What Variables Predict Playoff Success?

[seventieslord]

I believe I saw a ministudy previously that showed very little correlation between being hot down the stretch and winning in the playoffs. It showed that teams that tended to do best in the playoffs built up a huge lead in the standings early on, then sort of coasted towards the playoffs, maybe losing ground along the way. Makes sense to me.

There might be a difference between pre-lockout and post-lockout though, as there is more parity now, and the current NHL points system lengthens playoff races.

I believe I have read that, too.

 

[billybudd]

This may or may not be a starting point.

From HOH, the weakest team to win the Cup in recent memory: http://hfboards.mandatory.com/showthread.php?t=1195909




The recent Kings (terrible offense in the regular season) broke some major trends

The two outliers (Pitt and LAK) both had coach/system changes late-ish in the year (the one guy does note this about Pitt). I would think it best to throw them both out if looking for trends, reasoning that the hockey they played in the playoffs wasn't necessarily the same hockey that generated the sub-par numbers.

 

[Jason MacIsaac]

http://www.broadstreethockey.com/20...ferential-score-adjusted-fenwick-nhl-playoffs

Helped me win my playoff pool picking all LA and NJ.

Outside of dominating play, it's pretty hard to predict the playoffs because its plain dumb luck. Can anyone predict sh% and sv% in the playoffs because I know I can't. This is especially hard when trying to predict 7 game sample sizes.

 

[Yurog]

Perhaps it is just a coincidence. Semifinalist Tampa Bay is only twentieth. But there is a good correlation
2010-2011
Team Plus/Minus Differential

Team GF PPGF NetGF GA PPGA NetGA GoalDifferential
Boston 246 43 203 195 46 149 +54
Vancouver 262 72 190 185 45 140 +50
Philadelphia 259 49 210 223 54 169 +41
Pittsburgh 238 49 189 199 45 154 +35
NY Rangers 233 49 184 198 42 156 +28
Nashville 219 41 178 194 41 153 +25
Washington 224 46 178 197 43 154 +24
Phoenix 231 46 185 226 64 162 +23
San Jose 248 68 180 213 56 157 +23
Chicago 258 64 194 225 53 172 +22
Los Angeles 219 47 172 198 40 158 +14
Buffalo 245 54 191 229 51 178 +13
Detroit 261 67 194 241 53 188 +6
St. Louis 240 52 188 234 51 183 +5
Calgary 250 62 188 237 53 184 +4
Montreal 216 57 159 209 51 158 +1
Anaheim 239 67 172 235 57 178 -6
Dallas 227 55 172 233 55 178 -6
Carolina 236 55 181 239 51 188 -7
Tampa Bay 247 69 178 240 49 191 -13
Columbus 215 42 173 258 62 196 -23
Toronto 218 52 166 251 62 189 -23
Minnesota 206 53 153 233 53 180 -27
Florida 195 35 160 229 41 188 -28
New Jersey 174 34 140 209 40 169 -29
Colorado 227 49 178 288 75 213 -35
NY Islanders 229 52 177 264 52 212 -35
Atlanta 223 53 170 269 64 205 -35
Edmonton 193 44 149 269 74 195 -46
Ottawa 192 45 147 250 48 202 -55

Need to try out the statistics for last seasons

 

[Burke the Legend]

The two outliers (Pitt and LAK) both had coach/system changes late-ish in the year (the one guy does note this about Pitt). I would think it best to throw them both out if looking for trends, reasoning that the hockey they played in the playoffs wasn't necessarily the same hockey that generated the sub-par numbers.

Dismal underperforming first half offence was reformed with player acquisitions and style changes.

That said I believe it's almost impossible to consistently predict playoff performance based on regular season stats because the two biggest playoff variables IMO are 1) goaltending and 2) health.

Did any kind of statistical model predict J. Quick would deliver a .946 save % in the 2012 playoffs?

Health is fairly self exaplanatory. When key regulars start dropping like flies down the stretch and in the playoffs, any team stats acquired during the regular season become fuzzier and fuzzier as the team that actually created them begins to decompile.

 

[Hammer Time]

Dismal underperforming first half offence was reformed with player acquisitions and style changes.

That said I believe it's almost impossible to consistently predict playoff performance based on regular season stats because the two biggest playoff variables IMO are 1) goaltending and 2) health.

Did any kind of statistical model predict J. Quick would deliver a .946 save % in the 2012 playoffs?

Health is fairly self exaplanatory. When key regulars start dropping like flies down the stretch and in the playoffs, any team stats acquired during the regular season become fuzzier and fuzzier as the team that actually created them begins to decompile.

No, but any reasonable statistical model would have told you that if anyone were to go on such an amazing run, it would most likely be Quick, Smith, or Lundqvist (top 3 regular season sv%s).

 

[vanuck]

I've heard score-tied Fenwick or something similar may have been used to predict the Kings winning it all... any truth to this? Can't remember where I saw this.

 

[Jason MacIsaac]

I've heard score-tied Fenwick or something similar may have been used to predict the Kings winning it all... any truth to this? Can't remember where I saw this.

4 posts up...its adjusted Fenwick.

 

[Hammer Time]

Predicting the Winner of Playoff Series

Hockey is a game with both luck and skill elements. Due to that luck/randomness element, you can never predict 100% of the results in hockey (and that's a good thing, otherwise there would be no point watching the games). So here's my question:

In the long run, what algorithm will give you the most correct predictions? What is the maximum percentage of playoff series for which you could correctly predict the winner?

Comparison of Prediction Methods

These results are calculated from 2006-2013 (with Fenwick stats calculated 2008 onwards).

Traditional stats
Team with home ice: 69/120 (57%)
Better regular season record: 69/120 (57%)
Better regular season record, ignoring shootouts: 72/120 (60%)
Shot differential: 77/120 (64%)
Goal differential, ignoring shootouts: 81/120 (68%)


New stats
Better Fenwick tied: 51/90 (57%)
Better Fenwick Close: 55/90 (61%)
Better Score-Adjusted Fenwick: 58/90 (64%)

Goal differential has been the best predictor of playoff success in the salary cap era. Among the new stats, Score-Adjusted Fenwick so far has the highest predictive value.

Of course, a major caveat is that we only have 6 or 8 playoff tournaments which isn't a huge sample size.

Theoretical Limit

Here's a study from the Nations Network about the relative impact of luck and skill on results in different sports.

http://nhlnumbers.com/2013/8/6/theoretical-predictions-in-machine-learning-for-the-nhl-part-ii

The researcher found that actual NHL standings are indistinguishable from standings in a league where only 24% of games go to the "better team" and the other 76% of games are decided by a coin toss.

An implication of this is that in the long run, you can't correctly predict the winner of more than 62% (24+76/2) of NHL games. Parity is real.

The 62% number was calculated from regular season results. If that number also holds in the playoffs, then the binomial formula says that the "better" team wins 75% of playoff series.

http://nhlnumbers.com/2013/9/6/machine-learning-predictions-of-playoff-series

Here's a later study from the same author. After an iterative cycle of training and validating, he was able to reach 74% accuracy.

 

[Czech Your Math]

Hockey is a game with both luck and skill elements. Due to that luck/randomness element, you can never predict 100% of the results in hockey (and that's a good thing, otherwise there would be no point watching the games). So here's my question:

In the long run, what algorithm will give you the most correct predictions? What is the maximum percentage of playoff series for which you could correctly predict the winner?

Comparison of Prediction Methods

These results are calculated from 2006-2013 (with Fenwick stats calculated 2008 onwards).

Traditional stats
Team with home ice: 69/120 (57%)
Better regular season record: 69/120 (57%)
Better regular season record, ignoring shootouts: 72/120 (60%)
Shot differential: 77/120 (64%)
Goal differential, ignoring shootouts: 81/120 (68%)


New stats
Better Fenwick tied: 51/90 (57%)
Better Fenwick Close: 55/90 (61%)
Better Score-Adjusted Fenwick: 58/90 (64%)

Goal differential has been the best predictor of playoff success in the salary cap era. Among the new stats, Score-Adjusted Fenwick so far has the highest predictive value.

Of course, a major caveat is that we only have 6 or 8 playoff tournaments which isn't a huge sample size.

Theoretical Limit

Here's a study from the Nations Network about the relative impact of luck and skill on results in different sports.

http://nhlnumbers.com/2013/8/6/theoretical-predictions-in-machine-learning-for-the-nhl-part-ii

The researcher found that actual NHL standings are indistinguishable from standings in a league where only 24% of games go to the "better team" and the other 76% of games are decided by a coin toss.

An implication of this is that in the long run, you can't correctly predict the winner of more than 62% (24+76/2) of NHL games. Parity is real.

The 62% number was calculated from regular season results. If that number also holds in the playoffs, then the binomial formula says that the "better" team wins 75% of playoff series.

http://nhlnumbers.com/2013/9/6/machine-learning-predictions-of-playoff-series

Here's a later study from the same author. After an iterative cycle of training and validating, he was able to reach 74% accuracy.

Thanks for posting that. Interesting stuff and a some of it over my head. I'd like to see how regular season GF/GA ratios relate to playoff GF/GA ratios, and how each of those relates to individual PO game win%. Then could look at differences between actual and expected results.

Does 74.4% accuracy mean 25.6% by chance? So 51% chance, 49% skill, resulting in 74% correct predictions?

 

[hatterson]

The recent Kings (terrible offense in the regular season) broke some major trends

Coaching changes may also be a major variable that impacts the quality of a team.

2009 Pens -
Therrien: 27-25-7, 46.7% Fenwick
Bylsma: 18-3-4, 52.8% Fenwick

There was a significant improvement in the team after Bylsma's hiring. Using full season stats would probably have underestimated the Penguins' chances.

On the other hand, if you just use the stats for the 25 games after the coaching change, you would be using a small sample size with what is probably an unsustainable winning percentage.

Perhaps there's a middle ground, such as valuing games played after a mid-season coaching change more than those before it?

Trades also can make a large impact, exampled in the case of LA.

They traded a player who has been historically terrible in regards to puck possession for a player who is historically strong in that area and they played like a completely different team.

Using season long averages gives a completely different picture of that team.

 

[Hammer Time]

Updated after 2014 playoffs. This shows how often various statistics can predict the winner of a playoff series. Data is from 2008 through 2014 playoffs (105 series).

Method | Successes | Failures | % Correct
Home ice|57|48|54%
Better record|58|47|55%
Better record ignoring SO|59|46|56%
Goal diff, ignoring SO|68|37|65%
Better Fenwick Close|66|39|63%
Better Fenwick Tied|63|42|60%
Better Score-Adjusted Fenwick|70|35|67%
Better SOG Differential|68|37|65%

The LA Kings winning the Cup really gives a boost to all the metrics which measure shots.

 

[Xelebes]

Thank you HammerTime.

Question: are those stats pulled from regular season, playoffs or season and progressive playoffs*?

Progressive playoffs: numbers as the playoffs progress

 

[GKJ]

I don't know if this is accurate, so I have no proof, and it's been a week, but I think this 6 out of 7 years the best FF% Close L20 has produced the Cup winner.

 

[Hammer Time]

Thank you HammerTime.

Question: are those stats pulled from regular season, playoffs or season and progressive playoffs*?

Progressive playoffs: numbers as the playoffs progress

I'm using only regular season stats. Including playoff numbers introduces a lot of variation since some teams play significantly tougher schedules than others. Beating Chicago 5-0 is a lot more impressive than beating Columbus 5-0, but both will go into the books as a +5 goal differential. Even in the regular season, there are some teams which play tougher schedules than others, but the effect is not as pronounced since every team will play every other team in the league at least twice.

BTW, I've attached the raw data if anyone's interested. (Just for fun, I also included Maggie the Monkey's picks spinning the wheel on TSN)

 

[Brooklanders*]

Given a playoff series between Team A and Team B, what variables can be reliably used to estimate Team A's chance of winning the series (prior to the start of the series)?

You'll hear a lot of things bandied about by pundits (usually in early April). Some off-the-cuff thoughts:

Relative seeds?
Points earned in the regular season? (In the season prior? In the season prior to that?)
Goals for / goals against in the regular season? (In the season prior? In the season prior to that?)
Playoff success in the most recent N seasons?
Power play success / penalty-kill success?
Average age of the team?
Amount of prior postseason experience on the team?
Team A's record against Team B during the season?

Ideally, we'd be able to come up with a logistic regression of sorts that takes in all sorts of available information to predict how Team A and Team B would fare if they squared off in the playoffs. We could come up with an R^2 value (and other descriptive metrics) and improve upon those values.

Parallel to that, we could also look at what types of teams regularly outperform the predictions. Are these teams special in some way that would improve the model? Or are they just special?

This could also lead into insights about how to construct a team that's playoff-optimal.

Points earned in the regular season is the least important.
Its as useless as the Presidents Cup itself.

 

[ChibiPooky]

Playoff series are probably too short to make meaningful predictions. Over a lot of games I'd expect to be able to predict quite well using statistics, but seven games (at most) probably won't correlate all that strongly to any predictive model, at least with any consistency. What we've got now is likely the best we can hope for, or close to it. Still great work putting this together and definitely hope something can come out of it.

 

[torero]

predicting at 67% isn't bad !

i wonder if (very intuitively) a capacity to recover from a 2 goals trailing, 1 goal trailling situation ... could be relevant in determining the capacity to win it in PO.
Hence in a regression be a significant independent variable ?
a Score adjusted Fenwick and a capacity to recover from a trailing situation (captured under any form) could be an interesting independent variables mix.

 

[DTMAboutHeart]

Travis Yost had the most accurate method I have come across --- 68.2%

http://www.sportingnews.com/nhl/sto...rs-kings-score-adjusted-fenwick-2014-playoffs