Jump to content
You need to play a total of 5 battles to post in this section.
Bonfor

Win Rate by Ship Type & Tier - what can we conclude?

45 comments in this topic

Recommended Posts

115
[SRPH]
[SRPH]
Members
451 posts
958 battles

In discussing the whole issue with matchmaking, I decided to take the stats provided by WoWS Stats & Numbers as of August 28, 2018 and create a chart to average the Win Rate stats for each ship type by tier.  I used a weighted average by first calculating the number of wins for each ship then totaling those wins and dividing by the total number of battles for all ships of that type and tier.  The Win Rate for All Ships in the tier was calculated in the same way to come up with the weighted average by tier.  I started with T5 because most of the forum discussions seemed to be centered around T5 and up matchmaking.

I'd like to hear your thoughts on what we can conclude from this, or if I need to alter my methodology.  Obviously, the percentages are all pretty close to 50%, as they should be when averaged over many games and players.  However, because of the large sample size, even the smallest variances could mean something.

My own observations:

1. BB's start off well in T5 and T6, but lose more often as they go up in tier.

2. T6 Cruisers seem to be an outlier for that ship type, as well as T9 Cruisers, but most T9 cruisers gain Repair Party which is significant in T7/T8 matches.

3. CV's, like BB's, start off well in lower tiers but lose more often as they go up in tier.

4. T8's overall Win Rates actually fare better than T10's (whoa there, that's heresy!).

5. T5's Win Rate doesn't look as bad as people make it out to be, pretty close to T6 and somewhat better than T7.

6. Looking at the All Ships Win Rate, it's more difficult to win the higher you go, although T9's seem to enjoy a slight uptick in Win Rate thanks to Cruisers & DDs.

7. DD's don't have below a 50% Win Rate in any of the tiers listed...DD's Rule!

8. Your own mileage may vary...

 

Win Rate by Tier - Weighted Average        
Ship Type Tier V Tier VI Tier VII Tier VIII Tier IX Tier X
BB 51.42% 51.50% 51.19% 50.76% 50.34% 49.87%
CA/CL 50.88% 51.38% 50.87% 50.54% 51.19% 49.72%
CV 50.72% 50.70% 49.98% 49.89% 49.46% 47.15%
DD 51.50% 51.02% 51.17% 50.77% 51.14% 50.27%
             
All Ships 51.21% 51.30% 50.99% 50.65% 50.78% 49.89%
             
Memo Only:            
All Wins 20,499,740 21,738,213 21,533,444 20,177,450 9,525,583 9,113,877
All Battles 40,027,329 42,371,777 42,231,421 39,834,628 18,760,038 18,267,866
Edited by Bonfor
  • Cool 1

Share this post


Link to post
Share on other sites
1,758
[NGA-B]
Alpha Tester
9,366 posts
3,646 battles

3. Manual drop becomes a thing at Tier 6. So does strafing IIRC. Both should be removed.

  • Cool 2
  • Bad 3

Share this post


Link to post
Share on other sites
16
[WOLFB]
Beta Testers
79 posts
7,728 battles

My understanding is that WoWS Stats and Numbers only gathers data from the players that have been searched on the website.  Newer players and players that simply don't care about their stats are unlikely to visit or know about the site, thus their data would not be represented.  This would skew the data considerably and result in a higher win rate for all your calculations.

Share this post


Link to post
Share on other sites
444
[BOTES]
Members
1,895 posts
6,607 battles

In case anyone was wondering about the 50%+ averages for an entire tier (you can't have an average winrate across all ships that is more than 50%), WoWS Numbers only tracks data for people with more than 20 battles in that ship.

  • Cool 1

Share this post


Link to post
Share on other sites
507
[WOLFD]
[WOLFD]
Beta Testers
4,950 posts
1,487 battles

Aso WWS doesn't let you filter by recent games so you've got data from times when the game looked and played significantly differently mixed in, which makes the information the next best thing to usless.

  • Cool 1

Share this post


Link to post
Share on other sites
844
[SSG]
Alpha Tester
3,376 posts
7,971 battles
5 hours ago, Bonfor said:

1. BB's start off well in T5 and T6, but lose more often as they go up in tier.

Coincides at tier 8 when pretty much every DD has stealth torp options, and decent torpedo's aside from IJN's over nerfed ones. As well as higher point captains usually meaning IFHE which devastates them, the 203 mm cruisers that can use AP against them are more common, and thn you get to a ship like Des that spits 203 mm rounds every what 3-5 seconds? Which doesn't need IFHE to punch through most armour it'd hit against the angled BB's.

5 hours ago, Bonfor said:

2. T6 Cruisers seem to be an outlier for that ship type, as well as T9 Cruisers, but most T9 cruisers gain Repair Party which is significant in T7/T8 matches.

Has old data from when the Cleveland was tier 6, Aoba (3x2 203's), Now has Pensacola with it's 10 203 mm guns, the RU flamethrowers, Nurn, Leander, La Gal now with MBRB which against the lower tier ships, these things can wreck, against the higher tiers, they have a fighting chance. While 7 see's some further improvement on these ships, there's a major gap between some of the tier 8 and 9 ships. And it seems like 7 get's bottom tiered a bit more often. Then yeah, 9 you get repair party, which is that much more powerful when your king of the hill.

5 hours ago, Bonfor said:

3. CV's, like BB's, start off well in lower tiers but lose more often as they go up in tier.

Tier 5-7 AA is actually under-powered. It was fine, at times OP, when we had Biplanes going up to tier 7 ships. Where as Tiers 9 and 10 saw some trouble against jets and all, although it's AA was otherwise good. But the rework on planes, that altered speed and in places HP, botched it on both ends because while tiers 5-7 AA became too light, especially against higher tiers, AA of the higher tier ships became way too high because instead of fixing manual drop (damage nerf, forced to drop at longer range, or out right removed), they buffed AA, repeatedly, until they made USN DB's near useless and as we cried for those to be buffed decided to buff our already OP fighters ammo reserve that until some time a little over a year ago caused the official shift of a USN dominated meta to an IJN one, with the USN group rework the final nail in the coffin at 7-9.  The drop at tier 7 is the tendency of having to see tier 8 ships (start seeing some of the AA monsters) that of the ships don't have the most OP AA, but still a tad high, but more importantly the tier 9's tat just have god levels of AA vs a tier 7 Ranger or a Kaga with tier 6 planes. They need to add later AA hulls to tier 5-7 ships, give tier 4 CV's a shield from seeing them, and raise the HP on planes tiered 5-10, heavier on the lower tiers, less on the high tiers so that way, we can try and get a more uniform line of AA and plane HP so a +2 CV isn't waltzing through AA to obliterate things, and -2 CV's aren't banging their heads on a wall trying to attack the only lone ship hat still has AA to butcher planes. Also maybe consider moving Independence to tier 5 and make Bogue a premium, move Ranger to tier 6, and put Yorktown at tier 7. Against tier 4-7, you can get away with 36 planes, just barely against the 7's. But it is no where near enough against tier 8's. 

5 hours ago, Bonfor said:

4. T8's overall Win Rates actually fare better than T10's (whoa there, that's heresy!).

Power gap between tier 6-8 is greater than that of 8-10. Bismarck's that can run down the tier 6 and 7 BB's with ease, as well as Richelieu's, Balti that tank's like a BB, list goes on. Add in also Lex can blow through tier 6 AA for the most part with Cleveland gone, with AP bombs. Tier 8 may have issues against Midway (what dolt thought 2,2,2 being brought back when it was broken the first time was a good idea), but it's AA is still fairly good, some of the tier 8's are actually faster than the tier 10's, and still have similar armour. 

Tier 8's have Speed, Armour, and Firepower over tier 6, where as Tier 10 for the most part has Firepower, a bit more armour depending on ship and line, but speed is still really equal. 

6 hours ago, Bonfor said:

5. T5's Win Rate doesn't look as bad as people make it out to be, pretty close to T6 and somewhat better than T7.

21 knot BB with 10 356 mm guns and about 280 mm of armour, vs a BB that's around 30 knots, 6 380 mm guns or 9 283 mm guns, torpedoes, a secondary battery with range and the 1/4 rule meaning it can punch through all that 19-25 mm plate, and has about 350 mm of armour protection. 16x 340 mm guns vs an Omaha. When they go against still very similar tier 6 ships or the lower tiers it's not a thing, but few of them, other than maybe DD's, can really hold their own against the tier 7 fast battleships, and even the two slower ones, Nagato and Colorado, have heavy guns. Myoko's 5x2 guns, Atlanta, Schors, Algiere, more DD's that get stealth fire and the unspeakable things a tier 7 CV can do to tier 5 ships. It's better from battles against tiers 3-6 and ships like the Kami clones that are still extremely dangerous in a tier 7 battle, and Minekaze used to have torps just as good with a 10 km range that back in the day, Fujin, Kami and Kami R were considered "inferior" because they had shorter range torps for if I remember right slightly better guns at the time. Not to mention Radar and Hydro increases. 

6 hours ago, Bonfor said:

7. DD's don't have below a 50% Win Rate in any of the tiers listed...DD's Rule

This again falls to the fact you are pulling overall numbers. This counts before IJN's DD's were overnerfed. Before open water stealth fire, something USN DD's madea ton of use of, was removed. Before smoke changes were introduced. Before Radar existed. Before it and hydro became extremely wide spread. Before ship secondaries actually became good on multiple lines. Kami clones, PA DD's, Fletcher, German DD's tier 6+, RU DD's tier 9+ in the hybrid line, all have as fast or faster torps than IJN, some with similar or better range, reload, etc, with better spotted ranges (unless DWT, usually 1.3/4 km vs IJN's 1.6/7), that can launch from stealth, while also having better guns, protection, speed and HP. And even though Ognevoi has slower torps, as well as some USN ships, they have a spotted range of 1.1 km at 55 knots, while IJN systems with a mere 4 knots more have at best 1.5 km spotting. And without getting into boring calculations and digging up the chart someone made, I'll simply say this. Farragut's torps go from 56 knots to 64, double that sped difference, and only go from 1.1 to 1.3 spotted range. Fletcher, that goes from 55 knots, to 66 knots, 10.5 km, only goes up to 1.4 km spotted, while IJN's with .5 km less range, 1 knot faster, reload 6 seconds slower and have 1 less torp per tube, has a spotted range of 1.7 km. And Yugumo is supposed to be the torpedo boat, not Fletcher. What DD's can act as ninja's and Khab have helped keep DD's afloat, but they've been hit pretty hard with the nerf bat, on CV's have been hit harder than them.

 

Pull more recent data, the pictures may look very different. 

  • Cool 5

Share this post


Link to post
Share on other sites
115
[SRPH]
[SRPH]
Members
451 posts
958 battles

Hmm...nuts...well thank you all for the input so far.  Is there a better site to cull more recent data if WoWS Stats doesn't allow for filtering out older data?

Share this post


Link to post
Share on other sites
406
[R-F]
Members
675 posts
6,180 battles

Warships.today is great when it's working.  Which is like 10% of the time.  It seems to be tracking individual players fine right now but not overall ship data.

Share this post


Link to post
Share on other sites
1,470
[-K-]
[-K-]
Members
3,038 posts
12,524 battles
10 hours ago, Bonfor said:

3. CV's, like BB's, start off well in lower tiers but lose more often as they go up in tier.

This is 100% due to noise and people with too few games played/not tracked by the system. Every CV game is mirrored, which means that every CV game generates exactly one win and one loss in CVs of the same tier.

  • Cool 1

Share this post


Link to post
Share on other sites
Members
525 posts
168 battles

- Win rate is the most skewed stat. The only thing you can conclude from win rate is that it is an accurate representation of nothing.

- Performance data is all that matters. Kills... Damage... Survival rate... hit rate... k/d....etc

Share this post


Link to post
Share on other sites
49
[POP]
Members
173 posts
5,362 battles

How do CVs have a 47% win rate at Tier X?

Isn't there one on each team?

 

Share this post


Link to post
Share on other sites
Members
1,525 posts
672 battles
4 minutes ago, Carniolan said:

How do CVs have a 47% win rate at Tier X?

Isn't there one on each team?

 

I believe they are tracking back to the dawn of (WoWs) time.  There were a lot more draws than there are now.

Share this post


Link to post
Share on other sites
Members
705 posts
11 hours ago, Bonfor said:

In discussing the whole issue with matchmaking, I decided to take the stats provided by WoWS Stats & Numbers as of August 28, 2018 and create a chart to average the Win Rate stats for each ship type by tier.  I used a weighted average by first calculating the number of wins for each ship then totaling those wins and dividing by the total number of battles for all ships of that type and tier.  The Win Rate for All Ships in the tier was calculated in the same way to come up with the weighted average by tier.  I started with T5 because most of the forum discussions seemed to be centered around T5 and up matchmaking.

I'd like to hear your thoughts on what we can conclude from this, or if I need to alter my methodology.  Obviously, the percentages are all pretty close to 50%, as they should be when averaged over many games and players.  However, because of the large sample size, even the smallest variances could mean something.

My own observations:

1. BB's start off well in T5 and T6, but lose more often as they go up in tier.

2. T6 Cruisers seem to be an outlier for that ship type, as well as T9 Cruisers, but most T9 cruisers gain Repair Party which is significant in T7/T8 matches.

3. CV's, like BB's, start off well in lower tiers but lose more often as they go up in tier.

4. T8's overall Win Rates actually fare better than T10's (whoa there, that's heresy!).

5. T5's Win Rate doesn't look as bad as people make it out to be, pretty close to T6 and somewhat better than T7.

6. Looking at the All Ships Win Rate, it's more difficult to win the higher you go, although T9's seem to enjoy a slight uptick in Win Rate thanks to Cruisers & DDs.

7. DD's don't have below a 50% Win Rate in any of the tiers listed...DD's Rule!

8. Your own mileage may vary...

 

Win Rate by Tier - Weighted Average        
Ship Type Tier V Tier VI Tier VII Tier VIII Tier IX Tier X
BB 51.42% 51.50% 51.19% 50.76% 50.34% 49.87%
CA/CL 50.88% 51.38% 50.87% 50.54% 51.19% 49.72%
CV 50.72% 50.70% 49.98% 49.89% 49.46% 47.15%
DD 51.50% 51.02% 51.17% 50.77% 51.14% 50.27%
             
All Ships 51.21% 51.30% 50.99% 50.65% 50.78% 49.89%
             
Memo Only:            
All Wins 20,499,740 21,738,213 21,533,444 20,177,450 9,525,583 9,113,877
All Battles 40,027,329 42,371,777 42,231,421 39,834,628 18,760,038 18,267,866

tl;dr

Old data is old.

  • Cool 2

Share this post


Link to post
Share on other sites
1,672
[OO7]
Members
2,229 posts
9,694 battles
11 hours ago, LastoftheFallen said:

My understanding is that WoWS Stats and Numbers only gathers data from the players that have been searched on the website.  Newer players and players that simply don't care about their stats are unlikely to visit or know about the site, thus their data would not be represented.  This would skew the data considerably and result in a higher win rate for all your calculations.

Are you sure?  Often the folks being looked up are looked up because they did or said something that caused a player to confirm they aren't the sharpest knife in the drawer.  And 9/10 times, that confirmation is made to be true.  So what evidence do you have that the data is skewed specifically northward?

Share this post


Link to post
Share on other sites
1,594
[-K-]
[-K-]
Beta Testers, In AlfaTesters
5,751 posts
6,716 battles
1 hour ago, Edgecase said:

This is 100% due to noise and people with too few games played/not tracked by the system. Every CV game is mirrored, which means that every CV game generates exactly one win and one loss in CVs of the same tier.

Some people hide their stats. That fact alone makes searching for nuances in percent point differences a meaningless exercise, in my opinion. 

Share this post


Link to post
Share on other sites
1,470
[-K-]
[-K-]
Members
3,038 posts
12,524 battles
2 minutes ago, vak_ said:

Some people hide their stats. That fact alone makes searching for nuances in percent point differences a meaningless exercise, in my opinion. 

Would be a fun side project to estimate the population proportion of people who hide their stats based on the deviation of the CV winrate from 50%. I once tried to do this based on the differences between a Sub_Octavian post of internal Conqueror stats and the publicly available stats at the time. The rate wasn't high, but it did appear to be higher on the RU server than the NA one.

Share this post


Link to post
Share on other sites
18
[USFTW]
Members
94 posts
1,893 battles

The fact that this data goes all the way back rather then just most recent versions makes it somewhat useless for any assumptions about the game currently

 

Share this post


Link to post
Share on other sites
7,214
[NMKJT]
Beta Testers
21,196 posts
3,871 battles
12 minutes ago, vak_ said:

Some people hide their stats. That fact alone makes searching for nuances in percent point differences a meaningless exercise, in my opinion. 

Enough samples negates this, and I'm getting tired of people using that as a reason to throw out stats they don't like.

 

What I do draw exception to is that this data is all-time, and therefore also pulling from outdated performances from over a year ago. That's a significant portion of the sample being gathered under greatly different conditions, and makes the result suspect.

  • Cool 2

Share this post


Link to post
Share on other sites
1,594
[-K-]
[-K-]
Beta Testers, In AlfaTesters
5,751 posts
6,716 battles
20 minutes ago, Edgecase said:

Would be a fun side project to estimate the population proportion of people who hide their stats based on the deviation of the CV winrate from 50%

Hm. Yes, this could be interesting. However, wouldn't this work only if we assume what's the win rate for people that hid their stats?

11 minutes ago, KiyoSenkan said:

Enough samples negates this, and I'm getting tired of people using that as a reason to throw out stats they don't like.

It doesnt negate it. Sampling stops being representative when we draw from a sub-population with different means than an overall population. I think that common sense allows us to sssume that most people with hidden stats have a below average win rate. 

  • Cool 1
  • Bad 1

Share this post


Link to post
Share on other sites
7,214
[NMKJT]
Beta Testers
21,196 posts
3,871 battles
2 minutes ago, vak_ said:

It doesnt negate it. Sampling stops being representative when we draw from a sub-population with different means than an overall population. I think that common sense allows us to sssume that most people with hidden stats have a below average win rate. 

That's an assumption with no factual backing. Don't do that when you're examining statistics.

 

What it is, is "noise". And enough samples negates any  impact that these missing data points could possibly contribute.

Edited by KiyoSenkan

Share this post


Link to post
Share on other sites
1,594
[-K-]
[-K-]
Beta Testers, In AlfaTesters
5,751 posts
6,716 battles
7 minutes ago, KiyoSenkan said:

That's an assumption with no factual backing. Don't do that when you're examining statistics.

 

What it is, is "noise". And enough samples negates any  impact that these missing data points could possibly contribute.

It's an assumption based on common sense, i.e. most people that go through the effort of hiding stats are ashamed of them. It's also based on my examination of people with hidden stats back when it just happened, and warships today kept their fairly up-to-dare historic win rates. An absolute majority of players I looked at were either red or orange.

No, it doesn't. If you're sampling a sub-population with a different mean than the true mean general population, more samples won't get you closer to the true mean. I'm surprised this has to be explained.

Share this post


Link to post
Share on other sites
1,470
[-K-]
[-K-]
Members
3,038 posts
12,524 battles
2 minutes ago, KiyoSenkan said:

That's an assumption with no factual backing. Don't do that when you're examining statistics.

Actually, there is evidence to support this. If you add up all the wins and losses (%wins * number of games) across all ships on the server, the wins and losses should be exactly the same. What you'll find is that they aren't, and as of the last time I did this (back when WT still worked right), wins outnumbered losses. That means that the "missing" data was disproportionately from below-50% sources. Some of that is probably the games-played cutoff (people do worse the first few times they play a ship because it's stock and they don't know how to play it yet). Some of it probably does represent a bias in who hides their stats.

2 minutes ago, KiyoSenkan said:

What it is, is "noise". And enough samples negates any  impact that these missing data points could possibly contribute.

There are different kinds of noise (really, error). You're only considering truly random (or at least normally distributed) error, which is asymptotically unbiased. Vak is arguing that the sample itself is biased upward relative to the full population, which would not be negated by large samples, as all the sampled points contain the bias.

Share this post


Link to post
Share on other sites
7,214
[NMKJT]
Beta Testers
21,196 posts
3,871 battles
6 minutes ago, Edgecase said:

Actually, there is evidence to support this. If you add up all the wins and losses (%wins * number of games) across all ships on the server, the wins and losses should be exactly the same. What you'll find is that they aren't, and as of the last time I did this (back when WT still worked right), wins outnumbered losses. That means that the "missing" data was disproportionately from below-50% sources. Some of that is probably the games-played cutoff (people do worse the first few times they play a ship because it's stock and they don't know how to play it yet). Some of it probably does represent a bias in who hides their stats.

There are different kinds of noise (really, error). You're only considering truly random (or at least normally distributed) error, which is asymptotically unbiased. Vak is arguing that the sample itself is biased upward relative to the full population, which would not be negated by large samples, as all the sampled points contain the bias.

And you don't believe that the small number of poor players who hide their stats isn't offset by the substantially larger population of poor players who does not? This is what I mean by noise. Yes, a percentage hide stats, but it is nowhere near a majority, and therefore the lack of these data points is not actually a huge influence.

 

Also forgive me but I don't think anything that could be called an "assumption" belongs anywhere in an unbiased analysis. Because personal bias will always influence assumptions. It's about as useful as feelings and that's not useful at all when you're examining hard numbers.

  • Cool 1

Share this post


Link to post
Share on other sites
1,594
[-K-]
[-K-]
Beta Testers, In AlfaTesters
5,751 posts
6,716 battles
7 minutes ago, KiyoSenkan said:

And you don't believe that the small number of poor players who hide their stats isn't offset by the substantially larger population of poor players who does not? This is what I mean by noise. Yes, a percentage hide stats, but it is nowhere near a majority, and therefore the lack of these data points is not actually a huge influence.

Sigh. Okay, analogy time.

You have a bag with twenty marbles. Ten have a 0 written on them, ten have a 1 written on them. Let's "sample" that bag a thousand times by taking out a marble, recording the number, and putting the marble back to restore the initial state (we're assuming that marble drawing is completely random). Calculate and record the mean.

Now let's take out all the marbles except for one marble that's marked with a 0, and put them in a different bag. Sample that bag a thousand times. Calculate and record the mean. 

Do you think there will be a statistical difference (say, 99% CI) between these two means? Explain your answer. 

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.

×