Economics of philanthropy—evidence from health crowdfunding

The term out-of-pocket spending refers to medical costs that are not covered or reimbursed by insurance.

Other forms of crowdfunding include equity crowdfunding (see, e.g., Ahlers et al. 2015; Colombo et al. 2014), reward-based crowdfunding (see, e.g., Mollick 2014), real estate crowdfunding (see Schweizer and Zhou 2017), peer-to-peer lending (see, e.g., Lin and Viswanathan 2015), and Initial Coin Offerings (see Amsden and Schweizer 2018). For excellent literature overviews, see Moritz and Block (2016), Short et al. (2017), and Wallmeroth et al. (2018).

The platform takes its name from the town of Watsi in Central America, where one of the cofounders encountered a woman on a bus who was asking passengers for donations to pay for her son’s healthcare (see FAQ on Watsi.​org).

Watsi.​org features 4677 distinct patient campaigns, and has 5314 registered donors (see Bassani et al. 2019, for cross-platform evidence on healthcare crowdfunding).

In the context of this study, we follow Furnham et al. (2016), where altruism is defined as “a selfless exhibition of trading one’s personal resources to benefit another” (p. 359).

A potential concern is that donations in December may be driven by the tax effect. In subsection 6.2.5, “December Tax Effect,” we re-run our tests by deleting all the campaigns posted in December. The results remain highly consistent.

This differs from other crowdfunding platforms such as Indiegogo, where the number of campaigns launched is determined by demand from campaign creators, and is not managed by the platform. In other words, on Watsi, the choice set donors see during holidays is comparable to that on weekends or weekdays. Thus, there should not be an endogeneity problem driven by fluctuations in campaign supply.

We assume these two variables remain donor-specific during the entire sample period. In other words, we do not know whether donors change any of their profile characteristics between their first and last donations.

To control for donation dynamics, we also compare each PP Donor’s impact for a particular campaign relative to the PP Donors who donated before him. To do this, we define Public Profile Donor Reputation Higher Than Average as equal to 1 if the donor’s impact number is higher than the average of previous donors to the campaign, and 0 otherwise. In the example described above, David’s impact number is 10, which is lower than the average of the other nine donors, which was 26.37, so this dummy variable takes the value of 0. Similarly, we compare David with the PP Donor who made a donation right before him (using initial “B” (eighteenth in the queue) in panel C, Funder Sequence, in Fig. A1). In this particular case, the impact number of the previous donor is 9, so this dummy variable takes the value of 1.

See Ahlers et al. (2015) for different success measures in crowdfunding campaigns, as well as Galak et al. (2011) and Allison et al. (2013) for the relevance of funding speed as a dependent variable.

We classify a patient as a child in accordance with the United Nations Convention on the Rights of the Child as “a human being below the age of 18 years unless under the law applicable to the child” (see United Nations 1989). Infants (Latin word infans, meaning “unable to speak”) are commonly referred to as children up to 1 year of life, and thereafter usually referred to as toddlers (see, for example, Barker 2001, for a definition).

For some campaigns, it could be very difficult to distinguish infant girls from infant boys from the patient picture. However, certain signs may be gender-revealing, such as ear piercing, or color and style of clothes. Even if it is not possible to guess a child’s sex from a visual inspection of a picture, it is generally revealed in the patient story.

We are not able to meaningfully measure patient pictures along dimensions of “sadness.” This is because the pictures are not “randomly” selected, but are instead specifically chosen by Watsi’s medical partners to avoid facial expressions of “anger” or “sadness.” We found no patient pictures with the required confidence score for “sadness” of 0.9 or higher (see Fig. A1, panel B, for more details about the confidence scores). This clearly demonstrates how Watsi aims to avoid facial expressions with high levels of “sadness.” Therefore, the reference group for “happiness” is “neutral.”

In unreported results, we checked for the influence of natural catastrophes in countries such as Haiti, which was hit by Hurricane Matthew in 2016, or Nepal, which experienced a disastrous earthquake in 2015, on our results. Because these catastrophes increase media attention, they presumably affect potential donors. We find that our results remain qualitatively unchanged.

We are unable to control for both donation sequence fixed effects and donor fixed effects because we do not have enough within-case variation in donors. Our donation sequence fixed effects control partially for any systematic preferences among donors.

Note that Watsi holds the number of campaigns on its website constant at all times. However, campaigns tend to be funded more quickly during the holidays, which results in a higher number being funded at that time. This means there is no obvious difference in the choice set over time.

Related literature in reward-based and equity crowdfunding documented a positive “female effect” on campaign success (see Greenberg and Mollick, 2017 and Cumming et al. 2019b).

The hypothesis Religious Holiday coef. = Federal Holiday coef. and Religious Holiday coef. = Weekend coef. was rejected at the 1% level for all model specifications.

We find that incident rate ratios, calculated by exponentiating the Poisson regression coefficient for the ARI, CL, Gunning Fog, and FKG indices reported in Table 6, range from 1.05 to 1.14.

In an unreported robustness check on the subsampling, we check whether the results are sensitive to the number of allocations by Watsi on behalf of the donors who choose the “monthly” option. Similarly to the previous subsampling, we create a subsample of campaigns with a high level of support by Passive Donors. To define a “high” level, we use the upper quartile (75th percentile) of # of Monthly Donors, which corresponds to more than three contributions from the monthly donors. Therefore, our subsample with a “high” level of Passive Donors consists of campaigns with more than three Passive Donors; the subsample with a “medium-to-low” level of Passive Donors consists of campaigns with three or fewer monthly donors. Instead of using the upper quartile to define a “high” level of Passive Donors, we also tried using the median. Results for the upper quartile and median are available upon request, and are qualitatively similar to those when a single donation from a Passive Donor is classified as failed.

We also re-run our analyses by deleting all donations funded in December, and reach the same conclusion. The results are untabulated, but are available from the authors upon request.

We are unable to add donor fixed effects because of the small number of donations made per donor (i.e., there is no reason to treat all the anonymous donors the same.) Note that these two controls are constructed using the conditional mean. In the example described in the “Data” section, David’s impact number was 10, which was lower than the average of the nine previous PP Donors, which was 71. Therefore, this dummy variable takes the value of 0. Similarly, we compare David with the PP Donors who donated right before him (using initial “B” (eighteenth in the queue) in panel C, Funder Sequence, in Fig. A1). In this particular case, the impact number of the previous donor is 9, so this dummy variable equals 1.