Introduction

The dictator game has become a celebrated workhorse of experimental economics and social psychology (Engel, 2011; Güth & Tietz, 1990). In the standard version of the game an individual, the dictator, is given money, say $10, and chooses how to split this money between themselves and some other person, the receiver. A ‘selfish’ dictator would keep all the money for themself, but a significant proportion of individuals choose to share money with the receiver. Indeed, it is relatively common for the dictator to split the $10 (Camerer, 2011; Cartwright, 2018). By studying subtle variations in the framing of the dictator game, researchers have obtained critical insight on the underlying reasons why individuals are willing to give money to others.

In most dictator game experimental studies, the dictator and receiver are two subjects sitting in an experimental laboratory. A growing number of studies, however, position the receiver as a charity (e.g. Bekkers, 2007, Bachke et al., 2014). In this ‘charity version’ of the dictator game, the dictator must decide how much of the $10 to keep and how much to donate to charity. Simple though it may seem, this charity dictator game has been used in dozens of studies and provided critical insight on factors that influence individual’s willingness to donate to charity. For instance, comparing donations to local versus national charities. The game has also been used to test and evaluate interventions aimed at increasing giving, such as rebates, lotteries, matched funding and audio-visual primes.

In this paper, as part of the Special Issue on ‘Researching the Third Sector: Approaches, Methods, and Applications’ we provide an introductory overview of charity dictator games. Our objective is to inform interested researchers and practitioners on how to use the method as well as its main advantages and limitations. In Sects. 2 and 3, we introduce the charity dictator game and identify a range of experimental design issues. In Sect. 4, we summarize findings on the external validity of the game. In Sect. 5, we set out research questions that can be effectively studied using the game. In the supplementary material we provide additional background information.

Charity Dictator Games and Warm Glow

One of the key advantages of the charity dictator game is its simplicity. An individual is given an endowment of money and asked how much of that money they would like to donate to charity. Any money they donate is sent off to the charity and the remainder of the money is theirs to keep. Despite this simplicity there are still many possible permutations and variants of the game that can and have been analysed. Studying such subtle variants allow us to unpick the motives behind charitable giving. By way of illustrative example, we briefly look at how the charity dictator game can inform on warm glow giving.

We begin with the pioneering paper of Eckel and Grossman (1996) which, to the best of our knowledge, contains the first use of a charity dictator game. Their study had two laboratory based experimental treatments. One treatment replicated a version of the standard dictator game implemented by Hoffman et al. (1994). In this treatment, subjects were randomly assigned the roles of dictator or receiver. The dictator was given ten one dollar bills ($10), put their ‘donation’ in an envelope and left. The envelope was given to the receiver who was sitting in a room next door. In the second treatment, all subjects were dictators and any money donated was given to a local branch of the American Red Cross. In both treatments, an individual motivated purely by self-interest would simply keep the $10 and give $0 to the receiver. This is what many subjects did. On average, however, giving was positive and significantly higher when the receiver was the American Red Cross compared to another subject, average of $3 versus $1.50. This was interpreted as evidence for other-regarding preferences with donations higher for a more ‘deserving’ receiver.

Consider next the study of Crumpler and Grossman (2008) designed to distinguish warm glow from altruism. In their experiment the amount given to the charity was fixed at $10 irrespective of how much the dictator donated.Footnote 1 For instance, if the dictator donated $4 then they would end up with $6 and $10 would be given to the charity ($4 from the dictator and $6 from the experimenter). An individual motivated by pure altruism would have no reason to donate in this experiment because their giving does nothing to help the charity. Someone motivated by warm glow, by contrast, who gets utility from the act of giving, still has an incentive to donate (Andreoni, 1989). Crumpler and Grossman (2008) found that 57% of subjects who passed understanding checks donated to charity. This result suggests that warm glow can explain a large proportion of giving. Informally, with average giving of $2 in the study of Crumpler and Grossman (2008) compared to $3 in Eckel and Grossman (1996) warm glow seemingly accounts for two thirds of giving.

In a follow up study, Luccasen and Grossman (2017) confirm and extend the results of Crumpler and Grossman (2008). Specifically, they find that individuals give more when they have earned the endowment and less if they have an option to reduce the amount of money the charity will receive. This shows that warm glow can be sensitive to context, consistent with models of self-image (e.g. Andreoni & Bernheim, 2009; Bénabou & Tirole, 2006). For instance, an individual may feel a larger warm glow from donating money they have earned. In application, such results imply that taxation funded government spending will not sizably crowd out private charitable donations. Illustrating how the charity dictator game can inform on important policy debate (De Wit & Bekkers, 2017).

Variants of Charity Dictator Game experiments

In this section we provide further examples of the charity dictator game, with a particular focus on illustrating different experimental design choices. Consider first a study by Eckel and Grossman (2000, 2003). Participants were asked twelve separate decisions before rolling a 12-sided dice to decide which of the decisions would be enacted. The decisions varied in terms of endowment and the price of giving, replicating the effect of rebates and matched funding. Here are three examples of decision tasks, expressed in tokens worth $0.10 each: (a) Benchmark case—‘Endowment of 800 tokens; for every token you pass to the charity, the charity will receive one token’. (b) Matched funding—‘Endowment of 800 tokens; for every token you pass to the charity, the experimenter will match it with one additional token’. (c) Rebate—‘Endowment of 1200 tokens; for every token you pass to the charity, the experimenter will refund to you one-quarter token’.

The aforementioned experimental designs nicely illustrate the wide range of research questions on which charity dictator games can give insight. For example, Eckel and Grossman (2003) find that females and those actively engaged in religious activities gave relatively more. By varying the endowment and price of giving, they were able to estimate the income elasticity of giving (0.77–1.03), as well as the rebate and matching price elasticity (− 0.34 and − 1.07, respectively).Footnote 2 Moreover, they were able to directly compare a rebate frame against a matching frame, finding that a matching frame caused an effect on giving approximately three times larger than the rebate frame. Hence, matched funding appears more effective, ceteris paribus, than offering a rebate.

The approach used by Eckel and Grossman (2000, 2003), whereby an individual subject is exposed to twelve choice scenarios, is a within-subject design. This approach is valuable in terms of gathering lots of data, since each subject makes twelve decisions across different settings. Moreover, because the same subject is observed in multiple choice settings, this approach controls for unobserved heterogeneity. It does, however, have potential drawbacks. In particular, it increases the complexity of the experiment, may lead to confusion, and may cause experimenter demand effects where subjects make choices they think the ‘experimenter would want’ (Zizzo, 2010). For instance, subjects may mistakenly think that a 50% rebate is equivalent to a 50% match and adjust giving accordingly.

The alternative to a within-subject design is a between-subject design in which each subject is exposed to just one choice decision. This has the advantage of maintaining simplicity and avoiding experimenter demand effects. However, to make telling inferences on some issues, such as income and price elasticity of demand, a lot of experimental subjects may be required, which can be costly. Consequently, a hybrid approach may be appropriate. An illustrative example is Eckel and Grossman (2006). Half of subjects in their study were exposed to 12 decision tasks involving a rebate and the other half of subjects were exposed to 12 decision tasks involving matched giving. This allows a within-subject estimate of income and price elasticities together with a between-subject estimate on a rebate versus matched funding. Eckel and Grossman (2006) replicated their core finding that matched funding increases donations, ceteris paribus, compared to a rebate.

Another issue to consider in designing charity dictator games is the role of anonymity. Since the studies of Hoffman et al., (1994, 1996) there has been an interest in double-blind implementations of the dictator game, characterised by the experimenter not being able to observe what any particular subject has donated. The study by Fielding and Knowles (2015) illustrates the issue. Subjects were paid $19 for completing a survey and additional tasks. Subsequently, they were invited to put money in a charity donation box positioned at the exit of the room. Crucially, when subjects put money in the donation box they could tell that nobody was watching them. Thus, subjects were free to give without being observed, consistent with a double-blind protocol. In order that the experimenters could identify giving, each subject was given a unique set of coins distinguished by the year of issue written on the coin and whether the coin was polished on one or both sides. This allowed the experimenters to identify the exact distribution of individual giving based on the coins in the donation box.Footnote 3 As can be seen in this example, a double-blind protocol typically requires some ingenuity on the part of the experimenters. Ultimately, however, once a protocol is designed, the implementation need not be any more complex than a standard charity dictator game. In Fielding and Knowles (2015) study, for instance, subjects simply put money in a donation box.

All of the experiments we have discussed so far were lab based experiments involving students. An advantage of the charity dictator game approach, however, is that it can easily be taken into the field. One example is the study of Bekkers (2007). Subjects in the study were a large sample of the Dutch population (n > 1000) completing a national survey. For completing the survey, respondents ‘earned’ an amount (averaging €9) that depended on how long it took them to complete the survey. Subjects were then given the option to take the money in vouchers, air-miles or donate to one of three charities. This field environment is similar to the laboratory based setting of Fielding and Knowles (2015). We have though a more representative sample of the population in a natural setting. We will discuss differences between laboratory and field giving in the next section. At this point we highlight three aspects of the Bekkers (2007) experiment design worth considering.

The first is whether the money given to the dictator is a ‘windfall’ or ‘earned’. In some settings, e.g. Eckel and Grossman (2003, 2006), the endowment is a ‘windfall’ because it is given to the dictator for participating in an experiment. In other settings, e.g. Bekkers (2007), the money can be viewed as ‘earned’ through the completion of a non-negligible task. If money is ‘earned’, then subjects appear significantly less willing to give (Carlsson et al., 2013; Li et al., 2019). A second aspect to consider is whether the endowment is cash in hand, e.g. Fielding and Knowles (2015), a voucher, e.g. Bekkers (2007), or a delayed payment. There is evidence that giving is reduced when the endowment is more tangible, such as, cash in hand (Reinstein & Riener, 2012). A third aspect of the Bekkers (2007) design is that the dictator had an all-or-nothing decision. This compares to the more common version of the charity dictator game where the subject can donate a portion of their endowment to charity. Outcomes have been shown to depend on constraints to giving (Cartwright & Mirza, 2019).

A final design issue we will highlight is the size of endowment and whether the decision involves real or hypothetical rewards. A meta-analysis of standard dictator games (which excludes charity dictator games) found that giving in dictator games was less, but not by much, the higher the endowment. Studies, again on the standard dictator game, have also found subtle differences in giving with hypothetical rewards (Ben-Ner et al., 2008; Bühren & Kundt, 2015). In terms of all the experiment design issues we have discussed—between-subject versus within-subject, double-blind, constraints on giving and lab versus field ‘windfall’ versus ‘earned’ etc.—there is no prescribed way to implement the dictator game. Instead, there should be awareness that small variations can impact on the absolute amount of giving. The researcher, thus, needs to identify the design best suited to the relevant research question (Harrison & List, 2004).

External Validity of the Charity Dictator Game

The charity dictator game is simple to use and can be flexibly modified to study a range of issues. But how reliable are the results in modelling charitable giving? External validity concerns whether results from charity dictator experiments generalise beyond the specific setting of the study. We can say with some certainty that giving in the lab is noticeably higher than giving in the field (Carlsson et al., 2013). Indeed, subjects in the laboratory give an average of 30–50% of their endowment to charity. If such high levels of generosity were replicated in the field then charities would be rich indeed! It is ‘easier’ for a subject in the laboratory to donate a $10 ‘windfall’ than, say, a worker to donate ‘hard-earned’ cash (Li et al., 2019).

The generosity of subjects in the laboratory clearly raises questions of external validity. Note, however, the fact that charity dictator games overestimate absolute giving does not mean they cannot be informative on relative questions, such as whether giving is higher with a rebate than matched funding, or whether females give more than males. On this account, the evidence for the charity dictator game is much more positive. One line of support comes from comparing behaviour in the charity dictator game with self-reported giving. Bekkers (2007) found that those who donate to charity in the dictator game reported higher donations to charity in the past year (€376) than those who chose not to donate (€237). Moreover, the same factors that correlated with giving in the charity dictator game also correlated with self-reported giving. Specifically, giving was increasing in age, education, income, prosocial value orientation and generalized social trust.

For additional evidence on the robustness of charity dictator games consider the study of Benz and Meier (2008). Each semester, students at the University of Zurich are asked, when registering for their studies, if they would like to donate CHF7 to a fund which offers cheap loans to students in financial difficulties and/or CHF5 to provide support for foreign students at the University of Zurich. This same decision was replicated in a laboratory experiment allowing a direct comparison of a subject’s choice in the lab with that made in the four semesters before and after the experiment. Benz and Meier (2008) find that subjects are more generous in the laboratory than the field but giving in the laboratory correlates with giving in the field. Specifically they find a correlation between giving in the laboratory and field of 0.25–0.4 which, they argue, is strong correlation by the standards of the literature.

Consider next a study by Stoop (2014) that directly compared four treatments which range from a conventional laboratory experiment (students in a lab) to a natural field experiment (citizens at home, unaware they are taking part in an experiment). In all four treatments the subject ‘wrongly’ receives an envelope containing €10 and a thank you note, thanking someone for their voluntary services. The subject must decide whether to re-post the envelope to the correct recipient. Hence, we have a dictator game where the recipient is a volunteer. Comparison across treatments gives a good idea of whether behaviour in a conventional laboratory experiment is consistent with that in the field. Giving was almost identical across the four treatments at around 50%. Similar evidence is provided by Franzen and Pointner (2013).

The studies above illustrate the consistent evidence that giving in laboratory-based charity dictator games is an informative method of studying the factors which influence charitable giving, provided we focus on relative rather than absolute giving. We should, however, highlight some important caveats. First, it is desirable to have a representative subject pool. For instance, Carpenter et al. (2008) found systematic differences in the giving of students and community members.Footnote 4 Analysis suggests that these differences are driven by age and sex, with young male students being particularly ‘selfish’. Wherever possible, therefore, the researcher should seek a subject pool that is representative of the population being studied. A nice example is Carpenter and Myers (2010) who conducted a charity dictator game with volunteer firefighters to investigate the motives for volunteering.

Alongside external validity, Vazire et al. (2020) distinguish construct, internal and statistical conclusion validity. Construct validity concerns whether the measures used by a researcher capture what was intended, internal validity concerns the appropriateness of drawing causal inferences from results, and statistical conclusion concerns the appropriate interpretation of the data. On these criteria charity dictator games can be seen as relatively robust given that they directly measure charitable giving in a controlled environment (Mason, 2013). Findings are, for instance, likely to be more robust than those based on self-reported measures of giving (e.g. Stagnaro et al., 2020). A further advantage of charity dictator experiments is that they are readily replicable, allowing the robustness of studies to be independently tested. Replication is now increasingly recognised as essential in experimental work (Helmig et al., 2012).

Research Questions to Study with the Charity Dictator Game

In this section, we address the types of research question for which the charity dictator game is particularly well-suited (see also Bhati & Hansen, 2020). Given that donations in the laboratory are artificial high, attention naturally focuses on questions around comparative giving across settings. For instance, the literature has explored the influence on giving of:

  1. 1.

    Demographic, personality traits and beliefs, e.g. gender, religiosity, agreeableness.

  2. 2.

    Audio or visual primes, e.g. verbal versus written solicitations.

  3. 3.

    Information about the charity or their cause, e.g. information on quality or overheads.

  4. 4.

    The type of charity, e.g. local versus national.

  5. 5.

    Incentives such as matched funding, rebates and lotteries.

In the supplementary material we provide a more detailed list of questions with relevant references. Here we pick out a handful of examples to illustrate how the charity dictator game can inform research on government policy and the strategies used by fundraisers.

The previously mentioned study of Fielding and Knowles (2015) investigated whether a face-to-face verbal invitation to donate increased giving versus merely having a donation box on display. It made a huge difference. While only 8% of subjects put money in the box without a verbal invitation, 56% of subjects did so with a verbal invitation. Verbal cues can, therefore, be effective (Andreoni et al., 2017). The evidence on visual cues is more mixed. For instance, Ekström (2012) looked at the effect of ‘watching eyes’. When individuals recycle cans and bottles in a Coop supermarket in Sweden they are given the option of donating the money ‘earned’ to charity. The experiment involves putting eyes on some machines and a default picture on others. The watching eyes made very little difference, only causing a small increase in the proportion who donate on ‘quiet days’. Such studies show how controlled experiments can be used by fundraisers to evaluate the effectiveness of interventions.

Many studies have compared willingness to give across different categories of charitable cause. For instance, Li et al. (2011) compare: cancer research, disaster relief, education enhancement and parks and wildlife, at three different levels: national (USA), state (Texas) and local (Dallas), and across government versus private organisations. One of their key findings is that donations to government organisations were nearly as high as private charities (22 vs 27% of endowment). Bachke et al. (2014) systematically compare sixty different projects, varying according to recipient group (children, girls, boys, women and men), region (sub-Saharan Africa, South and South-East Asia, Middle-East, Latin American and Eastern Europe) and project type (education, health, peace and reconciliation, agriculture and business development). A key finding is that people donated most to projects benefitting those perceived as the most vulnerable and poor, e.g. children and women in sub-Saharan Africa. Such studies can inform government policy through identifying likely gaps in funding. They can also help fundraisers frame solicitations to best effect.

A final strand of literature we mention is that looking at the presentation of information and whether individuals are influenced by information about the charity. As an example consider the study by Exley (2020) where subjects are provided with varying performance metrics across charities. For instance, in one choice task they are shown three animal charities that differ in the proportion of animals ‘saved’ from 97 to 66%. The main finding is that information on performance appears to provide an excuse to not give to low rated charities. Brown et al. (2017) also find that low rated charities are given less. Such findings can inform policy on whether metrics are beneficial. In particular, to explore a potential trade-off between directing funds efficiently while not lowering the absolute amount given to charity.

Conclusion

In this paper we have provided a brief overview of the charity dictator game and identified settings where it can be usefully applied to study giving. Laboratory and field experiments, and randomised controlled trials provide a robust way to study giving (Bhati & Hansen, 2020; Mason, 2013) and the charity dictator game is an ideal tool to use in such experiments and trials. The game is of use to researchers to unpick the motives behind pro-sociality. It is also a tool that policy makers and fundraisers can use to understand giving and evaluate and test possible interventions. The game has many advantages—it is simple to use, directly measures charitable giving and has good external validity in measuring relative effects. Naturally the game also has some disadvantages—it has poor external validity in measuring absolute amounts of giving and findings can be sensitive to a range of factors, such as windfall gains and the subject pool. The researcher can balance these advantages and disadvantages depending on the research hypothesis. For instance, a laboratory based approach allows maximum control and can facilitate the collection of large amounts of data (e.g. if subjects are exposed to multiple choice scenarios), while a field based approach can have higher external validity.