Messin’ with data – vbaddict x tanks.gg x wn8

[ModderKraken 59]

bounce data fixed below - sorry

So after I deciding to grind the soviet LT line I rebought the T-34 to unlock the MT-25. I was shocked to realize that the Tier 5 T-34 gun had a DPM that overshadowed some of my Tier 8s.

I started to look at some data to see if there were some other outliers like the T-34, maybe at less blatant seal-clubbing tiers. Things quickly got out of hand from there.

Below are some of the more interesting data views I came up with by pulling data from various sources. I also attached the spreadsheet in case someone wants to geek out. All of this is for Tier 6 + and for tanks with at least 1k battles in the data set.

Disclaimer: obviously this data contains a significant skill skew as you go up the Tier ladder and as your look at reward tanks that are typically driven by some of the best players only.

All-in, no massive surprises, but I found it interesting.

DPM by tank

Raw DPM alone doesn’t tell the full story – crappy accuracy combined with poor gun handling and penetration can mess up an otherwise good DPM. So I compared theoretical DPM (average damage * rounds per minute) with “effective” DPM (theoretical DPM * average hit rate * average penetration rate).

Interesting are some of the outliers: T-44, ARL 44, T20 (average players struggle to convert their already low DPM). High-Tier Russian hover craft mediums and British TDs have the highest effective DPM and DPM yield.  Combined with hover craft mobility, that explains how Russian mediums can carry games better than others.

DPM, HP and survivability

So which tanks should you snipe with vs. which tanks should you brawl with? And what effect does armor and survivability have on this?

 

If you compare effective DPM with Health, you can see which tanks can both dish out and absorb damage, vs. which can only do one of the two well. T-44 still sucks, while top-tier autoloader heavies & big TDs rule supreme. But what about does that mean I should brawl with them? Which tanks can bounce incoming fire while still pushing out massive damage compared to their health?

 

In the bottom right you have tanks that put out tremendous amount of damage compare to their own HP, but can’t stop any incoming fire. When it comes to doing damage these guys should hang back in the second line and snipe.  Think Bromwell and  59-16. On the top left, you have tanks that can take a beating but need to have their guns hot for a long time to make a different – brawlers.  The guys in the bottom left are fucked – can’t handle incoming fire and so comparatively little effective damage. 

Per special request:

Effective HP vs. Eff DPM

With help of RichardNixon I was able to estimate "effective" HP - here defined as theoretical HP plus incoming damage that doesnt reduce HP (bounces, eaten by tracks etc).

T-44 still sucks

 

Stuff for missions:

Which of your tanks is good for HT-12 and TD-8 (bounce or do x-times your HP)?

NSFW – IF YOU ARE SENSITIVE TO STAT PADDING, DON’T READ ANY FURTHER.

So what if you are a shit lord like me and want to figure out tanks that are not only fun to play (high DPM), have low expectations to get a decent wn8 and offer some reasonable survivability?

Expected Damage vs. DPM

 

Tanks on top of this graph have very high expectations for damage compared to their effective DPM (Wn8 Exptected DMG / Eff. DPM), while tanks on the bottom of the chart have very low expectations. For example, for the VK28.01 even after accounting for average hit rate and penetration yield, you only need to keep your gun hot and on target for ~0.5 minutes to achieve the expected damage threshold.

Tanks on the left can’t bounce incoming fire, but tanks on the right can. So I am guessing that the tanks on the bottom are easier to pad (i.e. do well re wn8) in general. To do so, you need to avoid enemy fire better than the average for the bottom left (e.g. 59-16), or keep your gun on target more for the bottom right.

So lights in general offer opportunity to pad, So do SU-152, E2, and some of the Japanese Heavies - average players struggle to convert their DPM into damage and good players may keep their guns in the fight longer and cash in on those lowered expectations.

And the T-44 still sucks.

Data:

https://www.dropbox.com/s/54j4uiszuy3zs8k/dpmV3.csv?dl=0

All data from:

vBAddict (bounce, hit rate and pen rate data), wn8 expected value tables (expected damage) and tanks.gg (HP,DPM). ArmEff from RichardNixon.

Tx to @Orrie and @RichardNixon for tips on the data

 

Edited November 1, 2015 by ModderKraken
Fixed flawed data, added moar data cuts

[Sergeant_Fgt 479]

Stop that cyber bully.  
t-44 was my first t8 med and 2nd non prem t8, it is a gr8 vehicle to zoom around.

So.. basically.. vk 2801 still good to pad, y is that 52k battles guy still so terrible at it then ? Don't tell me he is padding assistant dmg  
_
Besides that, nothing new imo.

[ModderKraken 59]

9 minutes ago, Sergeant_Fgt said:

Stop that cyber bully.  
t-44 was my first t8 med and 2nd non prem t8, it is a gr8 vehicle to zoom around.

So.. basically.. vk 2801 still good to pad, y is that 52k battles guy still so terrible at it then ? Don't tell me he is padding assistant dmg  
_
Besides that, nothing new imo.

Agree nothing earth shattering. But tanks that suck can have tremendous potential in the right hands. For example, T20 which is nearly as bad as the T-44, is one of my most comfy carry tanks and something that I can shit wn8 with - ideal tank. If you can make shit tanks work, you win.

I actually ran into Taz a few nights ago, I think he played fairly defensively, sniping from second line.

[RichardNixon 835]

The bounce rate is interesting. What data was used for that?

15 minutes ago, Sergeant_Fgt said:

So.. basically.. vk 2801 still good to pad, y is that 52k battles guy still so terrible at it then ? Don't tell me he is padding assistant dmg

The VK2801's a bad WN8 padder because the expected values are still dominated by the derp HEAT version.

 

[Sergeant_Fgt 479]

7 minutes ago, RichardNixon said:

The VK2801's a bad WN8 padder because the expected values are still dominated by the derp HEAT version.

 

How long does it take for expected values to react to changes made to the game ? Is it based on a battle count 'pushing out' older results or does it take the stats as a whole ? If the latter is the case I imagine it could take longer as a lot of ppl prolly stopped playing it after the changes and it takes awhile to 'correct' the expected values.

[Mamas_Calvin 133]

34 minutes ago, RichardNixon said:

The bounce rate is interesting. What data was used for that?

The VK2801's a bad WN8 padder because the expected values are still dominated by the derp HEAT version.

 

VK2801 is currently ranked 42nd amongst tier 6 tanks for WN8 efficiency at VBAddict. It would seem that most players cannot convert the graphed potential into statisical success.

[Fabunil 1,597]

The VK 28.01 has one of the best bouncerates in the game? 

The VK 72.01 bounces less than the Type 64? 

Where the hell did you take these statistics from, the 3rd graphic doesn´t make any sense.

[RichardNixon 835]

3 minutes ago, Sergeant_Fgt said:

How long does it take for expected values to react to changes made to the game ? Is it based on a battle count 'pushing out' older results or does it take the stats as a whole ? If the latter is the case I imagine it could take longer as a lot of ppl prolly stopped playing it after the changes and it takes awhile to 'correct' the expected values.

Overall data doesn't contain any date information except account creation and last battle times. To work around that, the "classic" WN8 method assumes that any tanks in a player's top 50% (by WN8) are recent, and throws the rest away. Because most players improve over time, this usually works: Older tanks fall behind recent tanks, and so the newer results increasingly dominate. However, some older results stick around because not all players improve over time, and so the VK2801 will always be somewhat overrated by this method.

When a tank is nerfed extremely heavily, it can cause almost all recent performances to be discarded in favour of pre-nerf performances, and so the expected values remain near the old value. In WN8 this only applies to pre-8.6 artillery, notably the SU-26, M7 Priest, SU-5 and M41.

[CarbonWard 3,529]

2801 used to be great in the Fall of 2012 along with all the other 105mm caliber HEAT shells.

 

But we all know what happened to that.

[ModderKraken 59]

45 minutes ago, Fabunil said:

The VK 28.01 has one of the best bouncerates in the game? 

The VK 72.01 bounces less than the Type 64? 

Where the hell did you take these statistics from, the 3rd graphic doesn´t make any sense.

You right, me wrong. My bounce data column was jumbled for some reason. Results much more intuitive, fixed above. Sorry for confusion. 3rd and 4th charts changed, rest didnt.

2 hours ago, RichardNixon said:

The bounce rate is interesting. What data was used for that?

vBAddict - I think it is % SHOTS bounced not % DAMAGE bounced. Would have preferred the latter. Can't tell if a failtoon Luchs is circling a Maus and tickling him with he pew-pew or if he is bouncing bigger caliber. 

Data was flawed, fixed now.

[Heldar 10]

y u do numbers?  y u no say Heldar - pad this tonk?

[ModderKraken 59]

Can I get the HeldarPack with auto-padding?

[Orrie 1,017]

And I actually found the T-44 overall fun to play, _{^{when I wasn't bottom tiered.}}

Nice graphs though, can be of good use. :^)

[SFC_Storm 58]

I wrote a very long thing but it got erased.

I asked if you could fill in each graph for the top 25ish%.

Graph1:From AMX30b point to Vk2801 and all tanks above towards Vulcan Cannon

Graph2:From Conway down to Maus and everything above that.

Graph3: From Oho to ISU152 and all the tanks above

Graph4: Draw a line from Maus to Su152 and please label dots as those tanks.

I gave you lots of props and said this could catch on blah blah blah I totally could see this on Vbaddict etc.

I know its a bit of work but you already have graphs. I would have to chart it all and then label. Anyways let me know. And if you can only do a few do the Padding one and one more.

 

Also I wanted to know is there a way to make the "Hardness Rating" similar to Garbad. Where you take Raw HP and combine that with Bounce and get the adjusted or "EffectiveHP"? So for instance a IS7 bounces 50% more as much as a 215b or 100% more than AMX50b.

 

Then you could make the ultimate Effective DMG vs Effective HP or Eff DMG vs Eff HP vs Exp DMG ETC.

Please keep making these. Your on to something.

[ModderKraken 59]

4 hours ago, SFC_Storm said:

I wrote a very long thing but it got erased.

I asked if you could fill in each graph for the top 25ish%.

Graph1:From AMX30b point to Vk2801 and all tanks above towards Vulcan Cannon

Graph2:From Conway down to Maus and everything above that.

Graph3: From Oho to ISU152 and all the tanks above

Graph4: Draw a line from Maus to Su152 and please label dots as those tanks.

I gave you lots of props and said this could catch on blah blah blah I totally could see this on Vbaddict etc.

I know its a bit of work but you already have graphs. I would have to chart it all and then label. Anyways let me know. And if you can only do a few do the Padding one and one more.

 

Also I wanted to know is there a way to make the "Hardness Rating" similar to Garbad. Where you take Raw HP and combine that with Bounce and get the adjusted or "EffectiveHP"? So for instance a IS7 bounces 50% more as much as a 215b or 100% more than AMX50b.

 

Then you could make the ultimate Effective DMG vs Effective HP or Eff DMG vs Eff HP vs Exp DMG ETC.

Please keep making these. Your on to something.

Thanks! Will do, give me a couple of days on the graphs. Busy weekend.

I was thinking about effective HP, but the data isn't great for it. Let me play with it.

[TheQuagmire 30]

How does a Tog have bounces...

[Kitten 1,388]

Japanese heavy tanks a bit misleading since you're using the interim expected damage :p

[ModderKraken 59]

41 minutes ago, TheQuagmire said:

How does a Tog have bounces...

Tracks, tiny caliber vs. angled paper mache etc. All tanks have some random bounces (in this definition they are shots received that didnt cause HP reduction - so counting shots that get eaten by spaced armor in tracks etc).

23 minutes ago, Kitten said:

Japanese heavy tanks a bit misleading since you're using the interim expected damage :p

True dat. As always, pad at your own risk. 

Also, after hitting submit on the OP, I noticed that you just released new expected tables. 

5 hours ago, SFC_Storm said:

Also I wanted to know is there a way to make the "Hardness Rating" similar to Garbad. Where you take Raw HP and combine that with Bounce and get the adjusted or "EffectiveHP"? So for instance a IS7 bounces 50% more as much as a 215b or 100% more than AMX50b.

Then you could make the ultimate Effective DMG vs Effective HP or Eff DMG vs Eff HP vs Exp DMG ETC.

So I am thinking this - from vBAddict, I can get following data:

• Average damage received (HP reduced)
• Average shots % that didn't reduce HP

So I can calculate (with a risky assumption that shots bounced have same average damage potential per shot as penetrating shots):

Effective HP = Theoretical HP - Damaged Received + (Damage Received) / (Shot % that didnt reduce HP)

So I am saying that: Damage Received / Shot % that didnt reduce HP ~= average total incoming damage absorbed by tank

What do our math guys say to that? Better ideas for "effective HP"?

[stagnate 513]

I am not sure of a better way, but am 100% sure bounced shots have a lower average damage.  Bounces are going to be dominated by low tier low pen guns, damage dominated by higher tier high pen guns.

 

maybe find a tier average damage and use the -1 or -2 tier average damage for bounces.

[ModderKraken 59]

34 minutes ago, stagnate said:

I am not sure of a better way, but am 100% sure bounced shots have a lower average damage.  Bounces are going to be dominated by low tier low pen guns, damage dominated by higher tier high pen guns.

 

maybe find a tier average damage and use the -1 or -2 tier average damage for bounces.

Yes agree on your concern - that is why I flagged that as a major assumption. The issue is that I dont have data on DAMAGE absorbed without reducing HP. Any ideas where to get that would be stellar - in API?

That said, I would expect the effect to bigger on in the mid tiers, and less so in the higher tiers (much russian medium gold flinging going on). 

You idea is essentially develop a "discount" that I apply to the incoming damage value. Two ways to get at it:

a) Look at average per shot damage by tier - oh boy, that'll be a bitch to get in a usable form unless someone can scrape the wiki real good for me (plus I need to make assumptions about mix of top vs. stock gun, derps etc). From there I need to calculate/estimate the mix of MM placement per tank etc. Sounds very complex and still lots of assumptions.

b) Somehow develop a flat discount by tier that is based on penetration rates by tier (you are assuming that higher tier tanks pen more than lower tier tanks) and then assume some rule of thumb MM tier mix that each tank will face. More realistic to do, but moar assumptions.

[SFC_Storm 58]

52 minutes ago, stagnate said:

I am not sure of a better way, but am 100% sure bounced shots have a lower average damage.  Bounces are going to be dominated by low tier low pen guns, damage dominated by higher tier high pen guns.

 

maybe find a tier average damage and use the -1 or -2 tier average damage for bounces.

This would also depend on the tank I think. So a hard IS7 you can assume that bounced stuff is less DMG low tier guns. But vs a STB1 or Leo bounced will almost always mean poorly aimed shots as a T8 med will pen as easily as a Jagzilla.

Isnt there a number of avg DMG blocked? They have been tracking that since making tanked armor get XP right?

 

Either way STagnate has a good solution because with these it will always be safe to be on the lower side vs overestimating a vehicles Eff HP Better safe than sorry

5 minutes ago, ModderKraken said:

Yes agree on your concern - that is why I flagged that as a major assumption. The issue is that I dont have data on DAMAGE absorbed without reducing HP. Any ideas where to get that would be stellar - in API?

That said, I would expect the effect to bigger on in the mid tiers, and less so in the higher tiers (much russian medium gold flinging going on). 

You idea is essentially develop a "discount" that I apply to the incoming damage value. Two ways to get at it:

a) Look at average per shot damage by tier - oh boy, that'll be a bitch to get in a usable form unless someone can scrape the wiki real good for me (plus I need to make assumptions about mix of top vs. stock gun, derps etc). From there I need to calculate/estimate the mix of MM placement per tank etc. Sounds very complex and still lots of assumptions.

b) Somehow develop a flat discount by tier that is based on penetration rates by tier (you are assuming that higher tier tanks pen more than lower tier tanks) and then assume some rule of thumb MM tier mix that each tank will face. More realistic to do, but moar assumptions.

I think simply take the AVG alpha from forinstance T7 Lights, T8 TD, T8 heavies and T8 Meds and apply that as avg. SOmething very simple. We can get into more complicated stuff later. Like I said, error on the side of less Eff HP and all tanks might lose a few % HP over reality but that is better than overestimating HP. Most important thing is they are treated the same. More than the actual Eff HP is there ration of hardness to eachother.

[ModderKraken 59]

So the API offers damage blocked in addition to hits blocked.

https://eu.wargaming.net/developers/api_reference/wot/tanks/stats/

But at a per player vehicle level- so you would need to pull data for every player and each of their vehicles and then aggregate. That is above my skill / patience level.

I assume the dude behind vBAddict does exactly that, but not sure why he only publishes part of the data. Anyone here friendly with Marius? Maybe I can parlay the fact that I used to live 10km from his home into some goodwill?

http://www.vbaddict.net/about.php

[RichardNixon 835]

4 hours ago, ModderKraken said:

So the API offers damage blocked in addition to hits blocked.

https://eu.wargaming.net/developers/api_reference/wot/tanks/stats/

But at a per player vehicle level- so you would need to pull data for every player and each of their vehicles and then aggregate. That is above my skill / patience level.

I assume the dude behind vBAddict does exactly that, but not sure why he only publishes part of the data.

It's more likely that he uses data from replays & battle results. The problem with the tanks/stats values is that they're only averaged from data since 9.0, so they're valid for neither recent nor overall data.

I can put together some tanking factor averages from tanks/stats data, but bear in mind that if tanking factor has anything to do with skill then the data will be somewhat biased.

 

[ModderKraken 59]

17 hours ago, SFC_Storm said:

I asked if you could fill in each graph for the top 25ish%.

Graph1:From AMX30b point to Vk2801 and all tanks above towards Vulcan Cannon

Graph2:From Conway down to Maus and everything above that.

Graph3: From Oho to ISU152 and all the tanks above

Graph4: Draw a line from Maus to Su152 and please label dots as those tanks.

Here you go. Still working on the eff HP thing.

 

I also ran something on DPM Yield (Eff DPM/Theoretical DPM).

This explains why better players can pad lights so well ... aim better, avoid incoming fire better - live longer, do more dmg - harvest wn8. Also helps answer the salt vs. diamond question for E5.

29 minutes ago, RichardNixon said:

It's more likely that he uses data from replays & battle results. The problem with the tanks/stats values is that they're only averaged from data since 9.0, so they're valid for neither recent nor overall data.

I can put together some tanking factor averages from tanks/stats data, but bear in mind that if tanking factor has anything to do with skill then the data will be somewhat biased.

 

That would be awesome, thanks!

[RichardNixon 835]

Ok, this should be semi-legit, barring skill adjustment:

https://docs.google.com/spreadsheets/d/1NnIHlSfbMn4o-ehvGZSVj3Rr_wb5b6A68BhswG0xvkA/edit?usp=sharing