- Email: [email protected]
Social Influence in the Sequential Dictator Game
Journal of Mathematical Psychology 42, 248265 (1998) Article No. MP981213
Social Influence in the Sequential Dictator Game 1 Timothy N. Cason and Vai-Lam Mui University of Southern California
This paper introduces the sequential dictator game to study how social influence may affect subjects' choices when making dictator allocations. Subjects made dictator allocations of 840 before and after learning the allocation made by one other subject in the Relevant Information treatment, or the birthday of one other subject in the Irrelevant Information treatment. Subjects on average become more self-regarding in the Irrelevant Information treatment, but observing relevant information constrains some subjects from moving toward more self-regarding choices. We also find that subjects who exhibit more self-regarding behavior on their first decisions are less likely to change choices between their first and second decisions, and the use of the Strategy Method in this setting does not significantly alter choices. The relationships between our findings and the economic and psychological literature regarding how social influence operates are also explored. 1998 Academic Press
1. INTRODUCTION
This paper introduces the sequential dictator game to study how social influence may affect subjects' choices when making dictator allocations. In the dictator game, each subject is paired randomly and anonymously with another subject, and one is selected to ``dictate'' the allocation of a fixed amount of money. In the sequential dictator game, subjects make two dictator decisions, and they receive information regarding another subject's dictator allocation before making their second decision. When making these two decisions, subjects are not making two dictator offers to the same person. If a subject's only goal is to maximize her earnings, then she should allocate the entire surplus to herself. However, there is now overwhelming evidence that many subjects give away a substantial share of the surplus. 2 Different explanations have 1
We thank Matt Grund for research assistance and the USC Zumberge Faculty Research and Innovation Fund for financial support. We are also grateful to Gary Bolton, James Konow, Robert Slonim, two anonymous referees, and participants at the Fall 1996 Economic Science Association conference for valuable comments. We retain responsibility for any errors. 2 For two surveys of the early literature on the dictator game and the related ultimatum game, see Camerer 6 Thaler, 1995 and Roth, 1995; for more recent contributions see Hoffman, McCabe, 6 Smith, 1996; Bolton, Katok, 6 Zwick, 1997; and Cason 6 Mui, 1997. 0022-249698 25.00 Copyright 1998 by Academic Press All rights of reproduction in any form reserved.
248
SEQUENTIAL DICTATOR GAME
249
been advanced for this other-regarding behaviorfor example, the fairness hypothesis (Forsythe, Horowitz, Savin, 6 Sefton, 1994) and the observer effect hypothesis (Hoffman, McCabe, Shachat, 6 Smith, 1994)and these explanations typically employ the idea that subjects care about both their monetary earnings as well as about behaving in a socially appropriate way. 3 At a general level, these explanations are based on the idea that subjects are maximizing a utility function that includes both a monetary payoff and a social payoff, and this social payoff depends on the extent to which subjects' choices conform to or deviate from what they believe to be socially appropriate behavior. While a subject's belief regarding what constitutes socially appropriate behavior will depend on her personal characteristicsher experience, cultural background, personalityit is plausible that this belief may also depend on her estimate of others' beliefs regarding what constitutes socially appropriate behavior. Therefore, when a subject is exposed to socially relevant information that may lead her to revise her belief, her choice may change. This paper reports an experiment that provides some preliminary tests of this social influence hypothesis. More specifically, we investigate whether learning the dictator allocation made by another subject in the same experimental session affects subjects' choices. Other researchers have also conducted experiments in which subjects receive information about others' choices. For example, several recent public goods laboratory studies have examined whether group contribution levels change when subjects receive information about the previous contributions of all individual subjects in their group (Sell 6 Wilson, 1991; Weimann, 1994; and Croson, 1995). 4 These experiments capture natural settings in which social influence may be significant. However, in these experiments a subject's monetary payoff depends on both her own choice as well as others' choices. The presence of strategic interdependence creates two difficulties for our purpose of testing for the impact of social influence. First, even if one observes that a subject's behavior changes upon observing other subjects' choices, it may be difficult to discern to what extent these changes are due to social influence and to what extent they are caused by the fact that such information affects her strategic calculus when attempting to increase her monetary payoff. Second, when strategic interdependence is present, subjects' beliefs regarding what constitutes socially appropriate behavior may be influenced by the strategic environment itself. Roth (1995) reviews recent efforts in studying how social norms regarding what constitutes fair behavior in bargaining games can be created and enforced. He observes that ``(a)lthough subjects may have clear ideas about what is 3
According to the fairness hypothesis, subjects are trying to behave fairly by taking what they consider to be their fair share. This hypothesis implies that subjects' choices are sensitive to their beliefs regarding what is a fair allocation. According to the alternative observer effect hypothesis, subjects may be concerned about whether their allocations are judged as unfair by the experimenter, so the experimenter as observer could increase other-regarding behavior. This hypothesis implies that subjects' choices are sensitive to their expectation regarding what is perceived to be socially acceptable to the experimenter. 4 In a related public goods experiment, Andreoni (1995) finds that providing subjects with information regarding how their earnings rank in comparison to the other subjects in their session decreases cooperation. In a ``trust game'' experiment, Berg, Dickhaut, 6 McCabe (1995) finds that providing subjects with the full distribution of outcomes from previous sessions does not substantially affect their choices.
250
CASON AND MUI
fair in a variety of circumstances, and although these ideas about fairness may influence the strategic environment, the evidence suggests that subjects adapt their ideas about what is fair in response to their experience, in ways that may be heavily influenced by strategic considerations.'' (Roth, 1995, p. 271). Therefore, while social influence may arise naturally in many environments in which strategic interdependence is important, in order to better isolate the impact of social influence it is useful to ``go one step backward'' to study this issue in a simple environment in which strategic considerations are absent. Our experimental design includes two main treatments. In both treatments, subjects first chose an initial dictator allocation with no information other than the instructions. The Irrelevant Information treatment serves as a baseline control, in which subjects receive socially irrelevant information between their two dictator decisions. In the Relevant Information treatment, each subject learns the first dictator allocation chosen by one other subject. Each subject then specifies a single second dictator allocation to another subject who is different from the subject affected by his initial dictator allocation. Our major findings are as follows. Subjects on average become more self-regarding in the Irrelevant Information treatment, but do not become more self-regarding in the Relevant Information treatment. We also find that subjects who are more self-regarding on their first decisions are less likely to change choices between their first and second decisions.
2. EXPERIMENTAL DESIGN AND PROCEDURES
Each subject made two decisions sequentially that corresponded to a dictator allocation of 840. We employed the exchange framing for these decisions, in which the decision-maker is a seller choosing a price that ranged between 0 and 40 dollars (see Hoffman et al., 1994, for an experiment that examines the importance of this framing). The buyer was required to buy at that price. If the seller chose a price of P dollars, the buyer received 40-P dollars while the seller received P dollars, so the seller dictated the allocation of the 840 surplus. Notice that when a seller chose a higher price he is making a more self-regarding choice. In all cases price choices were restricted to two-dollar increments (i.e., 80, 82, 84, ..., 838, 840). Each session proceeded as follows: 1. All subjects chose an initial ``first price'' (P 1 ) with no information other than the instructions. In the Irrelevant Information treatment, subjects also indicated on their choice form the day of the month on which they were born (i.e., an integer between 1 and 31). 2. The experimenter drew bingo balls from a cage to assign subjects into groups of four. For ease of exposition, call a typical group of four subjects 1, 2, 3, and 4. 3. The experimenter transferred information within the odd-numbered and within the even-numbered subjects in each group. In the Relevant Information
SEQUENTIAL DICTATOR GAME
251
treatment, subject 1 learned the initial price chosen by subject 3, and subject 3 learned the initial price chosen by subject 1. Information was similarly transferred between subjects 2 and 4. In the Irrelevant Information treatment subjects instead learned the birthday of this other subject. 4. All subjects then chose a ``second price'' (P 2 ). 5. Again using the bingo cage, the experimenter randomly determined whether the odd-numbered subjects or the even-numbered subjects in each group would be sellers, as well as which of these two selected sellers would be the ``first seller'' and which the ``second seller.'' For example, if the odd-numbered subjects were selected as the sellers and if subject 1 was selected as the first seller, then subject 1 received his indicated first price P 1 while his paired buyer subject 2 received 40-P 1 . In this case subject 3 must be the second seller, so subject 3 received his indicated second price P 2 while his paired buyer subject 4 received 40-P 2 . There are several key points to notice from this procedure. Subjects know that one-half of them will be sellers and one-half of them will be buyers, but all fill out choice forms indicating their choices as sellers. They also know that the selection of which subjects play which role will be determined randomly only after they submit all their choice forms, and that there is a 25 percent chance that they will be the first seller and a 25 percent chance that they will be the second seller. Since the total amount to be divided is 840 (ex post) for each seller, ex ante the expected money at stake in each decision is 0.25*40=810. Ten-dollar stakes are commonly used in the dictator game literature, so our stakes and procedure generate monetary incentives that are roughly consistent with previous research. Notice also that the information exchanged before the second price choice is between two subjects who will either both be sellers or both be buyers. The instructions explain this and emphasize that if a subject ends up being selected as a seller, she exchanges information with a subject who cannot be a buyer; and that if a subject ends up being selected as a buyer, she exchanges information with a subject who cannot be a seller. In other words, subjects exchange information with someone who makes choices that do not affect their own earnings. Subjects were paid privately at the end of the session and never learned the identity of any other seller or buyer. A total of 100 subjects participated in the experiment: 40 in the Relevant Information treatment, 20 in the Irrelevant Information treatment, and 40 in a supplementary Strategy Method treatment described later in Section 4.3. Data were collected in five sessions. All subjects were recruited from large economics principles courses. Subjects received 85 upon arrival at the classroom used for the experiment. The instructions (available from the authors upon request) were read aloud by the experimenter while subjects followed along on their own copy. To ensure that subjects understood how their price choices determined their payoffs (and those of an anonymous subject) based on random draws at the end of the session, the instructions included a detailed example that included randomly drawn pairings. We were concerned that the subjects might use the example prices chosen by the experimenter as guides for what constitutes socially desirable
252
CASON AND MUI
behavior in this experiment. 5 In order to minimize this potential demand effect, the experimenter selected example prices randomly using dice thrown in full view of the subjects. Although subjects only required a few minutes to make their decisions, these extensive instructions (and time required to pay subjects) caused each session to last almost one hour. Including the 85 appearance fee, subject payments ranged from 85 to 845, with an average of 825. 6 3. HYPOTHESES
In our experiment, the majority (58 percent in the Relevant Information treatment and 70 percent in the Irrelevant Information treatment) choose a price less than 840 in their first price choice. This is consistent with previous findings in the literature. For those subjects who choose a price less than 840, it is reasonable to interpret this as reflecting that they are maximizing a utility function that has both a pecuniary and a social component. 7 For those subjects who are completely selfregarding and choose a price of 840, it is possible that some of them may still be maximizing a utility function that has both a pecuniary and a social component, but they decide that 840 is the choice that optimally balances the trade-off between pecuniary and social considerations. But it is quite likely that some of the subjects choose 840 because they are simply maximizing their monetary payoff, and they do not regard any social considerations to be important in this setting. For those subjects who care both about their monetary earnings as well as what constitutes socially appropriate behavior, socially relevant information may affect their beliefs and change their behavior. In particular, if subjects believe that the choice made by another subject gives them useful information, then the availability of information regarding choice made by another subject may change their beliefs regarding what constitutes socially appropriate behavior. This can in turn change their choices. 5 There is evidence in the psychological literature that people are more likely to conform to the behavior or opinions of an individual who appears to have expertise or authority in a given situation (Bushman, 1988). Since subjects were recruited from economics principles classes and knew that the experimenter is a professor in the economics department, they may therefore look up to the experimenter as a figure with authority, and hence use the illustrations chosen by the experimenter to guide their behavior. 6 There is a straightforward alternative approach to test the importance of social influence in the dictator game. One could first provide different groups of subjects with different information regarding how much another subject allocated in the dictator game. Then simply ask subjects directly what they would allocate hypothetically in this setting, and compare responses between subjects who learn that others had given less and who learn that others had given more. We do not adopt this hypothetical choice approach here because in environments such as the dictator game, subjects may have incentives to misrepresent their true preferences if they are not paid based on their choices. For example, a subject may believe that she will be perceived as a nice, fair person in the view of the experimenter by allocating onehalf of the total available stakes to herself. This social motivation is more likely to be an overriding concern if the subject is making a hypothetical choice (Forsythe et al., 1994, provide evidence on this point). 7 An alternative interpretation is that these subjects do not really understand the structure of the dictator game. However, given the simplicity of the dictator game and the observed behavior reported in numerous experiments, this alternative interpretation seems implausible.
SEQUENTIAL DICTATOR GAME
253
However, it is possible that subjects may not change their beliefs upon learning the choice made by one other subject, because they consider this choice as providing negligible information. Subjects may think that different people have different notions regarding socially appropriate behavior in this context. Since the choice made by another subject only provides information about a single individual's notion regarding socially appropriate behavior, it may not be sufficient to induce subjects to update their beliefs. 8 But a key finding in cognitive psychology is that individuals attach undue weight to salient events or vivid information (Nisbett 6 Ross, 1980). If subjects regard the choice of another subject in the same decision as salient or vivid, then subjects may still change their beliefs upon observing this information. One way to empirically test whether or not socially relevant information may change subjects' observable choices would be to compare directly subjects' second price choices to their first price choices in the Relevant Information treatment. However, this may not be an entirely legitimate approach. It is plausible that the mere fact that a subject receives some informationregardless of whether or not the information is socially relevantmay cause a change between her first and second choices. After receiving information, the subjects need to incur mental costs to process it and decide how to respond to it, including whether to ignore it altogether (Walker 6 Smith, 1993). There is also the possibility that when subjects are asked to make a second decision, they may change their choices because they have second thoughts. For example, suppose that a subject decides to take less than the whole surplus on her first choice. It is possible that when making her second choice, she may reconsider whether or not she wants to be as other-regarding in her second choice, which may cause her to make a more self-regarding allocation in her second choice. Of course, reconsidering the choice may also lead her to make a more otherregarding allocation on her second choice. Motivated by these observations, we introduced the Irrelevant Information treatment as a baseline control. 9 In the Irrelevant Information treatment, each subject receives only information about the birthday of the subject with whom he is paired (i.e., the day of the monthsimply an integer between 1 and 31). Since this information does not reveal the other subject's identity, nor does it contain any information regarding the other subject's age, gender, ethnicity, cultural, academic or economic background, we believe it contains no socially relevant information. This Irrelevant Information treatment allows us to control for the ``information-processing'' and ``second thought'' effects. In addition to these two effects, a social influence effect potentially operates in the Relevant Information treatment. Therefore, if the change between the second and first choice is different in the Irrelevant Information treatment than the Relevant 8 Note that even if subjects change their belief upon observing socially relevant information, they may not change their observed behavior. This can occur either because the change in their belief is too small to justify the required mental costs for making a behavioral change, or because subjects are constrained to make a discrete choice (i.e., in two-dollar increments between 80 and 840), which is too coarse to reflect desired changes. 9 We are grateful to Robert Slonim for insightful comments related to this issue, and for helpful suggestions that inspired the specific Irrelevant Information treatment adopted here.
254
CASON AND MUI
Information treatment, this difference should be attributed to social influence. Therefore, when testing whether socially relevant information will affect subjects' choices, we compare the changes in behavior in the Relevant Information treatment and the Irrelevant Information treatment, using the following null hypothesis: The Zero Social Influence Hypothesis. The change between subjects' second and first price choices in the Relevant Information treatment is not different from the change between the subjects' second and first price choices in the Irrelevant Information treatment. We also test an additional hypothesis that captures the idea that subjects' susceptibility to social influence may depend on the extent to which their first price choice is self-regarding. When a subject chooses a relatively self-regarding allocation in her first price choicethat is, something close to or even equal to 840this can either reflect that her estimate of what constitutes socially appropriate allocation is close to 840, or that she assigns a minimal (or zero) weight on the relative importance of the social component of her utility function, or both. In either case, this subject has a relatively strong preference for monetary earnings. Therefore, it is plausible that this subject is less likely to have second thoughts when making her second decision. Moreover, this subject may not find it worthwhile to incur the mental cost to process the information provided to her between her two decisions. In the Irrelevant Information treatment, such subjects may find it relatively easy to simply ignore the irrelevant information. In the Relevant Information treatment, these subjects may decide that it is not worthwhile to extract the social content embedded in the observed first price choice of other subjects and adjust behavior accordingly. Differential Social Influence Hypothesis. More self-regarding subjects are less likely to select a second price choice that is different from their first price choice. Finally, we test a third hypothesis that relates the likelihood that a subject changes his choice to the content of the information he receives about the allocation made by the other subject. It is mentally costly for a subject to process information and decide whether he should change his behavior. It is therefore possible that if he receives socially relevant information that is more dramatic, he is more likely to change his behavior. We shall classify information as more ``dramatic'' if the absolute difference between the subject's first price choice and the information he receives about the other subject's first price choice is large. Dramatic Social Influence Hypothesis. The larger the difference between a subject's first price choice and the information received about the other subject's first price choice, the more likely that the subject's second price choice will differ from his first price choice. 10 10 It is natural to consider testing a more refined version of this hypothesis: that a subject's second price choice moves in a direction toward the other subject's first price choice. We do not test this hypothesis because our data do not allow us to distinguish this directional shift from a simple regression to the mean. For example, suppose a subject takes all of the 840 on his first choice. It is therefore likely that he observes another subject's first choice that is more other-regarding. If he changes his choice at all, then he can only become more other-regarding on his second choice. The exact opposite pattern can be observed among those subjects who initially select other-regarding first choices.
SEQUENTIAL DICTATOR GAME
255
4. RESULTS
4.1. The Zero Social Influence Hypothesis Figure 1 presents the cumulative price choice distribution for the 20 subjects in the Irrelevant Information treatment and the 40 subjects in the Relevant Information treatment. Figure 1A displays the first price choice, and Fig. 1B displays the second price choice. In both treatments the modal price choice is 840, selected overall by 23 out of 60 subjects in their first decision (38 percent) and by 25 out of 60 subjects in their second decision (42 percent).
FIG. 1.
Cumulative price distribution on (A) first choice and (B) second choice.
256
CASON AND MUI
Our baseline for testing whether social influence matters is the Irrelevant Information treatment. The cumulative price choice distribution in the Irrelevant Information treatment exceeds the distribution in the Relevant Information treatment for prices greater than 86 on the first price choice. However, these two first price choice distributions are not significantly different (KolmogorovSmirnov (KS) p-value=0.51), nor are their means significantly different (t=1.37). The second price choice distributions and means in the two treatments are also not significantly different (KS p-value=0.99 and t=0.06). According to the Zero Social Influence Hypothesis, the change between subjects' second and first price choices in the Relevant Information treatment is not different from the Irrelevant Information treatment. We test this null hypothesis in two ways. First, we compare the frequency at which subjects change their price choices in both treatments. Second, we examine specifically how the first and second price choices differ within each treatment. This hypothesis receives considerable support in the first test that considers only the frequency of price choice changes. Sixteen of the 40 subjects (40 percent) change prices between their first and second decisions in the Relevant Information treatment, and they change their prices by an average of 811.75. Much to our surprise, a larger fraction of subjects11 of 20 (55 percent)change prices between their first and second decisions in the Irrelevant Information treatment, and they change their prices by an average of 810.27. However, the frequency at which subjects change choices in the Relevant Information treatment is not significantly different from the rate in the Irrelevant Information treatment (Fisher's exact test (two-tail) p-value=0.29). The Zero Social Influence Hypothesis receives significantly less support in our second test which examines how the first and second price choices differ within each treatment. The two panels of Fig. 1 describe the price distributions for the first and second choices. The same subjects select prices on the first and second choices, so we cannot compare the distributions from the two panels in the same treatment using standard distributional tests (such as the KolmogorovSmirnov test) that require two independent samples. However, we can test for a significant change across the two price choices by pairing the two prices made by each subject, and evaluating whether the distribution of the difference in the two prices has a mean that is significantly different from zero. As shown on Fig. 1, the mean price choice in the Relevant Information treatment falls from 831.1 to 830.7 between the first and second decisions. This difference (&80.4) is not significantly different from zero (t=0.25; (two-tail) p-value=0.81). By contrast, the mean price choice in the Irrelevant Information treatment increases between the first and second decisions from 827.1 to 830.6. This difference (83.5) is significantly different from zero (t=1.91; (two-tail) p-value=0.072), albeit at a marginal significance level. Since this shift toward self-regarding behavior occurs only in the Irrelevant Information treatment, it suggests that observing relevant information constrains some subjects from moving toward more self-regarding choices. These differencesno change in other-regarding behavior in the Relevant Information treatment and a decrease in other-regarding behavior in the Irrelevant
257
SEQUENTIAL DICTATOR GAME
Information treatmentare also reflected in a classification of how many subjects increase or decrease their price choices between their first and second price decisions. Table 1 shows that only 2 of the 20 subjects (10 percent) in the Irrelevant Information treatment become more other-regarding by selecting a lower price on their second decision, while 9 of the 40 subjects (23 percent) in the Relevant Information treatment become more other-regarding. Nine of the 20 subjects (45 percent) in the Irrelevant Information treatment become more self-regarding by selecting a higher price on their second decision, while only 7 of the 40 subjects (18 percent) in the Relevant Information treatment become more self-regarding. The average amount that the subjects change their prices in all four of these cases exceeds 810. This difference in the direction of change across treatments is statistically significant at the 8-percent level. Conclusion 1. A comparison of the frequency at which subjects change their price choices in the Irrelevant Information and Relevant Information treatments provides support for the Zero Social Influence Hypothesis. However, subjects' second price choices are more self-regarding than their first price choices on average in the Irrelevant Information treatment. This shift toward self-regarding behavior is not observed in the Relevant Information treatment, which provides evidence against the Zero Social Influence Hypothesis. Before conducting the experiment we expected that most subjects would not change their second price choice from their first price choice in the Irrelevant Information treatment, but that more would change their choices in the Relevant Information treatment. Much to our surprise, a (insignificantly) greater percentage of subjects changed their choices in the Irrelevant Information treatment. Recall that the second thought and the information processing effects are operating in both treatments. In the Irrelevant Information treatment the combined second thought and information-processing effects cause subjects to become more self-regarding. The result that subjects do not become more self-regarding in the Relevant Information treatment can be interpreted as follows. The introduction of socially relevant informationthrough its impact on subjects' beliefsinduces some subjects to shift toward other-regarding behavior in the Relevant Information treatment. But this social information impact appears just sufficient to offset the combined second thought and the information-processing effects. TABLE 1 Frequency of Shifts in Other-Regarding and Self-Regarding Direction by Treatment
Irrelevant information Relevant information Total (column)
Own second price
No change between first and second price
Own second price>Own first price
Total (row)
2 9 11
9 24 33
9 7 16
20 40 60
Note. Fisher's exact test (two-tail) p-value=0.08.
258
CASON AND MUI
Conclusion 1 indicates that the evidence regarding the Zero Social Influence Hypothesis is mixed. This could be due, in part, to subject heterogeneity; e.g., some subjects may respond to socially relevant information, while others do not. In the next subsection, we examine this heterogeneity directly, by identifying some characteristics of subjects who are more likely to change price choices between their first and second decisions. 4.2. Differential and Dramatic Social Influence Hypotheses According to the Differential Social Influence Hypothesis, more self-regarding subjects are less likely to select a second price choice that is different from their first price choice. Table 2 indicates that this hypothesis receives considerable support in both treatments. Subjects who select a first price of 840 are significantly less likely to change price on their second choice. For example, in the Relevant Information treatment only 3 of the 17 subjects who chose a first price of 840 changed their price on their second choice, while 13 of the 23 subjects who chose a first price less than 840 changed their price on their second choice. These differences are highly significant, and are also highly significant in the Irrelevant Information treatment. 11 Table 3 presents an alternative test of this hypothesis using a probit model. This model estimates the likelihood that a subject changes her price between her first and second price decisions as a function of her first price choice. The negative and significant coefficient estimates reported in columns (1) and (2) on the first choice variable indicate that more self-regarding first price choices reduce the likelihood that a subject changes her price between her first and second price choices. Conclusion 2. Subjects who exhibit more self-regarding behavior on their first price choice are less likely to change prices between their first and second decisions. Finally, we test the Dramatic Social Influence Hypothesis, which states that more dramatic information is more likely to cause subjects to change their price choices. Subjects learned the actual price choice made by another subject only in the Relevant Information treatment, so we test this hypothesis for this treatment only. Information is more dramatic if the difference between the subject's own first price choice and the other subject's first price choice is large. The median absolute difference between these two choices was 810, so we define information received by a subject as dramatic when the absolute difference between her first price choice and her paired subject's first price choice is larger than or equal to 810 (that is, |own first choice-paired subject's first choice| 810). Eleven of the 22 subjects (50 percent) who observe dramatic information change their price choice between their first and second decisions, and they change their price by an average of 813.50. By contrast, only 5 of the 18 subjects (28 percent) who observe non-dramatic information change their price, and they change their
11 Alternative definitions to classify subjects into self-regarding and other-regarding groups give very similar results; e.g., classifying a subject as other-regarding if he selected a price less than or equal to the median first price in that treatment.
259
SEQUENTIAL DICTATOR GAME
TABLE 2 Frequency of Changes in Price between the First and Second Choices, by Initial Choice Change price between first Do not change price between and second choices first and second choices
Total (row)
Relevant information a First price=840 First price<840 Total (column)
3 13 16
14 10 24
17 23 40
6 3 9
6 14 20
Irrelevant information b First price=840 First price<840 Total (column) a b
0 11 11
Fisher's exact test (one-tail) p-value=0.014. Fisher's exact test (one-tail) p-value=0.002.
price by an average of 88.00. The difference in the rate the these groups change prices is not quite statistically significant (Fisher's exact test (one-tail) p-value= 0.135). We also augmented the probit model in Table 3 to include the absolute difference variable |own first choice-paired subject's first choice|. As shown in columns (3) and (4), the coefficient on this variable has the predicted positive sign, but is not significantly different from zero. Conclusion 3. Subjects who receive dramatic information are not significantly more likely to change their price choice between their first and second decisions. TABLE 3 Estimates of Change Likelihood Model (Probit Model Maximum Likelihood Estimation) Irrelevant information (1)
(2)
First price choice (Standard error) | Own first priceother's first price | (Standard error) Intercept (Standard error)
&0.106 b (0.042)
&0.036 b (0.021)
3.062 b (1.209)
Observations Log-likelihood Restricted (slopes=0) Log-likelihood
Variable or statistic
Relevant information (3) (4)
0.027 (0.025)
&0.033 a (0.022) 0.015 (0.027)
0.864 (0.679)
&0.538 (0.337)
0.597 (0.826)
20 &8.97
40 &25.34
40 &26.34
40 &25.17
&13.76
&26.92
&26.92
&26.92
Note. Dependent variable=1 if change offer; dependent variable=0 otherwise. a Denotes significantly different from 0 at 10 percent level. b Denotes significantly different from 0 at 5 0 level (one-tailed tests, except for Intercept).
260
CASON AND MUI
4.3. A Strategy Method Hypothesis and Results We conclude this section with a brief presentation of a supplementary Strategy Method Hypothesis and results. Experimental economists have recently begun to use the strategy method to elicit subject choices (for example, Selten 6 Buchta, 1994; Straub 6 Murnighan, 1995). In this method, subjects specify a complete strategy, rather than only choosing an action in realized information sets. Therefore, in the Strategy Method treatment implemented here, subjects received no information between their two price decisions. After submitting their first price, subjects filled out a second form specifying a complete price choice strategy contingent on all 21 possible first choice prices selected by the subject with whom they were randomly paired. The instructions included an extensive example and emphasized that these contingent price choices were binding and that one would be carried out depending on the first price choice of another anonymous subject. Forty additional subjects participated in this treatment. We examined this strategy method approach in the sequential dictator game by evaluating the following null Strategy Method Hypothesis: Choices in the Strategy Method treatment are not different from choices in the Relevant Information treatment. Roth (1995) points out that there are at least two reasons why subjects might give different responses in the strategy and non-strategy methods. First, in the strategy method, subjects are required to simultaneously make all potential choices at the same time. When the game is sufficiently complex, for example, when it has many information sets and substantial strategic interdependence, the transformation to the strategy method amounts to a significant change in the representation of the strategic environment. Our sequential dictator game is extremely simple and strategic interdependence is nonexistent, however, so this effect is probably absent. Second, in the strategy method subjects are required to submit the complete allocation strategies contingent on every possible choice by another subject. Consequently, there is a reduced incentive to think carefully about every possible element of the strategy. This could lead to more variance in choices in the Strategy Method treatment (Walker 6 Smith, 1993). For example, subjects can reasonably expect that the likelihood that the first subject chooses a very other-regarding first price such as 82 is very low, so the expected payoff consequences for their price choice contingent on another first price of 82 is also very low. This second effect is likely to be important in the sequential dictator game. Besides these reasons, additional psychological considerations may cause subjects to behave differently under the two methods. Psychological research indicates that individuals attach too much weight to salient or vivid events and too little weight to non-salient events. Events that do occur are more real and immediate, and hence more concrete than the corresponding potential events that could have occurred. The greater salience of events that have actually occurred are more available for inference than events that have not occurred (Nisbett 6 Ross, 1980). Information provided in the Relevant Information treatment is based on choices that have in fact been made by another subject, rather than on choices that could have been made by others. This information is therefore more likely to have an impact on subjects' choices because of its vividness.
SEQUENTIAL DICTATOR GAME
261
We tested the Strategy Method Hypothesis by comparing the first price choices in both treatments. We also compared the second price choices in the Relevant Information treatment with the realized second price choices in the Strategy Method treatment (i.e., the realized second prices that were contingent uponand thus determined bythe actual first price choice of the randomly-paired other subject). In no cases are the distributions, means, or fraction of subjects choosing 840 significantly different, so we conclude that use of the Strategy Method in this setting does not significantly alter choices. 12 This finding suggests that it is appropriate to pool the data in the Strategy Method and the Relevant Information treatments when testing the social influence hypotheses. All of the results presented above are qualitatively unchanged after pooling the data. In some cases, the results become more statistically significant because of the larger sample size. For example, the result that subjects in the Irrelevant Information treatment shift more in the selfregarding direction (shown in Table 1) compared to the pooled Relevant InformationStrategy Method data is significant at the 2-percent level.
5. DISCUSSION
In this experiment subjects made dictator allocations of 840 before and after learning the allocation made by one other subject in the Relevant Information treatment, or before and after learning the birthday of one other subject in the Irrelevant Information treatment. Subjects on average became more self-regarding in the Irrelevant Information treatment, but did not become more self-regarding in the Relevant Information treatment. We also observe that subjects who exhibit more self-regarding behavior on their first decisions are less likely to change choices between their first and second decisions. Ever since the classic study on conformism by Asch (1951), psychologists have been concerned about discerning possible channels through which social influence occurs in different settings. In the Asch experiment, subjects were first shown a straight line and then three lines for comparison. Subjects were then asked to indicate which of these three comparison lines was closest to the first line in length. Each subject was placed in a group consisting of five to seven confederates of the researcher. The confederatesmost of whom provided responses before the subjectgave the same wrong answer. A significant number of subjects conformed to the incorrect answer provided by the confederates. Inspired by the Asch experiment and follow-up studies, two explanations have emerged as the principal mechanisms of social influence (Deutsch 6 Gerald, 1955). First, people may change their behavior upon observing how others behave in similar situations in order to obtain certain rewards or avoid potential punishments. For example, it is highly plausible that subjects in the Asch experiment conform to the view of the majority because they want to be accepted by the 12 The one relatively minor difference between the Strategy Method and Relevant Information treatments is the rate of price changes between the first and second choices. Twenty-seven of 40 subjects change prices in the Strategy Method treatment, while 16 of 40 subjects change prices in the Relevant Information treatment (Fisher's exact test (two-tail) p-value=0.03).
262
CASON AND MUI
majority or to avoid the discomfort for disagreeing with the majority. Besides this normative influence, social influence can also occur through informational influence. When individuals are not completely certain about what is the right behavior in a particular context, they may regard others' choices in similar situations as sources of useful information. 13 In our experiment, we believe that the main source of social influence is informational influence rather than normative influence. Since a subject's choice is not observed by other subjectsespecially the subject whose choice he learns in the experimenta subject does not need to be concerned about any potential embarrassment for not conforming to what he conjectures to be the ``appropriate'' behavior of others participating in the experiment. 14 This distinction between the normative and informational influences as different sources of social influence raises the natural question of whether social influence has permanent effect on people's behavior. As psychologists have long emphasized, when social influence operates mainly through informational influencethat is, when knowledge about others' behavior causes an individual to change her belief regarding what constitutes the appropriate or correct behaviorit is more likely that social influence will have a long-lasting effect on people's behavior (Aronson, 1995, Chapter 2). On the other hand, when social influence operates mainly through normative influencefor example, when an individual conforms to others' behavior in order to avoid embarrassmentthen it is less likely that social influence will have a permanent effect. 15 For example, in the Asch experiment, the incidence of conformism varies significantly across treatments when the level of privacy is varied. In the original treatment, having observed the choices of others, a subject provides his response verbally in the presence of the entire group. Later research discovered that in treatments in which subjects make their responses privately, the rate of conformism decreased considerably (Deustch 6 Gerald, 1955; Argyle, 1957). In contrast, in other studies of social influence in which arguably, social influence operates mainly through informational influence, many subjects exhibited permanent 13
In the Asch experiment, the lengths of the lines are such that it would be extremely easy for subjects to identify the correct answer. Given the simplicity of this cognitive task, it is quite unlikely that informational considerations can be the main source of social influence. However, when the tasks involved require subjects to make decisions that are (partially) affected by their belief regarding what constitutes socially appropriate behavior as in this experiment, or when the task is cognitively demanding, one would expect that informational considerations become more important. 14 We note, however, that since a subject's choice is observed by the experimenter it is plausible that a subject may be concerned about whether such behavior will be judged favorably or unfavorably by the experimenter (the observer effect). But the observer effect is also present when a subject makes his first price choice in all treatments, so any change that may occur in his second choice is unlikely to be explained by the observer effect alone. 15 Without going into details, we note that the difference in fact is more subtle than we characterize in the text. It is possible that an individual is motivated by normative influence to follow the behavior of others originally. However, having made the ``appropriate'' changes in his behavior, he may eventually adjust his own belief to make it consistent with his new choices. Once such internalization of beliefs has occurred, the new belief may become independent of its source and will become extremely resistant to change.
SEQUENTIAL DICTATOR GAME
263
changes in their beliefs (see, for example, Newcomb, Koenig, Flacks, 6 Warwick, 1967). In the past two decades, economists have come to recognize better the importance of social influence in affecting economic decisions. For example, Akerlof (1980) develops a formal model of social custom and uses it to provide an explanation for equilibrium involuntary unemployment. Drawing on the findings of organizational theorists and social psychologists, Jones (1984) and Levine (1991) have both argued that workers have propensities to choose their work efforts in conformism to the work norms in their work group, and explore the implications of this observation for the study of organizational design and wage dispersion. Banerjee (1992) and Bikhchandani, Hirshleifer, 6 Welch (1992) show how informational cascades can lead to conformity behavior arising from purely informational reasons. By contrast, Akerlof (1983) and Jones (1984, Chapter 5) present economic models that focus on how people's experience can change their beliefs permanently. 16 The above observations regarding the two different channels through which social influence operates suggest the need to investigate further the dynamic aspects of social influence in economic settings. For example, a person's choices with regard to how much to give to charity, or whether to behave more or less cooperatively in bargaining situations, are affected by both his belief regarding what constitutes appropriate behavior in these situations, as well as the economic and social incentives he faces in a particular situation. Society as a whole invests substantial resources in shaping people's beliefs through its educational, social and religious institutions, and some business organizations also expend resources to cultivate a corporate culture that they believe will enhance employee loyalty, cooperation and productivity. Given the importance of these social influence activities, it is important that economists conduct future theoretical and field studies as well as laboratory experiments to determine under what conditions these influences have a permanent effect on beliefs and economic behavior.
REFERENCES Akerlof, G. A. (1997). Social distance and social decisions. Econometrica, 65, 10051028. Akerlof, G. A. (1983). Loyalty filters. American Economic Review, 73, 5463. Akerlof, G. A. (1980). A theory of social custom, of which unemployment may be one consequence. Quarterly Journal of Economics, 95, 74975. Andreoni, J. (1995). Cooperation in public-goods experiment: kindness or confusion? American Economic Review, 85, 891904. Argyle, M. (1957). Social pressure in public and private situations. Journal of Abnormal and Social Psychology, 54, 172175.
16 For more recent work on the importance of social influence in the economics literature, see, for example, Akerlof (1997), Becker (1996), Bernheim (1994), Fershtman, Murphy, 6 Weiss (1996), and Kuran (1995). For a recent survey on the importance of social and psychological considerations in economic decision-making, see Rabin (1998).
264
CASON AND MUI
Aronson, E. (1995). The social animal. Seventh Ed. New York: Freeman. Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgements. In H. H. Guetzkow (Ed.), Groups, leadership, and men. Madison, NJ: Carnegie Press. Banerjee, A. (1992). A Simple model of herd behavior. Quarterly Journal of Economics, 107, 797817. Becker, G. S. (1996). Accounting for tastes. Cambridge, MA: Harvard Univ. Press. Berg, J., Dichhaut, J., 6 McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10, 122142. Bernheim, B. D. (1994). A theory of conformity. Journal of Political Economy, 102, 841877. Bikhchandani, S., Hirshleifer, D., 6 Welch, I. (1992). A theory of fads, fashion, and cultural change as informational cascades. Journal of Political Economy, 100, 9921026. Bolton, G. E., Katok, E., 6 Zwick, R. (1997). Dictator game giving: rules of fairness versus acts of kindness. International Journal of Game Theory, to appear. Bushman, B. J. (1988). The effects of apparel on compliance: A field experiment with a female authority figure. Personality and Social Psychology Bulletin, 14, 459467. Camerer, C., 6 Thaler, R. (1995). Anomalies: Ultimatums, dictators, and manners. Journal of Economic Perspectives, 9, 209219. Cason, T. N., 6 Mui, V.-L. (1997). A laboratory study of group polarization in the team dictator game. Economic Journal, 107, 14651483. Croson, R. (1995). Shirking and working: an experiment in team production. Mimeo, Wharton School of Business, University of Pennsylvania. Deutsch, M., 6 Gerald, H. B. (1955). A study of normative and informational social influence upon individual judgement. Journal of Abnormal and Social Psychology, 51, 629636. Fershtman, C., Murphy, K. M., 6 Weiss, Y. (1996). Social status, education, and growth. Journal of Political Economy, 104, 108132. Forsythe, R., Horowitz, J., Savin, N. E., 6 Sefton, M. (1994). Fairness in simple bargaining experiments. Games and Economic Behavior, 6, 347369. Hoffman, E., McCabe, K., Sachat, K., 6 Smith, V. L. (1994). Preferences, property rights and anonymity in bargaining games. Games and Economic Behavior, 7, 346380. Hoffman, E., McCabe, K., 6 Smith, V. L. (1996). Social distance and other-regarding behavior in dictator games. American Economic Review, 86, 653660. Jones, S. (1984). The economics of conformism. Oxford: Blackwell Sci. Kuran, T. (1995). Private truths, public lies. Cambridge, MA: Harvard Univ. Press. Levine, D. (1991). Cohesiveness, productivity, and wage dispersion. Journal of Economic Behavior and Organization, 15, 237255. Newcomb, T. M., Koenig, K. E., Flacks, R., 6 Warwick, D. P. (1967). Persistence and change: Bennington College and its students after twenty-five years. New York: Wiley. Nisbett, R. E., 6 Ross, L. (1980). Human inferences: Strategies and shortcomings of social judgement Englewood Cliffs, NJ: Prentice Hall International. Rabin, M. (1998). Psychology and economics. Journal of Economic Literature, 36, 1146. Roth, A. E. (1995). Bargaining experiments. In J. Kagal, 6 A. E. Roth (Eds.), Handbook of experimental economics. Princeton, NJ: Princeton Univ. Press. Sell, J., 6 Wilson, R. (1991). Levels of information and contributions to public goods. Social Forces, 70, 107124. Selten, R., 6 Buchta, J. (1994). Experimental sealed bid first price auctions with directly observed bid functions. Discussion Paper B-270, University of Bonn. Straub, P. G., 6 Murnighan, J. K. (1995). An experimental investigation of ultimatum games: Information, fairness, expectations, and lowest acceptable offers. Journal of Economic Behavior and Organization, 27, 345364.
SEQUENTIAL DICTATOR GAME
265
Walker, J. M., 6 Smith, V. L. (1993). Monetary rewards and decision costs. Economic Inquiry, 31, 245261. Weimann, J. (1994). Individual behavior in a free riding experiment. Journal of Public Economics, 54, 185200. Received: January 28, 1998
Social Influence in the Sequential Dictator Game 1 Timothy N. Cason and Vai-Lam Mui University of Southern California
This paper introduces the sequential dictator game to study how social influence may affect subjects' choices when making dictator allocations. Subjects made dictator allocations of 840 before and after learning the allocation made by one other subject in the Relevant Information treatment, or the birthday of one other subject in the Irrelevant Information treatment. Subjects on average become more self-regarding in the Irrelevant Information treatment, but observing relevant information constrains some subjects from moving toward more self-regarding choices. We also find that subjects who exhibit more self-regarding behavior on their first decisions are less likely to change choices between their first and second decisions, and the use of the Strategy Method in this setting does not significantly alter choices. The relationships between our findings and the economic and psychological literature regarding how social influence operates are also explored. 1998 Academic Press
1. INTRODUCTION
This paper introduces the sequential dictator game to study how social influence may affect subjects' choices when making dictator allocations. In the dictator game, each subject is paired randomly and anonymously with another subject, and one is selected to ``dictate'' the allocation of a fixed amount of money. In the sequential dictator game, subjects make two dictator decisions, and they receive information regarding another subject's dictator allocation before making their second decision. When making these two decisions, subjects are not making two dictator offers to the same person. If a subject's only goal is to maximize her earnings, then she should allocate the entire surplus to herself. However, there is now overwhelming evidence that many subjects give away a substantial share of the surplus. 2 Different explanations have 1
We thank Matt Grund for research assistance and the USC Zumberge Faculty Research and Innovation Fund for financial support. We are also grateful to Gary Bolton, James Konow, Robert Slonim, two anonymous referees, and participants at the Fall 1996 Economic Science Association conference for valuable comments. We retain responsibility for any errors. 2 For two surveys of the early literature on the dictator game and the related ultimatum game, see Camerer 6 Thaler, 1995 and Roth, 1995; for more recent contributions see Hoffman, McCabe, 6 Smith, 1996; Bolton, Katok, 6 Zwick, 1997; and Cason 6 Mui, 1997. 0022-249698 25.00 Copyright 1998 by Academic Press All rights of reproduction in any form reserved.
248
SEQUENTIAL DICTATOR GAME
249
been advanced for this other-regarding behaviorfor example, the fairness hypothesis (Forsythe, Horowitz, Savin, 6 Sefton, 1994) and the observer effect hypothesis (Hoffman, McCabe, Shachat, 6 Smith, 1994)and these explanations typically employ the idea that subjects care about both their monetary earnings as well as about behaving in a socially appropriate way. 3 At a general level, these explanations are based on the idea that subjects are maximizing a utility function that includes both a monetary payoff and a social payoff, and this social payoff depends on the extent to which subjects' choices conform to or deviate from what they believe to be socially appropriate behavior. While a subject's belief regarding what constitutes socially appropriate behavior will depend on her personal characteristicsher experience, cultural background, personalityit is plausible that this belief may also depend on her estimate of others' beliefs regarding what constitutes socially appropriate behavior. Therefore, when a subject is exposed to socially relevant information that may lead her to revise her belief, her choice may change. This paper reports an experiment that provides some preliminary tests of this social influence hypothesis. More specifically, we investigate whether learning the dictator allocation made by another subject in the same experimental session affects subjects' choices. Other researchers have also conducted experiments in which subjects receive information about others' choices. For example, several recent public goods laboratory studies have examined whether group contribution levels change when subjects receive information about the previous contributions of all individual subjects in their group (Sell 6 Wilson, 1991; Weimann, 1994; and Croson, 1995). 4 These experiments capture natural settings in which social influence may be significant. However, in these experiments a subject's monetary payoff depends on both her own choice as well as others' choices. The presence of strategic interdependence creates two difficulties for our purpose of testing for the impact of social influence. First, even if one observes that a subject's behavior changes upon observing other subjects' choices, it may be difficult to discern to what extent these changes are due to social influence and to what extent they are caused by the fact that such information affects her strategic calculus when attempting to increase her monetary payoff. Second, when strategic interdependence is present, subjects' beliefs regarding what constitutes socially appropriate behavior may be influenced by the strategic environment itself. Roth (1995) reviews recent efforts in studying how social norms regarding what constitutes fair behavior in bargaining games can be created and enforced. He observes that ``(a)lthough subjects may have clear ideas about what is 3
According to the fairness hypothesis, subjects are trying to behave fairly by taking what they consider to be their fair share. This hypothesis implies that subjects' choices are sensitive to their beliefs regarding what is a fair allocation. According to the alternative observer effect hypothesis, subjects may be concerned about whether their allocations are judged as unfair by the experimenter, so the experimenter as observer could increase other-regarding behavior. This hypothesis implies that subjects' choices are sensitive to their expectation regarding what is perceived to be socially acceptable to the experimenter. 4 In a related public goods experiment, Andreoni (1995) finds that providing subjects with information regarding how their earnings rank in comparison to the other subjects in their session decreases cooperation. In a ``trust game'' experiment, Berg, Dickhaut, 6 McCabe (1995) finds that providing subjects with the full distribution of outcomes from previous sessions does not substantially affect their choices.
250
CASON AND MUI
fair in a variety of circumstances, and although these ideas about fairness may influence the strategic environment, the evidence suggests that subjects adapt their ideas about what is fair in response to their experience, in ways that may be heavily influenced by strategic considerations.'' (Roth, 1995, p. 271). Therefore, while social influence may arise naturally in many environments in which strategic interdependence is important, in order to better isolate the impact of social influence it is useful to ``go one step backward'' to study this issue in a simple environment in which strategic considerations are absent. Our experimental design includes two main treatments. In both treatments, subjects first chose an initial dictator allocation with no information other than the instructions. The Irrelevant Information treatment serves as a baseline control, in which subjects receive socially irrelevant information between their two dictator decisions. In the Relevant Information treatment, each subject learns the first dictator allocation chosen by one other subject. Each subject then specifies a single second dictator allocation to another subject who is different from the subject affected by his initial dictator allocation. Our major findings are as follows. Subjects on average become more self-regarding in the Irrelevant Information treatment, but do not become more self-regarding in the Relevant Information treatment. We also find that subjects who are more self-regarding on their first decisions are less likely to change choices between their first and second decisions.
2. EXPERIMENTAL DESIGN AND PROCEDURES
Each subject made two decisions sequentially that corresponded to a dictator allocation of 840. We employed the exchange framing for these decisions, in which the decision-maker is a seller choosing a price that ranged between 0 and 40 dollars (see Hoffman et al., 1994, for an experiment that examines the importance of this framing). The buyer was required to buy at that price. If the seller chose a price of P dollars, the buyer received 40-P dollars while the seller received P dollars, so the seller dictated the allocation of the 840 surplus. Notice that when a seller chose a higher price he is making a more self-regarding choice. In all cases price choices were restricted to two-dollar increments (i.e., 80, 82, 84, ..., 838, 840). Each session proceeded as follows: 1. All subjects chose an initial ``first price'' (P 1 ) with no information other than the instructions. In the Irrelevant Information treatment, subjects also indicated on their choice form the day of the month on which they were born (i.e., an integer between 1 and 31). 2. The experimenter drew bingo balls from a cage to assign subjects into groups of four. For ease of exposition, call a typical group of four subjects 1, 2, 3, and 4. 3. The experimenter transferred information within the odd-numbered and within the even-numbered subjects in each group. In the Relevant Information
SEQUENTIAL DICTATOR GAME
251
treatment, subject 1 learned the initial price chosen by subject 3, and subject 3 learned the initial price chosen by subject 1. Information was similarly transferred between subjects 2 and 4. In the Irrelevant Information treatment subjects instead learned the birthday of this other subject. 4. All subjects then chose a ``second price'' (P 2 ). 5. Again using the bingo cage, the experimenter randomly determined whether the odd-numbered subjects or the even-numbered subjects in each group would be sellers, as well as which of these two selected sellers would be the ``first seller'' and which the ``second seller.'' For example, if the odd-numbered subjects were selected as the sellers and if subject 1 was selected as the first seller, then subject 1 received his indicated first price P 1 while his paired buyer subject 2 received 40-P 1 . In this case subject 3 must be the second seller, so subject 3 received his indicated second price P 2 while his paired buyer subject 4 received 40-P 2 . There are several key points to notice from this procedure. Subjects know that one-half of them will be sellers and one-half of them will be buyers, but all fill out choice forms indicating their choices as sellers. They also know that the selection of which subjects play which role will be determined randomly only after they submit all their choice forms, and that there is a 25 percent chance that they will be the first seller and a 25 percent chance that they will be the second seller. Since the total amount to be divided is 840 (ex post) for each seller, ex ante the expected money at stake in each decision is 0.25*40=810. Ten-dollar stakes are commonly used in the dictator game literature, so our stakes and procedure generate monetary incentives that are roughly consistent with previous research. Notice also that the information exchanged before the second price choice is between two subjects who will either both be sellers or both be buyers. The instructions explain this and emphasize that if a subject ends up being selected as a seller, she exchanges information with a subject who cannot be a buyer; and that if a subject ends up being selected as a buyer, she exchanges information with a subject who cannot be a seller. In other words, subjects exchange information with someone who makes choices that do not affect their own earnings. Subjects were paid privately at the end of the session and never learned the identity of any other seller or buyer. A total of 100 subjects participated in the experiment: 40 in the Relevant Information treatment, 20 in the Irrelevant Information treatment, and 40 in a supplementary Strategy Method treatment described later in Section 4.3. Data were collected in five sessions. All subjects were recruited from large economics principles courses. Subjects received 85 upon arrival at the classroom used for the experiment. The instructions (available from the authors upon request) were read aloud by the experimenter while subjects followed along on their own copy. To ensure that subjects understood how their price choices determined their payoffs (and those of an anonymous subject) based on random draws at the end of the session, the instructions included a detailed example that included randomly drawn pairings. We were concerned that the subjects might use the example prices chosen by the experimenter as guides for what constitutes socially desirable
252
CASON AND MUI
behavior in this experiment. 5 In order to minimize this potential demand effect, the experimenter selected example prices randomly using dice thrown in full view of the subjects. Although subjects only required a few minutes to make their decisions, these extensive instructions (and time required to pay subjects) caused each session to last almost one hour. Including the 85 appearance fee, subject payments ranged from 85 to 845, with an average of 825. 6 3. HYPOTHESES
In our experiment, the majority (58 percent in the Relevant Information treatment and 70 percent in the Irrelevant Information treatment) choose a price less than 840 in their first price choice. This is consistent with previous findings in the literature. For those subjects who choose a price less than 840, it is reasonable to interpret this as reflecting that they are maximizing a utility function that has both a pecuniary and a social component. 7 For those subjects who are completely selfregarding and choose a price of 840, it is possible that some of them may still be maximizing a utility function that has both a pecuniary and a social component, but they decide that 840 is the choice that optimally balances the trade-off between pecuniary and social considerations. But it is quite likely that some of the subjects choose 840 because they are simply maximizing their monetary payoff, and they do not regard any social considerations to be important in this setting. For those subjects who care both about their monetary earnings as well as what constitutes socially appropriate behavior, socially relevant information may affect their beliefs and change their behavior. In particular, if subjects believe that the choice made by another subject gives them useful information, then the availability of information regarding choice made by another subject may change their beliefs regarding what constitutes socially appropriate behavior. This can in turn change their choices. 5 There is evidence in the psychological literature that people are more likely to conform to the behavior or opinions of an individual who appears to have expertise or authority in a given situation (Bushman, 1988). Since subjects were recruited from economics principles classes and knew that the experimenter is a professor in the economics department, they may therefore look up to the experimenter as a figure with authority, and hence use the illustrations chosen by the experimenter to guide their behavior. 6 There is a straightforward alternative approach to test the importance of social influence in the dictator game. One could first provide different groups of subjects with different information regarding how much another subject allocated in the dictator game. Then simply ask subjects directly what they would allocate hypothetically in this setting, and compare responses between subjects who learn that others had given less and who learn that others had given more. We do not adopt this hypothetical choice approach here because in environments such as the dictator game, subjects may have incentives to misrepresent their true preferences if they are not paid based on their choices. For example, a subject may believe that she will be perceived as a nice, fair person in the view of the experimenter by allocating onehalf of the total available stakes to herself. This social motivation is more likely to be an overriding concern if the subject is making a hypothetical choice (Forsythe et al., 1994, provide evidence on this point). 7 An alternative interpretation is that these subjects do not really understand the structure of the dictator game. However, given the simplicity of the dictator game and the observed behavior reported in numerous experiments, this alternative interpretation seems implausible.
SEQUENTIAL DICTATOR GAME
253
However, it is possible that subjects may not change their beliefs upon learning the choice made by one other subject, because they consider this choice as providing negligible information. Subjects may think that different people have different notions regarding socially appropriate behavior in this context. Since the choice made by another subject only provides information about a single individual's notion regarding socially appropriate behavior, it may not be sufficient to induce subjects to update their beliefs. 8 But a key finding in cognitive psychology is that individuals attach undue weight to salient events or vivid information (Nisbett 6 Ross, 1980). If subjects regard the choice of another subject in the same decision as salient or vivid, then subjects may still change their beliefs upon observing this information. One way to empirically test whether or not socially relevant information may change subjects' observable choices would be to compare directly subjects' second price choices to their first price choices in the Relevant Information treatment. However, this may not be an entirely legitimate approach. It is plausible that the mere fact that a subject receives some informationregardless of whether or not the information is socially relevantmay cause a change between her first and second choices. After receiving information, the subjects need to incur mental costs to process it and decide how to respond to it, including whether to ignore it altogether (Walker 6 Smith, 1993). There is also the possibility that when subjects are asked to make a second decision, they may change their choices because they have second thoughts. For example, suppose that a subject decides to take less than the whole surplus on her first choice. It is possible that when making her second choice, she may reconsider whether or not she wants to be as other-regarding in her second choice, which may cause her to make a more self-regarding allocation in her second choice. Of course, reconsidering the choice may also lead her to make a more otherregarding allocation on her second choice. Motivated by these observations, we introduced the Irrelevant Information treatment as a baseline control. 9 In the Irrelevant Information treatment, each subject receives only information about the birthday of the subject with whom he is paired (i.e., the day of the monthsimply an integer between 1 and 31). Since this information does not reveal the other subject's identity, nor does it contain any information regarding the other subject's age, gender, ethnicity, cultural, academic or economic background, we believe it contains no socially relevant information. This Irrelevant Information treatment allows us to control for the ``information-processing'' and ``second thought'' effects. In addition to these two effects, a social influence effect potentially operates in the Relevant Information treatment. Therefore, if the change between the second and first choice is different in the Irrelevant Information treatment than the Relevant 8 Note that even if subjects change their belief upon observing socially relevant information, they may not change their observed behavior. This can occur either because the change in their belief is too small to justify the required mental costs for making a behavioral change, or because subjects are constrained to make a discrete choice (i.e., in two-dollar increments between 80 and 840), which is too coarse to reflect desired changes. 9 We are grateful to Robert Slonim for insightful comments related to this issue, and for helpful suggestions that inspired the specific Irrelevant Information treatment adopted here.
254
CASON AND MUI
Information treatment, this difference should be attributed to social influence. Therefore, when testing whether socially relevant information will affect subjects' choices, we compare the changes in behavior in the Relevant Information treatment and the Irrelevant Information treatment, using the following null hypothesis: The Zero Social Influence Hypothesis. The change between subjects' second and first price choices in the Relevant Information treatment is not different from the change between the subjects' second and first price choices in the Irrelevant Information treatment. We also test an additional hypothesis that captures the idea that subjects' susceptibility to social influence may depend on the extent to which their first price choice is self-regarding. When a subject chooses a relatively self-regarding allocation in her first price choicethat is, something close to or even equal to 840this can either reflect that her estimate of what constitutes socially appropriate allocation is close to 840, or that she assigns a minimal (or zero) weight on the relative importance of the social component of her utility function, or both. In either case, this subject has a relatively strong preference for monetary earnings. Therefore, it is plausible that this subject is less likely to have second thoughts when making her second decision. Moreover, this subject may not find it worthwhile to incur the mental cost to process the information provided to her between her two decisions. In the Irrelevant Information treatment, such subjects may find it relatively easy to simply ignore the irrelevant information. In the Relevant Information treatment, these subjects may decide that it is not worthwhile to extract the social content embedded in the observed first price choice of other subjects and adjust behavior accordingly. Differential Social Influence Hypothesis. More self-regarding subjects are less likely to select a second price choice that is different from their first price choice. Finally, we test a third hypothesis that relates the likelihood that a subject changes his choice to the content of the information he receives about the allocation made by the other subject. It is mentally costly for a subject to process information and decide whether he should change his behavior. It is therefore possible that if he receives socially relevant information that is more dramatic, he is more likely to change his behavior. We shall classify information as more ``dramatic'' if the absolute difference between the subject's first price choice and the information he receives about the other subject's first price choice is large. Dramatic Social Influence Hypothesis. The larger the difference between a subject's first price choice and the information received about the other subject's first price choice, the more likely that the subject's second price choice will differ from his first price choice. 10 10 It is natural to consider testing a more refined version of this hypothesis: that a subject's second price choice moves in a direction toward the other subject's first price choice. We do not test this hypothesis because our data do not allow us to distinguish this directional shift from a simple regression to the mean. For example, suppose a subject takes all of the 840 on his first choice. It is therefore likely that he observes another subject's first choice that is more other-regarding. If he changes his choice at all, then he can only become more other-regarding on his second choice. The exact opposite pattern can be observed among those subjects who initially select other-regarding first choices.
SEQUENTIAL DICTATOR GAME
255
4. RESULTS
4.1. The Zero Social Influence Hypothesis Figure 1 presents the cumulative price choice distribution for the 20 subjects in the Irrelevant Information treatment and the 40 subjects in the Relevant Information treatment. Figure 1A displays the first price choice, and Fig. 1B displays the second price choice. In both treatments the modal price choice is 840, selected overall by 23 out of 60 subjects in their first decision (38 percent) and by 25 out of 60 subjects in their second decision (42 percent).
FIG. 1.
Cumulative price distribution on (A) first choice and (B) second choice.
256
CASON AND MUI
Our baseline for testing whether social influence matters is the Irrelevant Information treatment. The cumulative price choice distribution in the Irrelevant Information treatment exceeds the distribution in the Relevant Information treatment for prices greater than 86 on the first price choice. However, these two first price choice distributions are not significantly different (KolmogorovSmirnov (KS) p-value=0.51), nor are their means significantly different (t=1.37). The second price choice distributions and means in the two treatments are also not significantly different (KS p-value=0.99 and t=0.06). According to the Zero Social Influence Hypothesis, the change between subjects' second and first price choices in the Relevant Information treatment is not different from the Irrelevant Information treatment. We test this null hypothesis in two ways. First, we compare the frequency at which subjects change their price choices in both treatments. Second, we examine specifically how the first and second price choices differ within each treatment. This hypothesis receives considerable support in the first test that considers only the frequency of price choice changes. Sixteen of the 40 subjects (40 percent) change prices between their first and second decisions in the Relevant Information treatment, and they change their prices by an average of 811.75. Much to our surprise, a larger fraction of subjects11 of 20 (55 percent)change prices between their first and second decisions in the Irrelevant Information treatment, and they change their prices by an average of 810.27. However, the frequency at which subjects change choices in the Relevant Information treatment is not significantly different from the rate in the Irrelevant Information treatment (Fisher's exact test (two-tail) p-value=0.29). The Zero Social Influence Hypothesis receives significantly less support in our second test which examines how the first and second price choices differ within each treatment. The two panels of Fig. 1 describe the price distributions for the first and second choices. The same subjects select prices on the first and second choices, so we cannot compare the distributions from the two panels in the same treatment using standard distributional tests (such as the KolmogorovSmirnov test) that require two independent samples. However, we can test for a significant change across the two price choices by pairing the two prices made by each subject, and evaluating whether the distribution of the difference in the two prices has a mean that is significantly different from zero. As shown on Fig. 1, the mean price choice in the Relevant Information treatment falls from 831.1 to 830.7 between the first and second decisions. This difference (&80.4) is not significantly different from zero (t=0.25; (two-tail) p-value=0.81). By contrast, the mean price choice in the Irrelevant Information treatment increases between the first and second decisions from 827.1 to 830.6. This difference (83.5) is significantly different from zero (t=1.91; (two-tail) p-value=0.072), albeit at a marginal significance level. Since this shift toward self-regarding behavior occurs only in the Irrelevant Information treatment, it suggests that observing relevant information constrains some subjects from moving toward more self-regarding choices. These differencesno change in other-regarding behavior in the Relevant Information treatment and a decrease in other-regarding behavior in the Irrelevant
257
SEQUENTIAL DICTATOR GAME
Information treatmentare also reflected in a classification of how many subjects increase or decrease their price choices between their first and second price decisions. Table 1 shows that only 2 of the 20 subjects (10 percent) in the Irrelevant Information treatment become more other-regarding by selecting a lower price on their second decision, while 9 of the 40 subjects (23 percent) in the Relevant Information treatment become more other-regarding. Nine of the 20 subjects (45 percent) in the Irrelevant Information treatment become more self-regarding by selecting a higher price on their second decision, while only 7 of the 40 subjects (18 percent) in the Relevant Information treatment become more self-regarding. The average amount that the subjects change their prices in all four of these cases exceeds 810. This difference in the direction of change across treatments is statistically significant at the 8-percent level. Conclusion 1. A comparison of the frequency at which subjects change their price choices in the Irrelevant Information and Relevant Information treatments provides support for the Zero Social Influence Hypothesis. However, subjects' second price choices are more self-regarding than their first price choices on average in the Irrelevant Information treatment. This shift toward self-regarding behavior is not observed in the Relevant Information treatment, which provides evidence against the Zero Social Influence Hypothesis. Before conducting the experiment we expected that most subjects would not change their second price choice from their first price choice in the Irrelevant Information treatment, but that more would change their choices in the Relevant Information treatment. Much to our surprise, a (insignificantly) greater percentage of subjects changed their choices in the Irrelevant Information treatment. Recall that the second thought and the information processing effects are operating in both treatments. In the Irrelevant Information treatment the combined second thought and information-processing effects cause subjects to become more self-regarding. The result that subjects do not become more self-regarding in the Relevant Information treatment can be interpreted as follows. The introduction of socially relevant informationthrough its impact on subjects' beliefsinduces some subjects to shift toward other-regarding behavior in the Relevant Information treatment. But this social information impact appears just sufficient to offset the combined second thought and the information-processing effects. TABLE 1 Frequency of Shifts in Other-Regarding and Self-Regarding Direction by Treatment
Irrelevant information Relevant information Total (column)
Own second price
No change between first and second price
Own second price>Own first price
Total (row)
2 9 11
9 24 33
9 7 16
20 40 60
Note. Fisher's exact test (two-tail) p-value=0.08.
258
CASON AND MUI
Conclusion 1 indicates that the evidence regarding the Zero Social Influence Hypothesis is mixed. This could be due, in part, to subject heterogeneity; e.g., some subjects may respond to socially relevant information, while others do not. In the next subsection, we examine this heterogeneity directly, by identifying some characteristics of subjects who are more likely to change price choices between their first and second decisions. 4.2. Differential and Dramatic Social Influence Hypotheses According to the Differential Social Influence Hypothesis, more self-regarding subjects are less likely to select a second price choice that is different from their first price choice. Table 2 indicates that this hypothesis receives considerable support in both treatments. Subjects who select a first price of 840 are significantly less likely to change price on their second choice. For example, in the Relevant Information treatment only 3 of the 17 subjects who chose a first price of 840 changed their price on their second choice, while 13 of the 23 subjects who chose a first price less than 840 changed their price on their second choice. These differences are highly significant, and are also highly significant in the Irrelevant Information treatment. 11 Table 3 presents an alternative test of this hypothesis using a probit model. This model estimates the likelihood that a subject changes her price between her first and second price decisions as a function of her first price choice. The negative and significant coefficient estimates reported in columns (1) and (2) on the first choice variable indicate that more self-regarding first price choices reduce the likelihood that a subject changes her price between her first and second price choices. Conclusion 2. Subjects who exhibit more self-regarding behavior on their first price choice are less likely to change prices between their first and second decisions. Finally, we test the Dramatic Social Influence Hypothesis, which states that more dramatic information is more likely to cause subjects to change their price choices. Subjects learned the actual price choice made by another subject only in the Relevant Information treatment, so we test this hypothesis for this treatment only. Information is more dramatic if the difference between the subject's own first price choice and the other subject's first price choice is large. The median absolute difference between these two choices was 810, so we define information received by a subject as dramatic when the absolute difference between her first price choice and her paired subject's first price choice is larger than or equal to 810 (that is, |own first choice-paired subject's first choice| 810). Eleven of the 22 subjects (50 percent) who observe dramatic information change their price choice between their first and second decisions, and they change their price by an average of 813.50. By contrast, only 5 of the 18 subjects (28 percent) who observe non-dramatic information change their price, and they change their
11 Alternative definitions to classify subjects into self-regarding and other-regarding groups give very similar results; e.g., classifying a subject as other-regarding if he selected a price less than or equal to the median first price in that treatment.
259
SEQUENTIAL DICTATOR GAME
TABLE 2 Frequency of Changes in Price between the First and Second Choices, by Initial Choice Change price between first Do not change price between and second choices first and second choices
Total (row)
Relevant information a First price=840 First price<840 Total (column)
3 13 16
14 10 24
17 23 40
6 3 9
6 14 20
Irrelevant information b First price=840 First price<840 Total (column) a b
0 11 11
Fisher's exact test (one-tail) p-value=0.014. Fisher's exact test (one-tail) p-value=0.002.
price by an average of 88.00. The difference in the rate the these groups change prices is not quite statistically significant (Fisher's exact test (one-tail) p-value= 0.135). We also augmented the probit model in Table 3 to include the absolute difference variable |own first choice-paired subject's first choice|. As shown in columns (3) and (4), the coefficient on this variable has the predicted positive sign, but is not significantly different from zero. Conclusion 3. Subjects who receive dramatic information are not significantly more likely to change their price choice between their first and second decisions. TABLE 3 Estimates of Change Likelihood Model (Probit Model Maximum Likelihood Estimation) Irrelevant information (1)
(2)
First price choice (Standard error) | Own first priceother's first price | (Standard error) Intercept (Standard error)
&0.106 b (0.042)
&0.036 b (0.021)
3.062 b (1.209)
Observations Log-likelihood Restricted (slopes=0) Log-likelihood
Variable or statistic
Relevant information (3) (4)
0.027 (0.025)
&0.033 a (0.022) 0.015 (0.027)
0.864 (0.679)
&0.538 (0.337)
0.597 (0.826)
20 &8.97
40 &25.34
40 &26.34
40 &25.17
&13.76
&26.92
&26.92
&26.92
Note. Dependent variable=1 if change offer; dependent variable=0 otherwise. a Denotes significantly different from 0 at 10 percent level. b Denotes significantly different from 0 at 5 0 level (one-tailed tests, except for Intercept).
260
CASON AND MUI
4.3. A Strategy Method Hypothesis and Results We conclude this section with a brief presentation of a supplementary Strategy Method Hypothesis and results. Experimental economists have recently begun to use the strategy method to elicit subject choices (for example, Selten 6 Buchta, 1994; Straub 6 Murnighan, 1995). In this method, subjects specify a complete strategy, rather than only choosing an action in realized information sets. Therefore, in the Strategy Method treatment implemented here, subjects received no information between their two price decisions. After submitting their first price, subjects filled out a second form specifying a complete price choice strategy contingent on all 21 possible first choice prices selected by the subject with whom they were randomly paired. The instructions included an extensive example and emphasized that these contingent price choices were binding and that one would be carried out depending on the first price choice of another anonymous subject. Forty additional subjects participated in this treatment. We examined this strategy method approach in the sequential dictator game by evaluating the following null Strategy Method Hypothesis: Choices in the Strategy Method treatment are not different from choices in the Relevant Information treatment. Roth (1995) points out that there are at least two reasons why subjects might give different responses in the strategy and non-strategy methods. First, in the strategy method, subjects are required to simultaneously make all potential choices at the same time. When the game is sufficiently complex, for example, when it has many information sets and substantial strategic interdependence, the transformation to the strategy method amounts to a significant change in the representation of the strategic environment. Our sequential dictator game is extremely simple and strategic interdependence is nonexistent, however, so this effect is probably absent. Second, in the strategy method subjects are required to submit the complete allocation strategies contingent on every possible choice by another subject. Consequently, there is a reduced incentive to think carefully about every possible element of the strategy. This could lead to more variance in choices in the Strategy Method treatment (Walker 6 Smith, 1993). For example, subjects can reasonably expect that the likelihood that the first subject chooses a very other-regarding first price such as 82 is very low, so the expected payoff consequences for their price choice contingent on another first price of 82 is also very low. This second effect is likely to be important in the sequential dictator game. Besides these reasons, additional psychological considerations may cause subjects to behave differently under the two methods. Psychological research indicates that individuals attach too much weight to salient or vivid events and too little weight to non-salient events. Events that do occur are more real and immediate, and hence more concrete than the corresponding potential events that could have occurred. The greater salience of events that have actually occurred are more available for inference than events that have not occurred (Nisbett 6 Ross, 1980). Information provided in the Relevant Information treatment is based on choices that have in fact been made by another subject, rather than on choices that could have been made by others. This information is therefore more likely to have an impact on subjects' choices because of its vividness.
SEQUENTIAL DICTATOR GAME
261
We tested the Strategy Method Hypothesis by comparing the first price choices in both treatments. We also compared the second price choices in the Relevant Information treatment with the realized second price choices in the Strategy Method treatment (i.e., the realized second prices that were contingent uponand thus determined bythe actual first price choice of the randomly-paired other subject). In no cases are the distributions, means, or fraction of subjects choosing 840 significantly different, so we conclude that use of the Strategy Method in this setting does not significantly alter choices. 12 This finding suggests that it is appropriate to pool the data in the Strategy Method and the Relevant Information treatments when testing the social influence hypotheses. All of the results presented above are qualitatively unchanged after pooling the data. In some cases, the results become more statistically significant because of the larger sample size. For example, the result that subjects in the Irrelevant Information treatment shift more in the selfregarding direction (shown in Table 1) compared to the pooled Relevant InformationStrategy Method data is significant at the 2-percent level.
5. DISCUSSION
In this experiment subjects made dictator allocations of 840 before and after learning the allocation made by one other subject in the Relevant Information treatment, or before and after learning the birthday of one other subject in the Irrelevant Information treatment. Subjects on average became more self-regarding in the Irrelevant Information treatment, but did not become more self-regarding in the Relevant Information treatment. We also observe that subjects who exhibit more self-regarding behavior on their first decisions are less likely to change choices between their first and second decisions. Ever since the classic study on conformism by Asch (1951), psychologists have been concerned about discerning possible channels through which social influence occurs in different settings. In the Asch experiment, subjects were first shown a straight line and then three lines for comparison. Subjects were then asked to indicate which of these three comparison lines was closest to the first line in length. Each subject was placed in a group consisting of five to seven confederates of the researcher. The confederatesmost of whom provided responses before the subjectgave the same wrong answer. A significant number of subjects conformed to the incorrect answer provided by the confederates. Inspired by the Asch experiment and follow-up studies, two explanations have emerged as the principal mechanisms of social influence (Deutsch 6 Gerald, 1955). First, people may change their behavior upon observing how others behave in similar situations in order to obtain certain rewards or avoid potential punishments. For example, it is highly plausible that subjects in the Asch experiment conform to the view of the majority because they want to be accepted by the 12 The one relatively minor difference between the Strategy Method and Relevant Information treatments is the rate of price changes between the first and second choices. Twenty-seven of 40 subjects change prices in the Strategy Method treatment, while 16 of 40 subjects change prices in the Relevant Information treatment (Fisher's exact test (two-tail) p-value=0.03).
262
CASON AND MUI
majority or to avoid the discomfort for disagreeing with the majority. Besides this normative influence, social influence can also occur through informational influence. When individuals are not completely certain about what is the right behavior in a particular context, they may regard others' choices in similar situations as sources of useful information. 13 In our experiment, we believe that the main source of social influence is informational influence rather than normative influence. Since a subject's choice is not observed by other subjectsespecially the subject whose choice he learns in the experimenta subject does not need to be concerned about any potential embarrassment for not conforming to what he conjectures to be the ``appropriate'' behavior of others participating in the experiment. 14 This distinction between the normative and informational influences as different sources of social influence raises the natural question of whether social influence has permanent effect on people's behavior. As psychologists have long emphasized, when social influence operates mainly through informational influencethat is, when knowledge about others' behavior causes an individual to change her belief regarding what constitutes the appropriate or correct behaviorit is more likely that social influence will have a long-lasting effect on people's behavior (Aronson, 1995, Chapter 2). On the other hand, when social influence operates mainly through normative influencefor example, when an individual conforms to others' behavior in order to avoid embarrassmentthen it is less likely that social influence will have a permanent effect. 15 For example, in the Asch experiment, the incidence of conformism varies significantly across treatments when the level of privacy is varied. In the original treatment, having observed the choices of others, a subject provides his response verbally in the presence of the entire group. Later research discovered that in treatments in which subjects make their responses privately, the rate of conformism decreased considerably (Deustch 6 Gerald, 1955; Argyle, 1957). In contrast, in other studies of social influence in which arguably, social influence operates mainly through informational influence, many subjects exhibited permanent 13
In the Asch experiment, the lengths of the lines are such that it would be extremely easy for subjects to identify the correct answer. Given the simplicity of this cognitive task, it is quite unlikely that informational considerations can be the main source of social influence. However, when the tasks involved require subjects to make decisions that are (partially) affected by their belief regarding what constitutes socially appropriate behavior as in this experiment, or when the task is cognitively demanding, one would expect that informational considerations become more important. 14 We note, however, that since a subject's choice is observed by the experimenter it is plausible that a subject may be concerned about whether such behavior will be judged favorably or unfavorably by the experimenter (the observer effect). But the observer effect is also present when a subject makes his first price choice in all treatments, so any change that may occur in his second choice is unlikely to be explained by the observer effect alone. 15 Without going into details, we note that the difference in fact is more subtle than we characterize in the text. It is possible that an individual is motivated by normative influence to follow the behavior of others originally. However, having made the ``appropriate'' changes in his behavior, he may eventually adjust his own belief to make it consistent with his new choices. Once such internalization of beliefs has occurred, the new belief may become independent of its source and will become extremely resistant to change.
SEQUENTIAL DICTATOR GAME
263
changes in their beliefs (see, for example, Newcomb, Koenig, Flacks, 6 Warwick, 1967). In the past two decades, economists have come to recognize better the importance of social influence in affecting economic decisions. For example, Akerlof (1980) develops a formal model of social custom and uses it to provide an explanation for equilibrium involuntary unemployment. Drawing on the findings of organizational theorists and social psychologists, Jones (1984) and Levine (1991) have both argued that workers have propensities to choose their work efforts in conformism to the work norms in their work group, and explore the implications of this observation for the study of organizational design and wage dispersion. Banerjee (1992) and Bikhchandani, Hirshleifer, 6 Welch (1992) show how informational cascades can lead to conformity behavior arising from purely informational reasons. By contrast, Akerlof (1983) and Jones (1984, Chapter 5) present economic models that focus on how people's experience can change their beliefs permanently. 16 The above observations regarding the two different channels through which social influence operates suggest the need to investigate further the dynamic aspects of social influence in economic settings. For example, a person's choices with regard to how much to give to charity, or whether to behave more or less cooperatively in bargaining situations, are affected by both his belief regarding what constitutes appropriate behavior in these situations, as well as the economic and social incentives he faces in a particular situation. Society as a whole invests substantial resources in shaping people's beliefs through its educational, social and religious institutions, and some business organizations also expend resources to cultivate a corporate culture that they believe will enhance employee loyalty, cooperation and productivity. Given the importance of these social influence activities, it is important that economists conduct future theoretical and field studies as well as laboratory experiments to determine under what conditions these influences have a permanent effect on beliefs and economic behavior.
REFERENCES Akerlof, G. A. (1997). Social distance and social decisions. Econometrica, 65, 10051028. Akerlof, G. A. (1983). Loyalty filters. American Economic Review, 73, 5463. Akerlof, G. A. (1980). A theory of social custom, of which unemployment may be one consequence. Quarterly Journal of Economics, 95, 74975. Andreoni, J. (1995). Cooperation in public-goods experiment: kindness or confusion? American Economic Review, 85, 891904. Argyle, M. (1957). Social pressure in public and private situations. Journal of Abnormal and Social Psychology, 54, 172175.
16 For more recent work on the importance of social influence in the economics literature, see, for example, Akerlof (1997), Becker (1996), Bernheim (1994), Fershtman, Murphy, 6 Weiss (1996), and Kuran (1995). For a recent survey on the importance of social and psychological considerations in economic decision-making, see Rabin (1998).
264
CASON AND MUI
Aronson, E. (1995). The social animal. Seventh Ed. New York: Freeman. Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgements. In H. H. Guetzkow (Ed.), Groups, leadership, and men. Madison, NJ: Carnegie Press. Banerjee, A. (1992). A Simple model of herd behavior. Quarterly Journal of Economics, 107, 797817. Becker, G. S. (1996). Accounting for tastes. Cambridge, MA: Harvard Univ. Press. Berg, J., Dichhaut, J., 6 McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10, 122142. Bernheim, B. D. (1994). A theory of conformity. Journal of Political Economy, 102, 841877. Bikhchandani, S., Hirshleifer, D., 6 Welch, I. (1992). A theory of fads, fashion, and cultural change as informational cascades. Journal of Political Economy, 100, 9921026. Bolton, G. E., Katok, E., 6 Zwick, R. (1997). Dictator game giving: rules of fairness versus acts of kindness. International Journal of Game Theory, to appear. Bushman, B. J. (1988). The effects of apparel on compliance: A field experiment with a female authority figure. Personality and Social Psychology Bulletin, 14, 459467. Camerer, C., 6 Thaler, R. (1995). Anomalies: Ultimatums, dictators, and manners. Journal of Economic Perspectives, 9, 209219. Cason, T. N., 6 Mui, V.-L. (1997). A laboratory study of group polarization in the team dictator game. Economic Journal, 107, 14651483. Croson, R. (1995). Shirking and working: an experiment in team production. Mimeo, Wharton School of Business, University of Pennsylvania. Deutsch, M., 6 Gerald, H. B. (1955). A study of normative and informational social influence upon individual judgement. Journal of Abnormal and Social Psychology, 51, 629636. Fershtman, C., Murphy, K. M., 6 Weiss, Y. (1996). Social status, education, and growth. Journal of Political Economy, 104, 108132. Forsythe, R., Horowitz, J., Savin, N. E., 6 Sefton, M. (1994). Fairness in simple bargaining experiments. Games and Economic Behavior, 6, 347369. Hoffman, E., McCabe, K., Sachat, K., 6 Smith, V. L. (1994). Preferences, property rights and anonymity in bargaining games. Games and Economic Behavior, 7, 346380. Hoffman, E., McCabe, K., 6 Smith, V. L. (1996). Social distance and other-regarding behavior in dictator games. American Economic Review, 86, 653660. Jones, S. (1984). The economics of conformism. Oxford: Blackwell Sci. Kuran, T. (1995). Private truths, public lies. Cambridge, MA: Harvard Univ. Press. Levine, D. (1991). Cohesiveness, productivity, and wage dispersion. Journal of Economic Behavior and Organization, 15, 237255. Newcomb, T. M., Koenig, K. E., Flacks, R., 6 Warwick, D. P. (1967). Persistence and change: Bennington College and its students after twenty-five years. New York: Wiley. Nisbett, R. E., 6 Ross, L. (1980). Human inferences: Strategies and shortcomings of social judgement Englewood Cliffs, NJ: Prentice Hall International. Rabin, M. (1998). Psychology and economics. Journal of Economic Literature, 36, 1146. Roth, A. E. (1995). Bargaining experiments. In J. Kagal, 6 A. E. Roth (Eds.), Handbook of experimental economics. Princeton, NJ: Princeton Univ. Press. Sell, J., 6 Wilson, R. (1991). Levels of information and contributions to public goods. Social Forces, 70, 107124. Selten, R., 6 Buchta, J. (1994). Experimental sealed bid first price auctions with directly observed bid functions. Discussion Paper B-270, University of Bonn. Straub, P. G., 6 Murnighan, J. K. (1995). An experimental investigation of ultimatum games: Information, fairness, expectations, and lowest acceptable offers. Journal of Economic Behavior and Organization, 27, 345364.
SEQUENTIAL DICTATOR GAME
265
Walker, J. M., 6 Smith, V. L. (1993). Monetary rewards and decision costs. Economic Inquiry, 31, 245261. Weimann, J. (1994). Individual behavior in a free riding experiment. Journal of Public Economics, 54, 185200. Received: January 28, 1998