Robert Kurzban

The Evolutionary Psychology Blog

By Robert Kurzban

Robert Kurzban is an Associate Professor at the University of Pennsylvania and author of Why Everyone (Else) Is A Hypocrite. Follow him on Twitter: @rkurzban

Are All Dictator Game Results Artifacts?

Published 22 May, 2013

You walk into a laboratory, and you read a set of instructions that tell you that your task is to decide how much of a $10 pie you want to give to an anonymous other person who signed up for the experimental session.

This describes, more or less, the Dictator Game, a staple of behavioral economics with a history dating back more than a quarter of a century. The Dictator Game (DG) might not be the drosophila melanogaster of behavioral economics – the Prisoner’s Dilemma can lay plausible claim to that prized analogy – but it could reasonably aspire to an only slightly more modest title, perhaps the e. coli of the discipline. Since the original work, more than 20,000 observations in the DG have been reported.

The first thing to be careful to mark about the DG is that running such a game does not, of course, in itself, constitute an experiment. The Dictator Game, in isolation, is better understood as a measuring tool. It measures, in some sense, generosity. If you like to think that people have “(pro-)social preferences,” – that people are happy when others are happy (as I am) – then measuring how much money a person gives to another individual in the carefully controlled setting of the DG is one way to assess the strength of this preference.

The DG can be, and has been, used as a control condition. So, for instance, early work used the DG as a control condition for the Ultimatum Game (UG), which is like the Dictator Game except that the first player’s proposed split can be rejected by the second player, in which case both get nothing. Comparing the DG and UG, so the logic went, allows the experimenter to measure how much more people propose in the UG because of the fear of the offer being rejected as opposed to, loosely, simply being generous. The DG, in this sense, controls for pro-social preferences as an explanation in the UG.

All but one implementation of the Dictator Game, as far as I know, until recently, have had something potentially important in common: the people playing the game know that they were in an experiment. This circumstance, of course, is very difficult to avoid, and obviously this claim can be made of the vast array of methods used across the experimental social science literature. To be sure, conducting experiments on subjects without them being aware of the fact that they are in an experiment is difficult, not to mention potentially unethical. Still, some of us have wondered for some time if there is something special about this fact when it comes to the Dictator Game. After all, we each are in Dictator Games every time we walk into the world with money in our pockets. We could, if we wished, split the cash we’re carrying with a stranger in the street. My casual observation of my own behavior suggests that if my pro-social preferences were measured this way, I would come across as stingy indeed or, as economists would have it, “rational.”

Further, there is something pragmatically odd about the “game.” (Even Wittgenstein might have balked at calling the Dictator Game a “game,” and he famously cast the net for this term broadly indeed.) Subjects in this experiment are faced with one decision, one lever to press, or not press, as it were. Subjects who give nothing are in the peculiar position of coming to a lab, reading instructions, and then, roughly, doing nothing at all and collecting their loot. It’s easy to imagine that subjects think there’s all something funny about this, and feel some urge to push on the single lever the experimenter has given them to play with, if only a little bit.

This raises the question: How much would participants in a Dictator Game give to the other person if they did not know they were in a Dictator Game study? Simply following me around during the day and recording how much cash I dispense won’t answer this question because in the DG, the money is provided by the experimenter. So, to build a parallel design, the method used must move money to subjects as a windfall so that we can observe how much of this “house money” they choose to give away.

And that is what Winking and Mizer did in a paper now in press and available online (paywall) in Evolution and Human Behavior, using participants, fittingly enough, in Las Vegas. Here’s what they did. Two confederates were needed. The first, destined to become the “recipient,” was occupied on a phone call near a bus stop in Vegas. The second confederate approached lone individuals at the bus stop, indicated that they were late for a ride to the airport, and asked the subject if they wanted the $20 in casino chips still in the confederate’s possession, scamming people into, rather than out of money, in sharp contradiction of the deep traditions of Las Vegas. The question was how many chips the fortunate subject transferred to the nearby confederate.

Figure 1 from Winking & Mizer, in press.

Figure 1 from Winking & Mizer, in press.

And the authors didn’t choose Las Vegas because they thought it would be a nice place to take a few trips to conduct field research. You couldn’t run the experiment easily just anywhere, with dollars, because the cover story wouldn’t work – dollars travel too well. The nice thing about Vegas is that casino chips might plausibly be given away in this fashion because the chips have value only in Vegas. (Probably a dedicated person could figure out a way to exchange the chips for cash through eBay or some such – or save them for a subsequent visit – but giving away forgotten chips when one is leaving Vegas is plausible enough.)

In a second condition, the confederate with the chips added a comment to the effect that the subject could “split it with that guy however you want,” indicating the first confederate. This condition brings the study a bit closer, but not much closer, to lab conditions, In a third condition, subjects were asked if they wanted to participate in a study, and then did so along the lines of the usual DG, making the treatment considerably closer to traditional lab-based conditions.

The difference between the first two treatments and the third treatments is interesting, but, as I said at the beginning, the DG should be thought of as a measuring tool. Figure 1 shows how many chips people give away in the DG in the three treatments. In conditions 1 and 2, the number of people (out of 60) who gave at least one chip to the second confederate was… zero. To the extent you think that this method answers the question, how much Dictator Game giving is due to people knowing they’re in an experiment, the answer is, “all of it.”

Of course this is one study one population and it would be hasty to made too much out of it. Still, if the results of this study in the West is telling us what people think they should do when they find themselves in a study on altruism, then it could be that the cross-cultural data using the DG are telling us what people across cultures think they should do when they find themselves in a study on altruism.

 Citation

Winking, J., & Mizer, N. (in press). Natural-field dictator game shows no altruistic giving. Evolution and Human Behavior

  • JaziZilber

    The cited study as well as the exmples from our usual not ghrowing money are equally irrelevant to DG.

    It needs to be a case that the money was given for both, and the dictator has the decision power only. Then you can ask how much he gives, because its obviously fair to give the other guy too.

    In the las vegas case, it is about giving money to strangers free of charge or reason. A different question altogether

    • rkurzban

      Jazi, Thanks for taking the time to comment. I’m finding it hard to craft a reply because I don’t understand what you’re trying to say. For one thing, it depends what you mean by the claim that “the money was given for both.” In typical DGs, the subject learns that they are “paired” with someone in the experiment. But subjects are typically not instructed that the money is “for” the two of them. Instead, they are given the option to give to the other person. I don’t understand the rest of your comment sufficiently well to reply to it.

      • JaziZilber

        First of all I define how I see it. Second, i will outline how I can see it in the experiments etc.

        1) DG does not try to check if people give their money away. even the name dictator indicates someone who decides how to split a common resource. the resource.

        Only morons (or Jesus) give away their money. This is common knowledge.

        DG takes a situation, where the natural inference is that the money is originally destined to both. The decider (dictator) does not have any superior rights to the money. He has, however, the decision role. He can decide.

        2) How this is reflected in the experiments.
        Here is the original text given to subjects in DG
        ““You will be matched at random with two other students, and you will get to share some money with one or both of them. If the two people made different decisions in the first stage (e.g. one of them took $10 and one took $18), then you must make a decision about how to allocate the money. Call the person who took $10 and gave the other one $10 student E (for even). Call the person who took $18 and gave the other one $2 student U (for uneven). Your choices are as follows: you may allocate $5 to yourself, $5 to student E, and nothing to student U; or you may
        allocate $6 to yourself, nothing to student E, and $6 to student U”.
        When two people enter a lab and there is a sum of money given to “split as you wish” it seems implied that the money is not destined to the decider to begin with. he is told to split this money as he wishes. i understand that the dictator is not told “this is your money, but you have the option to give part of it to another.”

        ((Kahneman, Knetsch et al. 1986:S290 f.)

        it is a matter of framing. and the framing seems to be of common funds, or funds destined to be splitted between both some way or another.

    • GoodsJL

      Hit the nail right on the head, strangers. Does not capture the condition of living in a community and the related reputation concerns. A good way researchers could tease out a result with DG is to expose test subjects to others at different frequencies. You would think the amount donated would vary with the probability you are going to have to get along with someone. Didn’t Steven Levitt in Freakonomics show that real-world DG donations depend on whether some is watching, i.e. depends on reputation?

  • http://popsych.org/ Jesse Marczyk

    Makes one wonder about the extent of punishment as well – both in the second and third party contexts – though I imagine crafting an experiment to deal with that would be…tricker

  • Pingback: Are All Dictator Game Results Artifacts? | Evo...

  • Nicolas Baumard

    Wonderful! That is such a useful reminder of the artificial nature of economic games!

  • carmiturchick

    Interesting. It strikes me though that the critical missing component is need. Altruism is not about giving to random strangers in a completely selfless way, it is about giving to others who are in need of help of some kind. Giving randomly is not altruism, so it is odd that DG and UG have been used to examine altruism, since they do not even if they are not artifacts.
    Run this same Vegas experiment with the second person on the phone loudly talking about how his wallet was just stolen and he does not know what to do, and then we will have evidence relating to altruism without the possibility of it being an artifact.

  • Pingback: A week of links - Evolving Economics

  • Victor Venema

    I agree with JaziZilber, this is a very different situation.

    In the DG you are *paired* with another person. This person also went to lab, knows what is happening and will go back home with a bad feeling if he does not get any money.

    In the Las Vegas experiment, the people are not paired. The Dictator expects that the that the other person does not even know what is happening and thus also does not expect any bad feelings by the second confederate. The framing is similar to giving money to a random person, which no one expects people to do.

    JaziZilber: “Only morons (or Jesus) give away their money. This is common knowledge.”
    carmiturchick: “Run this same Vegas experiment with the second person on the phone loudly talking about how his wallet was just stolen and he does not know what to do, and then we will have evidence relating to altruism without the possibility of it being an artifact.”

    People give to people in need all the time, just not to random people. In the framing of carmiturchick with someone in need, I would not be surprised if many people would give all the chips.

  • Wojtek Przepiorka

    Interesting indeed! But I would not conclude that all DG findings from lab experiments are
    artifacts. The main reason for my not sharing Robert’s conjecture is that results from a single field experiment (in addition, one in a very particular environment) will not be more externally valid than results from a lab experiment (see Falk and Heckman 2009, Science). Moreover, as Robert points out, the DG is often used to measure subjects’ pro-social preferences in a specific context (such as the lab) so it can be used as an explanatory variable for other behaviour measured in the same context. And that context and framing matter a great deal, we do know already for some time (see Liberman, Samuels and Ross 2004, PSPB). Finally, every tip left on the table of a highway diner is evidence for people’s unconditional generosity. Happy to discuss this further…

    • Lee_Kirkpatrick

      Tipping in places you never expect to revisit is an interesting phenomenon in its own right, but I think it can hardly be construed as “evidence for people’s unconditional generosity.” Tipping is very much conditional (on having received a service), so it seems more closely related to issues of social exchange or reciprocity. In the U.S., restaurant tipping is more or less equivalent to payment for service; indeed, some restaurants, under some conditions (e.g., large parties) automatically charge a “gratuity” to the bill. Failing to leave a tip at a restaurant in the U.S. is only a small step away from walking out without paying the bill when the cashier isn’t looking. In places like Europe where tipping is considered optional, it is still regarded as conditional on having received exemplary service (i.e., beyond what is expected in exchange for the amount paid for the meal). Now, if people commonly left tips in highway diners for waitpersons other than the ones who had served them, that would be pretty compelling evidence for unconditional generosity. But how often does that happen?

  • Pingback: Assorted links

  • Kanye West
  • Pingback: Dictator Game Results Artifacts? Duh? | Michael Kevane

  • Henz

    I am not an research psychologist. However, I regularly participate in psychology experiments (and surveys) online. These often have the incentive of a small cash prize or voucher.

    On completing one particular online survey (run each year by a well known UK University) participants are asked whether they would like to enter a draw for the full cash prize or offer the entire cash prize to a charity of their choice.

    In these cases the DG could be run rather simply, with large numbers of participants, over multiple surveys, by adding the option for the participant to split their cash prize with the charity. As described above the DG requires confederates and the participant(s) to be present in person, the survey version removes this requirement.

    There are obvious limits when compared with the standard DG: there is no ‘paired’ individual, the recipient is impersonal; the value of the ‘pie’ is unknown because the likelihood of winning is unknown; the ‘pie’ is not in the possession of the participant when they decide how to split it; it is unknown whether the participant will choose to split the ‘pie’ after the DG (i.e. if they choose to take the full amount and then win the ‘pie’). However, I expect someone with more experience in the subject area could put this to practical use.

  • rorysutherland

    Surely the very fact that the experiment took place in Las Vegas invalidates the result? Since we are repeatedly told that “what happens in Vegas stays in Vegas”, presumably there is no reputational carry-over from your behaviour in that city to the rest of your life. Perhaps that is the point of Las Vegas – it is a place you go to find anonymity and to escape the usual moral and ethical sanctions you experience everywhere else – to act like a homo economicus for a few days, in other words.

    And why was the second confederate on the phone? First of all, his wilful inattention disqualifies him from reward; second, especially in a place like Vegas, going up to random strangers and handing them money for what seems a highly contrived reason is likely to arouse suspicion as to your motives or sexual orientation.

  • Pingback: Some Links | Meta Rabbit

  • http://psychologydegreesx.com/ psychology degrees online

    Real good info

  • Pingback: From the bookmarks « Econstudentlog

  • Pingback: neuroecology

  • Kim

    Rob, I think this experiment takes place naturally thousands of times a day in real human life. People obtain winfalls. Do they keep them all for themselves — almost never. This is of course my casual observation, but an observation that I have made from modern Americans to hunter-gatherers. The disjunct between ethnographic reality and the results in Las Vegas should lead to deep thinking about the difference in experimental design. I once gave an Ache hunter-gatherer a sum of cash to spend because he had come with me to the city for the first time in his life. A half hour later I asked him to pull out his cash and pay for his lunch and he sheepishly replied that he didnt have much left. How could this be true I asked, we havent even been in a store yet. “I gave some to that guy back there on the street that was begging” was the reply. WTF! He was a random stranger not even from the same ethnic group. He just looked needy. I think your LV experiment is far from giving us a deep understanding of human prosociality.

  • Andreas Friedl

    Dear Sir,
    thank you for your article. I want to make two comments.

    To the study: the games (lab – field) are rather different.

    most important: in the Lab DG both participants know the rules of the game. If
    the other guy doesn’t know i am the
    dictator, and its likely that he will never know, i have less incentive
    to give. (not out of altruism but of following norms – what you don’t know won’t hurt you)

    to your conclusion: Subjects are not told that they are part of a altruism study. Rather they behave in a standardized and as abstract as possible environment.

    If i exchange “study on altruism” with “abstract situation” in your last paragraph this would read:

    “(These studies…) telling us what people think they should do when they find themselves in
    a abstract situation, then it could be that the cross-cultural data
    using the DG are telling us what people across cultures think they
    should do when they find themselves in an abstract situation.”

    A) no experimenter would claim anything different.
    B) don’t you think this is worth doing it?

    Best regards
    Andreas Friedl

  • http://gedcertificateclassesonline.com/ Gary Dean

    I have to agree with, using something other then money in this experiment would solidify the outcome. http://psychologydegreeclassesonline.com/

  • Brooke Oneill

    I’m late to this thread, and probably no one will ever read this, but I just have to get this off my chest, because the topic of this post seems to relate to something that has always nagged me about several well known experiments. E.g., take the ‘free will’ experiments:

    “In 2007, Haynes, a neuroscientist at the Bernstein Center for Computational Neuroscience in Berlin, put people into a brain scanner in which a display screen flashed a succession of random letters. He told them to press a button … whenever they felt the urge, and to remember the letter that was showing on the screen when they made the decision.” http://www.nature.com/news/2011/110831/full/477023a.html

    Nominally, the experimenter simply ‘informed’ the subjects that they were ‘free’ to push the button, if they got the urge to do so. But, in reality, given the experimental context, and given their status as subjects, the subjects would have understood the experimenter to be instructing them to push the button. But, of course, the experimenter qualified his instructions by as good as telling the subjects that he wanted them to delay pushing the button for an interval. However, he didn’t tell the subjects how long that interval should be. Rather, he told them to delay pushing the button until a time of their own choosing–i.e., whenever they got the ‘urge’.

    After delaying for an interval, the dutiful subjects formulated a subconscious intention to push the button several seconds before they were consciously aware of having decided to do so (or, consciously aware of having an “urge” to do so).

    “The results were quite a surprise…. The conscious decision to push the button was made about a second before the actual act, but the team discovered that a pattern of brain activity seemed to predict that decision by as many as seven seconds. Long before the subjects were even aware of making a choice, it seems, their brains had already decided. ”

    The researchers argue

    “that consciousness of a decision may be a mere biochemical afterthought, with no influence whatsoever on a person’s actions. According to this logic, they say, free will is an illusion. ‘We feel we choose, but we don’t,’ says Patrick Haggard, a neuroscientist at University College London.”

    I can’t understand how they reached that conclusion. There’s little doubt that the subjects wouldn’t have formulated a subconscious intention to push the button if they hadn’t first formulated a conscious intention to push the button.

    Indeed, has any subject ever NOT pushed the button? That is, has any subject ever reported, “Gee, I just never got the urge”? I doubt it. If I’m right, it’s because every single subject formulated a conscious intention to push the button when they were in the brain scanner long before they ever climbed inside. Their conscious decisions weren’t ‘afterthoughts.’ Rather, their subconscious decisions were. Their subconscious minds were just carrying out a conscious decision the subjects had made well beforehand. To put it another way, all I can conclude from the experiment is that the conscious mind enlists the subconscious mind to do its bidding (at least some of the time, anyway).

    Robert Kurzban writes in the post above:

    “… there is something pragmatically odd about the ‘game.’… Subjects in this experiment are faced with one decision, one lever to press, or not press, as it were. Subjects who give nothing are in the peculiar position of coming to a lab, reading instructions, and then, roughly, doing nothing at all and collecting their loot. It’s easy to imagine that subjects think there’s all something funny about this, and feel some urge to push on the single lever the experimenter has given them to play with, if only a little bit.”

    I think the subjects in the ‘free will’ experiment would feel even more odd about being in the experiment and doing nothing (i.e., not pushing the button). The subjects knew they were taking money from the experimenter in exchange for giving him some ‘data’–which, to their minds, would require them to DO something for him. Since they know very well that what he wants them to do is push the button … they push the button.

    The bottom line is that experimenters can’t simply abstract away the experimental context when they are interpreting their results. Subjects’ knowledge that they are in an experiment is a variable that can’t easily be eliminated–or ignored.

  • David Rand

    There is a beautiful new paper (which has been getting a lot less coverage than WInking & Mizer) by Franzen and Pointner that shows that the DG does predict real giving, in a really nice paradigm:

    The external validity of giving in the dictator game

    We investigate the external validity of giving in the dictator game by using the misdirected letter technique in a within-subject design. First, subjects participated in standard dictator games (double blind) conducted in labs in two different studies. Second, after four to five weeks (study 1) or two years (study 2), we delivered prepared letters to the same subjects. The envelopes and the contents of the letters were designed to create the impression that they were misdirected by the mail delivery service. The letters contained 10 Euros (20 Swiss Francs in study 2) corresponding to the endowment of the in-lab experiments. We observe in both studies that subjects who showed other-regarding behavior in the lab returned the misdirected letters more often than subjects giving nothing, suggesting that in-lab behavior is related to behavior in the field.

    http://link.springer.com/article/10.1007%2Fs10683-012-9337-5

    • Lee_Kirkpatrick

      This is a nice little piece of research (thanks for the heads-up, David), but I think it is largely irrelevant to the main points of discussion here. This study shows that responses to a laboratory dictator game have validity as a measure of individual differences in generosity or altruism (or something along those lines). But that’s a very different kind of question than the ones raised by the Las Vegas study and most of the discussion here, which concern whether the overall or average levels of altruism/generosity typically observed in laboratory dictator games is a valid reflection of people’s overall or average level of generosity/altruism in “real-life” settings — as well as questions about why these observed levels in the lab are as high as they are (e.g., whether they can be attributed to expected reputational benefits or demand characteristics). There is no reason why a given task couldn’t be a perfectly valid measure of individual differences on some trait, while at the same time being utterly invalid as an estimate of the average level of that same trait. Or to put it more technically, there is no reason why the correlation between two measures can’t be very high despite their means being very different.

  • Pingback: Are All Dictator Game Results Artifacts? | El Pla Subtil

Copyright 2013 Robert Kurzban, all rights reserved.

Opinions expressed in this blog do not reflect the opinions of the editorial staff of the journal.

Evolutionary Psychology - An open access peer-reviewed journal - ISSN 1474-7049 © Ian Pitchford and Robert M. Young; individual articles © the author(s)
Close


You're in!