Thursday, November 15, 2007

Is there Serial Correlation in World of Warcraft? Should There Be?

In World of Warcraft--and, I assume, similar games--players sometimes spend time hunting creatures for the loot they drop. Fire elementals, for instance, sometimes drop motes of fire; combined in groups of ten, the motes are used in making things.

If you are hunting fire elementals, it feels as if there are times when they drop a lot of motes, others when they hardly ever drop any. It feels, in other words, as though there is a probability of a drop that changes over time, slowly enough so that if you are doing well now you will probably be doing well five minutes later, and similarly if you are doing badly. My first question is whether the pattern is real or an illusion, the second is, whether or not it is real, whether it should be--whether such a pattern would make the game more enjoyable.

The reason for suspecting that the pattern may be an illusion is that gamblers often report similar patterns--sometimes the cards or dice are hot, sometimes they are not. In those cases, we know the underlying mechanics of the game. With rare exceptions, they imply that, unless someone is cheating, the pattern is an illusion. The probability that you will roll eleven is the same each time you roll, so a string of good rolls is evidence neither that the next roll will be good nor that it will not.

In the case of World of Warcraft, we do not know the underlying mechanics, or at least I don't. It would be perfectly straightforward to design the game with a drop probability that varies over time, with enough serial correlation so that current observations give you some information about what will happen in the near future. To find out whether that is how the game is designed I could keep track of a long series of tries, then do a statistical analysis to see if the results are consistent with the simple model--a fixed probability, the same each time. So far I haven't been sufficiently enterprising to do it; I don't know if anyone else has.

The second question is whether the game should be designed with serial correlation built in. My guess is that the answer is "yes." Human beings enjoy finding patterns, exercising skills. The fact that gamblers find patterns even when they do not exist in part reflects this. So why not make the game more interesting by building into it subtle patterns of the sort that players will look for?

Anonymous said...

I don't play WoW, so I have even less of a clue as to whether there's a nonrandom element to item drops.

I do play The Kingdom of Loathing, though, and there is a combination of randomness and determinism that is interesting. There are different categories of items. Common items are essentially random, but can be influenced in a variety of ways (skills, familiars, other items, some effects granted in various ways). Uncommon items are granted based on a turn-based counter, if you're doing the right thing at the right time. Rare items, as best anyone knows, come up randomly, but with a much lower likelyhood. There may be a deterministic factor involved, but conventionaly wisdom is that there isn't. "Ultra-rares" result from some deterministic behaviour, but it is clearly something really, truly strange, and there is likely also a significant random factor involved.

As far as your second question, I'm not sure I agree. People love building patterns that don't actually exist. Astrology is still with us, for instance. I don't know that games need to deviate from pure random behaviour to be ineresting. Discovering the underlying mechanism would seem to me to be only interesting to "spades" - see here for an analysis of how people relate to games (not that that analysis isn't controversial in the gameplaying community, but it does capture something). So perhaps a better way of looking at it is to break apart what various people like to do, and provide paths for each.

Anonymous said...

So far I haven't been sufficiently enterprising to do it; I don't know if anyone else has.

I don't play the game either. However I am reminded of a friend of mine, a long time ago before most of us had heard of identity theft, talking about the possibility. I said he was being paranoid. His reply was something that has always stuck with me. He said If I can think of how to do it, someone is already doing it.

I've played enough video games of one sort or another in my not too distant youth where I would've done exactly what you are proposing. Anything for an edge.

Unknown said...

I played the WoW demo, but didn't get into it enough to want to pay for the full thing. I used to be a programmer, and the studio I worked for was creating a MMORPG, and saw the design for their economy, and if WoW does something similar, it would explain such a phenomenon.

The studio I worked for wanted to keep their economy fairly "stable". If Blizzard wanted to keep the same type of stability for the various drops (in this case a fire mote), they would try to keep a stable level of motes in the environment. They would do this by dropping more motes when the current environment had few (i.e. many had been spent, or there was a surplus of new players), and they would drop fewer when the current environment has many (i.e. players had already looted the elementals, and not used up their stash). Not knowing their algorithm, I don't know if this is actually what is causing the phenomenon you are seeing.

Anonymous said...

To find out whether that is how the game is designed you could keep track of a long series of tries, or you could send an email to a Blizzard developer and ask it...

If just phisicists & economists would be able to do the same with god... :-)

David Friedman said...

Ryan suggests that drop rates might vary to maintain a stable level of motes. That's possible, but I don't think it could explain the sort of day to day variation that I seem to observe.

1. That's cheating.

2. My guess is that they wouldn't answer, that they don't make the internals of the game engine public.

Mike Huben said...

Just as a sanity check, I'd suggest that the internal workings are not much more complex than Nethack. Without even having looked at WoW, I'd use Nethack-style mechanisms as a first heuristic guess to explain the behavior.

(I never used to use spoilers for Nethack: I always preferred to read the code. :-)

Scott said...

I do notice something similar in FFXI (which, if we're being compeltely honest here, is so much cooler). A lot of game paramaters change with the phase of (in-game) moon, and the elemental of the (in-game) weekday, so that would be my first guess. Of course the pattern could be illusory--it's been a while since I've played sober.

William Newman said...

I enjoy finding patterns, but I enjoy it much more when the rules have an internal logic. For example, there are any number of 1-person games which have an explore/exploit theme, and I consistently enjoy the exploration and strategizing more when I'm doing it on maps randomly generated from some distribution I can learn rather than from maps handcrafted based on some designer's idea of what would be dramatic. (Exploring maps which are handcrafted to be pretty has its own appeal, not as part a challenge, just "ooh, shiny".)

Reverse-engineering the random number generator feels too much to me like guessing what some scenario designer will feel is dramatic. Also, it's one of those things where one person organization can figure it out and tell everyone else. I think if you want a challenge for players, it'd be much better to make some system where you publicize the algorithm, but not the values of the hidden variables on their server.

I also think it'd be better not to use such a hidden-parameter trick for general random number distributions like the ones which control your motes, but use it for a few specialized niches. E.g., pick a niche like fishing, and make the random variables be the semi-observable positions of schools of fish, and the unobservable positions of clouds of plankton that they feed on and follow, or something like that. I doubt your ordinary players would, on net, enjoy having their ordinary RNGs made more complicated and hard to understand, but I expect some minority of players would enjoy working with the more complicated inference problem puzzle subgame, and putting it into a subgame (fishing or whatever) would let them self-select for it.

Anonymous said...

I would say that Blizzard doesn't really need to put in any patterns - just as with gambling, players will see them regardless of whether they exist or not. Since as you point out determining the truth here would be very difficult (and any attempt would just make Blizzard more money anyway), why mess with things?

Anonymous said...

Maybe I dreamed it, but I think I remember reading that slot machines are programmed to have patterns (or behavior that is intended to be interpreted by players as patterns, even if it doesn't actually affect winning).

Anonymous said...

I don't know what kind of detail Blizzard went into when they designed the game's background, but if they gave the fire elementals some sort of "life cycle" where their bodies are richer in motes at certain points in the cycle, that would explain things.

Anonymous said...

I was very surprised on your first answer, that asking a game developer about game internals would be cheating. I'm using exclusively open source software for some time now, and in this completely free software world asking a developer about something is a daily practise. Even looking at the sources would be perfectly normal - but I'm not computer literate at such level, and asking a developer seems to me much more human. More on that, in the open source world there is no fundamental difference between a developer and a user - every user is a developer involved at different levels in developing. Also everyone has the option to ask - totally free market on questions :-) -, so I can't consider it cheating.

Cheating is altering the client for personal advantage. Communication, knowledge is not cheating!

Maybe what makes it cheating from your viewpoint is the (artificial!) difference between player and developer. Btw. that's what proprietary software is just about.

David Friedman said...

1. I don't use smileys--I think they are cheating too.

2. I didn't mean that I would be cheating other people. I meant that I would be cheating myself, in the same sense in which one could cheat at solitaire. Part of the game is figuring out the game--asking is cheating.

3. Unlike many WoW players, I don't use the webbed information that other players have created and shared--information on how to succeed at a quest and the like. To me, that is going too far out of the game world, hence makes the game less fun. I'm happy to use advice from other people, but it should be advice I acquired in-game--as one could acquire such advice if the game world was real.

I hope that makes it clearer.

John Fast said...

David, why don't you come to ICFA-29 and talk about Harald, and more generally about economics in sf and fantasy? I'm organizing a panel on politics and the Singularity, and the members include Vernor Vinge and Joe Haldeman (and hopefully Gregory Benford). You will be perfect for it, and you can talk about _Future Imperfect_.

("Stacked with anarcho-capitalists? I'm shocked, shocked!")

Raphfrk said...

In DAOC, they do give some information on the game mechanics. They have a 'grab bag' on their site every week or so. In this, they answer questions and often they are about specifics about how a certain ability works.

As an example:
http://www.camelotherald.com/more/3337.shtml

However, back in the 'good old days', players used to also carry out experiments to confirm that they were telling the truth.

Does WOW allow logging of the combat text ? In daoc, they would carry out lots of attacks and then parse the log files to extract the information.

One example, I can remember was that there was 2 dual wield type skills in the game depending on realm, and some players showed that they both game the same damage (despite being slightly different implementations).

Anonymous said...

Yes, it's much clearer now :-) !

- Mitch - said...

There are plenty of databases with drop rates like thotbott.com or my preference wowhead.com, that keeps track of drops. In my experience the drop rates can differ from day to day, but over a period of time it maintains a basically a set percentage.

If patterns for drops existed for benefiting players then scenarios of people running an instance 100+ times to get a certain item wouldn't happen.

Someone else mentioned combat logs. yes WoW lets you record combat logs to determine stats or combat mechanics. There are numerous forums that discuss this one, but imho Elitist Jerks guild forum has some of the more proven/tested data.

Roger W. said...

In my attempt to make gold in-game towards buying a flying mount, I farmed motes of mana for quite some time. And I played CONSTANTLY for about a month to make this happen. In addition, I play a DPS druid, so I was playing at about the maximum efficiency. This is all backstory to show that my experience takes out most of the variables, especially since I farmed fairly scientifically and at fairly random times of the day for quite a long time.

Point is, in my experience, no particular time of day and no particular life cycle (outside of level 69 elementals dropping motes more frequently than 68 ones) affected the number of motes dropped, but I sure did see motes drop far less frequently when there were many competing farmers in the area than when I was farming by myself, even though I was typically able to kill elementals at about the same rate. So I'm convinced that Blizzard has a hand in this and only allows so many motes to drop in a given time period. Then again, I expect it might just be a case of seeing patterns where none exist (because having to compete with ninjas upsets me) like scientist said.