News Article: Money Puck says the Oilers are ____ favourite to win the Cup.

[Mez]

I like this team but do I like them more than the Bruins, Leafs, Lightning, Avalanche.....only a mere 8 games ago people were talking about firing Woody...lol

Right? classic HFOil... we are terrible fire everyone! 7 games later... we are contenders!

 

[Rengorlex]

Moneypuck seems pretty prone to momentum swings with their model. Earlier in the season they had them as a near sure bet to miss. I don't take much stock in their playoff odds. Dom at The Athletic's model is not without it's flaws but seems more consistent with the way teams are rated for the remaining simulated games.

Yeah, it seems to be based entirely on team results over the past month or so. Good xgf% and dominating games for a month and a team will be at the top. It doesn't take into consideration the rosters and the players historical talent levels. The Athletic model is super high on Colorado, still, because it incorporates the players results from several years to form a prediction.

 

[Oilhawks]

Yeah, it seems to be based entirely on team results over the past month or so. Good xgf% and dominating games for a month and a team will be at the top. It doesn't take into consideration the rosters and the players historical talent levels. The Athletic model is super high on Colorado, still, because it incorporates the players results from several years to form a prediction.

I don’t mind how The Athletic does that too since it feels less “knee-jerk” but the downside is that some teams get overvalued for longer than maybe they should (took a while for Calgary to drop under 90% even after mediocre results after a decently hot start). No model is going to be perfect of course, but some seem better / more comprehensive than others

 

[Drivesaitl]

Once you make the bolded statement your argument goes completely down the tubes. This is really is basic math not rocket science. But your statement is exactly why most people struggle with many aspects of probability. You are ignoring aspects of the event space that are absolutely relevant but which seem not to matter. This is precisely how casinos make their money. This is for example why the Monty Hall Problem perplexes so many.

Tell me this. Suppose that the NHL changed there rules mid stream and said that only the first two teams in the division would make the playoffs. Would the Avs still have a better chance to make it than the Flames. What about if only the divison champ would make it. Would the elite Avs have a better chance to make it than say the Oilers??

Right now these models have a fairly significant error band because there are many games left and teams are still quite bunched. Varying the model slightly can make a big difference at the margins as well.

Now is I am not arguing that the model is perfect or even the best. Just that the results are reasonable and that they do not represent a ranking of the teams absolute strength. Hockey reference has the Flames at 60% top make the playoffs with the Avs at 81%, But their model is based on only 1000 simulations which is actually rather small. Sportsclub used 100's of thousands of simulations. Moneypuck runs 100000 simulations.

This site for example

NHL Playoff Odds

NHL team playoff odds tracked from the preseason to current day.

powerrankingsguru.com

has the odds for Avs to make it at 84.7% and the Flames at 73.0%. This is also reasonable. I say this because, as I said above, at this stage in the year playoff odds for teams on the bubble are extremely sensitive to the core assumptions the model makes.

You complicate things that are not complicated. The model somehow predicted that the Flames have an 84% chance of making playoffs. I've pointed out how ridiculous it is that the model is saying the Flames, who are BEHIND the AV's have a 15% better chance of making the playoffs than the AVs. I took further look at the model and how its contrived. But the first thing I noticed is this bolded statement, which is blatantly false.

By running a simulation of the rest of the NHL season 100,000 times we can create precise probabilities of the outcome of the season for each team.​

This quote is either careless wording or the authors don't recognize the difference between simulations and reality. The model DOES NOT precisely predict the outcomes of seasons for each team, The simulations, even if they ran a billion, predict what the model outputs based on what the imperfect model is.

So that I'll sit here and scoff all day at a model that is predicting the Flames have a 15% greater chance than Colorado of making the playoffs.

Its those kinds of model simulation predictions that should be fedback as some kind of indication the model needs some work if its spitting out predictions like that.

Its February, Only 32 games left for the Flames who again are currently OUT of the playoffs. I've never ever seen a prediction for a team out of the playoffs this late in the season that says they have an over 84% chance of making it. I don't know if the model or calculations ended up being wrong. My own take with the Flames is its closer to 50/50 they make it. They could easily miss. Considering they are not occupying even a wild card spot even granting them 50/50 is being generous.

 

[Fourier]

You complicate things that are not complicated. The model somehow predicted that the Flames have an 84% chance of making playoffs. I've pointed out how ridiculous it is that the model is saying the Flames, who are BEHIND the AV's have a 15% better chance of making the playoffs than the AVs. I took further look at the model and how its contrived. But the first thing I noticed is this bolded statement, which is blatantly false.

By running a simulation of the rest of the NHL season 100,000 times we can create precise probabilities of the outcome of the season for each team.​

This quote is either careless wording or the authors don't recognize the difference between simulations and reality. The model DOES NOT precisely predict the outcomes of seasons for each team, The simulations, even if they ran a billion, predict what the model outputs based on what the imperfect model is.

So that I'll sit here and scoff all day at a model that is predicting the Flames have a 15% greater chance than Colorado of making the playoffs.

Its those kinds of model simulation predictions that should be fedback as some kind of indication the model needs some work if its spitting out predictions like that.

Its February, Only 32 games left for the Flames who again are currently OUT of the playoffs. I've never ever seen a prediction for a team out of the playoffs this late in the season that says they have an over 84% chance of making it. I don't know if the model or calculations ended up being wrong. My own take with the Flames is its closer to 50/50 they make it. They could easily miss. Considering they are not occupying even a wild card spot even granting them 50/50 is being generous.

First let me say that I would also quibble with the use of the words "precise probabilities". In fact, I am not even sure what that means since in any stochastic process the results are by definition not precise.

As to the rest to your comments, I am sorry but they are simply not mathematically valid. You are exhibiting exactly the thought patterns that make the gaming industry so lucrative. And at the risk of sounding overly arrogant, you are in the vast majority of the population in that regard because your statement the "the Flames have roughly a 50/50 chance at making the playoffs" is the exact sort of "gut feeling" that the gaming industry exploits to make their billions.

You did not respond to my last scenarios that would have shown you why you previous claims were false, so I don't expect you to do so this time. But I'll ask anyway. Had Colorado lost their last game in OT rather than winning it would that have meant that their playoff chances would have been 50/50? If not what would you have said their chances would have been given that they would have been sitting out of a playoff spot this late in the season??

That said I'll try very briefly to comment on why Moneypuck's odds differ somewhat from the other models. They factor in Home vs Away games into their model in a significant way. Why does this matter? In the NHL statistically the home team wins 55% of the games historically. This might not seem like a big deal but if a gambler had bet only on the home team over the years they would have been quite successful. To put this into perspective in blackjack the house's odds over an average player is about 2%. For Roulette it is about 5.2%. And unlike blackjack in roulette skill won't change that balance. In the NHL right now the Flames have an advantage in the home vs away ratio over most of the teams ahead of them including. So while the impact on an individual game is small over the course of say 140 or so games it can easily be the difference that explains why one model would have the Flames at 84% and another at 73% given how close teh standings in the Pacific are right now.

Finally, I am not claiming that the individual odds are accurate because in a probabilistic sense that woud take significant analysis. I am however saying that the model is reasonable. Moreover, my initial post was simply to explain why the Oilers having a greater probability of winning the cup over the Avs or even over the Bruins would make sense.

 

[Drivesaitl]

First let me say that I would also quibble with the use of the words "precise probabilities". In fact, I am not even sure what that means since in any stochastic process the results are by definition not precise.

As to the rest to your comments, I am sorry but they are simply not mathematically valid. You are exhibiting exactly the thought patterns that make the gaming industry so lucrative. And at the risk of sounding overly arrogant, you are in the vast majority of the population in that regard because your statement the "the Flames have roughly a 50/50 chance at making the playoffs" is the exact sort of "gut feeling" that the gaming industry exploits to make their billions.

You did not respond to my last scenarios that would have shown you why you previous claims were false, so I don't expect you to do so this time. But I'll ask anyway. Had Colorado lost their last game in OT rather than winning it would that have meant that their playoff chances would have been 50/50? If not what would you have said their chances would have been given that they would have been sitting out of a playoff spot this late in the season??

That said I'll try very briefly to comment on why Moneypuck's odds differ somewhat from the other models. They factor in Home vs Away games into their model in a significant way. Why does this matter? In the NHL statistically the home team wins 55% of the games historically. This might not seem like a big deal but if a gambler had bet only on the home team over the years they would have been quite successful. To put this into perspective in blackjack the house's odds over an average player is about 2%. For Roulette it is about 5.2%. And unlike blackjack in roulette skill won't change that balance. In the NHL right now the Flames have an advantage in the home vs away ratio over most of the teams ahead of them including. So while the impact on an individual game is small over the course of say 140 or so games it can easily be the difference that explains why one model would have the Flames at 84% and another at 73% given how close teh standings in the Pacific are right now.

Finally, I am not claiming that the individual odds are accurate because in a probabilistic sense that woud take significant analysis. I am however saying that the model is reasonable. Moreover, my initial post was simply to explain why the Oilers having a greater probability of winning the cup over the Avs or even over the Bruins would make sense.

Glad you agree with the bolded. It is as I suspect yet another case where a modeling source doesn't differentiate between what their imperfect model predicts vs actual real world probability Models, fairly regardless of how many simulations are run will spit out numbers based on the parameters (faulty or otherwise) that are the inputs and weightings of that particular model. Basically what the wording means is that the practitioners have limited grounding in actual statistical research and methodology. A point I've made for decades.

As to your third paragraph theres one key factor that you, or moneypuck have not apparently considered. That Bettman pts greatly distort what a teams actual results are. For instance if we sum up Calgary losses they have 17+9=26. So that Calgary actually has 2 more losses than Wins. Calgary is only anywhere near a playoff spot because of all the OTL pts. Colorado on the other hand have few OTL and far less total losses. Colorado have 27W and 21 combined losses. Now you might find flaws in that argument, and maybe there are but Calgary are no juggernaut team by any account. The AV's ARE. I fully expect the AV's to be able to ratchet up a little intensity on back half to stake out a comfortable playoff spot. I think Flames could easily be out of it. Even the last 10GP of the respective clubs can point out some apparent trajectory. AV's are 7-3. A good record. Flames are 5-3-2. So flames still losing as many of their games as they win even if collecting loser pts.

Look, if the model said Calgary and Colorado had equal probability of making playoffs I wouldn't have engaged this discussion. That a team that was out of a spot, and that has a worse winning % than any team in a playoff spot, was allotted an 84% chances of making playoffs doesn't equate. To wit the model has granted Flames around 40% more chance of making playoffs than one could reasonably expect to be the case.

I gather what you are saying re: divisions and the Central being arguably harder to make, and thus Nashville probability being so low but I I wonder about Calgary having an 84 - 44 disparity in probability vs Nashville. The teams have a near identical win % and Nashville actually have more wins than combined losses. What you say would make more sense to me if Wild Card windows didn't exist. Two WC spots lessens the divisional exclusion factor, for lack of better words. This is why WC spots exist is to somewhat limit divisional disparity and unfairness of who makes it. Nor is there any impediment, or reason that one division can't have 5 playoff spots. Perhaps statistically its not common occurence but given where Calgary and Nashville sit, and the type of teams they are (clearly non elite) its close to 50/50 even between those two teams which gets in. If either do.

In anycase we've been discussing two different things. Throughout the discussion I've been questioning the applicability anyway of models to predict complex multiple variable events. Some of our difference is I think semantic. You find it "reasonable" that the model outputs the Flames of having a far better probability of making playoffs than the defending SC champs. I consider that conclusion, and how the site words their conclusion complete bunk.

cheers

 

[bone]

2nd. I think they are smoking something. It might even be illegal.

Their model has Edmonton 4th in their power rankings (while being closer to 2nd than to 5th). The next closest Western teams are Winnipeg and Calgary, so it's clearly liking the Oilers underlying stats and liking the easier path to the Cup.

NHL Power Rankings -MoneyPuck NHL Power Rankings

NHL Power Rankings Driven By Analytics

moneypuck.com

 

[bone]

Glad you agree with the bolded. It is as I suspect yet another case where a modeling source doesn't differentiate between what their imperfect model predicts vs actual real world probability Models, fairly regardless of how many simulations are run will spit out numbers based on the parameters (faulty or otherwise) that are the inputs and weightings of that particular model. Basically what the wording means is that the practitioners have limited grounding in actual statistical research and methodology. A point I've made for decades.

As to your third paragraph theres one key factor that you, or moneypuck have not apparently considered. That Bettman pts greatly distort what a teams actual results are. For instance if we sum up Calgary losses they have 17+9=26. So that Calgary actually has 2 more losses than Wins. Calgary is only anywhere near a playoff spot because of all the OTL pts. Colorado on the other hand have few OTL and far less total losses. Colorado have 27W and 21 combined losses. Now you might find flaws in that argument, and maybe there are but Calgary are no juggernaut team by any account. The AV's ARE. I fully expect the AV's to be able to ratchet up a little intensity on back half to stake out a comfortable playoff spot. I think Flames could easily be out of it. Even the last 10GP of the respective clubs can point out some apparent trajectory. AV's are 7-3. A good record. Flames are 5-3-2. So flames still losing as many of their games as they win even if collecting loser pts.

Look, if the model said Calgary and Colorado had equal probability of making playoffs I wouldn't have engaged this discussion. That a team that was out of a spot, and that has a worse winning % than any team in a playoff spot, was allotted an 84% chances of making playoffs doesn't equate. To wit the model has granted Flames around 40% more chance of making playoffs than one could reasonably expect to be the case.

I gather what you are saying re: divisions and the Central being arguably harder to make, and thus Nashville probability being so low but I I wonder about Calgary having an 84 - 44 disparity in probability vs Nashville. The teams have a near identical win % and Nashville actually have more wins than combined losses. What you say would make more sense to me if Wild Card windows didn't exist. Two WC spots lessens the divisional exclusion factor, for lack of better words. This is why WC spots exist is to somewhat limit divisional disparity and unfairness of who makes it. Nor is there any impediment, or reason that one division can't have 5 playoff spots. Perhaps statistically its not common occurence but given where Calgary and Nashville sit, and the type of teams they are (clearly non elite) its close to 50/50 even between those two teams which gets in. If either do.

In anycase we've been discussing two different things. Throughout the discussion I've been questioning the applicability anyway of models to predict complex multiple variable events. Some of our difference is I think semantic. You find it "reasonable" that the model outputs the Flames of having a far better probability of making playoffs than the defending SC champs. I consider that conclusion, and how the site words their conclusion complete bunk.

cheers

Where their system creates this strange scenario is that it doesn't consider current standings points in determining probability for who's more likely to win any given game but rather their power rankings. For example, their model would favor Ottawa over LA, especially in Ottawa, even though they are not even close in the standings. Currently, they have Calgary 7th and Colorado 15th. Couple that with Colorado likely having a more difficult schedule it likely overrates Calgary's chances and underrates Colorados.

This is compounded particularly when their power rankings seem to value a lot of the favorite predictability models of shot quality for and against, but doesn't necessarily factor in the talent level of each team.

Clearly, Colorado's best offensive players are more likely to outperform their expected results than Calgary's and to this point because Edmonton broke Markstrom and Calgary's actual results are lagging their expected results defensively.

I don't necessarily like this type of model as it is susceptible to rapid variations when a team goes on a XGF% heater like Edmonton has the last 8 games, and it largely ignores the talent level of teams, but it can be interesting to look at along with other models. The one positive I'm drawing from it is that their Power Rankings really like Edmonton, and Edmonton's a team you'd think can outperform their expected results particularly on offense, so I think it could bode well for Edmonton's chances down the stretch and into playoffs.

All that said, they appear to be trying to look at talent in their win predictability models based on this link MoneyPuck.com -About and How it Works , but obviously it needs a bit of work if it puts Carolina so far ahead of Boston, Toronto and Tampa and puts Calgary way ahead of Colorado. If puts Carolina further ahead of Boston, than it puts Boston ahead of their 13th ranked team (Ottawa). There's no way Carolina's gap on any playoff team is that big let alone a team on track to set a records for most points in a season.

 

[BoldNewLettuce]

I remember saying Colorado looked beatable but they are still so damned fast---im not sure what the strategy is for beating them.

 

[Fourier]

Glad you agree with the bolded. It is as I suspect yet another case where a modeling source doesn't differentiate between what their imperfect model predicts vs actual real world probability Models, fairly regardless of how many simulations are run will spit out numbers based on the parameters (faulty or otherwise) that are the inputs and weightings of that particular model. Basically what the wording means is that the practitioners have limited grounding in actual statistical research and methodology. A point I've made for decades.

As to your third paragraph theres one key factor that you, or moneypuck have not apparently considered. That Bettman pts greatly distort what a teams actual results are. For instance if we sum up Calgary losses they have 17+9=26. So that Calgary actually has 2 more losses than Wins. Calgary is only anywhere near a playoff spot because of all the OTL pts. Colorado on the other hand have few OTL and far less total losses. Colorado have 27W and 21 combined losses. Now you might find flaws in that argument, and maybe there are but Calgary are no juggernaut team by any account. The AV's ARE. I fully expect the AV's to be able to ratchet up a little intensity on back half to stake out a comfortable playoff spot. I think Flames could easily be out of it. Even the last 10GP of the respective clubs can point out some apparent trajectory. AV's are 7-3. A good record. Flames are 5-3-2. So flames still losing as many of their games as they win even if collecting loser pts.

Look, if the model said Calgary and Colorado had equal probability of making playoffs I wouldn't have engaged this discussion. That a team that was out of a spot, and that has a worse winning % than any team in a playoff spot, was allotted an 84% chances of making playoffs doesn't equate. To wit the model has granted Flames around 40% more chance of making playoffs than one could reasonably expect to be the case.

I gather what you are saying re: divisions and the Central being arguably harder to make, and thus Nashville probability being so low but I I wonder about Calgary having an 84 - 44 disparity in probability vs Nashville. The teams have a near identical win % and Nashville actually have more wins than combined losses. What you say would make more sense to me if Wild Card windows didn't exist. Two WC spots lessens the divisional exclusion factor, for lack of better words. This is why WC spots exist is to somewhat limit divisional disparity and unfairness of who makes it. Nor is there any impediment, or reason that one division can't have 5 playoff spots. Perhaps statistically its not common occurence but given where Calgary and Nashville sit, and the type of teams they are (clearly non elite) its close to 50/50 even between those two teams which gets in. If either do.

In anycase we've been discussing two different things. Throughout the discussion I've been questioning the applicability anyway of models to predict complex multiple variable events. Some of our difference is I think semantic. You find it "reasonable" that the model outputs the Flames of having a far better probability of making playoffs than the defending SC champs. I consider that conclusion, and how the site words their conclusion complete bunk.

cheers

This is hard to say without seeming insulting but that is not my intention. You are making a very common misinterpretation of what models are intended to do. This was a massive issue during coivd by the way. The goal is not to be right on 100% of your predictions for each team. It is to be mostly right most of the time for most teams given the data at hand.

You have picked out Calgary and applied your own bias (we all have these so this is again this is not intended to be overly critical) to try and come up with odds. To you this is "real world" stuff but in reality it is no more real world than what the models are. Buy that I mean that there is no "real world" information that tells you something different than the moneypuck model would see that itself makes your assessment more accurate. You are simply making different assumptions and in your case they are extremely simplistic relative to the immense complexity of this process. Again this is perfectly normal.

With respect to Bettman points I can understand why on the surface you might question not including them more directly in the model. In particular, the probability of an NHL game going to overtime is in the 20-25% range which is not trivial. But the distribution of "Bettman points" is such that 80% of the teams are within 3 points of the median number of Bettman points, and 93% of the teams are within 4 points of the median. Moreover there is little correlation between the distribution of Bettman points and relative standing As such I strongly suspect that eliminating these points from a model would have little impact on the reliability or accuracy, especially at this point in the season since theses extra points are for the most part going to almost randomly distributed. . I know you feel differently but there is almost no mathematical evidence to back this up.

For the record I stand by my claim that the model is reasonable in coming to the conclusion it does comparing Calgary and the Avs. Other models make different assumptions and come to different conclusions. They can still be reasonable at the same time even though they produce different results.

Finally, your comment about "far better" probability is again a bit of a misreading of what the model is about. The Flames and Avs are both teams that sit on the edge as far as the playoffs are concerned. for teams in their situation the probability assigned to their chances is not very stable compared with teams higher up or lower down the standings. Win two in a row vs lose two in a row and their numbers can change markedly. That is simply a reflection of how much parity there is in the standings. So when I look at these two teams I automatically take this into account.

 

[Drivesaitl]

Where their system creates this strange scenario is that it doesn't consider current standings points in determining probability for who's more likely to win any given game but rather their power rankings. For example, their model would favor Ottawa over LA, especially in Ottawa, even though they are not even close in the standings. Currently, they have Calgary 7th and Colorado 15th. Couple that with Colorado likely having a more difficult schedule it likely overrates Calgary's chances and underrates Colorados.

This is compounded particularly when their power rankings seem to value a lot of the favorite predictability models of shot quality for and against, but doesn't necessarily factor in the talent level of each team.

Clearly, Colorado's best offensive players are more likely to outperform their expected results than Calgary's and to this point because Edmonton broke Markstrom and Calgary's actual results are lagging their expected results defensively.

I don't necessarily like this type of model as it is susceptible to rapid variations when a team goes on a XGF% heater like Edmonton has the last 8 games, and it largely ignores the talent level of teams, but it can be interesting to look at along with other models. The one positive I'm drawing from it is that their Power Rankings really like Edmonton, and Edmonton's a team you'd think can outperform their expected results particularly on offense, so I think it could bode well for Edmonton's chances down the stretch and into playoffs.

All that said, they appear to be trying to look at talent in their win predictability models based on this link MoneyPuck.com -About and How it Works , but obviously it needs a bit of work if it puts Carolina so far ahead of Boston, Toronto and Tampa and puts Calgary way ahead of Colorado. If puts Carolina further ahead of Boston, than it puts Boston ahead of their 13th ranked team (Ottawa). There's no way Carolina's gap on any playoff team is that big let alone a team on track to set a records for most points in a season.

Thanks. Its incredible they aren't even doing the probabilities in relation to current pts and standings. Instead using their already wrong power rankings, which compounds the error that actual results have already proven wrong. In essence this is doubling down on error.

I took a look at the models description but got dizzy reading pages of "expected results". Seems too the model bought their own distortions about players. To wit they thought all of Huberdeau, Weegar, Kadri were elite players. I constantly felt that weegar was being hyped far better than he is and that Kadri is wildly inconsistent in performance one year to the next.

 

[Drivesaitl]

This is hard to say without seeming insulting but that is not my intention. You are making a very common misinterpretation of what models are intended to do. This was a massive issue during coivd by the way. The goal is not to be right on 100% of your predictions for each team. It is to be mostly right most of the time for most teams given the data at hand.

You have picked out Calgary and applied your own bias (we all have these so this is again this is not intended to be overly critical) to try and come up with odds. To you this is "real world" stuff but in reality it is no more real world than what the models are. Buy that I mean that there is no "real world" information that tells you something different than the moneypuck model would see that itself makes your assessment more accurate. You are simply making different assumptions and in your case they are extremely simplistic relative to the immense complexity of this process. Again this is perfectly normal.

With respect to Bettman points I can understand why on the surface you might question not including them in the model. In particular, the probability of an NHL game going to overtime is in the 20-25% range which is not trivial. But the distribution of "Bettman points" is such that 80% of the teams are within 3 points of the median number of Bettman points, and 93% of the teams are within 4 points of the median. Moreover there is little correlation between the distribution of Bettman points and relative standing As such I strongly suspect that eliminating these points from a model would have little impact on the reliability or accuracy, especially at this point in the season since theses extra points are for the most part going to almost randomly distributed. . I know you feel differently but there is almost no mathematical evidence to back this up.

For the record I stand by my claim that the model is reasonable in coming to the conclusion it does comparing Calgary and the Avs. Other models make different assumptions and come to different conclusions. They can still be reasonable at the same time even though they produce different results.

Finally, your comment about "far better" probability is again a bit of a misreading of what the model is about. The Flames and Avs are both teams that sit on the edge as far as the playoffs are concerned. for teams in their situation the probability assigned to their chances is not very stable compared with teams higher up or lower down the standings. Win two in a row vs lose two in a row and their numbers can change markedly. That is simply a reflection of how much parity there is in the standings. So when I look at these two teams I automatically take this into account.

I understand what models are devised to do. I also question their applications beyond what they are best at. I didn't apply my own bias in regards to Calgary, as @bone mentioned in another post the model applied its own expected results instead of just looking at the standings. So that the model isn't even realtime using the ACTUAL standings. Indeed the model having Ottawa better than LA, or Calgary much better than Colorado suggests the model is off.

Having said that I've maintained throughout that a model outputs only what is inputted in its metrics and weighting. But the problem often with models is that the input is off the moment of derivation. For instance one of the difficulties with the covid models was that some coefficients were off, and unlikely. Not that I want to re-enter that discussion.

 

[Fourier]

Where their system creates this strange scenario is that it doesn't consider current standings points in determining probability for who's more likely to win any given game but rather their power rankings. For example, their model would favor Ottawa over LA, especially in Ottawa, even though they are not even close in the standings. Currently, they have Calgary 7th and Colorado 15th. Couple that with Colorado likely having a more difficult schedule it likely overrates Calgary's chances and underrates Colorados.

This is compounded particularly when their power rankings seem to value a lot of the favorite predictability models of shot quality for and against, but doesn't necessarily factor in the talent level of each team.

Clearly, Colorado's best offensive players are more likely to outperform their expected results than Calgary's and to this point because Edmonton broke Markstrom and Calgary's actual results are lagging their expected results defensively.

I don't necessarily like this type of model as it is susceptible to rapid variations when a team goes on a XGF% heater like Edmonton has the last 8 games, and it largely ignores the talent level of teams, but it can be interesting to look at along with other models. The one positive I'm drawing from it is that their Power Rankings really like Edmonton, and Edmonton's a team you'd think can outperform their expected results particularly on offense, so I think it could bode well for Edmonton's chances down the stretch and into playoffs.

All that said, they appear to be trying to look at talent in their win predictability models based on this link MoneyPuck.com -About and How it Works , but obviously it needs a bit of work if it puts Carolina so far ahead of Boston, Toronto and Tampa and puts Calgary way ahead of Colorado. If puts Carolina further ahead of Boston, than it puts Boston ahead of their 13th ranked team (Ottawa). There's no way Carolina's gap on any playoff team is that big let alone a team on track to set a records for most points in a season.

I agree with a lot of what you say in terms of legitimate concerns about the model. But wrt to Carolina there is a simple reason as to why their numbers are so much higher. They play in a much easier division which means they have less competition to over come each step along the way.

 

[Fourier]

I understand what models are devised to do. I also question their applications beyond what they are best at. I didn't apply my own bias in regards to Calgary, as @bone mentioned in another post the model applied its own expected results instead of just looking at the standings. So that the model isn't even realtime using the ACTUAL standings. Indeed the model having Ottawa better than LA, or Calgary much better than Colorado suggests the model is off.

Having said that I've maintained throughout that a model outputs only what is inputted in its metrics and weighting. But the problem often with models is that the input is off the moment of derivation. For instance one of the difficulties with the covid models was that some coefficients were off, and unlikely. Not that I want to re-enter that discussion.

We all apply our own bias. That does not mean that it is not correct to do so. It is simply a reflection of what you or I think is more important.

The fact that they are not using present standings is a legitimate criticism. But I also see merit in their approach. Receny bias can certainly be justified in such a predictive model. One can argue thta it is more likely to cpature things such as roster changes due to trades or injuries for example. Again, this does not mean that it is the "right approach" but rather is a legitimate approach.

Arguing about the specifics of the model is fine. All models should be scrutinized. But the scrutiny must also be questioned. In the case of this discussion all I would say is that I have not seen anything that makes me reject the model as being reasonable.

 

[Drivesaitl]

We all apply our own bias. That does not mean that it is not correct to do so. It is simply a reflection of what you or I think is more important.

The fact that they are not using present standings is a legitimate criticism. But I also see merit in their approach. Receny bias can certainly be justified in such a predictive model. One can argue thta it is more likely to cpature things such as roster changes due to trades or injuries for example. Again, this does not mean that it is the "right approach" but rather is a legitimate approach.

Arguing about the specifics of the model is fine. All models should be scrutinized. But the scrutiny must also be questioned. In the case of this discussion all I would say is that I have not seen anything that makes me reject the model as being reasonable.

Thanks as always for the exchange. I'd argue that not referencing the standings is a worse probability play than banking on a very imperfect model. Even the models past success isn't all that great. In a standard season the average Win % of NHL teams hovers around 58%. In some years the model is pulling off numbers that barely get to that, other years they were under.

If a person were just guessing at start of season on their knowledge of teams and hunches a starting point of prediction success would be around 60%. Add knowledge of any particularly bad teams like Anaheim, Arizona, SJ, Ottawa continuing to be bad and this gives even better possible predictions.

For instance this is the current power rankings that the model uses to establish their playoff probabilities.

NHL Power Rankings -MoneyPuck NHL Power Rankings

NHL Power Rankings Driven By Analytics

moneypuck.com

Note that Calgary are somehow 7th best. Ahead of several playoff teams. Calgary is somehow ranked as an elite team. Thats their CURRENT ratings. Its obviously flawed.

One would wonder how even a season to season factorial does and whether just using previous seasons standings is more predictive of next year success without using any other coefficients. Just correlation. I haven't seen those posted anywhere.

Agreed with the bolded. So of course I welcome the scrutiny of my rebuttal while scrutinizing your semantic bias definition of what is "reasonable" as it applies to model. This stated tongue and cheek for a laugh.

 

[Tobias Kahun]

Theres hardly any discrepancy in divisional games vs other teams now. The divisional importance is not what it once was when teams were playing each other several times a season. Further the Canes do not play in an easy division.

Thanks as always for the exchange. I'd argue that not referencing the standings is a worse probability play than banking on a very imperfect model. Even the models past success isn't all that great. In a standard season the average Win % of NHL teams hovers around 58%. In some years the model is pulling off numbers that barely get to that, other years they were under.

If a person were just guessing at start of season on their knowledge of teams and hunches a starting point of prediction success would be around 60%. Add knowledge of any particularly bad teams like Anaheim, Arizona, SJ, Ottawa continuing to be bad and this gives even better possible predictions.

One would wonder how even a season to season factorial does and whether just using previous seasons standings is more predictive of next year success without using any other coefficients. Just correlation. I haven't seen those posted anywhere.

Agreed with the bolded. So of course I welcome the scrutiny of my rebuttal while scrutinizing your semantic bias definition of what is "reasonable" as it applies to model. This stated tongue and cheek for a laugh.

Playing in weaker divisions still absolutely matters as its easier to make the playoffs without having to battle 6 teams for 3-4 spots.

 

[Drivesaitl]

Playing in weaker divisions still absolutely matters as its easier to make the playoffs without having to battle 6 teams for 3-4 spots.

Yeah, I edited that part out and didn't post it. Then I posted another reply and because of the way the present board parameters are setup it includes a prior reply even though you've not posted it.

I had wrongly thought the reference was to regular season, not playoffs thus the wrong reply.

 

[bone]

I agree with a lot of what you say in terms of legitimate concerns about the model. But wrt to Carolina there is a simple reason as to why their numbers are so much higher. They play in a much easier division which means they have less competition to over come each step along the way.

Certainly that would impact Carolina's likeliness to win in the playoffs or even the remaining schedule.

My concern is how their model for generating their "power score" makes them out to better than Boston by a similar margin that Boston is better than Ottawa. That's ridiculous.

NHL Power Rankings -MoneyPuck NHL Power Rankings

NHL Power Rankings Driven By Analytics

moneypuck.com

Carolina's power score is 5.603 greater than 2nd place Boston's, while Boston is only 5.518 greater than 13th place Ottawa in thier power score. However, Boston is 7 points up on Carolina (despite playing in a tougher division) and also 32 points ahead of Ottawa. I appreciate models and results seldomly line up perfectly, and I don't expect them to, but that a pretty drastic swing that brings into question how good the model really is.

 

[bone]

Playing in weaker divisions still absolutely matters as its easier to make the playoffs without having to battle 6 teams for 3-4 spots.

True enough, but schedule wise it's all pretty balanced now.

62 games out of every teams 82 game schedule are identical league wide. Within the same Conference, 77 of the 82 games are identical. The only impact in terms of schedule difficulty is which 5 extra games a team gets vs. their own division so the worst case scenario (playing the 5 best within your division, vs. the 5 worst) likely isn't even a 4 point impact on the standings.

 

[Drivesaitl]

Certainly that would impact Carolina's likeliness to win in the playoffs or even the remaining schedule.

My concern is how their model for generating their "power score" makes them out to better than Boston by a similar margin that Boston is better than Ottawa. That's ridiculous.

NHL Power Rankings -MoneyPuck NHL Power Rankings

NHL Power Rankings Driven By Analytics

moneypuck.com

Carolina's power score is 5.603 greater than 2nd place Boston's, while Boston is only 5.518 greater than 13th place Ottawa in thier power score. However, Boston is 7 points up on Carolina (despite playing in a tougher division) and also 32 points ahead of Ottawa. I appreciate models and results seldomly line up perfectly, and I don't expect them to, but that a pretty drastic swing that brings into question how good the model really is.

Calgary Flames are the 7th ranked juggenaut in the NHL according to their power rankings. Far higher than say a dozen teams ahead of them in actual standings and including multiple first place clubs. One would think they already might discern their model isn't working too well. I guess not. Also of note these models do seem to be ever evolving but its mistaken to think that added levels of complexity will yield greater predictive validity. These models are still doing less well than one might think.

ps I said the models rankings were ridiculous a few times. heh.

I know I'm keying on Calgary a lot but its one of the teams they got most wrong, as well as Ottawa. In the case of Calgary they are 10rank ordinals off in a 32 team league where around 8 teams are tanking. So one could say they are off by 10 amongst the say 24 teams in the league even trying to be competitive. Thats really bad. In the case of Ottawa the power ranking ordinal is 13 and the standings is 22nd. Prehaps curiously they have Florida ranked 8th and they are lol 23rd. You would have to try to be more wrong..

Playing in weaker divisions still absolutely matters as its easier to make the playoffs without having to battle 6 teams for 3-4 spots.

Theres only one division in league though where thats the case. Just saying. 3 of the 4 divisions its pretty similar. So that Carolina are getting the same kind of shake that Pacific, and Central teams are. Sucks to be in the Atlantic division I guess. lol

 

[bone]

Theres only one division in league though where thats the case. Just saying. 3 of the 4 divisions its pretty similar. So that Carolina are getting the same kind of shake that Pacific, and Central teams are. Sucks to be in the Atlantic division I guess. lol

Yeah, the team with the rawest deal here is probably Florida drawing their extra 5 games vs. Boston, Toronto, Tampa Bay, Buffalo and Montreal. Whereas Washington is drawing Columbus, Jersey, Rangers, Islander and Philly.

Calgary Flames are the 7th ranked juggenaut in the NHL according to their power rankings. Far higher than say a dozen teams ahead of them in actual standings and including multiple first place clubs. One would think they already might discern their model isn't working too well. I guess not. Also of note these models do seem to be ever evolving but its mistaken to think that added levels of complexity will yield greater predictive validity. These models are still doing less well than one might think.

ps I said the models rankings were ridiculous a few times. heh.

I know I'm keying on Calgary a lot but its one of the teams they got most wrong, as well as Ottawa. In the case of Calgary they are 10rank ordinals off in a 32 team league where around 8 teams are tanking. So one could say they are off by 10 amongst the say 24 teams in the league even trying to be competitive. Thats really bad. In the case of Ottawa the power ranking ordinal is 13 and the standings is 22nd. Prehaps curiously they have Florida ranked 8th and they are lol 23rd. You would have to try to be more wrong..

Yeah, I focussed on Carolina but at least is reasonable that they are tops. My only beef was by how much. Certainly it's predictably on the teams you mention is what really messes it up and should be their area of focus if they want to improve on it, as I'm sure they do.

Even with Edmonton they are far off, going from 4th in Power to 12th overall in the league or LA is 21st with them, but 9th in the league (granted with quite a few more games played).

 

[Fourier]

Certainly that would impact Carolina's likeliness to win in the playoffs or even the remaining schedule.

My concern is how their model for generating their "power score" makes them out to better than Boston by a similar margin that Boston is better than Ottawa. That's ridiculous.

NHL Power Rankings -MoneyPuck NHL Power Rankings

NHL Power Rankings Driven By Analytics

moneypuck.com

Carolina's power score is 5.603 greater than 2nd place Boston's, while Boston is only 5.518 greater than 13th place Ottawa in thier power score. However, Boston is 7 points up on Carolina (despite playing in a tougher division) and also 32 points ahead of Ottawa. I appreciate models and results seldomly line up perfectly, and I don't expect them to, but that a pretty drastic swing that brings into question how good the model really is.

This is a legitimate issue. It certainly reflects recent game performance. And this is probably an outlier given Carolinas 9-0-1 record over the last ten compared with Boston's 7-2-1 and the fact that Carolina has a better record over the last 20. In that respect there is justification if you buy the merit of recency bias to have them ranked ahead of Boston but the margin certainly seems to be somewhat exaggerated.

 

[McBeastMode]

 

[AM]

Skating around the outside of the rink fast is not a hockey skill. Why would anyone want to do it.

 

[Mr Positive]

Skating around the outside of the rink fast is not a hockey skill. Why would anyone want to do it.

It's fun though. Arguably hardest shot isn't practical anymore too as most teams prefer well placed wrist shots through screens as those slappers tend to injure their own teammates. But hardest shot is a fun event