Halfway mark statistical analysis of top 20 point producers

[NiL8r87]

The "flaw", as we have discussed, is that the predictive model assumes PP production rate to continue at its current rate and EV production to regress to career average. Tavares has only 7 PPP so far and EV production that is much higher than his career average. Yes, the model is not perfect and some predictions, like that for Tavares, will be obvious in their questionable output.

You could probably use personal career shooting % to predict individual goal totals instead of oiSH%, for assists I don’t know. oiSH% is used moreso for team scoring projection isn’t it?

 

[TOGuy14]

Seems that way, if you put any stock in oiSH%. Or their offence may come from other players.

The problem with using career oiSH% as offensive indicators is that

1. We added an elite offensive weapon in Tavares this offseason which has improved our overall offence immensely

2. Most of our key players are young meaning they have not yet entered their prime. The expectation is that their performance will only grow from past seasons.

 

[42]

PSA

When it comes to hockey statistics, two decimal places will always suffice.

Haha, you're right except I didn't know how to do that with Google spreadsheet and didn't feel like spending the time to learn it. Sorry you have to look at all these decimals

 

[42]

Depending on how into it you want to get, I would take whatever cutoff you consider the top offensive players in the league and average out their oiSH this season and either regress to it or compare it to the same figure for last seasons top scorers and multiply everyone's career oiSH by the difference.

I'd also be curious as to how average shot distance factors in to oiSH regression.

Good suggestions. If I feel the inspiration to put in more time into this, I'll definitely consider them. Thanks.

 

[Martin Skoula]

Good suggestions. If I feel the inspiration to put in more time into this, I'll definitely consider them. Thanks.

Yeah this stuff would be a lot more fun if you were getting paid for it.

 

[lebdafor norris]

Well, I personally don't but I guess we'll see.

The way the season is going, I won't be surprise to see several Leafs players put on a career high ioSH% like what Marchand did last season or Stamkos did during his 10/11 season. Another thing to consider is that while the Leafs are likely getting lucky during ES, their PP (which isn't included by oiSH) on the other hand is going the opposite direction, their PP underlying numbers suggest they should have by far the best PP in the league; 36.15 high danger chances for per 60, the closest team to them is San Jose at 27.64 and we know that unlike Carolina, Leafs forwards actually has the shot and hands to finish their scoring chances.

 

[hatterson]

The fact that Huberdeau is the only player in that group whose current on ice shooting percentage is lower than his career average should tell you something.

It seems far more likely that something underlying has changed than that 18 of the top 20 scorers in the NHL at roughly the halfway point will experience a very significant slowing of their pace. (With McDavid staying equal)

 

[AvsFan29]

Mack and Rantanen still way off.

 

[AvsFan29]

PSA

When it comes to hockey statistics, two decimal places will always suffice.

He was more concerned about the accuracy of the numbers after the decimal point, then he was about the numbers infront of it.

 

[AvsFan29]

My man Johnny T with projected 68 points after 82 games, while he's 37gp and 42p as of now. Talking about a fast decline if I'm ever going to see one. And to think that this fella makes eleven big one's a year!

He has Rantanen scoring 41 points in his final 45 games while having 59 in his first 37 lol he's using oiSH% which is useless in most cases.

 

[lebdafor norris]

The fact that Huberdeau is the only player in that group whose current on ice shooting percentage is lower than his career average should tell you something.

It seems far more likely that something underlying has changed than that 18 of the top 20 scorers in the NHL at roughly the halfway point will experience a very significant slowing of their pace. (With McDavid staying equal)

League average goalie sv% is down 4-7% this year compare to the last 8 seasons, looks like the plan to increase scoring by shrinking the goalie equipment is actually working.

 

[Rich Nixon]

Career all-situations oiSH% stats are available: Combined, PP, EV, SH. They're all on Hockey Reference for every player. Every season, cumulative, whatever. They're not hiding. And, as with last time you did this, you used only EV and then made some weird, nonsense projection for PP production. That's not how stats work at all, whatsoever.

And that's not even the only thing wrong here. As with your quarter season one, I appreciate the effort, but this is just so laughably flawed that all you've done here is waste a lot of your own time. You really, really gotta go back to the drawing board with these.

Really, here's a simpler formula that could work for you: just run the oiSH% for all of the players in all situations based on their last 3 seasons (for better tailoring, since I don't think stuff like Kucherov or Grioux's rookie seasons are relevant to the players they are now). Then compare how they're currently pacing based on those. Then show us the deviations. That would be informative.

 

[lomiller1]

Nice work. IMO you are probably still overshooting on some of the younger players. Ideally you’d want at least 5+ years’ worth of oish% data because inevitable outliers could significantly skew the numbers for anything less. My expectation is that most of the players with less than 5 years in the league will ultimately end up with lower career oish% than they have now even if shooting % in general stay up where they have been so far this season.

 

[lomiller1]

You could probably use personal career shooting % to predict individual goal totals instead of oiSH%

If you used their individual sh% you'd need remove their own shots from their oish% or you are double counting their individual sh%. It could be done and would probably yield better results but it's more complicated.

oiSH% is used moreso for team scoring projection isn’t it?

It reflects directly on how many points the team scores with that player on the ice. If you assume the percentage of goals the player is involved in is relatively constant then point totals should scale with oish%.

 

[lomiller1]

Career all-situations oiSH% stats are available: Combined, PP, EV, SH. They're all on Hockey Reference for every player. Every season, cumulative, whatever. They're not hiding. And, as with last time you did this, you used only EV and then made some weird, nonsense projection for PP production. That's not how stats work at all, whatsoever.

And that's not even the only thing wrong here. As with your quarter season one, I appreciate the effort, but this is just so laughably flawed that all you've done here is waste a lot of your own time. You really, really gotta go back to the drawing board with these.

Really, here's a simpler formula that could work for you: just run the oiSH% for all of the players in all situations based on their last 3 seasons (for better tailoring, since I don't think stuff like Kucherov or Grioux's rookie seasons are relevant to the players they are now). Then compare how they're currently pacing based on those. Then show us the deviations. That would be informative.

You are the one who’s got it wrong. PP and ES are very different on ice situations that should be separated in any type of statistical model. PP oish% are much smaller sample sizes and are more heavily influenced by system and role, so they are less able to stand on their own, but this still doesn’t mean you just combine it with ES.

 

[Rich Nixon]

You are the one who’s got it wrong. PP and ES are very different on ice situations that should be separated in any type of statistical model. PP oish% are much smaller sample sizes and are more heavily influenced by system and role, so they are less able to stand on their own, but this still doesn’t mean you just combine it with ES.

I don't suggest doing that last part at all, I suggest using all of the available data. When you're talking about the production of points in the NHL, particularly among top-tier players, you're going to see substantial icetime and production coming from PP scenarios. They're not small sample sizes and I have no idea what you're saying with "system and role" because, no shit, every aspect of hockey is impacted by that.

John Tavares, for instance, has been in the NHL for 9 years and scored 663 points. Exactly one third of them, 221, have come on the power play. His career oiSH% at even strength is 9.6, his career oiSH% during the scenario where 33.33333333% of his points have occurred is 13.8. Career ES% is not a good indicator of their actual production, and it's not a good way to project point totals.

These analyses are junk stat work. I just don't want the OP to keep wasting his time with them. All of the data needed to do something like this correctly is available, it would just be a ton of work (calculating a unique correlation metric for points based on oiSH% and TOI for EV, SH, and PP scenarios for 20 different players then applying them as predictive tools through 82 games and adding it all up).

 

[Pierce Hawthorne]

Yeah I think there's far too much weight being given to a players career oish% in the current model, for a couple reasons...


1) oish% in general is way up right now. Scoring is up and scoring is especially up for the leagues elite players. Thanks to more enforced slashing rules allowing players a little more freedom to be creative in the offensive zone, especially the high skill guys. And then the goaltending equipment change that also has had an effect on scoring as well.

Because of this, guys who have been in the league for a very long time(Tavares one of them) who has played significantly more years in a lower scoring, lower oish% era will have lower stats here and thus more "expected" regression, when fact of the matter is he likely wont regress anywhere near that level. It's illustrated by the fact Tavares who has been in the league for almost 10 years now has a 9.6%, while Marner, Point, and McDavid who have been around for much less time are all up in the 10.5 range. Those 3 have almost exclusively played in the new NHL where scoring is much higher and it reflects in that oish% stat where they are almost a full percentage point higher then some of the older guys like Giroux, Landeskog, Tavares, etc.

I think a fair "average" shooting percentage to use across the board for these guys right now would be more along the lines of ~11.5%. I dunno what the overall league average in shooting is this year, but I'd venture to guess it's right around 9.5-10.0%. But given that all of these guys are the elite players in the league playing on top lines, they're oish% should be above the league average.



My thoughts on it anyway. Solid analysis all the same that gives a decently accurate representation of some expected point totals. But I think it favors a few young players in particular while hurting some of the guys who have been around for a while because of the era's they've played most of there respective careers in to this point.

 

[42]

Thanks for the feedback, guys.

A couple of comments: yes, I realize, and commented as such, that using career oiSH% is likely going to underestimate the totals for players - and a good point is made that this underestimation will affect the older players more - but it was the most readily available stat to use and short of applying some arbitrary adjustment factor to the oiSH% (or use last 3 years average, which was more work than I was wiling to undertake), it was what made the most sense to me. But at least gives a comparative prediction, even if it may underestimate some of the total values, with perhaps a greater bias towards the younger players.

As far as using a combined EV and PP stats, I am against it for the reasons already pointed out by lomiller1.
I've assumed PP production to continue on current pace but perhaps a better reference point would have been to use career PP production.

The model states its assumptions and shortcomings so use it with these caveats in mind.

 

[The Apologist]

It will be interesting to compare this to end of season actual numbers

 

[oooooooooohCanada]

There’s definitely a flaw based on Tavares’s calculations. He’s never scored so well at this point in the season so to assume he’d end the year 2-3 points away from his career low is crazy. He’s also averaging the highest shot output of his career.

I believe JT had more points at this point last year as he does this year. Kind of crazy tbh.

 

[jetsforever]

I was pretty surprised with Huberdeau but he has 42 points

 

[Sparksrus3]

I believe JT had more points at this point last year as he does this year. Kind of crazy tbh.

He had better linemates last year to carry him

 

[lomiller1]

I don't suggest doing that last part at all, I suggest using all of the available data. When you're talking about the production of points in the NHL, particularly among top-tier players, you're going to see substantial icetime and production coming from PP scenarios. They're not small sample sizes and I have no idea what you're saying with "system and role" because, no ****, every aspect of hockey is impacted by that.

Yes, it's small samples size. PP time at this point in the season ampunts to ~3-5 games worth of 5v5 ice time which is way to law to draw conclusions from. TBH you just seem to be griping for a model that makes your favorites look better.

John Tavares, for instance, has been in the NHL for 9 years

Different coach different PP system. These have a massive impact on who gets the goals/assists. As things stand now, for Tavares's PP point totals to go up 1 or more of Marner, Matthews or Rielly would have to see their totals drop. There is still only 1 puck.

Career ES% is not a good indicator of their actual production, and it's not a good way to project point totals.

These analyses are junk stat work. I just don't want the OP to keep wasting his time with them. All of the data needed to do something like this correctly is available, it would just be a ton of work (calculating a unique correlation metric for points based on oiSH% and TOI for EV, SH, and PP scenarios for 20 different players then applying them as predictive tools through 82 games and adding it all up).

5v5 is a better more stable and reliable indicator of overall offensive ability. You need the right role to generate points on the PP and both the numbers and the eye test say Tavaras isn't getting that on this team. The Leafs run their PP though Marner and Matthews. Rielly touches the puck a lot by necessity being the only D-man. That doesn't leave a lot left for the other 2 players on the ice. This could change drastically and rapidly of the Leafs change how they work their PP, but why would they?

These analyses are junk stat work. I just don't want the OP to keep wasting his time with them. All of the data needed to do something like this correctly is available

There are ways to do it better but you haven't suggested any of them.

 

[Rich Nixon]

Yes, it's small samples size. PP time at this point in the season ampunts to ~3-5 games worth of 5v5 ice time which is way to law to draw conclusions from. TBH you just seem to be griping for a model that makes your favorites look better.



Different coach different PP system. These have a massive impact on who gets the goals/assists. As things stand now, for Tavares's PP point totals to go up 1 or more of Marner, Matthews or Rielly would have to see their totals drop. There is still only 1 puck.



5v5 is a better more stable and reliable indicator of overall offensive ability. You need the right role to generate points on the PP and both the numbers and the eye test say Tavaras isn't getting that on this team. The Leafs run their PP though Marner and Matthews. Rielly touches the puck a lot by necessity being the only D-man. That doesn't leave a lot left for the other 2 players on the ice. This could change drastically and rapidly of the Leafs change how they work their PP, but why would they?



There are ways to do it better but you haven't suggested any of them.

This is inane. It doesn't matter if PP time is limited, we're looking at point shares here and a disproportionate number of points are accumulated on PP. If the whole goal is to project point totals, it's absolutely asinine to discount the value of the scenario where many of these guys put up a full 25+% of their scoring.

It's a badly constructed model. Has nothing to do with my feelings on the particular players (John Tavares isn't exactly a favorite of mine), I just don't understand the value of putting in lots of work and sincere effort into a premise that's just conpletely, irredeemably broken from the start. It's bad statistics. Also...I did suggest a better model, but looking at a single advanced stat (oiSH%) in the first place is a questionable foundation. But if you're going to base it on that metric, then you've got to go all the way with it and be thorough.

 

[lomiller1]

I think a fair "average" shooting percentage to use across the board for these guys right now would be more along the lines of ~11.5%. I dunno what the overall league average in shooting is this year, but I'd venture to guess it's right around 9.5-10.0%. But given that all of these guys are the elite players in the league playing on top lines, they're oish% should be above the league average.

League average is up from a little below 8 to ~8.5. Top offensive payers are normally ~1 point above league average with a handful of elite players or players in very good situations (like playing on a line with an elite shooter) may be as much as 2 points higher. IOW sustaining over 10% oish% would still be a rarity.