The Effect of Social Media on the Ethnic Dynamics in Donations to Disaster Relief Efforts

Abstract

Efficient resource-sharing via private transfers among households during natural disasters serves to foster a more resilient society. This study explores the effect of social media on private donations from households to natural disaster relief efforts. The donation pattern in a popular charitable crowdfunding platform exhibits inter-ethnic dynamics: Campaigns initiated by Hispanic fundraisers receive disproportionately fewer donations and lower donation amounts from non-Hispanic donors, relative to Hispanic donors. Moreover, we document a novel finding that social media statements from a political figure (President Donald Trump) influence the dynamic of private transfers across households from different ethnic groups. This finding provides a salient consideration for policymakers and government officials regarding the effects of their actions on society’s resilience and sustainability.

1. Introduction

In the face of growing concerns regarding the adverse effects of climate change on the environment and societies, resources around the world have been poured into initiatives that could improve households’ resilience to the adverse effects of climate change. Notwithstanding these efforts, natural disasters occur frequently and when faced with emergency events, governments typically act swiftly to provide funding and emergency services to the affected communities. In conjunction, recoveries and relief efforts from damages caused by these disasters are supported by donations from private donors, including households [1,2,3]. Efficient resource-sharing via private transfers among households during such disasters can serve to create a more sustainable and resilient society in the face of unpredictable environmental shocks.

Private transfers in the form of donations from households are traditionally collected by charitable organizations to be delivered to the intended beneficiaries. In concert with recent advancements in technology, crowdfunding platforms have also become viable channels for individual fundraisers to collect charitable donations. Charitable crowdfunding platforms enable us to observe household-to-household voluntary donations, a form of private transfers that are difficult to capture in other settings. Individual fundraisers on these platforms are typically soliciting funds from household donors for projects that are intended toward supporting specific household beneficiaries. Indeed, during the major 2017 Atlantic hurricane season—the costliest hurricane season on record with billions of dollars of damages [4,5], there are at least 4464 charitable crowdfunding campaigns on GoFundMe that are initiated by individual fundraisers to provide relief efforts associated with three major hurricanes during that season: Hurricanes Harvey (category 4), Irma (category 5), and Maria (category 5). Hurricanes in categories 4 and 5 are the most severe hurricanes that are likely to cause catastrophic damages [6].

The areas heavily affected by the three major hurricanes included in this study are among those with the highest percentage of the Hispanic population in the US. This study focuses on the pattern of donations received by Hispanic fundraisers as compared to the donation pattern received by non-Hispanic fundraisers. We observe a general effect of same-ethnic preference (i.e., ethnic homophily) in private donations that is consistent with findings from prior studies on human preference for other people who are more similar to themselves. We observe about one-fifth (20.7%) of total donation dollars received by Hispanic fundraisers came from Hispanic donors, whereas only about 1/20th (4.9%) of total donation dollars received by non-Hispanic fundraisers came from Hispanic donors. This observation is consistent with the findings of existing studies on homophily, which is a human’s tendency to associate with people who are similar to themselves. In particular, race/ethnicity-based homophily has been documented in various contexts, including marriage [7], social media [8], housing markets [9], hiring [10,11,12], access to credit [13], and sports [14].

The rise of social media has not left public officials behind, and they use their strong presence on social media to disseminate messages regarding political positions and official decisions, to engage with their electorates and the general public, and to raise awareness regarding public emergency situations—including health crisis and natural disaster events [15]. This study examines whether private transfers in the form of donations from household donors are influenced by statements from government officials, beyond their statements regarding public spending whose effects on private transfers have been extensively studied (e.g., [16,17,18,19]). Such private transfers are important since the recipients are in dire need of liquidity, thereby the resulting high marginal propensity to consume would directly facilitate the objective of helping the poor as well as stimulating economic activities.

The focus of this study is on social media statements from a government official (President Donald Trump) during natural disaster recovery processes. Specifically, we ask whether his Twitter statements regarding the hurricanes have a disproportionate effect on fundraising efforts initiated by Hispanic fundraisers. We hypothesize that statements from public officials could alter the observed pattern of ethnic homophily. The effect could theoretically go in either direction, but in the specific context of our analysis of President Trump’s tweets, we posit that more Hispanic donors contribute more to Hispanic fundraisers following his social media statements. This hypothesis is motivated by the documented effects of perceived threats on a group’s existential security, which refers to the condition in which the group’s survival is secure enough that it can be taken for granted [20]. As members of a particular ethnic group perceive that their group is threatened, they increase in-group solidarity and close ranks against outsiders to protect their group [20,21]. According to the annual National Survey of Latinos in the US, a large proportion of randomly surveyed Hispanic respondents view their situation to have worsened since President Trump took office in 2017, with many respondents attributing this to the policies set by the Trump administration [22,23] (the survey is available at https://www.pewresearch.org/topics/national-survey-of-latinos/, accessed on 16 May 2019). Assuming that the respondents of the National Survey of Latino are representative of the population of Hispanic donors, these donors are likely to have a pre-existing negative perception of President Trump and his administration, and therefore view President Trump’s statements as threats to their group.

The results from the empirical analyses support the hypothesis that ethnic homophily among fundraisers and donors is affected by social media statements from President Trump. The ratio of donations received by Hispanic fundraisers’ campaigns that come from Hispanic donors is higher following his tweets. Hispanic fundraisers receive 72% more donations from Hispanic donors following the tweets, consistent with these social media statements amplifying the ethnic homophily pattern we observed earlier.

This study contributes to the growing literature on the effects of social media on society’s cohesiveness and household sustainability. These analyses are important in view of the rapidly increasing use of social media by government officials. The findings from this study suggest that the use of social media by government officials could have substantial and immediate effects on the inter-ethnic distributions of private transfers among households. If the aim is to disseminate important information to the public, it may be optimal for public statements from government entities to be delivered after careful deliberations so that these essential messages can be delivered without resulting in unintended consequences.

The rest of this paper is organized as follows. The dataset used in this study is described in detail in the Section 2, along with the discussion regarding the baseline homophily in the sample. The Section 3 covers the theoretical rationale behind the hypotheses evaluated in this study as well as the empirical setting. This is followed by empirical results and additional discussions. We conclude this paper with a discussion of the implications for research, policy, and practice.

2. Data Description and Ethnic Homophily

This study utilizes a panel dataset of charitable campaigns on GoFundMe, which is the highest-ranked charity-focused crowdfunding platform based on site traffic as recorded by Alexa.com [24]. Among all crowdfunding platforms, it trails only Kickstarter, Indiegogo, and Patreon, which are not charity-focused crowdfunding platforms. Fundraisers on GoFundMe have raised more than USD 3 billion in total since the platform’s inception in 2010 [25].

The dataset contains campaigns that support the relief efforts associated with the three major hurricanes that hit the US and its territories in 2017: Harvey (August), Irma (September), and Maria (September). The dataset was collected from August to November 2017 and contains daily data from at least 50 days after the occurrence of each hurricane. The 4464 campaigns in the dataset raised a total of US $25.4 million in donations from 240,270 non-anonymous donors.

The choice of collecting crowdfunding campaign data related to the three hurricanes for this study is motivated by two research design considerations. First, the areas heavily affected by the three hurricanes are among those with the highest percentage of the Hispanic population in the US. Indeed, 99% of the population of Puerto Rico (affected by Hurricane Maria) is Hispanic, whereas the proportions of the Hispanic population in Texas and Florida (affected by Hurricanes Harvey and Irma) are 37.6% and 22.5%, respectively [26]. As a result, the dataset contains a significant proportion of fundraisers of Hispanic ethnicities, supplying sufficient statistical power for the relevant analyses.

Second, these hurricanes are the first major hurricanes affecting the US after President Trump took office in January 2017. This study employs his statements on the three hurricanes as shocks to private fundraising efforts associated with these hurricanes. Specifically, we examine how President Trump’s remarks posted on Twitter change the pattern of donations received by Hispanic and non-Hispanic fundraisers (President Trump’s tweets were obtained from http://www.trumptwitterarchive.com/).

The dataset also contains daily Federal Emergency Management Agency’s (FEMA) press releases and published news articles associated with each of the three hurricanes. The daily count of English published news articles associated with each disaster event was obtained from Factiva. These variables are included in the empirical model to control for potential confounding effects of public spending announcements (i.e., FEMA press releases) and the media coverage of each disaster event (i.e., the number of news articles).

Table 1. Dataset summary.

	Harvey	Irma	Maria	Total
Panel A. Crowdfunding Campaigns
Number of campaigns	1635	1465	1364	4464
Non-Hispanic fundraisers	1476	1352	875	3703
Hispanic fundraisers	159	113	489	761
Total donations (USD)	$13,807,829	$5,687,059	$5,993,032	$25,487,920
Non-Hispanic fundraisers	$13,062,686	$5,398,088	$4,093,958	$22,554,732
Non-Hispanic donors	$12,475,438	$5,251,150	$3,724,075	$21,450,663
Hispanic donors	$587,248	$146,938	$369,883	$1,104,069
Hispanic fundraisers	$745,143	$288,971	$1,899,074	$2,933,188
Non-Hispanic donors	$599,737	$239,845	$1,487,133	$2,326,715
Hispanic donors	$145,406	$49,126	$411,941	$606,473
Number of donors	128,202	51,450	61,218	240,870
Non-Hispanic donors	117,714	48,183	52,001	217,898
Hispanic donors	10,488	3267	9217	22,972
Average donation/donor	$107.70	$110.54	$97.90	$105.82
Panel B. Statements and News Information
Number of Trump’s tweets	26	20	41	87
Number of days Trump tweeted	8	9	11	28
Number of FEMA press releases	15	17	15	47
Number of news articles	37,307	41,558	17,987	96,852
Average number of news articles/day	454.96	506.80	219.35	1181.12

summarizes the dataset utilized by this study. Panel A of Table 1 summarizes the crowdfunding campaigns associated with the relief efforts for Hurricanes Harvey, Irma, and Maria in GoFundMe. Panel B of Table 1 summarizes the published public statements and news articles associated with those hurricanes.

In this study, the ethnicity of the fundraisers and donors is identified using each individual’s last name. We obtained the list of the most common last names for people who self-identified as Hispanics during the 2000 US Census (the list is obtained from the website: https://names.mongabay.com/data/hispanic.html, accessed on 11 December 2017). For the purpose of this study, only a last name with more than 50% of people with that particular last name self-identified as Hispanic is considered as a Hispanic last name. Anonymous donors are excluded from the dataset as their ethnicities cannot be determined. The identification of the fundraisers’ and donors’ ethnicities using their last names is suitable in this context as the distinct Hispanic last names provide the fastest and easiest way for donors to identify whether a fundraiser is Hispanic. This process of identifying Hispanic individuals using their last names errs on the side of under-identifying them as Hispanic. For instance, a Hispanic donor with a non-Hispanic last name (e.g., due to marriage) would not be identified as Hispanic.

Additionally, the language used in the project description is also used to determine whether a fundraiser is Hispanic. Natural Language Toolkit (NLTK) in Python is utilized to determine the language used in the project descriptions. We use the stopwords included in the NLTK library to detect the languages used in a given text. Stopwords refer to the most common words in a language that are usually filtered out in natural language processing [27] (the Python codes are provided by Alejandro Nolla and are available at the website: http://blog.alejandronolla.com/2013/05/15/detecting-text-language-with-python-and-nltk/, accessed on 25 March 2019). Fundraisers who write their project descriptions in Spanish or in more than one language with one of them being Spanish are considered Hispanics. The identification of fundraisers’ ethnicity using the language in the project description is consistent with the definition of ethnicity, which is a group of individuals with shared culture including language, ancestry, and belief [28].

Figure 1 depicts the fraction of Hispanic fundraisers associated with the relief efforts of each natural disaster event. The fractions are calculated using the numbers in Table 1. The solid (green) bar indicates the fraction of Hispanic fundraisers, whereas the doted (blue) bar indicates the fraction of non-Hispanic fundraisers. In aggregate, Hispanic fundraisers make up 17% of the fundraisers in this dataset. Campaigns associated with Hurricane Maria have the highest proportion of Hispanic fundraisers (35.9%) compared with campaigns associated with the other two hurricanes: Harvey (9.7%) and Irma (7.7%).

Figure 2 depicts the fraction of donations from Hispanic vs. non-Hispanic donors for campaigns administered by Hispanic vs. non-Hispanic fundraisers. The fractions are calculated using the numbers in Table 1. The solid (blue) bar indicates donations from Hispanic donors, whereas the stripped (green) bar indicates donations from non-Hispanic donors. Figure 2a shows the fraction of donations from Hispanic and non-Hispanic donors to crowdfunding campaigns associated with the relief efforts for all three hurricanes in the dataset, Hurricanes Harvey, Irma, and Maria. Figure 2b–d shows the same information for each of the hurricanes in the dataset.

Figure 2a shows that non-Hispanic fundraisers receive very small proportions of their funding from Hispanic donors (4.9%). In contrast, 20.7% of the funding received by Hispanic fundraisers comes from Hispanic donors. This pattern provides prima facie evidence of ethnicity-driven homophily among fundraisers and donors in charitable crowdfunding.

The homophily pattern is observed for each of the three natural disaster events in our sample, with fairly similar magnitudes across the three hurricanes, despite the relatively large variation in the fraction of Hispanics in the affected areas. In the case of Hurricane Maria that wreaked havoc in Puerto Rico—where the population is 99% Hispanic—21.7% of the donations received by Hispanic fundraisers came from Hispanic donors (Figure 2d). Similarly, 19.5% and 17.0% of the donations received by Hispanic fundraisers came from Hispanic donors in the case of Hurricane Harvey and Irma that heavily affected Texas and Florida—whose populations are 37.6% and 22.5% Hispanic, respectively (Figure 2b,c). This preliminary evidence is consistent with ethnic and racial homophily patterns that have been documented in both laboratory experimental settings and real-world settings, such as housing markets [9], hiring [10,11,12], access to credit [13], and sports [14]. Against the backdrop of potentially innate homophily, the one-shot or low-frequency natures of previous studies make it difficult to capture the dynamics of homophily. Our focus on social media and crowdfunding platforms allows us to examine these dynamics using relatively high-frequency data (daily), enabling us to detect changes in homophily in response to social media statements from government officials.

3. Research Design and Summary Statistics

We now turn our attention to exploring a potential source of impetus for the change in the donation pattern: statements from government officials. Statements from public officials that are directly related to government spending tend to affect the pattern of private transfers [16,17,18,19]. With elected officials having the power to influence public opinions, even statements that are not directly related to government spending may affect the pattern of private transfers.

Conceptually, the direction of the effect on the homophily pattern in private donations may be ambiguous as some statements may result in stronger homophily whereas other statements may result in weaker homophily. We posit that the direction of the effect would depend not only on the nature of the statements but also on the public perception of the government officials themselves. In this context, a large proportion of the Hispanic population in the US views their situation to have worsened during the Trump Administration, according to the respondents of the National Survey of Latino [22,23]. Specifically, 32% of Hispanic respondents of the National Survey of Latinos conducted during the first year of the Trump Administration (2017) said that their situation in the US had worsened relative to before he took office (i.e., their situation in 2016 during the Obama administration). This figure became even higher (47%) in the following year’s survey. This perceived worsening condition can be attributed to policies set by the Trump administration [22]. The 2018 National Survey of Latinos shows that more than half of Hispanic respondents (67%) view the policies set by the Trump administration as being harmful to the Hispanic population in the US. This figure is substantially higher than the same figure from surveys conducted during the Obama (15%, 2010) and Bush (41%, 2007) administrations.

Assuming that these respondents are representative of the population of Hispanic donors, Hispanic donors are likely to have a negative perception of President Trump and his administration. As such, Hispanic donors are likely to view President Trump’s statements as threats, consistent with the confirmation bias theory, which is a human’s tendency to interpret a piece of information in ways that agree with their prior beliefs [29]. Perceived threats to the survival of a particular ethnic group tend to lead to stronger in-group solidarity, with group members closing ranks against outsiders [20,21]. Therefore, we hypothesize that such perceived threats to strengthen ethnic homophily among Hispanic fundraisers and donors:

Hypothesis 1.

(The effect of public official statements). Ethnic homophily among Hispanic donors and fundraisers is stronger following statements from President Trump.

Equations (1) and (2) describe the empirical model used to test Hypothesis 1. The dependent variable in each regression model is the amount of donations received by campaign i on day t (in USD). To facilitate the examination of ethnic homophily, the donations are aggregated separately into two groups by the ethnicity of the donors: (1) the amount of donations received by campaign i on day t from Hispanic donors, $Hisp.Donation{s}_{i,t}$; and (2) the amount of donations received by campaign i on day t from non-Hispanic donors, $NonHisp.Donation{s}_{i,t}$. Additional analyses are performed using the number of donors contributing to campaign i on day t (Hisp.Donors_i,t and NonHisp.Donors_i,t), instead of the amount of donations received by campaign i on day t. As the distribution of the amount of donations received is heavily skewed, the dependent variables are logarithmically transformed to reduce the skewness.

$$\begin{gathered} \ln \left(Hisp.Donation{s}_{i,t}\right)= \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} {\alpha }_{1,H}\ast Hisp.Fundraise{r}_i+{\alpha }_{2,H}\ast DJ{T}_{t-1}+{\alpha }_{3,H} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} \ast Hisp.Fundraise{r}_i\ast DJ{T}_{t-1}+\mathsf{{\rm B}}\ast X+\mathsf{\Theta }\ast Z+\epsilon \end{gathered}$$

(1)

$$\begin{gathered} \ln \left(NonHisp.Donation{s}_{i,t}\right)= \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} {\alpha }_{1,NH}\ast Hisp.Fundraise{r}_i+{\alpha }_{2,NH}\ast DJ{T}_{t-1}+{\alpha }_{3,NH} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} \ast Hisp.Fundraise{r}_i\ast DJ{T}_{t-1}+\mathsf{{\rm B}}\ast X+\mathsf{\Theta }\ast Z+\epsilon \end{gathered}$$

(2)

There are three independent variables of interest in the empirical model. The main variable of interest is Hisp.Fundraiser ∗ DJT, the interaction term between Hisp.Fundraiser and DJT. The parameter estimates for this interaction term (${\alpha }_{3,H}$ and ${\alpha }_{3,NH}$) reflect the marginal effect of President Trump’s tweets on Hispanic vs. non-Hispanic fundraisers (Hypothesis 1). Hisp.Fundraiser_i is a binary variable that is set to 1 if fundraiser i is categorized as Hispanic and it is set to 0 otherwise. The parameter estimates for this variable (${\alpha }_{1,H}$ and ${\alpha }_{1,NH}$) reflect the baseline homophily in the absence of the public official’s statements. DJT_t is a binary variable that is set to 1 if President Trump posts at least one statement on Twitter regarding the relief efforts associated with the hurricane supported by campaign i on day t, and it is set to 0 otherwise. The parameter estimates for this variable (${\alpha }_{2,H}$ and ${\alpha }_{2,NH}$) reflect the effect of President Trump’s tweets on the average donors’ contribution to the charitable campaigns.

To mitigate potential endogeneity concerns, two sets of control variables are included in the empirical model. The first set of control variables is a vector of time-varying variables (denoted by X) capturing the characteristics of each event and each campaign that changes daily. This set includes the following variables measured at the beginning of day t: a binary indicator of whether FEMA publishes at least one press release associated with the disaster event supported by campaign i (FEMA), the number of campaigns supporting the relief efforts for the same event as campaign i (NumCampaigns), the number of news articles published on the disaster event supported by campaign i (NewsCount), a binary indicator of whether another disaster event has occurred in the last five days (NewEvent), the number of social medial mentions associated with campaign i (SMMention), the number of updates posted by the fundraiser (Updates), and the number of Facebook friends of the fundraiser (FBFriends). The inclusions of these variables work to mitigate the concerns that our estimates could reflect the time-series variations in the event’s situations and the campaign’s conditions.

The second set of control variables is a vector of variables containing the time-invariant characteristics of each campaign (denoted by Z), which capture the time-invariant heterogeneity across campaigns. These Z variables should mitigate concerns that any homophily pattern that we document reflects variations in campaign characteristics that may be correlated with ethnicity. This set of variables includes the initial funding goal set by the fundraiser at the beginning of the campaign (Goal), the number of words in the project description (NumWords), the number of videos posted on the campaign’s page (NumVideos), the date when campaign i was posted on the platform (StartTime), a binary indicator of whether the fundraiser is female (Female), and a binary indicator of whether the fundraiser is located in the state that is directly affected by the disaster event (FromAffectedArea). In the latter specifications, these Z variables are subsumed by campaign fixed effects.

An important consideration in comparing the parameter estimates for Hisp.Fundraiser ∗ DJT (${\alpha }_{3,H}$ and ${\alpha }_{3,NH}$) is that these fundraisers may be operating in a prevailing homophilic environment—as indicated by the Hisp.Fundraiser parameter estimates (${\alpha }_{1,H}$ and ${\alpha }_{1,NH}$). If the buzz created by President Trump’s tweets merely serves as reminders for potential donors to give, the parameter estimates for Hisp.Fundraiser ∗ DJT may only capture the change in the amount of donations due to the prevailing homophilic environment, rather than an increase in the intensity of homophily.

If Hispanic donors have a negative perception of President Trump and his administration and perceive his statements as threats then we posit that the effect of President Trump’s statements goes beyond creating buzz around the disaster relief efforts, that is, his statements can change the underlying pattern of the donations. To test this hypothesis, we employ an alternative specification to examine how President Trump’s tweets affect the proportion of donations from Hispanic donors. In this specification, the dependent variable is the ratio of the amount of donation from Hispanic donors over the total donation received by campaign i on day t (HispRatio.Donations_i,t).

$$\begin{gathered} HispRatio.Donation{s}_{i,t}={\alpha }_{1,R}\ast Hisp.Fundraise{r}_i+{\alpha }_{2,R}\ast DJ{T}_{t-1}+{\alpha }_{3,R}\ast \\ Hisp.Fundraise{r}_i\ast DJ{T}_{t-1}+\mathsf{{\rm B}}\ast X+\mathsf{\Theta }\ast Z+\epsilon \end{gathered}$$

(3)

If President Trump’s statements increase total donations without changing the distribution pattern of the donation sources, the proportion of donations from Hispanic donors should not change following his statements. In particular, the parameter estimates for ${\alpha }_{3,R}$ should not be (statistically) different from zero.

Table 2. Variables summary statistics.

Variable	Mean	Std. Dev.
Donations from non-anonymous donors (USD)	$126.01	$1711.33
Donations from Hispanic donors (Hisp.Donations)	$7.31	$111.54
Donations from non-Hispanic donors (NonHisp.Donations)	$118.71	$1662.97
Ratio of Hispanic donations (HispRatio.Donations)	0.10	0.10
Non-anonymous donors	1.13	17.28
Hispanic donors (Hisp.Donors)	0.10	1.71
Non-Hispanic donors (NonHisp.Donors)	1.03	15.81
Ratio of Hispanic donors (HispRatio.Donors)	0.10	0.10
Hisp.Fundraiser (binary)	0.18
DJT (binary)	0.15
PR.fundraiser (binary)	0.02
DJTPositive (binary)	0.13
DJTNegative (binary)	0.02
FEMA (binary)	0.33
NumCampaigns	2282.52	375.30
NewsCount (centered)	0.00	586.05
NewEvent (binary)	0.14
SMMention	8.55	97.55
Updates	0.06	0.32
FBFriends	740.33	995.84
Goal (USD)	$34,773.55	$695,664.50
NumWords	210.29	156.75
NumVideos	0.14	0.54
StartTime	6.06	5.49
Female (binary)	0.51
FromAffectedArea (binary)	0.45

Note: Please refer to Table A1 in Appendix A for the description of the independent variables included in the empirical models.

reports the summary statistics of the variables included in the empirical models in Equations (1) through (2). The dataset includes 59,062 campaign-day observations. All independent variables for each observation are measured by the beginning of day t, that is, before the measurement window of the dependent variable for that observation.

Our main empirical analysis consists of an empirical model estimated using a panel regression method, which allows us to control for various potential confounding effects. However, we perform a preliminary analysis using MANOVA to test whether there is any difference in the means of the two main dependent variables (Hisp.Donations and Non-Hisp.Donations) across subsamples identified using two dimensions: (1) Hispanic vs. non-Hispanic fundraisers, and (2) the days immediately following Trump tweets vs. other days. We find a strong statistical difference across these subsamples. In particular, the interaction term has a Wilkes’ Λ of 0.9970, corresponding to a p-value of <0.01), which indicates that the homophily pattern is different during days following Trump tweets, relative to other days in the sample.

4. Main Empirical Analyses

We perform our main empirical analysis using a panel regression method estimating the empirical models in Equations (1)–(3). Time period clustering is used in the regressions to control for within-time-period correlation. Event dummies are also included to control for the heterogeneity across the different disaster events.

Table 3. The effects of President Trump’s tweets on donations.

	(1)	(2)	(3)
Variables	Ln (Hisp.Donations)	Ln (NonHisp.Donations)	HispRatio.Donations
Variable of interest
Hisp.Fundraiser	0.187618 ***	−0.216810 *	0.013971 ***
	(0.055779)	(0.117227)	(0.002234)
DJT	0.058323	1.037171 **	−0.008660 ***
	(0.134921)	(0.506546)	(0.003265)
Hisp.Fundraiser ∗ DJT	0.543824 ***	0.114649	0.020504 ***
	(0.162154)	(0.365887)	(0.006675)
Event characteristics
FEMA	−0.027024	0.123326	−0.001311
	(0.047385)	(0.217099)	(0.001283)
Hisp.Fundraiser ∗ FEMA	−0.041114	−0.326521	0.001173
	(0.110445)	(0.214268)	(0.004858)
NumCampaigns	−0.001127 **	−0.001582	−0.000047 ***
	(0.000501)	(0.001526)	(0.000010)
NewsCount	0.000119	0.001449 ***	−0.000014 ***
	(0.000099)	(0.000301)	(0.000002)
Hisp.Fundraiser ∗ NewsCount	0.000607 ***	0.000136	0.000032 ***
	(0.000149)	(0.000241)	(0.000005)
NewEvent	0.150112	1.207786 ***	−0.004305 **
	(0.092415)	(0.275248)	(0.001920)
Time-varying campaign characteristics
SMMention	0.004473 ***	0.005725 ***	−0.000009 ***
	(0.000877)	(0.001434)	(0.000003)
Updates	0.446850 ***	1.614301 ***	−0.003344 *
	(0.080168)	(0.100970)	(0.001839)
FBFriends	0.000036 ***	−0.000043 **	0.000002 ***
	(0.000011)	(0.000018)	(0.000000)
Time-invariant campaign characteristics
ln(Goal)	0.137836 ***	0.528565 ***	−0.001463***
	(0.013855)	(0.036552)	(0.000324)
NumWords	0.000222 ***	0.001191 ***	−0.000004
	(0.000068)	(0.000159)	(0.000003)
NumVideos	0.002475	0.058440 *	−0.001057
	(0.019880)	(0.029610)	(0.000762)
StartTime	0.013524 ***	0.081294 ***	0.000057
	(0.002477)	(0.009134)	(0.000117)
Female	0.025686 *	0.065616 **	−0.000684
	(0.014559)	(0.030922)	(0.000708)
FromAffectedArea	0.091250 ***	−0.073495 **	0.004227 ***
	(0.021234)	(0.035776)	(0.000890)
Intercept	−5.469633 ***	−7.292873 **	0.214055 ***
	(1.198979)	(3.611264)	(0.024163)
Event fixed effect	Yes	Yes	Yes
Observations	59,062	59,062	59,062
R-squared	0.111	0.205	0.023

Notes: Please refer to Table A1 in Appendix A for complete description of other independent variables. All independent variables are measured by the beginning of day t, i.e., before the dependent variable is measured. Standard errors reported in parentheses are estimated using time error clustering. Asterisks (***, **, and *) denote statistical significance at 1%, 5%, and 10% levels, respectively.

reports the parameter estimates from the empirical models in Equations (1)–(3), respectively, each with a different dependent variable: (1) total amount of donations received by campaign i (in USD) from Hispanic donors at time t ($Hisp.Donation{s}_{i,t}$ in Equation (1); Column 1); (2) total amount of donations received by campaign i from non-Hispanic donors at time t ($NonHisp.Donation{s}_{i,t}$ in Equation (2); Column 2), and (3) the ratio of the amount of donation from Hispanic donors over the total donation received by campaign i on day t (HispRatio.Donations_i,t in Equation (3); Column 3). The independent variables of interest are whether campaign i’s fundraiser is Hispanic (Hisp.Fundraiser is set to 1 if the fundraiser is Hispanic), whether President Trump posts at least one statement on Twitter regarding the hurricane supported by campaign i (DJT is set to 1 if President Trump posts at least one statement on Twitter on day t) and the interaction term of Hisp.Fundraiser and DJT.

We first identify the baseline pattern of same ethnicity preference among donors and fundraisers that we observed during our data exploration by examining the parameter estimates for Hisp.Fundraiser in Columns 1 (i.e., ${\alpha }_{1,H}$in Equation (1)) and 2 (i.e., ${\alpha }_{1,NH}$in Equation (2)). The parameter estimates indicate that in the absence of the public official’s statements, Hispanic fundraisers receive 21% more donations from Hispanic donors and 19% less donations from non-Hispanic donors, relative to non-Hispanic fundraisers (the amount of donations received by Hispanic fundraisers from Hispanic donors is 1.21 (= exp(0.187618)) times the amount of donations received by non-Hispanic fundraisers from Hispanic donors. The amount of donations received by Hispanic fundraisers from non-Hispanic donors is 0.81 (= exp(−0.21681)) times the amount of donations received by non-Hispanic fundraisers from non-Hispanic donors).

These results are further confirmed by the positive parameter estimate for Hisp.Fundraiser in Column 3, which indicates that Hispanic fundraisers receive a larger proportion of donations from Hispanic donors as compared to non-Hispanic fundraisers.

After we examine the baseline donation patterns, we turn our attention to the parameter estimates for our main variable of interest, Hisp.Fundraiser ∗ DJT. The regression results indicate that ethnic homophily is more prevalent following President Trump’s tweets, supporting Hypothesis 1. The significant positive point estimate for Hisp.Fundraiser ∗ DJT in Column 1 (i.e., ${\alpha }_{3,H}$ in Equation (1)) indicates that the amount of donations from Hispanic donors to Hispanic fundraisers increases by 72% following tweets from President Trump (the amount of donations from Hispanic donors received by Hispanic fundraisers following President Trump’s tweets is 1.72 (= exp(0.543824)) times as high as in the absence of such tweets). In contrast, the trivial estimate for DJT in Column 1 (i.e., ${\alpha }_{2,H}$ in Equation (1)) indicates that the amount of donations from Hispanic donors to non-Hispanic fundraisers does not seem to change following his tweets.

The parameter estimates in Column 2 of Table 3 show that the tweets have a differential effect on the pattern of donations from non-Hispanic donors. The statistically significant, positive point estimate of DJT in this column corresponds to a positive effect of his tweets on non-Hispanic donors: they more than double their donations to non-Hispanic fundraisers following the tweets (total donations per campaign went up 2.82 (= exp(1.037171)) times following President Trump’s tweets). It is important to note that the amount of donations from non-Hispanic donors to Hispanic fundraisers also increases in similar magnitude, as indicated by the non-significant estimate of Hisp.Fundraiser ∗ DJT in Column 2.

The estimates in the first two columns of Table 3 indicate that the amounts of donations from both Hispanic and non-Hispanic fundraisers increase following President Trump’s statements. As these findings are consistent with President Trump’s statements serving as reminders to give, we next evaluate whether these findings merely reflect a general increase in donations in a homophilic environment due to the buzz created by his tweets, or whether his tweets increase the intensity of homophily.

We answer this question by examining the regression of donation ratio from Hispanic donors, which is reported in Column 3 of Table 3. The estimate for DJT (i.e., ${\alpha }_{2,R}$in in Equation (3)) is negative and significant, indicating that the proportion of donations received by non-Hispanic fundraisers that come from Hispanic donors decreases following the public official’s statements. In contrast, the estimate for Hisp.Fundraiser ∗ DJT (i.e., ${\alpha }_{3,R}$in Equation (3)) is positive and significant, indicating that the proportion of donations received by Hispanic fundraisers from Hispanic donors increases following the statements. In combination, these estimates suggest that ethnic homophily among Hispanic donors and fundraisers is stronger following President Trump’s tweets, likely reflecting an increase in solidarity among Hispanics. It is interesting to note that the count of news articles (NewsCount and Hispanic ∗ NewsCount) seems to affect the amount of donations in the same direction as President Trump’s tweets. It is possible that these variables partially absorb some of the effects of President Trump’s tweets as news media outlets heavily cover his tweets [30,31].

In contrast, while we find significant effects of President Trump’s tweets on the amount of donations, other parameter estimates show that announcements of government relief efforts through a more official communication channel—that is, FEMA press releases—do not significantly affect private donations. The estimates for FEMA and Hisp.Fundraiser ∗ FEMA are not significant in Columns 1, 2, and 3. This is likely because FEMA is perceived as a more neutral entity and therefore, does not elicit a response that strengthens ethnic homophily among Hispanic donors and fundraisers. This finding is also consistent with Okten and Weisbrod [2] that government grants do not crowd-out household donations.

5. Robustness Checks and Additional Analyses

We performed several sets of regressions to ensure the robustness of our results. First, we repeat regression analysis using alternative outcome variables that are based on the number of donors, instead of the amount of donations. Column 1 and 2 of

Table 4. Additional regression results.

	(1)	(2)	(3)	(4)	(5)
Variables	ln(Hisp. Donors)	ln(NonHisp. Donors)	ln(Hisp. Donations)	ln(NonHisp. Donations)	ln(Donations)
Hisp.Fundraiser	0.120221 ***	−0.136123 *			−0.13658
	(0.036110)	(0.075043)			(0.122465)
DJT	0.034754	0.666737 **	−0.180860 *	0.342281	1.026460 *
	(0.088205)	(0.328047)	(0.097017)	(0.300629)	(0.528759)
Hisp.Fundraiser ∗ DJT	0.365352 ***	0.077885	0.559375 ***	0.151852	0.294869
	(0.108297)	(0.237149)	(0.181048)	(0.276254)	(0.375600)
Event fixed effect	Yes	Yes	No	No	Yes
Campaign fixed effect	No	No	Yes	Yes	No
Observations	59,062	59,062	59,062	59,062	59,062
R-squared	0.117	0.213	0.333	0.472	0.209

Note: Please refer to Table A1 in Appendix A for complete description of the independent variables. All independent variables are measured by the beginning of day t, that is, before the dependent variable is measured. Standard errors reported in parentheses are estimated using time error clustering. Asterisks (***, **, and *) denote statistical significance at 1%, 5%, and 10% levels, respectively.

reports the results of two regressions, each with different dependent variables: (1) the number of Hispanic donors contributing to campaign i (Column 1) and (2) the number of non-Hispanic donors contributing to campaign i (Column 2). The independent variables of interest are: (1) whether campaign i’s fundraiser is Hispanic (Hisp.Fundraiser is set to 1 if the fundraiser is Hispanic), (2) whether President Trump posts at least one statement on Twitter regarding the hurricane supported by campaign i (DJT is set to 1 if President Trump posts at least one statement on Twitter on day t), and (3) the interaction term of Hisp.Fundraiser and DJT. Table 4 only reports the parameter estimates of the independent variables of interest, with parameter estimates from control variables suppressed to conserve space.

We observe similar patterns using the number of donors in Columns 1 and 2 in Table 4. In the absence of the public official’s statements, 13% more Hispanic donors give to Hispanic fundraisers as compared to the number of Hispanic donors who give to non-Hispanic fundraisers. In contrast, 19% fewer non-Hispanic donors give to Hispanic fundraisers as compared to the number of non-Hispanic donors who give to non-Hispanic fundraisers. Following the official’s tweets, the number of Hispanic donors who give to Hispanic fundraisers increased by 44% (following President Trump’s tweet, the number of Hispanic donors who donated to Hispanic fundraisers is 1.44 (= exp(0.365352)) times the number of Hispanic donors who donated to non-Hispanic fundraisers.).

Second, to capture various time-invariant campaign characteristics that may influence campaigns’ performance, we ran the regressions again with campaign fixed effects. Other studies have documented campaign characteristics as important determinants of campaign success in raising funds in the crowdfunding context (e.g., [32,33]). Columns 3 and 4 of Table 4 report the results from regressions estimating the empirical model in Equations (1) and (2) with campaign fixed effects included in the regressions. The regressions were estimated twice, each with a different dependent variable: (1) the amount of donations received by campaign i from Hispanic donors (Column 3) and (2) the amount of donations received by campaign i from non-Hispanic donors (Column 4). The effect of President Trump’s statements on ethnic homophily documented in Table 3 remains robust with campaign fixed effects as indicated by the significant positive point estimate for Hisp.Fundraiser ∗ DJT in Column 3 of Table 4.

Lastly, as shown in Table 1, Hispanic donors make up a minor proportion of total donors in the dataset. Only 10.5% of donors in the dataset are identified as Hispanics. This leads to a question of whether minority fundraisers operate at a disadvantage due to the inherent ethnic homophily among donors and fundraisers. To answer this question, we run an additional regression with the total donations from all non-anonymous donors (identified Hispanic and non-Hispanic donors) as dependent variables. The results are shown in Column 5 of Table 4. The dependent variable is the total amount of donations from Hispanic and non-Hispanic donors received by campaign i (Donations). The non-significant estimate of Hisp.Fundraiser in this column indicates that Hispanic and non-Hispanic fundraisers appear to have received equal support from the donors.

The results also show that President Trump’s statements seem to increase the total amount of donation received by Hispanic fundraisers and non-Hispanic fundraisers equally, indicated by the significant positive estimate of DJT and the non-significant estimate of Hisp.Fundraiser ∗ DJT in Column 5. While Hispanic fundraisers do not seem to operate at a disadvantage in terms of their ability to raise funds in this setting, the overall patterns suggest that the donations they receive come from a more concentrated group of donors, that is, their own ethnic group, following the tweets from public officials. This increased segregation in the distribution of donations has the potential to lead to a less resilient society if the minority group has more limited resources [34]. In this context, we find that the average donation size from Hispanic donors ($74.46) is lower than the average donation size from non-Hispanic donors ($109.12).

Despite our careful and painstaking daily collection of the crowdfunding platform data, our usage of alternative empirical methodologies (including MANOVA and regression analyses), and our multitudes of robustness checks, we should mention several remaining caveats associated with our analysis. First, our sample is limited to social media statements from a single public official. The choice of this public official is done carefully to optimize the statistical power of our analyses of his social media statements, given his large number of social media followers. Extending the analysis to social other public officials with (much) smaller sets of social media followers may be rife with statistical power issues as their statements may not lead to similar public reactions and statistically detectable effects.

Second, President Trump, whose statements are the focus of this study, is quite unique in his controversial stand on many issues. Public statements from other public officials may not have the same effect on public transfers. Our objective in this study is not to identify the reaction to each public official’s statements, perhaps as a function of their reputation and/or political position, but rather to document the existence of such effect in at least one specific setting.

Third, our analysis focuses on three large-scale natural disasters in 2017. With climate change potentially resulting in a higher occurrence and intensity of natural disasters, our analysis highlights both the relevance of extending our analysis to a larger sample of natural disasters and the unfortunate increase in the availability of such data. We expect to see more analyses and findings regarding the effect of private transfers on the resilience of society in the face of such disasters.

6. Summary

This study contributes to the growing literature on the effects of social media on society’s cohesiveness and household sustainability that cuts across multiple disciplines. Given the rapid increase in the use of social media by government officials, it is important to understand the effects of social media usage by public officials. This study documents how social media statements issued by public officials can change the pattern of private transfers among households in the context of charitable giving to natural disaster relief efforts.

We first document ethnic homophily in private donations on a popular charitable crowdfunding platform. We observe about one-fifth (20.7%) of total donation dollars received by Hispanic fundraisers on the platform came from Hispanic donors, whereas only about 1/20th (4.9%) of total donation dollars received by non-Hispanic fundraisers came from Hispanic donors.

We then test our main hypothesis that social media statements from public officials could alter the observed pattern of ethnic homophily. We document that social media statements from a particular public official, President Trump, increase the proportion of donations coming from Hispanic donors to Hispanic fundraisers, reflecting the increased solidarity among Hispanics.

The findings from this study suggest that the use of social media by government officials could have substantial and immediate effects on the inter-ethnic distributions of private transfers among households. The resulting distributional effect could lead to a less resilient society if the minority group has limited resources. If a public official aims to disseminate important information to the public, the choice of communication channel should be deliberated carefully to reduce potential negative side effects.