Wednesday, January 2, 2013

Surprising statistics: where are the PvP-ers?

I wrote a post to debunk the commonly recited claim that WH people are both extremely rich and that WH space is extremely dangerous. I used regional statistics from killboards to prove that WH ship losses aren't that expensive. They claimed that because they go to PvP in null (because they are so awesome that no one attacks them in their WHs). So I went to look for more data. Someone who wished to remain anonymous downloaded me 2868915 API verified kills from EVE-kill.net from 2012 (until Dec. 20.) for a nice sum of ISK. Using this data I could track pilots and not places.

I identified 197164 individual pilots who had at least one 50M loss in 2012. I discarded the rest as cyno/scout alts, lost newbies and such. I mean a single 50M+ loss isn't too much to ask if you PvP. Then I classified these pilots into "Highsec", "Lowsec", "Nullsec" and "WH" pilots based on which zone they had the most losses. It's pretty straightforward that you defend your home with more expensive ships than you roam with. Some people might place himself in another group, for example nullsec JF pilots who has most losses in low and highsec, but they are wrong. Such hauler couldn't exist without low- and highsec to haul to. His whole existence is bound to low- and highsec. With the above I found the following amount of pilots who lost ships in PvP
  • 18231 WH pilots
  • 87353 Nullsec pilots
  • 39806 Highsec pilots
  • 51774 Lowsec pilots
After that I collected all the losses of these pilots and calculated the average loss/pilot. The chart below shows that, with different colors for the different regions. You can see that while theoretically one can be a WH pilot with only 25.1% losses in WH (and 24.9 in low, high, null), pilots of all zones have vast majority of their losses in their home regions.
As you can see, the roaming (and hauling) losses of WH people outside WHs are only 18% of their total losses. For the record: 84% of the total WH kills are done by "WH people" so even if I misidentified some WH-ers because of a highsec hauler loss, their contribution to WH PvP is minimal.

As next move I split the people into 4 groups depending how the number of their kills relate to the average of their own zone. The averages themselves are 7.0 losses/year for WH, 14.6 for Null, 7.1 for High and 11.3 for Lowsec people. Now let's see how the lost ISK is distributed between the four groups:
Highsec ISK losses are dominated by those who lost less than half the average amount of ships. They are ganked PvE players and haulers who lost just a few expensive ships. Highsec and lowsec are dominated by lot of smaller losses of active PvP-ers. WH on the other hand seems to be pretty rounded, most losses were of "average" people and the "carebears" and "PvP-ers" are both a minority. Why the amount of losses is important?
As you can see, for each type of pilots the average loss values are the highest among those who lost only a few ships, most obviously they got their pimped PvE ship ganked. Please note that it's the value of one loss, the yearly total loss increases with PvP activity (obviously). For this reason the WH-ers lose less money on PvP (first figure) despite their single losses are more expensive.

Now let's look at the loss value pattern! The chart below shows the percentages of various value ranges in the total losses (1x10B is the same as 100x100m).
Similar to the previous: the low and nullsec are dominated by smaller losses of the busy PvP population, the medium losses dominate WH and highsec has the largest part of large losses. Freighter ganks beat even the supercapital losses of nullsec!

Finally let's look at how dangerous systems are! While the feeling of danger is subjective, the real danger is very objective: the chance of loss. The chart below show how much ISK was destroyed in a year in an average solar system:
Now look at that! Not only the mighty C6s were beaten by highsec systems but the lowsec came out on top by large! Also, C5 and C6 is dominated by expensive losses and we already established it to be a sign of PvE. If we check the number of kills for trivial (0-30M) and non-trivial ship losses in the various system classes, we get the same, lowsec is the most dangerous place to be, while WHs see much less kills than highsec due to their low population.
PS: If you can get a more complete database and send me, I gladly analyze that too!



The reason why I'm at -4 sec status after only 2 days.

22 comments:

Anonymous said...

Forgive me if I am wrong here, but are you not making the assumption that wormhole pilots are losing pimped PvE ships here based entirely on value?

What change to your conclusions do you get if you assume that they are losing their pimped ships to PvP (...say for example the Hard Knocks fight, where it is clear that the attackers in that fight were flying fairly pimped, especially in capital assets)

Gevlon said...

Look at the third graph. For EVERYONE, including WH-ers, those lose the most expensive ships who only lose 1-2 ships. I doubt if they are PvP-ers.

I know that there was the Hard Knocks fight. But it null there are supercapitals and still 60% of the null losses are cruisers and below. Expensive loss of PvP ships is a rare exception.

Anonymous said...

did you look at the pilots who have 1-2 ship losses KILLS to see whether they are PvPers?

I just see the PvE assumption as an enormous assumption - one which your entire analysis is then built up on. Take that assumption away and the whole thing topples over. Given how important an assumption it is on your conclusion I would have expected more effort in verifying that these are in fact PvE Ganks.

The cheapest PvP ship values in fleets I fly in in a C6 (one of the larger C6 corps) are around 1.5b. It is not uncommon to roll out with 2 or 3 billion isk ships on a daily basis. We kill a lot more than we lose (and some of our pilots do indeed have only 1 or 2 losses on their 'hardcore' pvp toons for the year).

Also, on your numbers did you normalize the values for population size? Obviously more kills happen in nulsec, it is more populated - did you normalize this against the lowsec and WH populations?

Gevlon said...

One thing is for sure: the loss pattern of WH and Highsec is the same. WHs have large 1-2-3 loss population. 1-2-3 loss people lose expensive ships while many-loss people lose cheaper ones.

Yes, for a perfect data I should look on killers. Care to make me a 2-5GB sized database with
Loss value, region, sec status, victim name, Killer1 name, Killer1 damage, Killer 2 name, Killer2 damage ...

Of course not. And I won't pay multi-billion of ISK just to prove which is already obvious.

I don't question that there are PvP-ers in WHs. There are PvP-ers in highsec, I've just joined such group. But it won't change the fact that the characteristics of Highsec is PvE.

Anonymous said...

Nice study with a lot of data, i find it way more conclusive than your previous analysis about the subject.

Anonymous said...

The characteristics of highsec are PvE, most certainly. It is where the vast majority of mission runners live. It is where the newbies start out. It is where most of the manufacturing of eve takes place (proven plenty of times, by your and other analysis)

To say the characteristics of WHs are PvE as well because the loss pattern is the same..well correlation does not indicate causation.

Yes, we in wormholes do a lot of PvE, in terms of PvE value. In terms of time working on sites we clear them very quickly and have very high isk/hr. Ok, if you take into account the fact that we have to scan/roll/probe/scout/haul to highsec that isk/hr is much lower than it many claim it to be the fact remains that the isk/hr in a wormhole is relatively high. Certainly higher than any general highsec activity (I count your activity in highsec to be a spectacular outlier, not the norm for the vast majority).

You may even try to suggest that all the scouting/scanning/rolling/etc time should be counted towards PvE. Let me assure you that all of the wormhole alliances I know about always take PvP over PvE. An "op" may start as looking for some red crosses to shoot but may very quickly morph into a 2 hour cloaked camp to try and trap the other guy's gang and escalate into a large fight.

To state we are "PvE" based on an apparent correlation with a highsec pattern is dishonest.

Basically you cannot use "isk killed" as the only metric here to determine player behavior. You are grossly oversimplifying something which you don't totally understand.

A lot of the pilots in the alliance I am in are former power block pilots and FCs who will attest to the fact that those wormhole alliances that will fight are ruthless. I'd suggest you reach out and interview a few of them to temper your results with real world experience. Better yet, join one and experience it for yourself.

The graphs are interesting though - the highsec kill distribution is particularly surprising. Question: do you count dead freighter pilots as PvE pilots (big expense, low number of losses per year)?




Jim said...

"If we check the number of kills for trivial (0-30M) and non-trivial ship losses in the various system classes, we get the same"

Who flys a ship worth less than 30mil in WH space? Aside from my Venture and Heron, the cheapest ship I fly is a Vigilant....

Sugar Kyle said...

There are entire low sec and fw groups that stay in frigs, dessies and cruisers and never break the 50 mil mark. Our latest fleet comps are running 15-20mil pet ship because we can be effective at that price.

I put my cynabal aside in place of a jaguar last week because it was the better ship for what I was doing.

RubyPorto said...

"Then I classified these pilots into "Highsec", "Lowsec", "Nullsec" and "WH" pilots based on which zone they had the most losses."

"pilots of all zones have vast majority of their losses in their home regions."

Umm.... that's Tautologically true, but not exactly a useful conclusion.

If you classify everybody based on where they have the most losses, then of course they're going to have the most losses in the place you've classified them as belonging to.

I would think that a better way to do it would be to classify pilots based on where they have the most kills, then look at where they lose their ships to see if WH players really do make excursions into Null or if they're primarily active in Null and just make their ISK in WHs.

Gevlon said...

@Sugar: Zkillboard says http://zkillboard.com/character/91512622/ you have 11 battlecruiser losses, that's all above 50M. Your 20 total losses places you to the TOP 5% PvP-ers!

This is another example why personal experience worth little: everyone considers his close surroundings norm when it's not. The Hard Knocks people think that everyone else also runs around in 5B dreads, the Pandemic Legion people think that everyone multiboxes supercapitals and you think that you are a noob when you are one of the most active PvP-ers of New Eden. If you don't believe me, just look at the right side of that Zkillboard page: All time rank: 5390. Out of like 200000.


@RubyPorto: while it's obvious that if I classify people WH-ers by having the most losses in WH they'll have lot of losses in WH. However what I wanted to say is that they fit to this container very purely. What I mean is that an average person has 80%+ losses in ONE zone.

Anonymous said...

What I mean is that an average person has 80%+ losses in ONE zone.

and that is why you are wrong about wormholes.

The above *may* be true if you live in deep nul, or you've got a low sec status and are stuck in lowsec, or you are a major highsec carebear. It is true because you are more or less geographically bound to your region. You've chosen where you live and your likely live and hunt (or mine) in that geographic location.

Wormholes are utterly and completely different. The geography changes and moves. We can find a nulsec exit on one side of the map in one chain - collapse that and find another and find a lowsec on the other side of the map - collapse that and find a high sec and go gank some freighters or whatever we want to do.

We also use this geography to insert into enemy wormholes which can lead to kills outside of wormholes. Because C5s are difficult to roll into, one strategy is to get a scout into the target C5, scan a route out to nulsec, and do the same from home system. The wormhole chains are connected by a pipe through nulsec. Moving an expensive PvP fleet through nulsec may result in impromptu fights.

Pick an active wormhole alliance or 2 and have a detailed look at their killboards. You'll see that their kills are spread all over the map. Lots of losses are made all over the map. The top wormhole alliances are probably going to lose as many if not more ships outside of wormhole space simply because we're not as good in K-Space and because we use K-Space to move fleets around.

The biggest problem you have with this analysis is that you have exactly no understanding of how wormholes work both in terms of mechanics and the way that we have adapted to exist within those mechanics. You have therefore built your argument on what you see to be reasonable assumptions which have very little or no basis in reality.

Such is the failing of the armchair expert.

Druur Monakh said...

@Gevlon "And I won't pay multi-billion of ISK just to prove which is already obvious."

And yet you should: the seemingly obvious is where our own personal biases (we all have them) reach out and strangle us. Especially when trying to argue a contrary position.

IO said...

"There are PvP-ers in highsec, I've just joined such group."

Ganking gunless Mackinaws is not PvP. Really.

Oursbern said...

Good effort but a little incomplete.

Let's just take the second graph (frequency of lost ISK)

There are some very important (and obvious) points that are not taken into account in your analysis.

1. Killing supercapitals and freighters (null and high sec) create the same behavior of the dependencies near zero. And as far as I know one can build supercapitals in WH. Which makes it impossible to say what is actually killed in WH.

2. (and more important) The average value of each kill in WH is much higher than that in other space. This is probably due to the restrictions in transport that make everyone in WH (both for PvP and PvE) fly a bigger ship. Which means that what is considered as a trivial ship in WH and in other space is very different. That makes your sub-sequential graphs like a comparison between apples and oranges.

To do the analysis right one have to use the data from the second graph to estimate the mean value of a trivial ship in each case or establish the value of a 1 ISK invested in ship. Then re-normalize all data to those values and use only the re-normalized values to draw conclusions.

Good luck!

Oursbern said...

p.s. I forgot to mention:

I do not know what are the reasons of the particular ISK intervals you used for graph 3 but they seem a little arbitrary which is enough reason to attack your conclusions.

They should be established on base of some in game values like: minimal price of T1 cruiser -> minimal price of T2 cruiser -> minimal price of battleship etc.

Or one should use some justified mathematical dependence.

Mabrick said...

One thing I need to point out about your starting numbers. If you divide the number of WH pilots by the total of the other three you get .101887... etc. That tells me 10% of your sample are WH pilots. However, by CCP's last released numbers only a bit over 5% of all capsuleers lived in WHs. Your WH population is over represented by almost double. That's a big problem that ripples down through the rest of your statistics. Can you do this justified to CCP's last released population numbers?

Gevlon said...

NOTE: In yesterday's post I replaced one table with a chart and recognized the importance that there are not just "ganked PvE-ers and PvP-ers", but there are "average" people in the middle. Using this the outcome became much less offensive to WH PvP-ers. While the ruthless hunters are a minority, the average people, who are not "carebears" are in large supply in WHs. Also removed the last table since the CCP data was for accounts and mine for pilots so same-account pilots compromised it.


@Mabrick: my statistics don't show those people who didn't lose a single ship. Please note that all my pilots are 197K while there are surely more active pilots (characters) in EVE. Most of them are missing from highsec, artificially decreasing the % of highsec, increasing the % of others.

Anonymous said...

Your changes make the analysis more balanced. There is a way you can fix this:

WHs see much less kills than highsec due to their low population.

and that would be to normalize the results for population. Average losses per person in that region etc. This I think would be interesting. I am not sure what the result will be but it would potentially show how risk averse a region is.

You could take it further, average isk per loss per person living in that region to see just how much is being risked (not just checking any risk). Could you perhaps include both of these metrics in your analysis?

Anonymous said...

Carebears love this post! They like the fact that high sec is more dangerous than null sec and that will cause them to continue to be more extreme in pressuring to make EVE safer in high sec.

Anonymous said...

Will post the same comment I had on EN24:

Gevlon,

Appreciate your stat analysis. However I think the problem really is that how "dangerous" a class of systems is doesn't quite correspond to the average amount of isk loss per system AT ALL. You said that objectively danger = chance of loss, however average isk loss per system / total isk loss doesn't represent the chance of loss. If you take a deeper thought into it you'll see that it doesn't make sense. All your statistics are well done but the inference here is flawed.

The reason is that chance of loss is independent of ship cost (isk) in almost every case but tremendously dependent of other attributes. take the WH as example, the chance of loss is different, for any kind of ship, depending on who you are. If you're a resident with a solid group, it will be different from someone who lives off highsec just coming in for some rats. If you really want to create a probability distribution on the chance of losing a ship in every different class of systems, you to actually model the probability based on many different "latent variables" instead of just isk loss. This inter-dependency of variables is very complex and takes more advanced statistical methods than the graphs you have shown.

This explains why many people feel your conclusion isn't correct. While sometimes the correct answers are counter-intuitive, but guts-feeling or intuition isn't necessarily incorrect all the time.

Cyras DeValera said...

The data you paid for, any chance a copy of that is for sale? Everything has value..

Anonymous said...

It doesn't make sense to split up W-Space into C1-6. Yes, that way you get to point at WHs and yell "told you so!". But then you'd completely ignore the fact, that WH-people (need to) visit adjacent holes regularly or get visitors from one of those ajacent systems.

You didn't divide low or nullsec by realsec status either, did you?

Subscribe to the goblinish wisdom