Data Analysis: what is the actual average player's score per mode? (and how long will it take you to do this event)

Game Discussion General Discussion
59

To put it a better way, Ill use the Israel BR 10.0 graph having “7”

If we assume that is player count right, how did the youtube seperate the values.

Because lets say those 7 accounts have the Mk2D and a Nesher in there lineup, so on paper its 10.0 but has a 9.7 Tank, did he exclude that metric from 9.7 if it got downtiered? or does he still place that player count metric in the 10.0 cell.

Now do you see why this graph more than likely cannot be player count and if it was there are bound to multiple entries in multiple BRs that will surely intersect one another based on the BR of that lineup if its up or down tiered

60

Gaijins own peak concurrent player count has never even gone above 260k :D so you claiming 260k factually isnt possible at this point.

What does any of this have to do with peak player count? It’s counted over several days as well.

this youtuber who I might add NEVER claimed it was player count, he never made the claim or how hes filtered out the figures.

From the replays he clearly has all the information about the players, their score, their nation, their vehicles being used, so it’s not just battles.

image
image1204×184 59.8 KB

it 100,000 random replays from random accounts or based on the top 100,000 players or etc. or did he pull multiple replays from multiple accounts at once and cross reference the data.

From all the replays from several days as they come by.

Because lets say those 7 accounts have the Mk2D and a Nesher in there lineup, so on paper its 10.0 but has a 9.7 Tank, did he exclude that metric from 9.7 if it got downtiered? or does he still place that player count metric in the 10.0 cell.

There’s always going to issues with determining up and downtiers because it is determined by the highest BR vehicles and people can have a top BR vehicle in a lineup but not use it, and it can end up being listed as an uptier.

Either way with a sample size this large I have no concerns about anomalies, which is exactly what those ‘7’ could be.

61

Care to put a bet on that.

Spoiler

https://www.reddit.com/r/Warthunder/comments/197zkly/analyzing_unofficial_server_statistics_uptiers/ this is where the image you posted came from.

Spoiler

https://www.reddit.com/r/Warthunder/comments/197zkly/comment/ki3wfxu/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

this is the github code he used to compile the graphs and if you bothered to do some digging in that code you would find the mega link to the .csv data that was used to compile it.

Spoiler

WT_K2_data_analysis/data_k2.ipynb at main · untitled1048576/WT_K2_data_analysis · GitHub

Interestingly it has over 1.2 million replay entries/battle entries and ironically 3.4 million player entries which seems to fall right in line with what I said about if its 100,000 replays with 32 player match lobbies then the total player data set is about 3.2 million which means your bound to have intersecting player names within BR ranges.

So if you think this data set is accurate when I think ive done the bare minimum to prove that its bound to have multiple same entries within intersecting BRs then im done explaining how your claim couldnt be anymore inaccurate based on this set of results because unless the user filtered out duplicate entries which going of the code used I dont see anyway or how he/she has done this then your claim literally is unconfirmable

I dont doubt the stats the user has compiled, but unless his code is making use of code to exclude “same accounts” when it comes to specific BR entries then its simply not possible to confirm the player count let alone confirm its player count or battles

62

It’s interesting that according to my calculations and observations, the average score (from the last month) you see in your profile is very consistent with the time needed to complete the event. So Bruce’s calculations actually make a lot of sense.

I usually spade vehicles or play strange vehicles that barely anyone uses and my average score in Ground Arcade is only 1162 from the last month (1257 overall).

During this event I decided to not change my plans and just play Strv-103-0, which is a terrible vehicle if you need good average scores.

The first star took me 41 battles and almost 7 hours. This is consistent with my observations about one Ground Arcade battle takes about 10 minutes at average. You can also do easy math here and considering I played at rank 5 (multiplier 1.0) and Ground Arcade (0.93), my average score per battle was 1180. Which is actually very consistent with my regular average score. If you used the method Bruce used, you could get these numbers from pure calculations, and that’s pretty impressive actually!

The problem with this method is, it works very well only if the player stay in battles from the beginning until the end.

The second star I grinded differently. Because I wanted to compare leaving battles early to staying in battles, I again played Strv-103-0, but I played more aggressively, leave the battle after first death and it took me 63 battles, but less than 6 hours. So my average score per battle was only 768, but doing this I “manipulated” the time value.

I know this all sounds weird, but it’s very consistent with my previous observations. If you think about this, when the battle starts, you know exactly where the enemies will be, so you can just go straight to some good position and get a few quick kills (whole enemy team is alive, there will basically always be some targets for you). But when you die, you respawn at the spawn zone and the situation on the map is completely different. Your team is occupying the best positions already, there are less enemies and you have to play much more defensively, because you don’t know which positions enemies took on the map.

It’s not really a question if it’s better (from the score perspective) to leave battles early, because it’s obvious it is better. The real question is: is it better to leave battle early and wait in queues and watch loading screens more often or stay until the end of every battle? This is what I wanted to test, and according to my test, score-wise it’s still much better to leave battles after 1 death (even if you consider loading screens and queue times).

If someone still don’t understand my way of thinking. During the first star grind, I had the battle that shows exactly why it’s better to be aggressive at the beginning and then just leave that battle:

Spoiler

Battle example

In this battle, I started very aggressively, and after 3 minutes and 40 seconds I already had 2252 score! But it was my first star, so I respawned. With my second life, after additional 5 minutes and 52 seconds I only gained 1038 score. Scoring wise, it was totally not worth respawning.

I did similar test in the past at rank 1, and the results were the same. You could gain more score by playing aggressively at the beginning and then changing the battle.

It could work slightly differently in Ground Realistic. I don’t play there, but I assume if you are a good CAS player, it would make most sense to spawn one tank, play aggressively, die, spawn a plane or a helicopter, use all bombs/missiles, die, change the battle. I’m pretty sure this would give you the best score per minute.

PS: I’m not saying everyone should become one-death-leaver during events. I’m only saying it’s definitely a quicker way to complete the event, especially if you know maps well.

BTW: Don’t use Strv-103-0 to grind the event, it’s a terrible vehicle. Maybe it’s pretty durable, but you lose a lot of time repairing. It’s a much better idea to just use META vehicles.

63

Well if you want, you can log in excel time when you start battle, time when you leave battle and the score you had in the moment of leaving

Then you can send that to me, I will put it into excel table i made and showed in the other thread, and we will have something to compare.

Just a suggestion.

64

If you count the time from pressing “To battle!” to seeing a battle summary screen, then it can be misleading. And you don’t really need to do this.

Imagine different players having a different computers at different locations. For example, I have a fast computer and I play on EU/RU servers, which is pretty much the best case scenario. It almost always takes me below 10 seconds to find a battle during events. When someone from e.g. Africa wants to grind the event, he could have a completely different experience with waiting times in the queue.

The important statistics is the time you played in the battle. Loading screens and queue waiting times can be determined separately and will be personal. For example, my queue times take usually below 10 seconds, my map loading screens take 20 seconds, my battle summary screens take 5 seconds to appear. You also have to include 25 seconds waiting time before the battle starts. So in total I can count my time “between battles” as 1 minute. And this time is actually very consistent for me. For other players this time can be different.

But if you want do the proper math, you don’t want to include players computers performance or players locations (that affects waiting times). From the math point of view, this adds more variables that are misleading, if you want to figure out the time that is required to grind the event using different methods. Just base your calculations on the time played, the score and (when needed) add some specific “time between battles” to your calculations, which could be something like 1.5 minutes. This is the only proper way to have a comparable results between different players.

65

Not pressing to battle.

The moment game starts loading into the game.

Queue times, stretching just talking with friends and bathroom breaks, those i count under “time between battles” and i count them as time between game loading back into menu and loading into next game.

Why not? It shows the average time between battles, that can be used with average match length to make estimates as to how many games you can play in a hour, or how long does it take to grind 22.5k daily/45k total.

Average time between battles i recorded was 2mins30 seconds. Total extra time spent doing things like stretching, toilet etc. was 41 minutes. That time needs to be added to time you actually spend playing.

66

Ok, so basically your “round start” and “round end” is: the time played + loading screens time + battle start waiting time (it’s 25 sec if your map loading screen is very short, less if it’s longer). Why not just look at the time played or battle time and ignore loading screens? It’s much simpler this way, you only note one value.

I’m just saying you lose a lot of accuracy this way. All personal things like breaks, waiting times etc. should be excluded from calculations. Why? Because they will be different for every single person. More than that, they will be different even for you every time you complete a star! So these additional factors are not even consistent for one player.

Of course this can be showed more like an interesting fact that is very personal and not important from the calculations point of view. But it definitely shouldn’t be compared between different players.

Let’s say someone will add his data to your topic. He will write his “round start” and “round end” and you will do calculations based on these values. But how that person would understand these values? The “round start” can be understood as pressing “To battle!” button, first loading screen, the moment battle loaded, first vehicle spawn and whatever someone understands as “round start”. The same goes for the “round end”. I actually understood your “round start” as the first spawn of the vehicle and your “round end” as the moment the round ends and you see your total score for the battle. So I thought it’s the mission duration - 25 seconds (or the time played if you didn’t die in the battle).

67

I would arrive at the (almost) same result, but I might use those values in a way i have yet not thought of. Better to have that data than not i guess.

Solution to this wouldnt be to exclude them at all but rather increase sample size from one person (me) to more players to get the average value.

Moreso, I know which server i play on, and what rank/br/nation, and i dont switch lineups/countries when im recording.

so I can calculate the average dependant on those conditions (ie. whats the average X for 5.7 for example).

This can be rectified by simple explanation i will add later to the original topic, and moreso can be explained to person who would volunteer to add his data.

68

I may just be understanding this wrong. You are saying it takes six hours of play to get a rank 7 in the tech tree? And that this is the same time it takes in events? So to get an event vehicle, it is supposed to take about 6 hours?

69

It’s better for the reasons you just said, and especially cause queue times are basically non existent in Ground AB .
I play very agressively at the start and get a lot of score very fast in most of my matches, I also play to survive the entire match but there are matches where I die. What I usually do is either I leave to get another match fast if the batlle is already lost or I respawn a SPAA cause that will give you a lot of targets faster than driving a tank from spawn to the middle of the map. I also use the fighters to kill the enemy planes as the air strike begins before I jump back to the SPAA to finish the rest, if I just sat on the SPAA I would risk other players using the fighters getting them first and end with no targets to kill. If you have a good SPAA on your lineup, just try it. ;)
Also from experience, it doesn’t matter to keep getting kills cause they seem to give you less score the more you get, so instead of going just for ground kills try to diversify what gives you the score, like get a cap, kill a plane, get an assist. My biggest scoring matches are usually not the ones where I get more kills but the ones where I do a bit of everything.

70

So just an addendum to this, as I think it might be getting misinterpreted a bit in other threads.

The score vs placement comparison only works within the same mode (like ground AB). Obviously, an 80 per cent (top-fifth) player in one mode might not be the same level in another.

What you can infer from it is that Gaijin is luring the better ground AB players to try more RB with this kind of score gradient they’ve established from events. A top-third player in AB will get the same score as a top half player in RB. So if they can keep their relative position, they will make more score and do the event faster.

You can also infer from this that Gaijin feels the ground AB player base is less skilled overall than the RB one, which has (arguably) some more complex mechanics and probably a more experienced player base on average. I don’t think this should come as a real surprise to anyone.

What the data also shows though, is why the reverse strat, jumping from RB to AB for an event for an artificial skill boost, likely won’t work, because the best players in AB really don’t have that much more score to get, with raw average game scores of 2100 for the AB top player as opposed to 2000 for RB. The amount you’d gain for even a just somewhat above-average RB player would get, even assuming their average game placement would improve as much as Gaijin assumes, still wouldn’t be worth it.

Note also that ground and air have comparable skill gap based multipliers (about a 40% differential for both ground and air) as opposed to 15% for naval, suggesting Gaijin, at least, thinks the skill delta in that mode between AB and RB players is significantly less than in ground (the two air modes are so different in structure that there could be a few other obvious reasons for a bigger gap than the other modes too, of course.)

2 Likes
71

Would it be possible from the data you have at hand to make out some comparison, like whats the average score, winrate and K/D, for example on rank III vehicle or vehicles, for both arcade and realistic?

72

An 80% player is a lot better than top fifth, probably closer to top 0.1% of players.

image

The 80% player would be at the very end of the graph where only a tiny population is.

73

Ordinarily yes, but this is ordinal data, not quantitative.

“80%” here means “my average score and game position is greater than 80% of all the other players in matches with me (and less than the other 20%).”

There’s definitely still going to be some belling (because as the number approaches 100 it gets harder and harder to keep up) but it’s harder for me to say what the shape of that curve would be.

Just looking at the distribution of points on the graphs also suggests a greater concentration of points in the middle and less at the ends. Larger player samples would give a better idea what the distribution is. I’m the sample I used there were only 2 dots above 80 in ground AB and none in ground RB.

1 Like
74

Working on something like that at the moment actually, when the grind permits…

In the meantime, another interesting inference one could draw from the data is, if people accept the lines on the scatterplots are a reasonable first guess (yes, more data would be better, granted), then plotting your own dot on the graphs provided (from your ARP and score on your service record) says something about the natural efficiency of your play as opposed to what it should be for events. It helps that a lot of this last month was also an event, in this regard, so if you played it that could be useful to know.

The biggest factor in that, especially if you’re below the line, is going to be BR, but if you’re above there’s likely also an element of doing things that allow you to really run up absurd scores in those few games when it all comes together for you too.

In the sample used there wasn’t a lot of up and down variation for the ground modes, suggesting there aren’t a lot of killer BRs in ground with those kinds of “natural” efficiencies. There’s still the artificial disincentivization Gaijin is applying against playing lower ranks, though.

In air that’s different: a person who does the “Ar-2 arbitrage” strat for events will likely place well above the line and have an easier time of it (as one example).

And if you’re under that line and planning to do an event in that mode, yeah, either switch BRs or identify those “run the table” kinds of strats. Note this is separate from strats that just make you a better player generally, which would just move you up and down the line, not vertically offset from it. If you’re above the line by a lot you’re doing something right that only clicks in when you’re at the top of the team (or as I suspect with naval, but also air to an extent, there’s actually a couple of performance lines in the data for the different BR ranges, not just one, and you’re on the “good” line).

2 Likes
75

Could you please mention me if you put it together? I would be interested in the results.

76

The data you used to get the average scores was from normal gameplay or from playing the event? I’m just asking cause I took the time to calculate the average score on my matches while doing the 3rd mark and I’m getting a slightly higher average score than you but that could be explained for the fact that during events some players play “harder” or optimize score.

I got an average score for Ground AB of 1190.

77

I’d say there is little point in using normal gameplay data if you want to estimate event duration. I most certainly play score optimized when doing events, unless I don’t need to worry about time. And I don’t think I am alone with this.

1 Like
78

So your service record gives you your average score and average relative position by mode for the last 30 days (as well as lifetime), if you’ve played 100 games or more in that mode. That’s all I was using for this sample of players. As far as we know, it’s good data, although the amount of lag on it is unclear.

People are going to optimize, sure. In this case, “last 30 days” includes a lot of the previous event, so that helps. There are also limits on how much optimizing you can actually achieve when everyone else is also optimizing against you. The flip side of this is the 1200 point player saying they can “turn it on” and become a 2000 point player for events if they feel like it. My experience is that, generally, they can’t and they have a very inaccurate idea how long this event will take them, and others.

The graphs actually show this for ground. If you could just “turn it on and off” statistically there would be more vertical variation. You could make that case here on the data for air and naval, I think. But if all the dots are all in a line, as it is with ground with this sample, that means score and ARP are more tightly tied and all you can do to get more RAW score per hour faster, really is:
A) reduce average match time by quitting faster, as has been suggested already, or
B) do better than you do on average position in your team than you do normally.

Which, again, is going to be non-trivial if the rest of your team is doing the same thing as you. If you’re a top-third player, and you go to a lower BR and an optimized lineup, you still have to actually BECOME a top-quartile player by doing so to actually see event performance gains.

Your point also raises the question, now that the format has changed and there’s always an event, if there’s ever any non-optimized time again.

1 Like