Jump to content


Blitz statistics (Since update v5.4)

Statistics

  • Please log in to reply
48 replies to this topic

jylpah #1 Posted 03 February 2020 - 06:41 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

Hi all,

 

I have generated the first set of statistics based on my stats database. It covers Blitz updates since v5.4. At this point, the stats cover player WR distribution, and battle counts, but also how the players have on average improved (yes, mostly) over the time. As surprising as it may sound, it seems the below average players improve consistently in terms of their WR.

 

Enjoy: https://jylpah.github.io/blitz-analysiz/

 

Here are few examples:

 

The graph below tells how in average players have performed during the update vs. their Career WR before the update. This one compares WR at a tier to career WR at the tier. The color grading/contour means the number of players in that particular area.

 

 

Plus more. Comments?


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


gallabru #2 Posted 03 February 2020 - 07:34 AM

    Senior Sergeant

  • Player
  • 17980 battles
  • 641
  • [LAGER]
  • Member since:
    09-01-2016

Very nice !

Looking at your first chart, it looks like pruning and nerfing the lower tiers did not eliminate sealclubbing. The proportion of 60+% wr players is large there. Probably via reroll, but still. Ok, maybe it was worse before.

On the second, it is interesting to notice that the average is above 50%.

 

Looking at the distribution of number players, we can imagine how the 5k limit was set. 


Edited by gallabru, 03 February 2020 - 08:10 AM.

My team stats in real time  : http://forum.wotblit...ats-repository/

Carlsberg clan family is recruiting. Contact us.


gallabru #3 Posted 03 February 2020 - 07:40 AM

    Senior Sergeant

  • Player
  • 17980 battles
  • 641
  • [LAGER]
  • Member since:
    09-01-2016
Second comment on the first chart, above tier 5, the average WR of the players slightly increases with tier. But not that much and there is nearly the same amount of bad players.

Wow, you had to find a colour for the sub 30%wr ?

My team stats in real time  : http://forum.wotblit...ats-repository/

Carlsberg clan family is recruiting. Contact us.


jylpah #4 Posted 03 February 2020 - 08:22 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

Yes, it is very clear the average WR of the set is above 50%. There are multiple possible explanations for this:

  1. Selection bias. Some of the graphs filter players according to numbers per tier (min 100 battles per update per tier). This could cause significant bias, but on the other hand provides more accurate assessment of players’ performance. 
  2. Selection bias in the database. I collect account_ids mainly from WoTinspector replays. This causes bias in two ways: 1) replay posters are likely better than average players (1/14th of the players) and 2) new players are added to the database only when found in replays: these missing players can cause bias. 
  3. Bots. These are likely to cause rise of average WR
  4. Issue with the current calculations. Due to lack of RAM (16GB), I cannot join the data in a way that includes player/tank records found in update’s stats, but not in the career stats. Money will fix this soon once my RAM upgrade arrives. I doubt this causes bias, but it causes delays of getting data of new tanks.

 

btw. I would appreciate ideas for analysis topics. I will be adding stuff. These are on my task list:

  • Tank specific performance graphs (e.g. average damage + kills vs WR)
  • Best/worst tank per tier

BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


Just_merpug #5 Posted 03 February 2020 - 09:19 AM

    First Sergeant

  • Player
  • 52324 battles
  • 9,526
  • [AFK]
  • Member since:
    03-05-2013
On point 2: more replays needed, then. Bulk uploading them would be nice. Is there something that I could use on my IPad to just haul them out of whatever folder the thing hides them in and sling them at wotinspector.com? Even then, the games I rage quit because the other "players" are having trouble finding their backsides with both hands and a map would be excluded.

Denoobing member of the Active Frontline Klan - and proud of it. Be nice to me..https://www.blitzstars.com/player/eu/Chairman_merpug


gallabru #6 Posted 03 February 2020 - 05:42 PM

    Senior Sergeant

  • Player
  • 17980 battles
  • 641
  • [LAGER]
  • Member since:
    09-01-2016

(slightly childish presentation of the) Evolution of the cumulative battle count with updates (5.4-6.4). Your data.

Interesting evolution...

Posted Image

Would look like player retention is good but having new players in, not so much. Well, it depends on the total player count obviously.


Edited by gallabru, 03 February 2020 - 05:45 PM.

My team stats in real time  : http://forum.wotblit...ats-repository/

Carlsberg clan family is recruiting. Contact us.


stubbo66 #7 Posted 03 February 2020 - 06:35 PM

    First Sergeant

  • Player
  • 74769 battles
  • 1,712
  • [49DIV]
  • Member since:
    06-29-2014

I don't put much stock in career win rates, what would be much better would be to take 30, 60 or 90 day winrates and see what they look like.

 

You could start of with a career WR vs 90 day WR (and/or 30 day WR).

Then have copies of the charts above with the career WR and the periodic WR so we can get a better idea of how accurate career WR is of players in battle as I think it lies!

 

As an example, my career WR is 61.1% but I play at around 67%.....that must be true of a lot of players in the game, and as career WR changes so slowly the more battles you have it isn't very indicative of the players real stats.

 

I wish WG would have both WR in game, we know they have 100 day WR for their charts on the portal, so why not make use of that data in game as well.



jylpah #8 Posted 03 February 2020 - 10:17 PM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

View Poststubbo66, on 03 February 2020 - 08:35 PM, said:

I don't put much stock in career win rates, what would be much better would be to take 30, 60 or 90 day winrates and see what they look like.

 

You could start of with a career WR vs 90 day WR (and/or 30 day WR).

Then have copies of the charts above with the career WR and the periodic WR so we can get a better idea of how accurate career WR is of players in battle as I think it lies!

 

As an example, my career WR is 61.1% but I play at around 67%.....that must be true of a lot of players in the game, and as career WR changes so slowly the more battles you have it isn't very indicative of the players real stats.

 

I wish WG would have both WR in game, we know they have 100 day WR for their charts on the portal, so why not make use of that data in game as well.

 

Yeah, such an analysis is definitely something interesting. I am still struggling how to automate the site generation. I have managed to script out the analysis page creation, but need next learn how to programmatically build a site with proper cross-linking and run such a "trends between the updates" analyses.

 

At the moment I do not enough RAM to ran the analysis, but maybe I get the upgrade DIMMs (=> 32GB) in week or two.


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


Mjr_Eazy #9 Posted 03 February 2020 - 11:11 PM

    First Sergeant

  • Player
  • 31860 battles
  • 7,275
  • [BT_SV]
  • Member since:
    04-26-2017

You should ask the chaps at WG, as according to Canny & Strummer et al, WG run statical analysis on every shot in every battle to work out which players need their WR buffed or nerfed ;)

 

Interesting stuff Jylpah, as you say it would make sense there’s a bias to wins in the replay data as people more likely to upload wins/aces/good battles...

 

@chairman, afaik there’s no way to do batch of replays on iOS other than selecting them all in game, tapping share button and you can send multiple to wotinspector.com in one email but never tried 100 at a time, I send em to dropbox then load from my mac via Jylpah’s awesome script loveliness! 


547146227.png

 


jonty_2014 #10 Posted 03 February 2020 - 11:58 PM

    First Sergeant

  • Player
  • 72078 battles
  • 5,144
  • [EMRS]
  • Member since:
    09-23-2014

View Postjylpah, on 03 February 2020 - 08:22 AM, said:

Yes, it is very clear the average WR of the set is above 50%. There are multiple possible explanations for this:

  1. Selection bias. Some of the graphs filter players according to numbers per tier (min 100 battles per update per tier). This could cause significant bias, but on the other hand provides more accurate assessment of players’ performance. 
  2. Selection bias in the database. I collect account_ids mainly from WoTinspector replays. This causes bias in two ways: 1) replay posters are likely better than average players (1/14th of the players) and 2) new players are added to the database only when found in replays: these missing players can cause bias. 
  3. Bots. These are likely to cause rise of average WR
  4. Issue with the current calculations. Due to lack of RAM (16GB), I cannot join the data in a way that includes player/tank records found in update’s stats, but not in the career stats. Money will fix this soon once my RAM upgrade arrives. I doubt this causes bias, but it causes delays of getting data of new tanks.

 

btw. I would appreciate ideas for analysis topics. I will be adding stuff. These are on my task list:

  • Tank specific performance graphs (e.g. average damage + kills vs WR)
  • Best/worst tank per tier


 

something on hit rate and calibre perhaps.  The outliers will be the interesting ones ....


Empros is recruiting! Message me in game to apply.


Just_merpug #11 Posted 04 February 2020 - 12:09 AM

    First Sergeant

  • Player
  • 52324 battles
  • 9,526
  • [AFK]
  • Member since:
    03-05-2013

View PostMjr_Eazy, on 03 February 2020 - 11:11 PM, said:

You should ask the chaps at WG, as according to Canny & Strummer et al, WG run statical analysis on every shot in every battle to work out which players need their WR buffed or nerfed ;)

 

Interesting stuff Jylpah, as you say it would make sense there’s a bias to wins in the replay data as people more likely to upload wins/aces/good battles...

 

@chairman, afaik there’s no way to do batch of replays on iOS other than selecting them all in game, tapping share button and you can send multiple to wotinspector.com in one email but never tried 100 at a time, I send em to dropbox then load from my mac via Jylpah’s awesome script loveliness! 

I'm obviously being stupid, but I can't for the life of me see how to select more than one at a time. 

IPadOS is horrible in its lack of transparency. Were it anything else, I'd be able to find the filepath to the replays folder... 


Denoobing member of the Active Frontline Klan - and proud of it. Be nice to me..https://www.blitzstars.com/player/eu/Chairman_merpug


Mjr_Eazy #12 Posted 04 February 2020 - 12:30 AM

    First Sergeant

  • Player
  • 31860 battles
  • 7,275
  • [BT_SV]
  • Member since:
    04-26-2017

View PostChairman_merpug, on 04 February 2020 - 12:09 AM, said:

I'm obviously being stupid, but I can't for the life of me see how to select more than one at a time. 

IPadOS is horrible in its lack of transparency. Were it anything else, I'd be able to find the filepath to the replays folder... 

it’s not you, as you say on any other device you could browse and find the files, not iOS as they’re considered part of the app and it’s not a file based system, anyhoo, just tap the die in the top right corner of the replay (or shield if ratings) on the replays screen...

All the selected ones get a different border...

Then tap the share button bottom left - up arrow icon and if you do mail for example it will create an email with all the selected replays attached and they can be sent to replays@wotinspector.com in one email...

HTH.


Edited by Mjr_Eazy, 04 February 2020 - 12:31 AM.

547146227.png

 


corsos_buddy #13 Posted 04 February 2020 - 02:14 AM

    Junior Sergeant

  • Player
  • 7494 battles
  • 176
  • Member since:
    08-22-2019
Interesting is, that the average is above 50%, for when i play at night, i must be lucky to have one player above 45% in my teams

jylpah #14 Posted 04 February 2020 - 06:21 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

View Postcorsos_buddy, on 04 February 2020 - 04:14 AM, said:

Interesting is, that the average is above 50%, for when i play at night, i must be lucky to have one player above 45% in my teams

 

Yes, in the data set the average WR (total wins/ total battles) is 52.4%. This tells that there is selection bias and not all the players are being captured. I think this is due to churn (lot of people play just little time and quit before ending up the data base) and bots (no stats given). And I am really surprised of the finding and need to double-check is there some selection bias. I was expecting a different result.


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


jylpah #15 Posted 04 February 2020 - 06:21 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

Minor update. The WR change vs Career VR graph has an average line now

 


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


gallabru #16 Posted 04 February 2020 - 08:15 AM

    Senior Sergeant

  • Player
  • 17980 battles
  • 641
  • [LAGER]
  • Member since:
    09-01-2016

The selection bias is not easy to understand, though. If a green was happy enough to upload a replay, it is likely the reds would not have.

Could you extract data from premium tanks only ? I do not know how present that, but I am wondering whether there is a correlation between the number of battles and the proportion of premium tanks played. The obvious one is that with more battles you have more techtree tanks. But is there more that would justify a sub5k battle count protection ?

Did the tier 1-5 nerf promote buying premium tanks ? 


Edited by gallabru, 04 February 2020 - 08:16 AM.

My team stats in real time  : http://forum.wotblit...ats-repository/

Carlsberg clan family is recruiting. Contact us.


jylpah #17 Posted 04 February 2020 - 10:06 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

View Postgallabru, on 04 February 2020 - 10:15 AM, said:

The selection bias is not easy to understand, though. If a green was happy enough to upload a replay, it is likely the reds would not have.

Could you extract data from premium tanks only ? I do not know how present that, but I am wondering whether there is a correlation between the number of battles and the proportion of premium tanks played. The obvious one is that with more battles you have more techtree tanks. But is there more that would justify a sub5k battle count protection ?

Did the tier 1-5 nerf promote buying premium tanks ? 

 

Hi, let me clarify how the data fetching works as this might be the source of the bias:

 

1) I periodically run scripts to fetch replays uploaded to wotinspector.com and I parse account_ids from those to the DB

1b) Sometimes I also also fetch account_ids from BlitzStars.com via custom API awesome SockRobber has built for me

2) At the night of Blitz update, I run another script to fetch battle stats via WG API (https://developers.wargaming.net/reference/all/wotb/tanks/stats/)

3) After this I export stats from the DB to R - one set of cumulative stats (latest stats per account_id/tank_id ) and one set of periodic stats (stats per account_id/tank_id of battles during the update)

 

So the selection bias happens in both 1) and 1b):

1) WoTInspector/replays:

  • Posters of replays are likely more advanced players. They represent max 1/14 of the account_ids fetched but I assume (have not checked the data), the replay posters post multiple replays therefore further reducing the bias here. I do not think this is the source of bias
  • I suspect, posters of replays could be biased, and over-sample games played on certain times (e.g. after work in the evenings when all grumpy baby boomers and gen X enter the game), thus causing selection bias.
  • I suspect there are plenty of players who churn out relatively quick of the game and they won't be captured in any of the replays. Therefore the players who stick with the game can have above 50% WR since the churners take the losses

1b) Blitzstars.com

  • I suspect only more active players end up to Blitzstars, since some one has to check their stats first before BlitzStars start to record their stats .

 

And then there is the possibility of bots who are not shown in stats.


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


jylpah #18 Posted 04 February 2020 - 10:44 AM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014

Continuing... Gallabru, you listed lot of interesting analysis topics. Sub 5k players are definitely one. Also share / ownership of premium tanks is an interesting one. Especially the question how the playerbase is divided based on premium tank use/ownership and whether there has been any change over the time.


Unfortunately I started  systematic data acquisition only on update 5.4. I have collected data before that occasionally, but haven’t merged that data with the database yet. Also Face, the dev behind now-closed wotbstars sent me his stats DB backup so once I have time I should be able to merge that data with my main DB. BIG thanks to him!! 
 

Btw. All the R code is available at GitHub: https://github.com/Jylpah/blitz-analysiz


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


Mjr_Eazy #19 Posted 04 February 2020 - 12:05 PM

    First Sergeant

  • Player
  • 31860 battles
  • 7,275
  • [BT_SV]
  • Member since:
    04-26-2017

Jylpah, we are not worthy!  Awesome work!:medal:

 

Your explanation of how the data fetching could be biased makes sense.  Does the WG API data break it down by player?  I was thinking maybe you could calculate how many players have stopped playing with less than 100 battles between updates as that would give an idea of churn rate.  Or even do it for new players, ie account id in release x not in x-1 or x+1...

 

Who knew? Stats are interesting and fun :)

I’m way behind you tho, only got an ‘O’ Level in Maths (High School 16 year old level) and a partially read stats for dummies book so I appreciate you and Gallabru and t’other true mathematicians / statisticians putting up with my noob comments and thoughts... 


Edited by Mjr_Eazy, 04 February 2020 - 12:09 PM.

547146227.png

 


jylpah #20 Posted 04 February 2020 - 01:01 PM

    First Sergeant

  • Player
  • 22275 battles
  • 5,124
  • [AFK]
  • Member since:
    08-05-2014
Yes, this churn rate is something I will work in near future, but I need first more RAM and would like to fetch old stats from Blitzstars before those get pruned out forever. 

BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  






Also tagged with Statistics

1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users