The Determinants of Altruism: Evidence from an Online Peer-To-Peer Wealth Redistribution Platform

This study seeks to investigate the drivers of altruism that peer-to-peer wealth transfer platforms can leverage to raise resources for workers with lost income during the Covid-19 crisis. In particular, we partner with Bagirata, a Jakarta-based online wealth-redistribution platform, to collect comprehensive information from Bagirata donors and beneficiaries. Several treatment interventions would be implemented involving varying the number of recipient potential donors would observe when they access the platform, the number of other donors who have donated, and the share of the targeted donated amount that has been collected.

We will also conduct a randomized experiment on Bagirata's platform to explore factors influencing donor's decision to give and whether donors are susceptible to psychic numbing/choice overload when making their donation decision. Specifically, we randomize the number of potential beneficiaries and the beneficiaries displayed to donors and study the accompanying donation size for every web page visit. We will also vary the information about recipients shown to donors. In one condition, we will provide information on the sum of donations obtained so far, and in the other condition, we will provide information on the number of donors donating while keeping all else equal. We will compare these two conditions with a control condition where such information is absent.

This study will contribute to the experimental evidence on altruism and charitable donation in a developing country setting. This study also aims to contribute to the literature on cash transfers and altruism. Existing research shows cash transfers are a useful anti-poverty tool, and various government administrations have embraced this strategy to mitigate the impact of Covid-19. Nevertheless, investigations of disaster response that do not consider private and individual altruism responses cannot provide a complete picture. This study could provide evidence on their relative performance, targeting accuracy, and impact by comparing individual donations and government transfers. Furthermore, by combining data from user activities on the website and the donor survey, the evidence from this 'in-the-wild' interaction setting would also improve our understanding of altruism determinants (see Gee and Meer, 2020; Sudhir et al. 2016).

We want to investigate the following research questions:

1. What is the impact of choice set size on donors' decision making? Donors are randomly assigned to view either 3, 8, or 10 recipients at a time upon their visit to the website. Using this variation, we can analyze if donors are susceptible to choice overload/psychic numbing when they make a decision to donate.

2. What is the impact of diversity and competition among possible beneficiaries of the donations on donors' decision making? Donors see a random draw of potential beneficiaries from the Bagirata database. Each draw will vary in gender composition, occupation, social status, and other salient characteristics that would influence their decision to give. Using this variation, we can investigate the most salient drivers of altruism among donors.

3. Does providing information on the number of donors who have donated and the donations collected positively impact donors' incentive to donate, all else being equal?

4. Are there some other socio-demographic factors that influence the decision to donate?

2020-10-01

2021-06-30

The donation amount and the likelihood of receiving a donation.

We are interested to find out how donor behavior would respond to variations in choice set size? In particular, do donors behave differently when faced with a smaller choice set? If so, what are the reasons for the differences in donor behavior?

Deservingness index of our beneficiaries and some the saliency of information about beneficiaries. Also evaluation whether donations are driven by in-group bias behavior (if any).

We are also interested in investigating how the saliency of beneficiary characteristics might be a driving mechanism behind any observed differences in donor behavior. We also attempt to test and distinguish the saliency effect from two other hypotheses: deservingness and in-group bias.

For the saliency analysis; we perform the following steps. Donors encounter a variety of beneficiary traits based on random database selections. For example, in the 3-set beneficiary treatment, a donor might see up to three beneficiaries with a unique characteristic like a specific name or location. Refreshing the page within a three-hour period provides new sets of three beneficiaries, ensuring diverse trait mixes for donors viewing multiple sets. On average, donors viewed about 3.4 sets, typically refreshing 2-3 times, which allows us to analyze random characteristic variations at the set level. Salience is determined at the set level. A beneficiary is deemed to have a 'salient' characteristic 'x' if they are the sole individual in that set with that specific trait.

For the analysis of deservingness, we surveyed donors regarding how likely they would be to donate to beneficiaries possessing various characteristics. We will also derive the deservingness index using "keyness statistics" method to classify and construct a deservingness index specific to each beneficiary narrative seen on the website.

Lastly, donors might prefer giving to individuals within their own identity group. This could be due to the shorter social distance fostering greater trust and sympathy among group members. Alternatively, it could be a way for donors to show loyalty to their group, leading them to give more to members of their own identity groups. We test for an effect of group ties on donation by pairing our beneficiary data with
demographic information about our donors from the survey.

The following treatments are implemented in the experiment.

1. Donors are presented with three recipients at a time upon their visit to the website.
2. Donors are presented with eight recipients at a time upon their visit to the website.
3. Donors are presented with ten recipients at a time upon their visit to the website.

Donors are presented with a menu/choice set of randomly selected beneficiaries of varying characteristics ranging from gender composition, occupation, social status, and other salient characteristics. In this treatment, donors would see ten recipients at a time upon their visit to the website.

Donors are also presented with information on the number of other donors who have donated to the same beneficiaries. Donors are presented with information on the magnitude of donations that beneficiaries have received so far.

The first treatment will serve as our control treatment. Donors will first undergo treatment 1, 2, or 3: Donors are randomly assigned to view either 3, 8, or 10 recipients at a time upon their visit to the website. Using this variation, we can analyze if donors are susceptible to choice overload/psychic numbing when they decide to donate.

After treatment 1 to 3, they will be shown varying characteristics of potential beneficiaries. Donors see a random draw of potential beneficiaries from the Bagirata database. Each draw will vary in gender composition, occupation, social status, and other salient characteristics that influence their decision to give. Using this variation, we can investigate the most salient drivers of altruism among donors.

After the treatments have ended, we plan to do a follow-up phone survey on beneficiaries to collect more information on demographics, asset ownership, receipt of government assistance, use of donation received, health behaviors, and recipients' well-being. We can compare Bagirata targeting performance by the overlap between its beneficiary database with government assistance receipt.

The Bagirata website also prompts its users to fill an online survey on the research team's altruism. Our primary outcome variable is a continuous variable that measures the donation amount.

The following treatments are implemented in the experiment.

1. Donors are presented with three recipients at a time upon their visit to the website.
2. Donors are presented with eight recipients at a time upon their visit to the website.
3. Donors are presented with ten recipients at a time upon their visit to the website.

Donors are presented with a menu/choice set of randomly selected beneficiaries of varying characteristics ranging from gender composition, occupation, social status, and other salient characteristics. In this treatment, donors would see ten recipients at a time upon their visit to the website.

Donors are also presented with information on the number of other donors who have donated to the same beneficiaries. Donors are presented with information on the magnitude of donations that beneficiaries have received so far.

The first treatment will serve as our control treatment. Donors will first undergo treatment 1, 2, or 3: Donors are randomly assigned to view either 3, 8, or 10 recipients at a time upon their visit to the website. Using this variation, we can analyze if donors are susceptible to choice overload/psychic numbing when they decide to donate.

After treatment 1 to 3, they will be shown varying characteristics of potential beneficiaries. Donors see a random draw of potential beneficiaries from the Bagirata database. Each draw will vary in gender composition, occupation, social status, and other salient characteristics that influence their decision to give. Using this variation, we can investigate the most salient drivers of altruism among donors.

After the treatments have ended, we plan to do a follow-up phone survey on beneficiaries to collect more information on demographics, asset ownership, receipt of government assistance, use of donation received, health behaviors, and recipients' well-being. We can compare Bagirata targeting performance by the overlap between its beneficiary database with government assistance receipt.

The Bagirata website also prompts its users to fill an online survey on the research team's altruism. Our primary outcome variable is a continuous variable that measures the donation amount.

Randomization is done by the algorithm behind the Bagirata online platform.

We varied the choice set size in our study by randomly assigning potential donors to one of three experimental treatments, each featuring either a 3-set, 8-set, or 10-set of beneficiaries. Upon entering and navigating beyond the landing page, each donor has an equal chance of being assigned to one of the three treatments. In our study, donors assigned to the 3-set beneficiary treatment saw three beneficiaries on their screen. Those in the 8-set and 10-set treatments saw eight and ten beneficiaries, respectively. This assignment lasted for three hours, meaning that if donors refreshed the page or revisited the Bagirata platform on the same device within this window, they continued to see the same set size. However, refreshing the browser or reaccessing the platform presented a new mix of beneficiaries within the set size. Donors also had the option to trigger a new draw of beneficiaries using a button at the bottom of the page.

The beneficiary section of the platform features each individual as a compact card containing standardized information. This includes their name, occupation, residence area, and social media presence (Instagram, Facebook, Twitter). Each card also presents a brief narrative about the impact of COVID-19 on the beneficiary, the need for monetary assistance, the minimum required amount, and the duration of assistance needed. Additionally, it shows the total donations received as a percentage of the requested amount and lists the available e-payment channels for donations.

On both desktop and mobile versions of the website, beneficiary cards are shown to donors in a vertical sequence. The random selection from the beneficiary database for each card ensures a random order of display. This randomness allows us to analyze how the sequence affects donations, specifically whether beneficiaries displayed at the top receive different donation outcomes compared to those lower down in each sequence.

Yes

Donors visiting the platform will face a random choice set size and will remain in the same choice set size for a three hour duration. A web refresh action would give the donors the same choice set size with different beneficiary compositions.

Our unit of analysis is at the beneficiary–display level. The number of beneficiaries within each display varies based on the choice set that donors are presented with in the different treatment conditions. We target to have more than 2000 beneficiaries. Each time a donor refresh the page, the same choice set size of beneficiaries would be displayed with different compositions of beneficiaries.

We hope to have more than 2000 unique beneficiaries displayed in varying choice set sizes seen by donors, and more than 2000 unique donor sessions. Our unit of analysis is at the beneficiary-display level. The number of beneficiaries within each display varies based on the choice set size that donors are presented with in the different treatment conditions. In total, we hope to have more than 50,000 beneficiary displays.

The Bagirata operation has granted the research team access to their datasets that includes beneficiary roster, session trace, and donation trackers, which will be the backbone of our analysis. While the website has been operational since its launch in April 2020, they have only started collecting session trace data that allows the analysis in the last three months. Since then, the pandemic fatigue has led to a decline in donor activities despite a persistent rise in beneficiary sign-ups. In the six weeks of October-November, there were only ~500 active donating visitors to the website, a substantial decline from its May Day peak activities.This constraint in our sample size significantly curtails our statistical power and makes sample expansion necessary for the planned analysis in our study. For example, Table 1B, row 1 below, describes the necessary addition of ~2,000 new individuals to have 80% power in testing the equality of means between our three treatment groups if the treatments have effects of 0.05*sigma and 0.15*sigma, respectively.