Just a short(ish) one this time… please see Data Analysis #3 for a more detailed conversation, that goes on and on in the comments, on what Statshark has meant for players getting a better sense of how this game works than they had before.
This is just about the fact that K/D (Kills over Deaths) is actually a pretty simple and unrevealing stat. While it’s simple to calculate, and Statshark is allowing us to see our own K/Ds and those of vehicles general better than before, you can’t tell by just looking at the one number, how much of the value of it is in the numerator (kills) or denominator (deaths). Is a vehicle tanky, but not able to kill much (low denominator), or a “glass cannon” that kills a lot and dies a lot? (high denominator).
This is a question that has been asked before about real combat. The field of operational research is all about simplifying all the conditions of individual battles in order to ask larger strategic questions. A key value for operational researchers is kills per unit of time ( normally expressed as α for the red force, β for the blue force). Can we get a similar value from Statshark values and what would it tell us?
For this purpose we’ll just look at a small set of vehicles currently the focus of some discussion, the battleships introduced in the Leviathans update. We take the Statshark data from the last six days of June, and add a couple new derived values:
So the first five columns are just straight Statshark, for RB above and AB below. Number of spawns, naval kills (NK), air kills (AK), total kills (TK) and deaths (D). I’ve thrown in the two new premiums, Gneisenau and Sevastopol, even though they’re not “top tier”… it doesn’t really work to put in other ships as BR decompression on June 25 will likely have significant effects on this data, so we can really just look at the new ships.
The next three columns are what’s of interest, and are color-coded for ease of reference (blue is good, red is bad). The first standard K/D is as shown on Statshark. You can then decompose this, however, into the next two columns… first measuring the chance a vehicle has of surviving a spawn (1-deaths/spawns), giving you their survivability/tankiness value, and then, multiplying that to give how many kills they would have gotten on average if they didn’t die in that spawn (regular TK divided by survivability). A low number here means they wouldn’t have killed very much if they’d just been left alone the whole game, basically.
What this shows in the case of Battleships is that Sovetsky Soyuz is both a defensive and an offensive problem. It has both high survivability and high destruction, which is why it’s dominant just now. This means fixing it will likely require both offensive and defensive nerfing.
Here’s a scatterplot of the same data.
A couple other conclusions one could draw from this:
-
The Yamato’s problem currently is basically survivability; in AB and RB it’s killing when left alone just fine.
-
The Vanguard is surprisingly tanky in early Leviathans play, which is elevating its stats above Bismarck or Richelieu.
-
The Roma is problematic both in terms of survivability and damage-dealing. Even if you left it alone it wouldn’t kill very much currently.
-
Also interesting is that AB’s overall curve is steeper than RB. In operational research this would be reflect the greater “intensity” of AB battleship play over RB, in that more things are dying faster.
Of course these are all battleships, so they’re all relatively tanky and clustered on the right side of the scatterplot. PT boats and the like with low survivability could be clustered on the other side.
It’s important to not this is very much dependent on the vehicles the vehicle you’re measuring is currently facing. A BR shift up will likely move any vehicle both to the right and up somewhat on the scatterplot. As such this could suggest a way to look at a vehicle’s poor performance and determine if it could be solved with a nerf/buff (producting movement on the scatterplot in only one axis, greater or lesser survivability or damage) or you need a BR change (which would move you diagonally up or down).
The value of kills per spawn if left alone, the last column in the chart, equates to kills in a fixed unit of time, which is a key value for operational research calculations (α or β). Lanchester’s Square Law, for instance, states that in modern warfare, the comparison of two values here in terms of who will win a fight is as to the square. So if the square law held for War Thunder, in an even fight a Soyuz (11^2 or 121) would have a 50-50 chance against two Iowas (8^2 is 64, times 2 = 128) in RB (or 2v4, 3v6, etc.). This doesn’t mean that’s a true statement for War Thunder, but what you’d do for real wartime operational research is use real-world performance to “tune that in” to come up with an exponent less than 2 that gave good predictions… in some research they raise α/β to an exponent of 1.5, or even 1 to get good predictability out of Lanchester equations.
The next step past Lanchester is so-called “salvo” combat models, which incorporate a value for defensive toughness, and is seen as more appropriate for punctuated battles with distinct you-go/I-go phasing, where as Lanchester is more for warfare with “continuous” damage. The survivability factor here could give us a method to allow us to start introducing salvo-based analysis to War Thunder as well. But that gets us deep into the weeds. It’s not so much that operational research could help us understand this game better, really (that would take a much deeper dive to confirm) but Statshark data and a little math does allow us to know some of the variables it would be looking for. Which is kinda cool, I think anyway.
Also of note, Statshark has allowed the addition of custom columns for your personal records (not collective records yet). So if you ever wanted to determine what vehicles are your tanks, and what are your glass cannons, consider adding a deaths/spawn value there yourself. For me it produced some interesting, if perhaps unsurprising conclusions (my strike aircraft and coastal small boats die, A LOT… but on the other hand I’ve clearly been doing something right with keeping my StuG III G alive, which has not been a go-to TD for me before this, so there you go).
Previously: Data Analysis #3: The arrival of Statshark answers some old questions
