Not to worry; this won't last says Myrtle

[Resolute 14]

Originally Posted by Jay Random

Not necessarily. Winning percentage is another stat where the standard deviation tends to decrease as sample size increases. What matters is not the absolute numbers of PDO, but the relative rankings.

What we really ought to be watching, it seems to me, is the z-score of PDO rather than the number itself. That would adjust for sample size and make the numbers between, say, a 19-game stretch and a full season strictly comparable.

Agreed. That said, I really can't see the Flames staying around 3rd. We are not going to maintain a crazy ES shooting percentage of 11.5% any more than Chicago will be stuck with a crazy low percentage of 7.0%. I do think that we are going lose more ground relative to many other teams. That is, however, why this hot start is so important. The farther ahead we get, the harder it is to lose that ground.

Very probably true. Though there is still the problem of measuring proxies instead of actual events. I mean, for instance, that the number of DZ faceoffs is not necessarily an accurate indicator of the percentage of time spent in the DZ. There are situations in which a team may be coached deliberately to try to get stoppages in play while hemmed in their own zone; in which case the coach's tactical decisions will contaminate the data. It's a pity that we don't have actual direct measurements of time of possession and time in zone. No doubt that will come with the new tracking technology. Having RFID chips on every player (and ideally the puck too, and just forget about FoxTrax) will make it possible to collect much more direct data instead of mucking about with all these proxies.

In general, I try not to look at one stat exclusively for this reason. However, the faceoff percentages, zone starts and possession stats combined indicate that we spend too much time in our own zone.

And the NHL's RFID program would be a major boost to the advanced stat world. Shame they are so focused on keeping it privileged. There is a veritable army of people that would do yeoman's work on their behalf - all for free. However, that work becomes publicly available for everybody, which would snatch away any advantage from teams spending money on such analysis.

This is an important point. 'If nothing changes' is hardly ever a valid assumption in sports (or in life generally). The Flames have been lucky to do as well as they have with all these injuries; but they weren't lucky to have all the injuries.

Yup. This, I think, is one key point that many stat haters tend to miss. Their predictive value relies on stability. A series of injuries, a big trade, a player taking a major step forward/backward from year to year, or any number of things can change the picture. Mark Giordano's injury last year being a major example.

"All things being equal..." should be an assumed starting point for any of these arguments. And that, of course, is a reason why deviation decreases over time. The farther we get into the season, the fewer such wildcards are likely to occur.

[Street Pharmacist]

Originally Posted by GioforPM

This. They are useful for a team that says "Hey - we are losing. What can we improve on?" They are not useful to say "Hey - we are losing. How can we say we are doing well anyway".

Is also useful though to look behind the results. The best team in the league will lose more than 30% if their games in a season. If you only look at score and won/lost you can't necessarily see if the fundamentals are there. Shot totals and such will help you understand if you at least controlled what you could. If the fundamentals are strong, you'll win more often than lose.

I wish we'd stop calling them "advanced stats" too. They're just stats. Really, +/- is just as advanced as corsi or Fenwick. Maybe "non-traditional" stats it my personal favourite: "stats". Each one is just another tool adding a different layer of context.

[Tron_fdc]

Originally Posted by _Q_

So what you're saying then is the crash will be extra rough when it happens because the Flames have been playing above their head for 2/3rd of a season now.

Get ready boys and girls, the regression to the mean is gonna hurt and it's gonna hurt real bad. The advanced stats have spoken.

I totally know you're kidding, but I can use this as an example (sorta)

There's no way that the Flames will regress EXACTLY to where a stat predicts them to be. They can fall on either side of the mean, and predicting which side depends on how good of a team you think they are. Sure, there are probably stats for that, but I kind of fall on the other end of the spectrum as far as that goes. I hate stats, and I don't need them to tell me that there are ebbs and flows to a season, and that the Flames are likely playing over their heads right now. They will drop in the standings at some point this year; how far is anyone's guess.

I would hope that anyone who uses advance stats is using it as a reference, and not as gospel. Sure it strengthens an argument, but it certainly doesn't cement it.

IMO this team took a step forward this year vs last year. I've been saying for a while that they are not a bottom 5 team, but neither are they a top 5. They should challenge for 14th-16th overall, in a playoff position, which is probably where the statisticians put them.

I will say that as long as we take a step forward from last year I'm happy.

[Five-hole]

Originally Posted by Tron_fdc

I totally know you're kidding, but I can use this as an example (sorta)

There's no way that the Flames will regress EXACTLY to where a stat predicts them to be. They can fall on either side of the mean, and predicting which side depends on how good of a team you think they are. Sure, there are probably stats for that, but I kind of fall on the other end of the spectrum as far as that goes. I hate stats, and I don't need them to tell me that there are ebbs and flows to a season, and that the Flames are likely playing over their heads right now. They will drop in the standings at some point this year; how far is anyone's guess.

But why?

Why isn't it the case that maybe our PDO will regress to the mean rather than our winning percentage?

Only because we're the Flames, and everyone expects us to regress towards the mean because our #1 center is 20 years old and we've been irrelevant for 5-6 years now. The analysis is backward. We're proving a narrative with statistics, not creating one.

[Textcritic]

Originally Posted by New Era

Commensurate understanding of statistics? I was polite enough to explain to you why these stats are crap, and identified the very reasons they are junk, but the best you have is to come back and insult someone by telling them they don't have commensurate understanding of the subject, without any explanation of your own. That takes some real brass ones. I suspect you don't have the comprehensive knowledge of statistics you like to advertise and the claim of people not understanding them is nothing more than your defense mechanism to prevent that disclosure. If this isn't accurate, please prove otherwise by giving me that 600 level class in advanced statistics I must have missed in grad school. Save the obnoxious response and finally try to explain the validity of the work you place so much faith in.

If you feel "insulted" by what Tinordi has responded to you with, then I would suggest this may in fact be a direct component of how you all too frequently present yourself on this forum, this discussion included.

You set yourself up in your first post in this thread as someone who is much less interested in actual dialogue, and primarily seeking to confirm your already strongly entrenched assertion that "these stats are crap." I encourage you to actually stop and take a look at the link provided by Tinordi here, because it outlines fairly precisely why your "explanation" is a moot point. Perhaps he was somewhat unfair in his accusation that you lack a commensurate understanding of the topic, but you pretty fairly brought this on yourself. You have done so by the brash claim at the outset about the COMPLETE invalidity of the advanced stats that Mirtle is employing, and then by responding with an explanation that turns out to be practically negligible.

The fact of the matter is that PDO is useful under certain conditions, but useful nonetheless. It has proven to be a relatively good—if imperfect—predictor of what might happen at the end of the season. By this qualification, would you not agree that this is actually a fairly well grounded method for making general predictions? Or at minimum, for showing areas in which some teams will need to improve in order to either begin to achieve or to maintain success?

[GirlySports]

If we win tomorrow this all changes

[Street Pharmacist]

Originally Posted by Five-hole

But why?

Why isn't it the case that maybe our PDO will regress to the mean rather than our winning percentage?

Only because we're the Flames, and everyone expects us to regress towards the mean because our #1 center is 20 years old and we've been irrelevant for 5-6 years now. The analysis is backward. We're proving a narrative with statistics, not creating one.

If your PDO regresses it means a higher percentage of shots against are going in your net and/or fewer of your shots are going in. Last time I checked, goals for and against strongly correlate with winning percentage.

Mirtle has no bias here. He wrote these same things about other teams

[EldrickOnIce]

Originally Posted by Five-hole

Only because we're the Flames, and everyone expects us to regress towards the mean...

No. It's because everything statistically significant than the mean will regress toward the mean.
On average, all other things being equal, something has to give (regarding the Flames).
It doesn't mean the win/loss record has to 'give' though, but it's more than fair to speculate it might.

[Five-hole]

Originally Posted by Street Pharmacist

If your PDO regresses it means a higher percentage of shots against are going in your net and/or fewer of your shots are going in. Last time I checked, goals for and against strongly correlate with winning percentage.

Mirtle has no bias here. He wrote these same things about other teams

Okay, good point. I'm pretty fuzzy on these stats. So, if our PDO regresses that means we'll probably win less games / X games than we are now. Makes sense.

But why is everyone so convinced our PDO will regress? No one is talking about Pittsburgh's PDO regressing. (This point has been made in this thread already.) Because we're playing better than people expected. The argument, if I understand it, is that it doesn't seem like we're winning more games than we should based on PDO. It's that our PDO isn't what it should be because people expected us to be bad.

And if that's the argument, then the analysis is backward.

[_Q_]

Originally Posted by Textcritic

The fact of the matter is that PDO is useful under certain conditions, but useful nonetheless. It has proven to be a relatively good—if imperfect—predictor of what might happen at the end of the season. By this qualification, would you not agree that this is actually a fairly well grounded method for making general predictions? Or at minimum, for showing areas in which some teams will need to improve in order to either begin to achieve or to maintain success?

But see the thing about PDO is that it really isn't predictive as some writers claim. There's nothing "unsustainable" about the 1.027 PDO the Flames currently have. It's been done before by other teams, there's no reason the Flames can't sustain that number.

All these advanced stats arguments always boil down to at the end of the day is "the Flames numbers are at the top of the league and are in the same boat as the best teams from last year. The Flames are obviously not one of the best terms in the NHL so those numbers will regress. " but seriously though, does that really say anything predictive? Or are these writers claiming the Flames aren't very good and using stats to write a certain narrative? Why is it that the Flames can't maintain their current clip?

[Street Pharmacist]

Originally Posted by Five-hole

Okay, good point. I'm pretty fuzzy on these stats. So, if our PDO regresses that means we'll probably win less games / X games than we are now. Makes sense.

But why is everyone so convinced our PDO will regress? No one is talking about Pittsburgh's PDO regressing. (This point has been made in this thread already.) Because we're playing better than people expected. The argument, if I understand it, is that it doesn't seem like we're winning more games than we should based on PDO. It's that our PDO isn't what it should be because people expected us to be bad.

And if that's the argument, then the analysis is backward.

Pittsburgh's PDO will regress unless they have a season that is unprecedented. However, with their goal differential, it likely means less to their winning percentage. One more goal against here and there and one fewer for them here and there isn't going to make a much of a difference to their winning percentage when they out shoot and outscore opponents by as much as they are. Plus, they've proven that a high save% and high shooting% is actually the norm for them (though nowhere near the height they're currently at).

[Jay Random]

Originally Posted by Five-hole

The argument, if I understand it, is that it doesn't seem like we're winning more games than we should based on PDO. It's that our PDO isn't what it should be because people expected us to be bad.

And if that's the argument, then the analysis is backward.

That definitely is a backwards analysis, I agree.

What puzzles me is that a few people seem to want to use PDO for two different things. If a good team has a high PDO, they have a high PDO because they're a good team, and that is good. If a team they don't like has a high PDO, then suddenly that is unsustainable because everybody's PDO ought to be exactly 100.

Put that way, it sounds stupid, mostly because it is. But when anybody says that a team's PDO has to regress towards the mean (i.e. 100), that is, in fact, the argument they are making: that any difference between a team's actual PDO and 100 is purely a matter of luck, and there are no other reasons why a team might be better or worse than 100. If the same person then turns around and gives reasons why (for instance) Pittsburgh ought to have a PDO above 100, or Buffalo ought to be below it, then you can tell they are just moving the goalposts at will to please themselves.

[Street Pharmacist]

Originally Posted by EldrickOnIce

No. It's because everything statistically significant than the mean will regress toward the mean.
On average, all other things being equal, something has to give (regarding the Flames).
It doesn't mean the win/loss record has to 'give' though, but it's more than fair to speculate it might.

The only way the flames winning preventative didn't take a hit as PDO regresses is if they start out shooting their opponents. We're starting to see some improvement in that area

[Robbob]

Originally Posted by Tinordi

Who knows, the big problem with shot quality is that it gets highly subjective. It would introduce a much larger problem than the one you're trying to correct.

My feeling is that yes some shots are meaningless, but really, think about it, how many are actually meaningless? How many shots on goal a game are to get a line change? Generously 1 a game, on a sample of 60 to 70 shots.

Doing a quick look at shots for games yesterday, these were the results of shots from the neutral zone or defensive zone:

Jets vs Wild: 6
Sharks vs Cans: 3
Panthers vs Ducks: 5
Hawks vs Stars: 1
Habs vs Wings: 2
Oilers vs Yotes: 1

I just think there may be improve things by setting a standard on what data points to use. That is all I am trying to say. Is 3% over a course of a year worth overlooking (based on 82 of these events over shots 2500 SA for the season). Completely just ball park numbers and completely change if you just factor 5v5.

[_Q_]

Originally Posted by Street Pharmacist

Pittsburgh's PDO will regress unless they have a season that is unprecedented. However, with their goal differential, it likely means less to their winning percentage. One more goal against here and there and one fewer for them here and there isn't going to make a much of a difference to their winning percentage when they out shoot and outscore opponents by as much as they are. Plus, they've proven that a high save% and high shooting% is actually the norm for them (though nowhere near the height they're currently at).

But see the Flames are 6th in the league in goal differential. Why is it that a drop in PDO will affect Calgary but not Pittsburgh?

I'm actually looking forward to April when are analyst pulls out the "is anyone really surprised this team has made the playoffs? Their PDO is at the top of the league, I'm actually surprised they didn't finish higher in the standings." then everyone will think he's a genius.

[Lanny_McDonald]

Originally Posted by Textcritic

You set yourself up in your first post in this thread as someone who is much less interested in actual dialogue, and primarily seeking to confirm your already strongly entrenched assertion that "these stats are crap." I encourage you to actually stop and take a look at the link provided by Tinordi here, because it outlines fairly precisely why your "explanation" is a moot point. Perhaps he was somewhat unfair in his accusation that you lack a commensurate understanding of the topic, but you pretty fairly brought this on yourself. You have done so by the brash claim at the outset about the COMPLETE invalidity of the advanced stats that Mirtle is employing, and then by responding with an explanation that turns out to be practically negligible.

the problem with Tinordi's link is it shows a weak attempt at normalizing data that is already acknowledged as being faulty. If sampling problems are a result of methodology consistencies the data is considered unreliable. You cannot normalize data that should be excluded from the data set as a result of collection errors. If you do it just further compromises your data.

The fact of the matter is that PDO is useful under certain conditions, but useful nonetheless. It has proven to be a relatively good—if imperfect—predictor of what might happen at the end of the season. By this qualification, would you not agree that this is actually a fairly well grounded method for making general predictions? Or at minimum, for showing areas in which some teams will need to improve in order to either begin to achieve or to maintain success?

If you believe the data to be accurate, then yes, you may attempt to make an inference. But if the data is questionable, then any inference made is just a guess. Do I really have to explain how bad data can misrepresent the facts and present a skewed picture of the facts?

[opendoor]

The big one that stands out to me is shooting percentage. The Flames are shooting at 11.5% and have 5 guys over 20% right now. Obviously that's not sustainable long term. So either they're going to need to generate more shots (they're currently 28th in the league in that regard) or their offense is going to dip.

Yeah sometimes if you build up a huge cushion in the standings you can handle a regression later on like Colorado and Toronto have in the past, but the Flames are only 2 points up on 5th in the division so without an increase in shots they're probably not going to survive the inevitable dip in shooting percentage.

But does this even really matter? Most people picked the Flames to finish at the bottom of the league. People should be happy at how well key young guys are performing and be laughing at the Oilers rather than freaking out because some journalist is poking holes in the Flames' record.

[CaptainCrunch]

You could argue that they also have players with unusually low shooting percentages, ie Backlund, Colborne and others that in theory could balance out the players with the highest shooting percentages if they slump.

[Textcritic]

Originally Posted by New Era

the problem with Tinordi's link is it shows a weak attempt at normalizing data that is already acknowledged as being faulty. If sampling problems are a result of methodology consistencies the data is considered unreliable. You cannot normalize data that should be excluded from the data set as a result of collection errors. If you do it just further compromises your data.

So the solution then is to disregard the interpretations of all the data? I think the fairly simple alternative here would be to recognise the rather infinitesimal margin of error when it comes to data collection for shots from one building to the next. Yes, there are flaws in the data collection system, but what the linked beech does help to show is that with regards to this particular item the variances are likely to have no perceivable impact on the results.

Originally Posted by New Era

If you believe the data to be accurate, then yes, you may attempt to make an inference. But if the data is questionable, then any inference made is just a guess. Do I really have to explain how bad data can misrepresent the facts and present a skewed picture of the facts?

No, but what you should be willing to demonstrate is the degree to which the data is actually questionable. How badly skewed are the numbers, and how have you determined this? If you can establish a good basis for your scepticism, then I will concede the point. But so far, you have provided nothing more than anecdotal claims about the degree to which the data is bad, and how dramatically this affects the outcome.

[_Q_]

So what's a reasonable shooting percentage then? Anywhere between 9 and 10%?

We were 13th last season in shooting percentage at 9.2%. We're obviously a better team this season, so can we expect to reasonably be closer to 10% and finish in the top ten in shooting percentage? And really, assuming we maintain a 26 shots on goal average that we're at right now, a drop in shooting percentage of say 1 or 1.5% only means a drop of two or three goals every 8 games. Is that a big enough drop to make a difference?