lot of points.

 

[MNNumbers]

This site explains it well.

How To Calculate the Elo-Rating (including Examples)
In my ranking system every team starts with 2500 and I use a K constant of 100

Basically, a higher ranked team is expected to win against a lower ranked team.

The numbers in this example are not accurate at all but just to demenstrate

Say a team has a score of 2600 goes plays a team with score 2400
An expected win score is calculated (see site above). Expected score could be something like 70% or 0.7

A win is scored as 1 and a loss is scored as 0

If the team wins then they score 1, exceeded the expected score by o.3. Multiply that by 100 (K), that team earns 30 points to become 2630
The other team's expected score was 0.3, they scored 0, so they lose 30 points and become 2370.

If the lower team won then they exceeded expectations by 0.7 and earn 70 points: 2470
loser team loses 70 points: 2530

Beating a higher ranking team earns you a lot of points. Beating a lower ranked team earns you fewer points
Similarly losing to a higher ranked team costs you fewer points and losing to a lower ranked team costs you a lot of points

If this is true, then all games in the ELO system are worth (1.0). In other words, the expectation in your example, is... 0.7 + 0.3. Correct? Then, what needs to happen for shootout games is that a win needs to be worth (0.75 or 0.667 or 0.6, but since they won, more than 0.5), and a shootout loss the equivalent up from zero. {EDIT....I see you are using 0.5 for each and effectively treating them as ties.}

Now, here is the thing that bothers me about the ELO system:
1- Why is K=100? Is there some mathematical reason for this? Note that anyone has ever explained to me (And another poster here has his own website and does as well. I think he is 'morehockeystats' or something like that. EDIT: Here is his site:NHL Teams Elo ratings for each season) The K=100 is purely a guess.
2- Does this take home ice into consideration? There is definitely a home ice factor. No question.
3- At this point, Elo (someone explained that it's a man's last name, so it isn't ALL CAPS. I don't know if that's true or not) has so little information to go on that, while it's certainly valid in some sense, it's a little premature.
4- What happens if some teams which are right now "good" end up showing themselves to be actually (fair - to - bad) by the time we get to 26 games. Then, teams who have beaten them now get far more benefit than they deserve.

What to do about all of this is a big question, of course.

Last year I created a purely mathematical ranking system with no assumptions. It works off of the Bradley-Terry model of rankings. It can be modified to include home-ice, which I did. And, by the time the season was over, the results of that system so closely mimicked the actual standings, that I actually gave up the project. By March 4, it wasn't yielding any information that wasn't accessible from the standings.

 

[Straight Fire]

 

[willmma]

If this is true, then all games in the ELO system are worth (1.0). In other words, the expectation in your example, is... 0.7 + 0.3. Correct? Then, what needs to happen for shootout games is that a win needs to be worth (0.75 or 0.667 or 0.6, but since they won, more than 0.5), and a shootout loss the equivalent up from zero. {EDIT....I see you are using 0.5 for each and effectively treating them as ties.}

I had to make decisions in order for this to work. 1 for a win. 0 for a loss is standard. OT and SO games are close so I scored them at 0.75 and 0.25 for an OT/SO win and loss respectively.

Now, here is the thing that bothers me about the ELO system:
1- Why is K=100? Is there some mathematical reason for this? Note that anyone has ever explained to me (And another poster here has his own website and does as well. I think he is 'morehockeystats' or something like that. EDIT: Here is his site:NHL Teams Elo ratings for each season) The K=100 is purely a guess.

The value of K is usually 32, but it is irrelevant. I use 100 because it allows for a larger spread. If we use 32 or 50 or 100 or 768, it just creates a larger spread. The difference is in the fact that sometimes, like in chess, the value of K changes over time. For this ranking each team plays 82 games a year and for me it did not merit changing K.

2- Does this take home ice into consideration? There is definitely a home ice factor. No question.

No, not taken into consideration. 41 home games and 41 away games balance out, imo. Determining how much home ice favors a team and how much that impacts their expected score is very subjective and can skew results.

3- At this point, Elo (someone explained that it's a man's last name, so it isn't ALL CAPS. I don't know if that's true or not) has so little information to go on that, while it's certainly valid in some sense, it's a little premature.

You're right. But I'm used to saying ELO now.

4- What happens if some teams which are right now "good" end up showing themselves to be actually (fair - to - bad) by the time we get to 26 games. Then, teams who have beaten them now get far more benefit than they deserve.

Yes. Scoring is based on team scores at the time they play. A team can go on a tear and win several games in a row and then go on slump and lose several in a row. The first team to beat them and put them on their slump gets the most points. Beating a team when they are hot is harder to do and earns more points. It's a cold mathematical process and doesn't consider "fair-to-bad" or retroactively update scores.

What to do about all of this is a big question, of course.

Last year I created a purely mathematical ranking system with no assumptions. It works off of the Bradley-Terry model of rankings. It can be modified to include home-ice, which I did. And, by the time the season was over, the results of that system so closely mimicked the actual standings, that I actually gave up the project. By March 4, it wasn't yielding any information that wasn't accessible from the standings.

That's awesome, please share.