War Thunder

Stats and the real value of the winrate chart.

warthunder 4 - Stats and the real value of the winrate chart.
Loading...

The Value of Data

Warning: Wall of text, lots of stats and numbers. TL;DR at bottom.

Most of you probably saw this post the other day.

The guy who made it didn't comment on it with any info about how he extrapolated that data, or what meant what, but curious nonetheless, I went on a crusade to find out more. I should point out that everything from here on in is centred around Ground RB.


The Creation of the Dataset

First, I created a god-awfully messy python scraper, which I repurposed from an old script I was using to parse some of the datamine files. I won't post it here but if you're interested PM me. It basically pulled all the Thunderskill monthly stats as current on 22/05/2020 (or the 21st for you guys living in the past). Then I imported that all into Excel, where I used a formula to weight the value of a winrate (basically, count for a vehicle divided by sum of all vehicles at a BR for a nation times winrate). This let me weed out the weird cases, like the Youtube Cup T-80U which has a 100% winrate over 29 games.

Finally, that left me with some useable numbers. So, first things first, here are the results. I've done what my best assumption is for what the last chart had done to it; averaged out any interim BR holes. I also avoided averaging a BR range, because that's inaccurate as we're looking at the data for BRs specifically (and don't want them weighted due to changes a full BR above them).

zaEM17H - Stats and the real value of the winrate chart.


Average Winrate by Nation & Battlerating

Then, to follow up, I grabbed the next best stat I had: kills per death. Same as before, weighted average and then charted it up:

mPP3us2 - Stats and the real value of the winrate chart.


Average KDR by Nation & Battlerating

What does the Data mean?

Cool. We've got two charts, laid out in the same way, taking two different numbers for the same battlerating range. But many of you will notice that there's a bit of a weird thing going on here – there's no rigid correlation between kills per death and winrate. Well, there are a couple reasons for that.

First, Thunderskill's dataset isn't all-encompassing. It's largely comprised of the players who are most interested in their stats, and the people they look up – within the last month. It's the best source we have, but Gaijin please, public APIs?

Secondly, they shouldn't be guaranteed to correlate. More games are won through control of caps than dominance (if any of you have ever tracked your stats, you'll notice that), so winrate is likely something that you could calculate given a team's likelihood of objective control and their kills per death.

So what can we take from these charts?

Well, it's not just about winrate. Looking at the 10.3 examples, the margin between winrates and the related KDR shows that despite having a nearly 20% lower winrate, the American teams actually have a 0.09 higher KDR than the Russians, while the British KDR is 0.37 higher. Meanwhile, the French with their average KDA over 1.2 higher than the Russians only have a 1.62% higher winrate. At least a portion of this rides on who plays with who. You could argue that the Germans carry the Italians and Russians… Except the Russians play with just about anyone (by my count, with the Germans about 58% of the time) so in theory their winrate should average out between the Axis and Allied teams.

Meanwhile, if you look at some battleratings, you'll see that there are these really weird numbers that appear to be out of nowhere. A good example is the British 6.0 – only a 29.57% winrate, but a 7.60 average KDR. This means something! It means the Black Price is really good, but way too slow to play ring around the rosie with cap points.

Загрузка...

The Results

Regardless, what we can tell from these numbers:

  • Thunderskill is not a perfect source, we end up with a lot of extremes.
  • Winrate is not directly correlated to a vehicle's or lineup's performance. Some teams just can't or won't play objectives.
  • Because of our limited data source, we should not assume that these numbers are necessarily representative, rather take them as an indication. It's pretty safe to say that overall, the Germans have a higher winrate, whether due to better vehicles or better players. More than likely, this effect would be less pronounced in a complete dataset.
  • Good players with rare premiums get some seriously scary average KDRs.
  • Winrate is influenced by how complete of a lineup you can create. This is why everything's even-ish at 6.7 – everyone can build a decent, complete lineup at that BR (except for Russia, the poor bastards).

Fun Facts

I pulled a shitload of data from thunderskill by ripping apart HTML responses (sorry, whoever owns the server). That gave me a few cool bits and pieces of info some might find interesting:

  • 100% Winrate:
    • AMR.35 ZT3 with 35 games.
    • H.35 with 8 games.
    • Ro-Go with 23 games.
    • Pz.38(t) F with 7 games.
    • T-80U Youtube Cup with 29 games.
  • Highest winrates with 1000+ games (all premium):
    • KV-1B – 76.28% with 7176 games.
    • Sd.Kfz.234/4 (Pak Puma) – 73.23% with 13981 games.
    • Brummbar – 71.78% with 1101 games.
    • B1 Ter – 71.62% wiith 13052 games.
    • Pz.Kpfw. Churchill – 70.61% with 6297 games.
  • There are 27 tanks which have been played with a 0% winrate.
  • 10 tanks have not been played at all.
  • The most played vehicle in the last month was the CV 90105 TML, closely trailed by the KPz-70, with 46760 and 44764 games respectively.
  • The Lvrbv 701 is the most KDR-efficient AA, with 1.58 air kills per death. The FlaRakPz 1 was the most KPB efficient, with 1.11 kills per battle. Maybe FlaRak drivers adventure into the battlefield more, I dunno.
  • The best AA for killing tanks by a huge margin was the Pvlvv fm/42, with 2.03 kills per death. For the curious, the OTOMATIC only had 0.72 and the ADATS had 0.25.
  • The most effective non-AA vehicle against planes was the Sd.Kfz. 140/1.

There's heaps more stuff in here, but this largely covers the interesting bits.


TL;DR

  • Data is beautiful, but consider the faults of your source. I'm not a statistician but pulling numbers without explaining how and why they exist is relatively valueless.
  • zaEM17H - Stats and the real value of the winrate chart.
    Average Winrate by Nation & Battlerating
  • mPP3us2 - Stats and the real value of the winrate chart.
    Average KDR by Nation & Battlerating
  • Germany definitely doesn't suffer, but realistically it doesn't appear that anyone else does, at least not to any significant margin, seeing as KDR stats are pretty damn consistent in most cases.
  • Play the objectives rather than hunting kills, and your winrate can be higher with the same KDR.
  • Some guy in a Black Price and another in a Chi-He are going ham on people.
  • I spent way too long playing with this.
  • I regret ever playing French reserve.
  • If you want the script I used to scrape this all or the excel spreadsheet, PM me.

That's all, wish everyone luck with their War Thunder games – and don't let the thunderskill stats get you down!

Source: Original link


Loading...
© Post "Stats and the real value of the winrate chart." for game War Thunder.


Top 10 Most Anticipated Video Games of 2020

2020 will have something to satisfy classic and modern gamers alike. To be eligible for the list, the game must be confirmed for 2020, or there should be good reason to expect its release in that year. Therefore, upcoming games with a mere announcement and no discernible release date will not be included.

Top 15 NEW Games of 2020 [FIRST HALF]

2020 has a ton to look forward to...in the video gaming world. Here are fifteen games we're looking forward to in the first half of 2020.

You Might Also Like

Leave a Reply

Your email address will not be published. Required fields are marked *