Eavesdropping in SL – The Unbearable Weight of Erroneous Assumptions

by Pixeleen Mistral on 24/02/07 at 11:46 am

by Fiend Ludwig

Thisbejohn_william_waterhouse1909According to Nick Yee, of Stanford University, and his colleagues, Second Life, and by extension other virtual environments, is an ideal place to test paradigms of real-life human social interaction. In the paper The Unbearable Likeness of Being Digital: The Persistence of Nonverbal Social Norms in Online Virtual Environments, Yee asserts that:

Overall, our findings support our overall hypothesis that our social interactions in online virtual environments, such as Second Life, are governed by the same social norms as social interactions in the physical world. This finding has significant implications for using virtual worlds to study human social interaction. If people behave according to the same social rules in both physical and virtual worlds even though the mode of movement and navigation is entirely different (i.e., using keyboard and mouse as opposed to bodies and legs), then this means it is possible to study social interaction in virtual environments and generalize them to social interaction in the real world.

Although it sounds plausible on the surface, Yee’s conclusions are based on a raft erroneous assumptions. Of the five variables that Yee observed during avatar interaction – gender, interpersonal distance, mutual gaze, talking, and location – only interpersonal distance and location can be accurately measured by simply observing avatars while they communicate.

For all of the others, Yee has missed the boat entirely. Observations were made in-world by research associates who used a script to collect data. He reports, “When triggered by a designated key press, the script would collect the name, Cartesian coordinates (x, y), and yaw of the 16 avatars closest to the user within a 200 virtual meter radius. The script would also track whether the avatars were talking at that given moment. The script would then store the information as a text file.” This text file is called a ‘snapshot’ in the study. The snapshots were then analyzed to isolate dyads (pairs) of avatars who were talking to one another. No indication is given whether the snap-shotted avatars were asked whether or not they wanted to participate in this study.

After the snapshot was recored, Yee’s assistants determined avatar gender. He notes “In many cases, however, the gender of avatars was unable to be determined, as users chose to be androgynous or non-human. Each dyad was then coded as male-male, female-female, or mixed.” Although it is generally possible to determine the gender of the avatar by examining their appearance in SL, it is impossible to know the gender of the real person controlling that avatar. Therefore, Yee’s conclusions based on avatar gender and their relationship to real world observations are invalid. An observed female-female SL dyad could easily be the result of two male real life Second Life participants – there is a least one well known relationship where two real-life men have female avatars who are ‘partnered’ in SL.

Yee’s mutual gaze measurements are also problematic, as the user behind the avatar must accomplish a sequence of deliberate keystrokes to direct his/her avatar to look at the avatar with whom he/she is speaking. It is a personal observation that there is very little importance ascribed to maintaining eye contact during chat, as user attention is usually directed to the UI chat window anyway. And since avatars don’t automatically emote (although emote animations can be triggered by the user) there is no body language feedback to be gained from maintaining eye contact during SL chat. Circumstance is more likely to create a mutual gaze – simply walking toward another avatar can cause spontaneous eye contact, as an avatar’s eyes automatically look in the direction of travel.

Additionally, Yee states, “If users were in this ‘is typing’ mode, they were coded as ‘talking.’” Although it is true that the default state of an avatar is to show the ‘typing’ animation when chatting, there are numerous animation overrides that are commonly used by experienced SL residents that suppress the typing animation, overriding it with a different, customized action. Therefore, two avatars who are standing adjacent to one another, but are not showing the typing animation, may indeed be chatting. Yee also excluded dyads that were observed as being more than 3.7m apart, as this was shown in real world studies to be the distance further than which social interaction does not occur. Again, however, this assumption is false in SL, where chat can be ‘heard’ for 30m and use of the ctrl-alt keys allow the avatar’s camera (the view the avatar’s user sees on his/her computer screen) to zoom in on the chat partner without physically moving, allowing for effective chatting outside of the 3.7m range. Yee’s observations also ignore the use of SL instant messaging, which is commonly used by avatars to chat privately, and is invisible to any external observation or scripted eavesdropping.

There may be patterns of social interaction observable in Second Life, these patterns may give the observer some insight into the behavior of avatars, and this insight may lead to generalization about the social norms of virtual online environments as a whole. But, without detailed information about the people who are driving these virtual interactions, drawing parallels between real world norms and observed ‘virtual norms’ seems both erroneous and premature.

[picture: Thisbe, a painting by John William Waterhouse, 1909]

25 Responses to “Eavesdropping in SL – The Unbearable Weight of Erroneous Assumptions”

  1. Prokofy Neva

    Feb 24th, 2007

    I am so glad you systematically took this on Fiend, good job. I remember when Nick Yee’s studies first began to be celebrated over at the egghead station, Terra Nova, I asked in bewilderment but wait, how can you measure “avatar gaze”? Avatars can’t look at each other. That is, there is some keystrokes you can do, is it tab or something? that makess one avatar turn toward another. Who the hell ever bothers with that, unless they are making a machinima movie? And all those other things you’ve brilliantly noted that I had never thought of, like the camera zoom or IM chat or whatever — people don’t necessarily tend to land, come close to each other, and talk as if they had ears that need to hear something closer.

    I’m wondering then how I “just know” which avatars who appear female are transgender, and why some dragons or furries with androgynous names seem to manifest as “male”. You’d have to spend a lot of time parsing these cues. I *think* it has something to do with territorial marking. I watch A LOT of avatars being a landlord — it seems to me, the males tend to walk around much more rapidly, and try to walk through the circumference of their property several times hurriedly. It’s like a little territorial pissing thing I guess. They walk up and down fast. Some of them that are older, have animation overrides, weapons, or are in IMs, will tend to just hover in the air a bit above ground; they don’t land.

    Females, however, either land on the ground and stay in one place, or they tend to sit on something or sort of “nest” more. Or they may be busy clicking on stuff, a rental box, to get a card, or a house, to see how many prims it has. These ancient codings of homestead maker and hunter are probably fairly hard wired.

    But I don’t see that there are absolute statistical norms, and I think that’s because gender isn’t an absolute anyway, and Second Life proves that in spades. So I might get a female who walks up and down rapidly who in fact is a real female, not cross-dreser, or a male who becomes interested in clicking on the about-land menu or something. Generally I can tell the people who are cross-dressers, but I don’t know how I know this, I have to think about it.

  2. Lugh/Samantha

    Feb 24th, 2007

    As an Anthropology major at a major US University, I’ve often thought of using SecondLife to do studies regarding real world behavior and situations. But I’ve found more often than not, that it is literally impossible to do such studies in a precise and accurate manner. Secondlife allows for some freedoms or some exemptions from certain types of behavior that are required to function in real life. Eye contact and body language being two of the more important and pertinent behaviors. You can get around those things in Secondlife and you don’t necessarily have to be looking at someone in order to talk to them and then there’s instant messaging which adds a whole new level in regards to communication between people.

    SecondLife can be used to studies ideology though. We can study what types of ideas are most prevalent throughout SeconndLife and then compare them to those that are found in real life. This data could be helpful to psychologists or ethnographers or even the business world because SecondLife is all about fulfilling our deepest desires and fantasies and by looking at what’s present in Secondlife, manufacturers or even marketing executives or psychologists could find out what we really want and produce a product or service that we need.

    But then again, not everyone is into Secondlife and while some might claim that there’s an even spread of the real world population in Secondlife, demographically speaking, there’s still no definite way to measure that.

    SecondLife is a giant canvas coated with numerous brush strokes of various colors. Sure you can make some general assumptions and guesses and statements about it, but nothing is truly definite or concrete. It’s a changing world that you cannot accurately make any claims about in terms of the behavior of its citizens. You can make some claims about their creations and then guess at the motivation behind the creation of them but you cannot pin down definitively why they did it.

    There are far too many variables in SecondLife that cannot be measured accurately enough to produce solid socially scientifically valid findings.

    After all, I play a male and a female avatar and you’d never have guessed either was the same person.

  3. Prokofy Neva

    Feb 24th, 2007

    What I find uber annoying about this is that Nick Yee, just because of a pre-existing credentialed reputation, can immediately put out a paper, captivate the media looking for stories like this, have every single major outlet related to gaming and even general news cover his *gasp* dramatic discover that males and females have inherent characteristics they port in to virtual worlds with them *ungasp* and then sit back and bask in the media limelight and the concommitant invitations for all-expense-paid lectures, attendance at game conference seminars, etc. etc. and nobody ever checks if it is true, except a couple of people on a forums like the Herald.

  4. Panda

    Feb 24th, 2007

    Just a few things to point out about Fiend Ludwig’s article. The chat-limit is currently 20m, not 30m, if I’m not mistaken, and the “typing” avatar state is separate from any animation or AO active on an avatar, so they would indeed be able tell what avatars are currently typing.

  5. Fiend Ludwig

    Feb 24th, 2007

    The SL Knowledge Base says the chat range is 30m (http://secondlife.com/knowledgebase/article.php?id=116) which is where I got my information. In the past, I have chatted with numerous residents who have suppressed the typing animation with an AO, however this may have changed with the most recent update. If anyone can clarify this it would be appreciated.

    Additional note regarding the article:

    Prok’s first comment alerted me that this topic had been discussed on the Terra Nova blog, and in checking I found that it had indeed been hashed over in a fair amount of detail here: http://terranova.blogs.com/terra_nova/2006/08/the_prison_of_e.html.

    I first saw this story a couple of days ago when it was reported by the CBC (http://www.cbc.ca/technology/story/2007/02/22/tech-2ndlife.html), presumably because Yee’s paper has just been published in the February 2007 issue of CyberPsychology & Behavior (http://www.liebertonline.com/toc/cpb/10/1?cookieSet=1).

  6. Panda

    Feb 24th, 2007

    Ludwig, you can indeed override the typing animation, but the typing “state” is separate from the animation. You can read this state with a script.

  7. Michael Seraph

    Feb 24th, 2007

    And how many times has each of us found ourselves staring straight at some avatar who is busy typing away while IM’ing somebody else 10 sims away? LOL.

  8. Nick Yee

    Feb 25th, 2007

    I’m really glad some of these issues are being brought up because we weren’t given enough number of words in the journal article to go over some of these concerns. I’d like to address some of Fiend’s points here.

    1) Fiend is concerned that we never sought explicit permission from SL users during the study. While this is true, we also never publicize users’ identities. Before we conducted the study, we had approval from our own institutional review board as well as Linden’s own review process for academic research. In both cases, our study fell into the “public observation” category which does not require explicit consent. Of course, there are gray issues in VW research that haven’t been ironed out yet, and this may well be one of them, but at the moment, institutional review boards and Linden Labs do not require explicit consent for this kind of data collection.

    2) Fiend points out that we don’t know the gender of the users themselves. While this is true, in the paper, we are actually making the more subtle point that users conform to the social norms of their avatar’s gender (regardless of their RL gender). Although our data supports this hypothesis, we can’t rule out the alternative explanation that perhaps no one gender-bends in SL and thus it is everything to do with real gender. And since we couldn’t collect RL gender information, we didn’t belabor this point. But we do have more recent experimental data that show users conform to various attributes of their avatars, and that’s really what we were trying to suggest with the SL gender data: http://www.nickyee.com/pubs/Yee%20&%20Bailenson%20-%20Proteus%20Effect%20(in%20press).pdf

    3) Fiend also points out that the avatar’s eye gaze in SL doesn’t correspond with what the user is actually paying attention to. While we would agree that there probably isn’t a 100% correspondence between the two, we in fact get into a much stranger situation if we argue that there is very little correspondence. This is because it would mean that avatars in SL follow many social norms *without* direct user intervention. Thus, we would argue that while some degree of non-correspondence occurs, and this lowers the correlations in our data, that the norms seem strong enough to come through even in a situation where other factors are washing out the effects a little.

    4) Fiend also mentions that users in SL can use IM to communicate. While it is true we didn’t study IM communication in SL, I would like to point out that the use of cell phones in RL (and that psychology) is different from face-to-face interactions, and that it is possible (and makes sense) to study the two separately. And in our study, we were interested in body space and mutual gaze so we focused on close-range interactions. The fact that people chat and interact differently on cell phones has no bearing on the research on personal space. That’s comparing apples and oranges. And that’s also the case here. We were studying personal space, not IM.

    Again, I’d like to thank Fiend for providing an opportunity for us to respond to some of these concerns. As we had strict word limits for our paper, we were not able to address these issues as fully as we would have liked in the paper itself.

    I’d also like to point out that what the media picks up on (and how they hype it) is so beyond our control. I have about 15 published journal papers and this was the first one they chose to go bonkers over. If you have time, check out our other work: http://www.nickyee.com/cv.html

  9. Fiend Ludwig

    Feb 25th, 2007

    @Nick – thanks for responding to my article. I appreciate your clarification, but would like to follow up on your responses:

    1. I respect the ruling of your (by that I am assuming Stanford’s) ethics review board in this matter (not so much LL’s, but that is beside the point). However, if this type of study were conducted on people (as it obviously has) I wonder if the freedom from requiring consent of the the subjects would be equally applied – I would seem difficult to measure IPD, gaze angles, etc., from a real person without consent because presumably videos or photographs of the subjects interacting would be required. If people (or avatars) know they are being observed, their behaviors change. And if parallels between social research in SL are to be applied to real world societies, surely the parameters of the study must match.

    2. If, as you point out, your more subtle point about gender is that “users conform to the social norms of their avatar’s gender,” the question then becomes – how does a male user even know how to ‘pretend’ to conform to female norms with their female avatar, and vice versa. Gender specific social behavior is deeply ingrained in every person. The ability to simply adopt the social norms of the opposite gender at will when ‘playing’ an avatar seems wildly improbable except, as Prok points out, in cases in which the user moves between gender roles in real life (such as a transgendered person perhaps). You note that you “didn’t belabor this point”, but it seems to me the crux of the study.

    3. You note regarding eye gaze that “…we in fact get into a much stranger situation if we argue that there is very little correspondence. This is because it would mean that avatars in SL follow many social norms *without* direct user intervention.” In a way, this is exactly what happens, the SL UI does (and probably was specifically designed to) follow social norms without user intervention. By simply walking toward another avatar, your avatar is, by default, looking at that ‘target’, if this is not the desired gaze, the user must specifically take steps to avoid it.

    4. Fair enough, but my point is that even though an avatars may be gazing at one another (by default or otherwise), there is no way to tell whether or not they are really mimicking the real world social norms of chatting dyads, or if they are also, as Michael Seraph says above, “…staring straight at some avatar who is busy typing away while IM’ing somebody else 10 sims away?” Unlike in real life, where you can see someone using a cell phone and make assumptions based on that behavior, this is impossible in SL, therefore direct correlations are more tenuous to draw.

    I believe that there are ways to correlate VW and real world research, but useful data will only be available once more effective means of collecting it are established. It is likely that, like a number of other pursuits, social research in SL will prove more profitable when based on an in-world, cooperative model, rather than an arm’s-length, traditional model. In others words, it is my opinion that using surreptitious scripts to collect information (however innocuous) about SL users without their consent or knowledge without the ability to compare those data to corresponding real user info will lead only to superficial conclusions about the true nature of virtual world societies.

    However, I do agree that the MSM does froth about these topics to some degree, and I certainly do not blame you for their coverage of your work.

  10. Aertisan Fei

    Feb 26th, 2007

    One of the things that struck me about Nick Yee’s observations about social norms within Second Life, something that Fiend Ludwig mentions, is that the distance for a conversation within Second Life is essentially meaningless. Two avatars could be having a chat conversation yet for whatever reason stand 10 yards apart – something that would not happen in real life, not without shouting.
    Proximity for real life conversation, and the social norms that are associated with it, is partially dictated by the vocal levels used (i.e., whisper, “indoor voice”, shouting). Two people whispering to one another are typically huddled close together while another couple talking at normal decibel levels might be 3-10 feet apart. Both cases are typical for real life conversation and fall into the social norms of proximity.
    Temporarily ignoring IM’s, there are 2 levels within Second Life for chat – normal and shouting. While I think most would agree that avatars having a conversation would rarely shout at one another, the fact that avatars can engage in normal conversation at a distance of up to 30 meters negates the need for any sort of close proximity while chatting.
    Besides which, anyone within 30 meters of the speaker can read and take part in the chat. While real world conversations can be overheard as well, most would be inaudible at 30 meters.
    This brings up the use of IM. IM’ing one another within Second Life allows for private conversations, therefore the distance between the conversing avatars becomes pointless.
    Relative proximity for conversation within Second Life is a choice and not a necessity. Whether or not two avatars standing a couple meters apart and conversing with one another is to mimic real life social norms is up to the individual. Social norms might exist within Second Life but the real world definition doesn’t always apply.

  11. Prokofy Neva

    Feb 26th, 2007

    Nick, I just find what you are writing completely unpersuasive.

    Avatars can’t see. They don’t have eyes. They don’t have a gaze. They can’t look. So how can you tell where they are looking? The human typist behind them has the eyes and does the looking — but they could be looking at anothe window on another tab, at an IM, in the distance, at something in RL — you just have no way of correlating that to the avatars’ movements. I don’t know how you can possibly make any solid judgements about it.

  12. Maria LaVeaux

    Feb 27th, 2007

    I read of the Study in MSN a Few days ago, and Laughed. Now i’ve read a More in depth analysis of it here Including Input by the Study’s author.
    I’m afraid i also side with Fiend, and Prok. Nick, you simply disregarded FAR too many of the On Line Variables, and SL mechanics to make any Valid conclusions here. I have frequently been standing next to someone while conversing out loud with someone twenty meters away.
    Also, Many of the Avs On Line Myself included have Not disabled the Function that automaticly attracts my gaze to the person who has been typing last. I have also been in Lengthy conversations with people while my Camera is locked on the Floor above where i am Constructing something. Simply Put, you Cannot guage with any accuracy what-so-ever where an avie is Looking at any Given moment. Your assumption that the Player, regardless of RL gender is conforming to the Mannerisms of the Avi gender they are playing assumes that the Player is on some level conscious of that Gender difference (In standing distance for example) and is making a decision, Concious, or otherwise to Over ride thier natural Inclinations, and conform to a standard that they would, in fact, be completely Oblivious of (Unconscious mind says “Ok, you are playing a woman so stance location needs to be adjusted by .XX closer, or farther away”).
    What you have in your study Nick is conclusions based Not upon Imperical Observation, but upon supposition, Preconception, and Pure Guesswork.


  13. Andy

    Feb 28th, 2007


    Fiend Ludwig point number 2, post 25 February:

    2.if as you point out, your more subtle point about gender is that “users conform to the social norms of their avatar’s gender” the question then becomes – how does a male user even know how to ‘pretend’ to conform to female norms with their female avatar, and vice versa. Gender specific social behavior is deeply ingrained in every person. The ability to simply adopt the social norms of the opposite gender at will when ‘playing’ an avatar seems wildly improbable except, as Prok points out, in cases in which the user moves between gender roles in real life (such as a transgendered person perhaps). [..]

    Is this science? In which discipline?

    As Wittgenstein said, to paraphrase roughly, there are things we can talk about and there are things for which we must remain silent (or at least keep quiet about till we have answered a few others!) In other words, some of these questions can be answered by recourse to the literature of relavant academic discplines or, if not, by designing fresh experiments.

    The primary question in gender role playing is: Can people – not how does – a male pretend to conform to social norms? How? What?, the mechanism? Tricky. SL might help to answer this elementary question. First you ask people if they can guess which sex a number of avatatrs are, (controls include RL -> RL (not using SL), Rl -> avatar, avatar -> avatar) then you do some stats, then you divulge who is which sex.

    You could use the RW/VW opportunity to try to answer such questions as “Can people pretend to be the opposite sex as avatars?” (Even Q.Can they do it better as avatars than just pretending in less complex scenarios).

    But you could also ask, “Can people pretend to be the opposite sex?” using a standard Turing Test set-up. These are the fundamental questions which people have been asking for over 60 years in the AI arena!

    There are much more simple but important questions to get straight. Such as ? Well, are your looking for analogues or homologues?

    By analogy, many of the best answers about how the brain works have been obtained by looking at damaged brains. In social psychology, a good way [sometimes the only way] to explain what a group is and how it functions is through what happens when a group [or anything] fails, i.e. Groupthink.

    By this token it might be a quicker route to many of the question about whether VR is just RL or has some added ingredient (meta-phor) to look for the weak links. To do this it would be important to set out the analogues and homlogues of each. What are the commonalities?

  14. Fiend Ludwig

    Feb 28th, 2007

    @Andy – Totally agreed with your whole post. As I said in my article regarding the gender observations, – “You [Yee] note that you ‘didn’t belabor this point’, but it seems to me the crux of the study.” I also said “I believe that there are ways to correlate VW and real world research, but useful data will only be available once more effective means of collecting it are established.” You have pointed out several much more relevant ways to establish these connections.

  15. Fiend Ludwig

    Feb 28th, 2007

    @Andy – correction: the quotes I noted above are from my response comment to Nick Yee’s feedback, not the original article.

  16. Andy

    Feb 28th, 2007

    Yes, thanks Fiend.

    Guess there is a bandwagon tendency in a lot of this stuff. Often they are just muddying the waters rather than bringing clarity.

    People seem to be walking right out of their own disciplines into areas where they know little or nothing and asking lots of questions without getting the background thinking basics, which may be well established though not necessarily within the terminological sphere they are use and understand. That is, they may need, for example, crash courses in social psychology, game theory and so on.
    I think everyone should read the Extended Phenotype!
    In my view it is the philosophy first: what can you ask? Then the science. The logic and programming just go on all by themselves! If you can create the software and it works, people use it often in imaginative ways the designers hadn’t thought of. An example: plagiarism software is used by many higher education establishments to deal with the cut and paste culture of the modern student. But it can also be used by anyone with online content to see how his/her stuff is spreading online.
    The debate above has highlighted the fact that a cyberworld is a restricted model of RW. There may come a time when RW=VR. That’s when they won’t need Hollywood superstars!

    The business of whether we can lump together RW and online world (are there any things in OLW which are extra, over and above RW; is it a metaphorical web or not? ,and so on, need to be carefully analysed.)
    There are many other areas which could help such as value networks, and aspects of game theory.

    This 1996 piece: Identity and Deception in the Virtual Community

    < http://smg.media.mit.edu/people/Judith/Identity/IdentityDeception.html#30835 >

    looks as if it might answer many of the half-formed questions buzzing around…..

    Even ideas like :

    Twelve Virtues of Rationality

    < http://yudkowsky.net/virtues/ >

    has a few pointers such as the Saint-Exupery quote:

    “Perfection is achieved not when there is nothing left to add, but when there is nothing left to take away.”

  17. Nick Yee

    Mar 3rd, 2007

    Fiend, Andy, and Maria touch on the “mechanism” question on the gender-bending issue. This is something we explore in greater detail in the paper I linked earlier, repeated again here:

    What we’ve found in these experimental studies (as opposed to the observational study reported here) is that people do conform to traits of their avatars very quickly (within a 1-3 minute timeframe).

    Of course, it’s impossible to elucidate all aspects of personal space and virtual gender-bending with one study, and this wasn’t the purpose of this paper. And much as Andy is suggesting, we are approaching this issue from many angles, with several methodologies. Seldom does one study provide anything definitive. It takes time and a series of studies to figure out what is going on and you can get a sense of that by looking through other studies we have done: http://vhil.stanford.edu

  18. Nick Yee

    Mar 3rd, 2007

    I guess the other thing I want to point out is that oftentimes we may not know the exact mechanisms that drive observed behaviors, but that doesn’t mean the observed behaviors don’t exist.

    In other words, while it is true that avatars in SL can IM with others 30 feet away and their avatars may not be facing what they’re paying attention to, we’re still left with these questions among others:

    - Why is it that male avatars stand further away from other male avatars than from female avatars?
    - Why is it that male avatars maintain less eye contact with other male avatars than with female avatars?

    And while some of the factors others have brought up clearly add to the noise in the data, and we clearly don’t understand everything about personal space and gaze in Second Life, that the most parsimonious epxlanation for our findings is that some social norms from our physical world transfer into virtual worlds.

  19. Economic Mip

    Mar 3rd, 2007

    Well Nick I can try to answer several of those questions:
    1. Why do guys stand farther away from each other?
    - It is not as much about personal space as it is a desire not to be in a situation where the other individual feels threatened…
    2. Why don’t guys look the other people in the eyes while talking? This one is easier. Try to get a male avatar to look where you want it too using only a mouse. Chances are it will occur 10% of the time, and then only after sending you into several animations that look like epileptic shocks, or worse. Part of it could very well be that the avatars come with preset natural reactions to certain things. Guys bump into something, they take a step back. And part of it is just the imperfections of a virtual world.

  20. Urizenus

    Mar 3rd, 2007

    Nick, I think that Fiend should put his point this way:

    Yes, it is a really interesting question as to why male avatars establish greater distance than female avatars and an equally interesting question as to why male avatars maintain less eye contact. But how do we get from those very interesting questions to the very strong conclusion that we can extrapolate results about the behavior of different genders (or other groups) in RL from the analogous presented differences online?

    The questions that Fiend is raising all go to the following point: the underlying causal mechanisms involved in RL gaze and distance are radically different from the underlying causal mechanisms involved in the avatar gaze and distance. So extrapolating strong conclusions here just doesn’t fly — particularly extrapolating conclusions that generalize to classes of behavior that we haven’t even studied in a virtual setting yet.

    It seems to me that there are numerous explanatory possibilities that haven’t been explored yet, not least of which is that distance and gaze are easily manipulated elements of gender roleplay (as, I suppose, voice pitch modulation would be). There are many other facts about RL gender difference that are not so obvious and easily mimicked, so it would be careless to think that online gender differences will encode those facts, that is all.

  21. Nick Yee

    Mar 6th, 2007

    Mip – Your answer to the distance issue is also partly why men stand further apart in real life. This is congruent with our claim that norms transfer from RL to SL. Your answer to the gaze issue might explain a general gaze difference in male avatars, but what we found was a gendered gaze difference. Why do male avatars maintain gaze less with other MALE avatars than FEMALE avatars?

    Urizenus – I think your comments are a rephrasing of the observed findings vs. mechanics issue. What we found was that many RL norms seem to reappear in SL. We actually don’t argue for why this happens. Now, this may be because social norms are engrained, because people are role-playing, or other factors, but we don’t make any claims to the causal mechanism. Our main claims from the paper, supported by a large data set, was that several social norms related to personal distance and gaze transfer to SL. We don’t claim that all RL norms transfer. We don’t claim we know precisely why this happens. And this was what the press reported on. Male avatars in SL stand further away from other male avatars than they do from female avatars (among several other findings).

    You say that these classes of behavior haven’t even been studied in virtual worlds yet, but that’s actually what we did. We were studying personal space and gaze in SL. The findings could have turned out very different from RL norms, half-half, or produced findings that required a new theory of social interaction, but they didn’t. They came out congruent with RL norms.

    Your comments also point to the most interesting aspect of the findings. SL users use the mouse and keyboard to move their avatars, so there’s no reason why SL avatars would follow RL norms, and yet we still find from a large data set that SL avatars behave as if they were following RL social norms. We didn’t extrapolate strong conclusions. We simply stated what we found. Personal space norms in SL seem to mirror RL norms.

  22. Urizenus

    Mar 6th, 2007

    Nick, the problem is when you move from the observation that social norms of distance and gaze transfer to SL to claims like this:

    “our findings support our overall hypothesis that our social interactions in online virtual environments, such as Second Life, are governed by the same social norms as social interactions in the physical world. This finding has significant implications for using virtual worlds to study human social interaction. If people behave according to the same social rules in both physical and virtual worlds even though the mode of movement and navigation is entirely different (i.e., using keyboard and mouse as opposed to bodies and legs), then this means it is possible to study social interaction in virtual environments and generalize them to social interaction in the real world.”

    That is the quote that Fiend led with above and I guess I agree with him that it is *way* too strong given the evidence on the table, which just speaks to distance and gaze. How do we get from that to the more general claim that

    “our social interactions in online virtual environments, such as Second Life, are governed by the same social norms as social interactions in the physical world”


    “this means it is possible to study social interaction in virtual environments and generalize them to social interaction in the real world”?

    It doesn’t mean anything of the sort. That would be a grand generalization from a narrow class of facts relating to gaze and distance. There is no prima facie reason to expect similar results with respect other social facts (especially since you concede that the causal mechanisms are different). It’s perfectly fine to offer a hypothesis that this pattern will recur with other social facts, and it will be interesting to test it, but it is a mistake to conclude that “it is possible to study social interaction in virtual environments and generalize them to social interaction in the real world”. It is way too early to *conclude* that, however.

    Finally, we don’t actually know that the gaze and distance results are because of the same norms at all. The online behavior might not be norm goverened at all, or the relevant norm might be be “mimick RL as much as you can” — in effect the same norm that applies when people make their guns and prim hair look as realistic as possible, which is not the same norm as “respect personal space.”

  23. Nick Yee

    Mar 6th, 2007

    Urizenus – Thanks for your clarification. I’m much more willing to agree to the strong claims critique and indeed in many of the media interviews I’ve noted that more foundational studies are needed to see what does and doesn’t transfer.

    On the other hand, I think the norm transfer hypothesis is the most parsimonious explanation for our set of findings currently. Sometimes, people interact in SL the ways they are used to interacting in RL. The problem with the “mimicry” hypothesis is that most people aren’t consciously aware of personal space and eye gaze norms. Also, saying that people “mimic RL norms” is incredibly close to saying people “follow RL norms” in SL. There will always be alternative explanations for every observed phenomenon – whether this is global warming, or interpersonal distance in SL. Only future studies can test out the alternatives, and unfortunately science seldom provides instant gratification or conclusive answers.

  24. Helge Staedtler

    Mar 10th, 2007

    I recognized the discussion here for a while now. Reflecting on the arguments I have read, I come to the conclusion, that the link Nick made (SL –> RL) for behaviours is a rather weak one.

    I think the opposite is quite more appealing to be researched. I think I would try to create some experiment where I show people Screenshots of people in SL standing around and let them evaluate and describe the situations seen. I think the same way as Prokofy Neva who stated “Avatars can’t see!”. I think this is really the point here: People are looking towards a screen and they arrange some figure of a human. So they place this figure according to their specific understanding of what kind of proximity might be adequate.

    This is the closest you can get to evaluating behaviour of people in SL, I think. There is no “Me having a valid gaze.” there is in the best case a Screen on which I arrange an avatar to match certain criteria I have on my mind to reach certain things.

    As of the gender issue… I also wrote some words over on my blog mentioning “the Turing Game”.
    details here: http://www.ifeb.uni-bremen.de/wordpress_staedtler/?p=108

    It might be possible, that you could resolve the gender-issue completely by just analyzing the chat-transcripts. So the gender issue might perhaps be completely resolved, as experiments in the Turing Game have showed people are pretty good at finding out about the others’ gender, so why should a group of scientists e.g. be worse at this question.

    regards, Helge

  25. Population of One

    Aug 15th, 2007

    The unbearable likeness of being digital

    I came across a discussion of the paper The unbearable likeness of being digital: The persistence of nonverbal social norms in online virtual environments, by Nick Yee, Jeremy N. Bailenson, Mark Urbanek, Francis Chang and Dan Merget, published in Cyber…

Leave a Reply