The New Meta: An Analysis on Discipline

[Nykonax 1,671]

The New Meta: An Analysis on Discipline
AKA the Gustav is Wrong Thread

Some of you may know me for meta in s79-s82, and I’m here to bring you an analysis on the new meta, and something I haven’t really seen talked about by in-depth by anyone else, and in fact it’s mainly talked about in a negative light by a lot of prominent members in the community.

 

 

 

@Victor@Spartan @Gustav  (for shame)

 

And I, in fact, have been saying it’s good for years now. I will also give a shoutout to @Enorama who hasn’t necessarily been saying it’s good (as far as I can tell), but hasn’t been on the DI is useless train and does recognize it for what it is. I’ll also give a shoutout to @DMaximuswho said this:
 

On 9/8/2020 at 9:58 AM, DMaximus said:

I'd like to see an analysis of PIM and Goals Against and how that fits into all this. It seems like the average PP is around 20%. You could roughly equate that to 10 PIM (5 penalties) = 1 Goal Against. So someone who records 100 more PIM than normal because they have their CK at 80 and DI at 40 is causing their team 10 goals against.

 

I didn’t realise he said until I re-read Gustav’s article in the process of writing this, and I’ve used this metric before in the Prague thread, so great minds think alike. I’ve used similar metrics to this before in arguing for awards here
 

I’m sure some of you have also seen my comments about discipline in other threads (specifically the Prague thread) but here’s a concrete manifesto of my whole thoughts on it.
 

1. What does Discipline actually do?

I feel there is a large misconception about what discipline does. Intuitively it would make sense that CK increases how much your player hits, or how hard your player hits, or some explanation with total hits. Likewise, it would make sense that discipline is almost an attribute of how “clean” your player hits. One would expect that increasing CK would result in more hits, and increasing DI on top of that would result in the same increase in hits but less penalties off of it.
 

However, that is not the case. The quick punchline here is that CK increases how often your player hits, DI decreases how often your player hits, it’s basically negative CK.
 

Lets take a look at some numbers. I have 28 seasons of player stats and attributes from S70 - S97. We’ll first explore some basic relationships between the variables of interest (CK, PIM, DI) and then get into some regression models.

 

 

From the charts we can see very obvious strong positive correlations between CK, Hits, and PIM.



This is a super similar chart to what Gustav had, we have a super high population at 40 DI (mostly due to people thinking it’s worthless) and some sparse data points above that with a weak negative correlation. Like Gustav also said, this chart doesn’t really prove much, since it doesn’t tell us anything about CK. The people at 40 DI could have 99 CK or 40 CK for all we know, same with any other point.
 

Let’s then look at the gap between DI and CK, so that we can get information on both.

Now we see a negative correlation. As DI increases compared to CK, we see a decrease in hits. We can further analyze this plot and make some improvements to it to build an even better understanding, since there’s still some points that can be made about this one.

We’ll get a bit more colour in here to make it look a little fancier. Here I’ve coloured points green if Discipline = 40, and red if Discipline isn’t 40. Let’s dive deeper into this plot now.

We can use the 0 point on the x-axis as a marker of sorts. Anything to the left of that point is super similar to CK vs. HIT’s plot we made earlier and we still see that strong correlation. This is further backed up by a lot of these values having DI of 40, which suggests that this is just purely caused by increases in CK. If we look to the right of the 0 marker, we see all the players who have upgraded DI higher than CK. We can “zoom-in” on this.

This does look a little random, but there is a negative trend line, and we see the right end having much less hits than the left end. I’ve coloured the graph by whether CK is 40 or not this time, and we can see that a lot of the upper hits numbers are a result of increased CK on top (which we will see later).

 

I can see why the conclusion that DI isn’t that impactful was made, especially since Gustav was just looking at s72 data (a much smaller sample size). It’s also hard to isolate DI from CK, and the correlation between Hits and CK is very small. If you are questioning my results right now I totally understand.

From this though we can certainly conclude that CK, PIM, and HITs are all positively correlated with one another, and there may be a dubious relationship between DI and HITs.

 

2. Deep Dive With Statistics

We’ll do a regression of CK and DI on number of hits.

Here is some R code output. I’ll interpret it for you if you have no idea what you’re looking at.

The coefficients section is basically how much each variable affects the outcome. The estimate is the main one we want to look at. We can see that CK has an estimate of 4.41 and DI has an estimate of -2.52.
 

The regression is like a linear line, where Number of Hits = 4.41 * (CK Value) - 2.52 * (DI Value) - 10.46
 

So each point in CK is about 4.41 more hits, each point in DI is about 2.52 less hits. We can ignore the intercept of -10.46 since that’s not really relevant (it’s the number of hits you’d have if you had 0CK and 0DI). We also see that the P-value is very significant.
 

We can do the same regression on PIMs as well.

We see that each point increase in CK is about 2.7 more PIM, while each point in DI is about -1.12PIM.
 

From both of these regressions, we see that DI is about half as “powerful” as CK, in that 1 point in DI reduces about half the hits that 1 point in CK gains, similarly 1 point in DI reduces about half the PIM that 1 point in CK gains.

We however see an R^2 of 0.6, which means this regression explains 60% of the variance in the dataset. Since this is a sim engine, this means that 40% of your hit total cannot be explained by attributes, and is probably a factor of team/lines/minutes.

 

Interlude:

I will also just share that throughout the course of writing this I’ve been slowly adding more and more seasons. This started as just S83-S97 but the final version for now is S70-S97 (before S70 my function to scrape doesn’t work due to index formatting differences and I don’t want to fix it right now). However, as I’ve added more data, especially pre-hybrid, the estimate of DI has gone up, while the estimate of CK has gone down. If we just use S83-S97 data the CK estimate is 3.11 and the DI estimate is -0.93. I think this is just due to a bias in the sample, in that a lot of players keep DI low (due to a lot of people thinking it’s a useless stat). We can visualize the distribution.

 

We can see a huge spike at 40 DI, with very few players actually choosing to upgrade DI (In-fact 71% of players in the dataset have DI at 40). After hybrid attributes the number of players choosing to upgrade DI is even smaller than pre-hybrid, which is why including more seasons before S83 is increasing the “power” of DI, since we are able to sample more players actually updating it.

 

 3. Why this is important.
 

So you may be asking, what’s all the fuss about discipline, maybe it works, maybe it doesn’t, why is it good? I’ll tell you why it’s good.
 

Hits are bad.
 

As we saw before, hits are directly correlated with penalties. Penalties are correlated with power plays for the other team, and power plays for the other team are correlated with goals against your team.
 

I’ll show you what I’m talking about. Consider a player with 300 hits. From our previous plot, we’d expect them to have about 160 PIM. 160 PIM is about 80 powerplays for the other team. If the other team has an average power play of 18%, that’s about 14 goals against your team. For comparison, that’s about equivalent to the difference between a .930 goalie and a .924 goalie.
 

We can further extend this to say each hit is about 0.53 penalty minutes. Which means every 4 hits results in a powerplay. With an average power play of 18%, about every 20 hits results in a power play goal against your team.

If we furthermore use the regressions, 1 point in CK is equivalent to 2.7 penalty minutes. If each powerplay has a 18% chance of being scored on, 2.7 * 0.18 = 0.486. So every point in CK results in almost 0.5 goals scored against your team
 

Similarly, 1 point in DI is -1.1 PIM. So every point in DI results in 0.2 less goals against your team.
 

One flaw in the metric (that also helps my case), is that if you are a player who has high DF and is on the PK, say for example a team’s top defenseman, hitting is even WORSE for you, since those powerplay averages will presumably go up against your team if the top defenseman is the one in the penalty box.
 

We can conclude though that hits are bad, but why not just avoid updating CK and not touching discipline? Well there does come a point where it is better to upgrade discipline.
 

4. When is upgrading discipline useful

The main problem with discipline is that it’s effect is kind of weak as seen above. In this article I’m not proposing that everyone should just be rushing DI to 99, it’s not that good. In order to properly value it we need to consider the TPE opportunity cost of DI vs. other attributes. Obviously it is much better to upgrade DF to 50 before you upgrade DI to 50, but at what point does it become worth it to upgrade DI?
 

One thing that I’ve seen pointed out is that at some point you have to upgrade BC as it’s the cheapest way to get STHS DF. However, this comes with the drawback of getting 0.06DF but also 0.36CK. So maybe there is a way to also upgrade Poise in-line with Body-Checking to cancel out the CK and just get the DF, while also being cheaper than the next best source of DF. Time to do more math.
 

The most efficient source of DF is PC (Poke Checking, but I always think this is Puck Control for some reason), which gives 0.24 DF and 0.18 ST.
 

At a value of 80 it starts costing 5 tpe to upgrade, so each tpe is worth 0.24/5 = 0.048 DF. Once PC is 80 you start getting more DF per TPE from BC than PC. (0.06 vs. 0.048). But the cost of that cheaper DF means to cancel out the CK increase you basically need 1 point in Poise. So it actually costs 2 tpe for that 0.06 DF to be purely beneficial.
 

We can use an estimate that 1 point in CK is cancelled out by 2 points in DI (0.36CK vs. 1.08DI for upgrading 1 BC and 2 PO respectively), which results in an overestimate [less DI is needed to cancel], and overestimating here is fine.
 

So we need to find a point where 0.06/3  = 0.02DF/tpe (since you need 1 tpe for BC and 2 tpe for the PO to cancel it) is a better alternative to other DF sources.
 

We find that for PC this will never happen, as at the maximum 0.24DF/10tpe = 0.024 DF/TPE, but for the other DF-granting attributes this does happen.

For DK which gives 0.12DF, this happens when DK = 90 (0.12/8) = 0.015DF/TPE
For DC which gives 0.18DF, this happens when DC = 95 (0.18/10) = 0.018DF/TPE
 

So once DK and DC get to the 90/95 breakpoints respectively, it becomes more efficient to upgrade BC and cancel out the CK increase with PO than to continue upgrading them. Which if you are only upgrading defense, will happen around 580 player TPE.

5. Takeaways, Further Theories, and Conclusion

Alright, so we’ve concluded that DI is not in fact useless as many have claimed, and does result in a reduction in Hits and PIM.
 

The main reason why DI is useful though is twofold, hits in general are just bad and you want your player to avoid them as they cost your team goals against. DI does slightly reduce hits and PIM (although it’s easier and more effective to not get hits and PIM by just avoiding updating CK). However, as seen above there does come a breakpoint where to efficiently get DF you need to also upgrade CK, discipline does allow you to effectively cancel this out.
 

However, throughout this analysis we’ve just seen that DI has a somewhat weak effect. I think that this is just due to how STHS works, and a lot of the further variance in hits for players is just due to roster composition and STHS giving a bunch of hits to people high CK already on the team. So for instance if you have a player who is 40CK 40DI, they’ll get about the same amount of hits as someone who is 40CK 60DI just because on the team there is someone with 80CK who is taking all the hits, and not leaving many to be assigned.
 

A basic example to explain what I’m talking about is imagine a line of 3 players, one has 80CK, one has 40CK 40DI, the other has 40CK 60DI. If STHS decides to give that line 100 hits, suppose the 80CK player gets 80 of those 100 hits, and then the remaining 20 hits are to be given to the other players. That distribution may look like 12 hits to the 40DI player and 8 hits to the 60DI player. The 60DI player gets “less” hits but the effect of it is so small just because the 80CK player is absorbing all of it.
 

I think that if you have a team of low CK players, say for example a line of 3 40CK 40DI players, STHS may now evenly assign like 80 hits. However, if those players are now 40CK 60DI, STHS may now assign only 60 hits.

What I’m saying is that DI becomes strong on a overall low CK team, as there’s no hits going to the team already, and reducing the hits even more is beneficial (since hits are bad).
 

I saw this anecdotally with test sims with Vancouver in the S80’s. We had a low CK team, and when I ran test sims with everyone at 60DI our winrate increased by like 5% or so.
 

Also please don’t try (or do try) to make the argument that hits lead to takeaways which lead to goals. I might write another article on this but I just don’t think it’s true, or isn’t true at the rate it needs to be true. For the takeaways to be better you’d need a ~5% chance of scoring as a direct result of each hit, which there’s just no way that’s true.

TLDR Results:

Over 25+ seasons of VHL data:

It’s slightly hard to analyze DI just because so little people upgrade it
1 point increase in CK results in 2.7 more PIM and 4.41 more hits
1 point increase in DI results in 1.1 less PIM and 2.52 less hits
Hits are bad because they lead to penalties -> powerplays -> goals against your team
DI allows for efficient upgrading of DF in BC by cancelling out CK
Overall low CK teams probably benefit from DI more.
and most importantly, @Gustav is wrong.  

 

 

2700 words

[Enorama 2,055]

24 minutes ago, Nykonax said:

I will also give a shoutout to @Enorama who hasn’t necessarily been saying it’s good (as far as I can tell), but hasn’t been on the DI is useless train and does recognize it for what it is.

 

Vidot had 70 DI before the hybrid attribute rollout and 75 after!

[Spartan 4,839]

1 hour ago, Nykonax said:

community.

 

 

 

@Victor@Spartan @Gustav  (for shame

1. Speed is OP in DDSPF so I'm actually saying DI is OP in STHS

 

2. That screenshot has to be fake I'd never say something that optimistic about DI

[jacobcarson877 2,678]

I don't hate the take that reducing penalty minutes can be a better investment than other things at a certain cost point, but that cost point is definitely VERY far down the TPA scale. I also agree that hits are probably not actually efficient for driving offense relative to penalty rates, but of course there surely are goals that stem from hits, so paying to reduce a little good and a middling amount of bad always feels bleh.

 

The real perk of DI is that it doesn't depreciate! So you truly do only pay for it once, which only Leadership can compete with, and well...

[Nykonax 1,671]

4 minutes ago, jacobcarson877 said:

I don't hate the take that reducing penalty minutes can be a better investment than other things at a certain cost point, but that cost point is definitely VERY far down the TPA scale. I also agree that hits are probably not actually efficient for driving offense relative to penalty rates, but of course there surely are goals that stem from hits, so paying to reduce a little good and a middling amount of bad always feels bleh.

There are goals that stem from hirs for sure but there's more goals against than goals for from hits. Another part of the analysis is that on PK your team also is probably not scoring a goal, so the expected value of a hit is even worse.

 

I think DI is a worthwhile update at some high point, like 40->50 di is probably better than 90->91 df or something like that. Especially if you can get a whole team low CK with some additional discipline it adds up. I want to do more modeling on some kind of expected points metric factoring attributes hits penalties stuff like that down the road.

[JardyB10 4,892]

2 hours ago, Nykonax said:

2. Deep Dive With Statistics

We’ll do a regression of CK and DI on number of hits.

Is this STHS source code? If so, could you perhaps rip it and make it better?

Edited yesterday at 08:47 PM by JardyB10

[gorlab 4,375]

damn, your university classes are paying off nyko, good job.

[Gustav 6,634]

I have a few snarky things to say, so I'll get something out of the way right up top: this is a really good article and your analysis is valid. I'll forever regret not flipping around the attribute names in the title of my old article and making it "Gustav's Big Fat DI/CK Manifesto," but everyone makes mistakes. I think that was the biggest mistake about it, though, because I'd disagree that I was necessarily wrong about any of it. This is just a better version.

 

I think that essentially the same conclusions are here as have always been there: that DI reduces hits and CK increases them. If I remember correctly, my original article only looked at 4-ish players who had both upgraded substantially. All of those players had unimpressive hit totals and weren't throwing any "cleaner" hits than anyone else (the PIM rate wasn't any lower than league average). It's obvious to anyone who cares to think about it that that isn't enough of a sample size, but something I concluded that I think is backed up here is that if you want to put up hits, it's pretty much useless to also try to upgrade DI with CK. You also seem to back this up with the analysis stating that DI works about half as well as CK both ways. If you wanted to target specific totals for either hits or PIM, then it would be smart to upgrade one of those only.

 

4 hours ago, Nykonax said:

 

I really like this part because it illustrates basically what I said here--visually speaking, it doesn't appear that there's much difference when upgrading both attributes. But something I don't remember concluding (likely because of small sample size) is that upgrading DI only does appear to be mildly helpful if your objective is to avoid PIM.

 

 

I'm also going to inject this comment with a bit of a traditionalist opinion. We've basically always known that upgrading CK and putting up hits is heavily correlated to PIM, which is bad for the team (though I appreciate that you took the extra step of quantifying this and looking at when upgrading DI might be helpful; that part was cool too). Adding PIM can also cut into your ice time, and it's going to be an opportunity to build more efficiently with respect to TPE if you ignore CK entirely. Pretty objectively, it's not a smart build. That said, my traditionalist opinion is that hits are fun and I like seeing my player record them. I don't think I've ever noticed a single-game PIM total of mine unless it's in the double digits (at which point, I scroll to look for fights, which I also think are fun), but I check how many hits I had in every single sim and I'm happy when I have a lot of them. If your goal is to have a TPE-efficient build and be your GM's best friend, then this isn't the first time that it's been clear that you probably shouldn't try to build a physical player. I've known this across multiple player careers, but I upgrade CK anyway simply because I personally enjoy it. That above all else should be the strongest factor in determining how one's player build should progress.

 

I hope I can say that while still making it clear that I really liked reading this. I'm looking forward to more!

[Nykonax 1,671]

2 hours ago, Gustav said:

because I'd disagree that I was necessarily wrong about any of it. This is just a better version.

Oh yeah I absolutely agree. I don't think you said anything really wrong in your thread, I just quoted one line of yours taken wildly out of context for better content lol. The point you make about there not being a reason in upgrading CK and DI is absolutely true at least in the pre-hybrid era, and is basically still true now. The only argument to updating both CK and DI is in my point 4 when you are hyper min-maxing your player.

 

The point of this article was to mainly just share that discipline is not a useless attribute as many claim or think, and also have a central source and evidence on what it actually does. Certainly DI just reducing hits was around for a while, but I think it was almost like a folklore result, and that actually measuring its true effect is something useful.

 

It's also just to share some deeper VHL/STHS analysis that I personally think is interesting in min-maxing the sim and exploring the deeper effect of things from an analytics side.

 

I absolutely agree that checking is super fun and people should build it if they want to have fun with their players. It definitely feels good lol.

 

 

[BOOM 9,144]

Holy shit.

 

MIT wants its advanced research back.

 

Reading this made me realise that I'm a bit of an idiot.

[Daniel Janser 2,336]

2 hours ago, BOOM said:

Reading this made me realise that I'm a bit of an idiot.

I feel you brother

[LucyXpher 2,010]

Double posted, idk what happened 

Edited 5 hours ago by LucyXpher

[LucyXpher 2,010]

From what I've gathered reading the STHS forums, and SimonT's replies specifically, DI definitely reduces PIMS by reducing hits since that is primarily where PIMS come from.  It also seemed suggested too that DI would likely reduce the number of fights or non-physical penalties your player might take (random game misconducts).  I'd be kinda curious to see a DI/PIM graph since I don't think you had one above as you focused primarily on hits as the main culprit for PIMS and I wonder if DI has some unaccounted value in limiting other various types of PIMS.  

 

Also, anecdotally, my DI rating in S97 seemed to do very little to prevent me from hitting and taking penalties, and interestingly, the seasons where I hit more and took more penalties resulted in higher personal points totals and team hardware.  Obviously the sample size is wayyyy too small to make any conclusions, but it does kinda baffle me since I'd be inclined to invest in DI if I hadn't already tried that with a result that went completely against what I'd expect.  What I think we might take from this, though, is that there are numerous other factors that contribute to high hits/pims totals, including relative CK rating to teammates and opponents.  I honestly have no clue, but my own personal results leave me a little more confused than anything about the value of DI and hits.

 

S98 Prague: 65 CK, 40 DI - 119 hits, 48 pims, 31.12 AMG, 63 points (rerolled)

S97 Malmo: 67 CK, 61 DI - 193 hits, 115 pims, 28.79 AMG, 92 points, Victory Cup

S96 Riga: 67 CK, 40 DI - 73 hits, 85 pims, 29.61 AMG, 79 points

S95 Riga: 67 CK, 40 DI - 150 hits, 119 pims, 30.20 AMG, 86 points, Continental Cup

 

Not to try and grasp at straws, but is it possible that hitting provides some kind of morale boost in addition to a turnover, which might help boost the value of a turnover caused by a hit?  That's total speculation...

Edited 5 hours ago by LucyXpher

[Doomsday 4,201]

Hits definitely have a purpose, but since our version of STHS doesn't track takeaways resulting from hits, their positive impacts can't really be measured. The more hitters you have, though, does lead to more penalties, so teams need to be careful with how many heavy hitters they roster. 

 

Overall, if a player's build doesn't want to be a hitter, they absolutely should invest in DI.