Let Me Think About It: How More Choices Boost Charitable Giving
Description
We ran an online experiment (N=3,200) investigating the effect of choice on charitable giving. Participants were endowed with 200 tokens which represented £2.50 in real money (1 token = £0.0125). They can then choose whether to donate the money to charities, or whether they will keep it for themselves. They are allowed to donate as much or as little money as they like, and the money donated was sent to the charities after the experiment was completed. Each participant also received a flat £10 participation fee. Participants are allowed to drop out before the experiment proper, an option which 200 participants took. They keep all the money without taking part in the rest of the experiment at all, save for a short demographic questionnaire and a few follow up questions asking why they left. Participants were reassured throughout the experiment that they were allowed to keep all the money for themselves and the experiment was not investigating selfish behaviour. Even when in the experiment proper, donating 0 is always an option. The experiment proceeds as follows. Each participant has two rounds of donations, for each of which they have 100 tokens. You cannot carry tokens over to the second round. In each round, participants are randomly allocated between having a ‘high’ number of charities to donate to (40) and a ‘low’ number of charities to donate to (5). In addition, participants are randomly allocated to one of four treatments which remain the same across the whole experiment (for them): (1) The 1Choice arm, where they can choose one out of a list of charities to donate to. (2) The MChoice arm, where they can split their donation between multiple charities. (3) The 1ChoiceDefault arm, where they can choose only one charity and a default option is pre-selected for them. (4) The Perceived Importance arm, where they see how many charities are devoted to a given cause before they pick it. The Perceived Importance arm is slightly different to the others because there is only one round as the subjects choose the cause out of four categories. There are multiple specific charities and donation purposes within each cause. The causes are: Animal Welfare, Housing and Hunger, Children & Youth, and Health & Medicine. The number of options (5 or 40) within each of these four causes is displayed in the Perceived Importance arm only. After the experiment, participants are asked a series of demographic questions, as well as questions about their choices including their satisfaction, how difficult the choice was, and other candidate moderators/mediators.
Files
Steps to reproduce
Stata 18, including the packages estout and violinplot. We reshaped the data into a panel structure, with generated user ids as the factor operator and the rounds as the time operator. The variable Category shows which of the four charitable causes were chosen. The variable Treatments shows which of sixteen paths the participant was on. This takes the form R1R2 Arm, where R1 is the amount of charities the participant sees in Round 1, R2 is the same for Round 2, and Arm is which of the four arms of the experiment they are in. For example, 540 1DChoice indicates that the participant saw 5 charities in Rund 1, 40 in Round 2, and was in the arm of the experiment with a default option. task1playerdonation_total and task2playerdonation_total show the donation for rounds 1 and 2, respectively. We generated dummy treatment variables for: - Whether the participant had 5 (0) or 40 (1) charities to choose from in each round; - Whether the participant was in the MChoice arm (1) vs the 1Choice arm (0) - Whether the participant was in the 1ChoiceDefault arm (1) vs the 1Choice arm (0) - Whether the participant was in the Perceived Importance Treatment (1) vs the rest of the experiment (0). We created simple bar plot comparison of means and ran t-tests for each of these dummy variables. We ran regressions of the amount donated on each of the treatment variables, clustering standard errors by individual id. For some specifications, we included fixed effects and/or demographic covariates. We also used lagged variables for the high versus low number of charities in some regressions, including the interactions between rounds. We recoded the subjective variables to make the categories clear and in order (either ascending or descending) then summed them up for a total for each category eg satisfaction, regret, difficulty. We added potential mediators and moderators, including the demographic variables and the subjective variables. We used the totals for the subjective variables but investigated the choice difficulty one question-by-question as it was the only one which was a significant mediator. We also used satisfaction and regret as dependent variables, with the amount donated included as a mediator. We ran a Kolmogorov-Smirnov test testing whether participants chose similar categories of charity between the Perceived Importance arm and the rest of the experiment. The regressions to check if the default treatment were used had to include separate data from DefaultRegsAdditional, merging 1:1. This uses task1dec and task2dec, which both measure whether the participant chose the first charity on the list (for rounds 1 and 2, respectively), which was always the default if the default was present. The violin plots were coded differently because the package does not work with a panel structure.