«Nicholas Bardsley National Centre for Research Methods University of Southampton ESRC National Centre for Research Methods NCRM Working Paper Series ...»
Dictator Game Giving: Altruism or Artefact?
National Centre for Research Methods
University of Southampton
ESRC National Centre for Research Methods
NCRM Working Paper Series
Dictator Game Giving: Altruism or Artefact?
National Centre for Research Methods
University of Southampton
This is the final version, forthcoming in Experimental Economics.
The original publication is available at www.springerlink.com.
First Version April 2005 Accepted April 27 2007 Dictator Game Giving: Altruism or Artefact?* Abstract Experimental dictator games have been used to explore unselfish behaviour. Evidence is presented here, however, that subjects’ generosity can be reversed by allowing them to take a partner’s money. Dictator game giving therefore does not reveal concern for consequences to others existing independently of the environment, as posited in rational choice theory. It may instead be an artefact of experimentation. Alternatively, evaluations of options depend on the composition of the choice set. Implications of these possibilities are explored for experimental methodology and charitable donations respectively. The data favour the artefact interpretation, suggesting that demand characteristics of experimental protocols merit investigation, and that economic analysis should not exclude context-specific social norms.
Keywords altruism, artificiality, experiments, methodology JEL classifications C91, C70, D63, D64 Experimental economists have investigated unselfish behaviour with a familiar set of simple games (Camerer 2003, p43-100). The simplest is the dictator game, in which subjects have to decide how much, if any, of an endowment to give to another subject, typically in a one-shot anonymous setting. Usually transfers are plentiful; it is common for over 50% of subjects to give money away (Camerer 2003, p57-58). Several models of unselfishness have been put forward within the parsimonious framework of rational choice theory. These theories compete over data from dictator games and other experiments, and include models of inequality aversion (Bolton and Ockenfels 1998, Fehr and Schmidt 1999), altruism (Andreoni and Miller 2002), egocentrism (Cox et al. 2002) and Rawlsian “social welfare” preferences (Charness and Rabin 2002). Such models continue to attract intensive research effort.
Whilst dictator giving has proven hard, though not impossible, to eradicate in the lab, some experimentalists still express unease about the results.1 A common concern is that people could always make anonymous donations to random strangers in everyday life, for example by mailing cash to persons sampled from the telephone directory, but few if any choose to do so. Transfers are made instead to family members, specific organisations or face-to-face to people requesting money.
Since we face the dictator game all day every day, it could be argued, the experimental design would appear to score highly on conventional standards of ex-ante external validity. It presents a familiar set of opportunities in a task which makes minimal cognitive demands. Yet the results appear to be misleading about the extent of faceless interpersonal altruism.
Giving is also notoriously sensitive to experimental treatments. This adds to the uncertainty about how the results from dictator games generalise to naturally-occurring contexts. See, for examples, Eckel and Grossman (1996) on anonymity and charities versus individuals as recipients, Hoffman et al. (1996) and Bohnet and Frey (1999) on social distance, Cherry et al. (2002) on earned versus unearned income, and Haley and Fessler (2005) on visual suggestions of observation.
The evidence has tended to be that giving is sensitive to factors for which theory makes no prediction, or which can be explained by theory. Intriguingly, theoretically-motivated sensitivity tests reported by Andreoni and Miller (2002), found that rational choice theory fits dictator game data surprisingly well.
On the other hand, there is evidence from related games that behaviour is also sensitive to factors which are irrelevant according to consequentialist theories, such as intentions. For example, in ultimatum games Falk, Fehr and Fischbacher (2003) find that respondents’ behaviour depends on payoffs to the proposer in avoided subgames. Similar results are offered by Brandts and Solà (2001). In a variant on dictator games, Dana et al. (2005) find that introducing uncertainty over the recipient’s outcomes reduces egalitarian choices even when the uncertainty can be costlessly resolved. This result is discussed further in section 3 below and seems to indicate that much giving is not motivated by a desire for fairer outcomes for their own sake, but for a further motive. Because of this kind of data, orthodox accounts of social preferences might be better seen as attempts at reasonable approximations of specific motivations, for limited domains, than at literal general truths.
The above synopsis of the existing evidence suggests that dictators’ giving is probably more volatile than a literal interpretation of social preference theories might suggest, and does not translate straightforwardly into insights about behaviour in the field. However, the possibility still appears open that dictator game data is explained by subjects’ underlying tastes, values, goals and so on, imparting value, whether directly or indirectly, to the other’s payoff. The experiments reported below investigate directly whether dictator games measure social attitudes, in this broader sense. Much, perhaps even most, of the dictator games literature supposes that subjects arrive at the lab with such dispositions, which then determine their satisfaction with the initial allocations. If this process explains giving, subjects should also want to give if taking is possible. The experiments compare dictator games to “taking” games which introduce taking options; the proportion of subjects willing to give should be unchanged. This prediction is quite intuitive, but is also an implication of rational choice theory, as shown in section 1 below.
In contrast with many studies of dictator games, but in common with Andreoni and Miller (2002), the present study is a theoretically-driven sensitivity test. It focuses on a subject’s willingness to give, in abstraction from the amount given. The treatment manipulation is ineffective assuming that the endowment is consistently regarded as either too selfish, too generous or just right along budget lines which implement the same price of transfers. In contrast to previous studies, therefore, it tests this basic qualitative proposition implied by common readings of the design. The bold claim that a given social preference theory is confirmed by the data requires this postulate and more, for there subjects’ attitudes determine a well-behaved indifference map over outcomes. The test conducted is lenient towards social preferences explanations of giving, since it allows gifts to vary in size, as predictably occurs with any change in the environment, without rejection of the null hypothesis. The null is only rejected if there is evidence that some subjects’ willingness to give can be completely eliminated by taking options.
1 Design Three experiments were conducted, with two treatments in each. The design compares a dictator game (treatment 1; T1) to a “taking game” (treatment 2; T2), which introduces taking options. The treatments are represented in figure 1 below. Let (a, b) represent an allocation of £a to a dictator and £b to her partner, B denote a dictator game budget line and e the endowment point. B has the same gradient and endowment point as the budget line in T2, denoted B'but B', extends to the south east of B.2
Social preference theories posit convex indifference curves in (a, b) space. Points on a budget line further away from the optimum in a given direction lie on progressively lower indifference curves.
Suppose a subject starting at e, facing B, gives money because her optimum is to the northeast of e.
Then she should also like to give facing B' since an indifference curve can be tangential with a, budget line, of arbitrary length, at one point only. In experiments 1 and 2 the treatments provide identical giving options, that is, B ∩ B’ = B. The population proportions giving should therefore be equal. In experiment 3, in contrast, subjects cannot give in T2 so B ∩ B’ = e. So there, if the optimum represents a positive transfer, e should be chosen from B’. The proportion of givers in T1 should therefore not exceed that choosing e in T2. In each case the null hypothesis is that of equality of the relevant proportions, with a two-sided alternative hypothesis.
Common procedures were as follows. The experiments were run as pen-and-paper exercises, with dictators choosing payoffs from a table. Subjects were mixed-gender undergraduates. Separate samples were used for each treatment to eliminate deliberate responses to experimental manipulation. T1 and T2 were run at different times of day for organisational reasons, with the order varied across experiments. Subjects in one room were randomly matched with a partner in another. Subjects in the receiver role did not have any task to perform but were fully informed; they simply awaited the outcome. To preserve anonymity, dictators were seated apart and made their decision privately. The experimenter issued instructions, available on the journal website, which were also read aloud, but was not present when decisions were made.3 Assistants then administered and made payments, who were uninformed about the design, so the experimenter and assistants were blind to individual behaviour. The resulting double anonymity between subjects and experimenters was emphasised in the instructions. T1’ s instructions abridged T2’ s, making only the changes required to alter the choice set. As conventionally conceived, this holds the framing constant across treatments. Basic details of each experiment are given in sections 2.1-2.3; for full results and statistical significance see figure 2 and table 1 of section 2 respectively.
2 Experiments 1-3
2.1 Experiment 1 Experiment 1 implemented equal endowments (£6 each) and scaled-up transfers, following Cox et al. (2002) who provided initial evidence of dictator taking. In T1 dictators could transfer £(0 - 4), in increments of £0.33p. Transfers were doubled. In T2 dictators could make the same gifts, or take up to £2 at the same rate: £1 received cost their partner £2. Subjects also received £2 show-up fee. Experiment 1 was conducted in June 2004, with 110 subjects. T2 was implemented before T1.
In T1 the proportion of subjects giving, p1, was 6/26. In T2 this proportion, p2, was just G1, is significantly different from zero (p.05).4 The 1/29. The difference in proportions, p1-p2 extent of the inconsistency is unclear, however, since few subjects gave in T1. This problem motivated experiment 2, using more normal dictator game parameters. Also, experiment 1 suggested that giving might be extremely fragile; experiment 2 made taking options a smaller proportion of the choice set to test this.
2.2 Experiment 2 Experiment 2 used transfers at a rate of £1:1 and unequal endowments. Subjects were informed they had a show-up fee of £4 each. Dictators in both treatments were given an additional £7. In T1 they could give £(0 - 7). In T2 they could give £(0 - 7) or take £(0 - 2) of their partner’ s show-up fee. Transfers were made in increments of £0.50p. The experiment was conducted over two dates, in November and December 2004, with 130 subjects. T1 was implemented before T2.
Results were in the same direction as for experiment 1, with 23/33 subjects giving in T1 and 15/32 giving in T2, but only significant at the 10% level. However, taking options in T2 constitute a smaller proportion of the choice set. A sterner test of dictator giving was therefore conducted next, removing strictly positive transfers from T2.
2.3 Experiment 3 Experiment 3 also implemented £1:1 transfers and unequal endowments. Dictators were informed they had an endowment of £10 and their partner £5. In T1, dictators could give £(0 - 3). In T2, dictators could take £(0 - 3), but could not give. Transfers were in increments of £0.25p.
Experiment 3 was conducted in February 2005 with 120 subjects. T2 was implemented before T1.