COVID-19 and pro-sociality: How do donors respond to local pandemic severity, increased salience, and media coverage?

Has the COVID-19 pandemic affected pro-sociality among individuals? After the onset of the pandemic, many charitable appeals were updated to include a reference to COVID-19. Did donors increase their giving in response to such changes? In order to answer these questions, we conducted a real-donation online experiment with more than 4200 participants from 149 local areas in England and over 21 weeks. First, we varied the fundraising appeal to either include or exclude a reference to COVID-19. We found that including the reference to COVID-19 in the appeal increased donations. Second, in a natural experiment-like approach, we studied how the relative local severity of the pandemic and media coverage about local COVID-19 severity affected giving in our experiment. We found that both higher local severity and more related articles increased giving of participants in the respective areas. This holds for different specifications, including specifications with location fixed effects, time fixed effects, a broad set of individual characteristics to account for a potentially changing composition of the sample over time and to account for health- and work-related experiences with and expectations regarding the pandemic. While negative experiences with COVID-19 correlate negatively with giving, both approaches led us to conclude that the pure effect of increased salience of the pandemic on pro-sociality is positive. Despite the shift in public attention toward the domestic fight against the pandemic and away from developing countries’ challenges, we found that preferences did not shift toward giving more to a national project and less to developing countries. Supplementary Information The online version contains supplementary material available at 10.1007/s10683-022-09753-y.

. * p < 0.10, * * p < 0.05, * * * p < 0.01. Note: Robust errors. All columns include time fixed effects, and location fixed effects. For baseline, financial, and health controls see note to Table 2. Other socioeconomic controls include place of living dummy (big city, small city, suburbs), employement status dummy (employed, unemployed, student, apprentice, retired), number of children in the household, and primarily source of news dummy (high quality, medium quality). Work change controls include work change since COVID-19 dummies (lost permanently, lost temporarily without pay, lost temporarily with pay, hours reduced), number of days commuting before COVID-19 and since COVID-19, and remote work dummies (fully, partly). * p < 0.10, * * p < 0.05, * * * p < 0.01. Note: Robust errors. All columns include the following controls location fixed effects, slider initial position, age, dummy born in the UK, female dummy, socioeconomic status, number of household members, and session dummies (time fixed effects). * p < 0.10, * * p < 0.05, * * * p < 0.01.  Note: See note to Table 6. The sample consists of first-stage donors and non-donors. * p < 0.10, * * p < 0.05, * * * p < 0.01.

B Timeline of the main experiment
After the development of the idea and programming of the experiment, we ran a technical pilot with six participants on June 5, 2020, followed by a first rough preregistration on June 8, 2020 (https://osf.io/23sc4/). This was followed by two pilots with 25 and 26 participants on the afternoons of June 8 and 10, 2020, to calibrate the payments. A final preregistration with a pre-analysis plan was completed on June 15, 2020 (https://osf.io/h5syz/). Following this, we ran several waves of the experiment on Monday evenings until the end of August 2020, starting with a larger initial sample and reducing the sample over time. Further sessions were run in October and November 2020 in order to capture the second wave of the pandemic.

C Exclusion criteria
We excluded participants who fulfilled three or more of the following criteria: • Time taken for completing the experiment below 5 minutes or above 25 minutes, • Estimated number of COVID-19 cases in UK below 30,000 or more than 30,000,000, • Estimated number of cases in local area larger than the estimated number of cases in the UK/10.
• Number of household members (children plus adults) more than 8, • Expectation that the poverty rate in the UK or in developing countries will decrease below 10%, • Expectation that the GDP growth rate in the UK or in developing countries will increase above 10%, • Inconsistencies between the answers reported to Prolific and answers in our survey: 1 -Area of residence, -Household income, -Number of household members, -Employment status.

D Prolific pre-screening criteria
Our only pre-screening criterion was the current area of residence, which needed to be in England. However, in order to secure baseline sociodemographic information, we required that the following variables have no missing values: gender, age, country of birth, household size, household income (including "prefer not to say" category for sample size reasons), and socioeconomic status, see Table D14. We chose those variables for their relevance, but excluded other variables that would result in a large reduction of the available Prolific subject pool.

E Additional analysis
In the preregistration, we specified a number of supporting hypotheses and tests on which we comment here. As pre-specified, we apply Bonferroni correction for multiple hypothesis testing, assuming 20 tests. In the following, barely any test is confirmed. Many of the tests concern, however, the outcome being the share of donations to the UK program and interactions with the treatment dummy for which the direct effect has been shown not to be significant in the main analysis. The score variables were created following a preregistered protocol.
SH0 Interaction effects of the two main explanatory variables: The coefficient on the interaction between treatment dummy and local severity is not significant (and very small) (see Table A7 in the main article).
SH1 COVID-19 skeptics will decrease giving in the treatment condition: The interaction term (as well as the direct coefficient on COVID-19 skeptics score) is not significant.
SH2 Those who follow rules and recommendations regarding COVID-19 will increase their giving in the treatment condition: The interaction term (as well as the direct coefficient on rule followers score) is not significant.
SH3a-c Regarding the impact of reporting in the media on giving to the local program versus the global program, there was not enough variation across sessions to test those hypotheses.
SH4a-c The relative amount of giving to the UK program versus the global program will reflect the perception of how negatively the UK will be impacted relative to developing countries. In a regression analysis, the following explanatory variables are looked at: GDP growth in the UK versus developing countries, poverty in the UK versus in developing countries, dummy UK more affected by COVID-19 (subjective statement), and the interactions with the treatment. For the direct effects, see Table 5 and description in the main text. Regarding the interaction effects, only the coefficient on the interaction between the dummy UK more affected by COVID-19 and the treatment is positive and significant.

SH5
a Individuals whose economic situations have been negatively affected since the spread of COVID-19 and those fearing such negative consequences will donate less than others: We confirm this hypothesis.
b Individuals whose health status has been negatively affected since the spread of COVID-19 and those fearing health deterioration will donate less than others. The coefficient on the health score is not significant (the reason is likely an inverted u-shaped pattern of giving in health, on which we comment in the main text and which seems to not be well reflected in the created health score variable).
c Individuals with less distancing opportunities will donate less than others. The coefficient on the distancing score is not significant.
SH-Other Individuals might donate less in the treatment condition if they think that they have contributed sufficiently to prevention and mitigation of the consequences of COVID-19: Coefficient is not significant.
SH-Other COVID-19 individual contribution and level of empathy: We confirm a positive correlation between empathy and giving in the experiment.

F Additional survey experiment F.1 Design
We designed an additional survey experiment to better understand the mechanism behind the results of our main experiment, where we found higher giving in the treatment group compared to the control group. In addition, the survey aimed at informing us about a potential experimenter demand effect arising in the main experiment. Following the design of the original experiment, we recruited 220 participants on Prolific who indicated their area of residence to be in England. We used the same pre-screening (see Section D) and exclusion (see Section C) criteria as for the main experiment. The latter resulted in the final sample of 172 participants used in the analysis. The survey was not incentivized, and the participants received a fixed amount of £2 after the completion of the survey. Similar to the main experiment, in the control group, the participants read a donation ask for Save the Children. In the treatment group, the participants read the same donation ask with the additional paragraph about COVID-19. Next, on each page, participants were asked to "think of an average Prolific participant from the UK who might receive this donation appeal" and answered a number of questions regarding how they think the donation appeal would affect such a person. The additional survey experiment was preregistered on OSF (https://osf.io/rw86z/) prior to the experimental sessions at the end of April, 2021. The preregistration contains further details of the survey experiment, the hypotheses, and screenshots of the experimental instructions.

F.2 Results
Next, we show the results of various tests of differences between treatments. First, we asked participants to answer how strongly they expected the appeal to evoke different emotions in the average Prolific participant. We asked the question separately for all 20 emotions that are part of the Geneva Emotional Wheel (GEW, see Scherer, 2005;Scherer et al., 2013). We took the average over the positive and over the negative emotions. Both variables range from 0 to 100. Table F1 shows results from OLS regressions. We see that the treatment evokes less positive emotions (marginally significant at p<0.1) 2 and more negative emotions (not significant). Note: OLS regressions; robust errors. Baseline controls include age, UK birth dummy, female dummy, socioeconomic status dummies, and household size. * p < 0.10, * * p < 0.05, * * * p < 0.01.
Next, we tested, whether participants expect the money to be spent sooner in the treatment condition. Table F2 presents the results which show no significant differences in the expected timing of relief in both treatments. We asked participants to name the goals that they expected the donations collected in the appeal to be spent on. They entered text into an open text field. We opted against providing a multiple-choice list as this could have influenced their responses. We classified the words used in the responses into major categories including COVID-19 (participants having included words such as pandemic, corona, or coronavirus). While in the control treatment, no one mentioned COVID-19, 16% in the treatment condition did so, and the difference is statistically significant, as can be seen in Table F3. However, this compares to, altogether, 51% mentioning education, 38% protection, 29% health, 22% support, 16% poverty, and 13% hunger. 3 This means that COVID-19 relief was not perceived as the main goal of the project. Next, we asked participants to compare the perceived importance, effectiveness, and urgency of the donation to Save the Children's appeal with a donation to (i) Transparency International, (ii) the World Wildlife Fund, and (iii) the Alzheimer's Society. The participants answered by using a slider on a scale from less important/effective/urgent to more important/effective/urgent. For each participant, we computed an average over the three charities and used this score for the final comparisons. The score ranges from 0 to 100. The results are presented in Table F4. There are no significant differences in how important, effective, or urgent participants perceive giving to Save the Children in the treatment versus the control condition.  Table F1. * p < 0.10, * * p < 0.05, * * * p < 0.01.
Next, we studied whether the treatment condition might exert on participants more pressure to give. In the literature, it has been long recognized that more (social) pressure results in higher giving (see, among others, Andreoni et al., 2017;DellaVigna et al., 2012). Moreover, anecdotal evidence suggests that fundraisers actively use such techniques. We asked a randomly chosen 50% of our sample (equally distributed among the treatments) to judge the following statement: "The person would feel pressure to donate when receiving such a donation request in a letter by the Royal Mail." Participants answered by using a slider on a scale from "not at all" to "a lot," coded 0-100. The results in Table F5 show that the difference is not significant. Next, we checked for a potential unintended experimenter demand effect in our main experiment. We asked the remaining 50% of the sample the following question: "The person would feel pressure to donate when receiving such a donation request in a study on Prolific." The participants answer using a slider on a scale from "not at all" to "a lot." Using a difference-in-difference approach, we study whether the additional pressure in the treatment condition in our experiment is different from that which the participants believe to experience in real life. Table F6 shows coefficients on treatment, dummy for the group that judged the pressure to give on Prolific (versus real life), and the interaction between the two. The coefficient on the interaction term is not significant, meaning that the pressure in the treatment condition is not different from that which would arise in real life (level coefficients are also not significant). In Figure F1, we present the averages in perceived pressure by treatments.