Inequity in Washington state covid-19-related crowdfunding

Abstract

The first seven months of the US COVID-19 pandemic saw a massive increase in COVID-19-related crowdfunding campaigns. Despite their popularity, these campaigns were rarely successful in reaching their monetary goals, with nearly 40% of them not receiving a single donation. Previous research has indicated that crowdfunding has increased inequities and disparities in wealth, and this study set out to examine the situation in Washington State, an area greatly divided socio-economically, culturally, and geographically. Using GIS-based Multi-Criteria Evaluation (MCE) models with Quantile classification, we created geospatial representations of composite independent variables representing the impacts of COVID-19 and social marginalization on Washington state counties. We then examined the relationships between these variables with the dependent variables, campaign prevalence and outcome (campaign median amount raised), and assessed for relationships through ANOVA tests. These representations allowed us to assess the possibility that both COVID-19 impact and social marginalization may be amplifying already present inequities in Washington state by influencing crowdfunding monetary outcomes and prevalence. Our research indicates that county-wide social marginalization is associated with COVID-19-related crowdfunding campaign prevalence, generally pointing towards counties with lower social marginalization producing more campaigns. We also found that COVID-19 impact is, surprisingly, not associated with campaign prevalence, and that there was no relationship between campaign outcome and marginalization or COVID-19 impact. These findings indicate that inequitable access to, and use of, technologies like crowdfunding may be further contributing to COVID-19 inequities within Washington state. Policy-makers should be advised to use crowdfunding information with extreme caution when considering it as a useful tool in assessing community needs within counties and allocating resources, because those with the highest need likely are not being represented in this data due to lack of access.

1. Introduction

1.1. Overview of the COVID-19 situation in relation to social marginalization

COVID-19 has been a ubiquitous crisis, disrupting peoples’ social support systems, jobs, sources of food and medicine, childcare, and other resources (Chakraborty & Maity, 2020). It profoundly affected the global economy, with the United States seeing a record unemployment rate of 14.8% in April 2020 (Falk, Carter, Nicchitta, Nyhof, & Romero, 2021). Unemployment was not equitable in the U.S., with part-time workers, people without college degrees, young people, and racial and ethnic minorities experiencing much higher rates of unemployment throughout the pandemic than white, college-educated workers (Falk, Carter, Nicchitta, Nyhof, & Romero, 2021). COVID-19 infections, morbidities, and mortalities have also disproportionately affected Black, Indigenous, and other people of colour and people living in rural areas in the US (Iyanda, Boakye, Lu, & Oppong, 2021).

Historically, socially marginalized groups have been inordinately affected by disasters particularly in places where austerity measures damaged the functionality and availability of necessary public services and social support systems (Basu et al., 1982). Thus, disasters can exacerbate socioeconomic and health disparities and worsen health outcomes in general, while simultaneously reinforcing neoliberal and austere systems of care that have harmful economic and health impacts (Klein, 2008; Pfeiffer & Chapman, 2010; Stuckler & Basu, 2013). Increasingly, people who are unemployed, lacking adequate monetary or healthcare support, or struggling with basic needs are turning to crowdfunding platforms such as GoFundMe for help, soliciting charitable donations from their social networks online (Bassani et al., 2019; Snyder, 2016). Through this paper, we aim to leverage detailed socio-economic and COVID-19 impact data to explore geospatial intersections of COVID-19 repercussions, existing social marginalization, and crowdfunding outcomes in Washington State.

1.2. Crowdfunding as a response to crises

Crowdfunding is popular among those seeking funding largely due to its pervasiveness and low barriers to access (Kenworthy, 2019; Snyder, 2016). However, despite its accessibility, crowdfunding is rarely equitable, and previous research has shown that it can exacerbate socio-economic and racial inequities, with privileged participants experiencing the highest rates of success in raising funds (Barcelos, 2020; Duynhoven, Lee, & Michel, 2019; Igra, Kenworthy, Luchsinger, & Jung, 1982; Kenworthy et al., 2020; Lukk et al., 2018). For instance, Duynhoven et al. (Duynhoven, Lee, & Michel, 2019) found that individuals with higher rates of education and incomes were more likely to turn to crowdfunding and consequently have higher rates of success when faced with health-related needs. Cookson et al.’s research shows similar outcomes, with high-income beneficiaries benefitting from personal and professional network advantages, and generally receiving 28% more support than those with lower incomes (Berliner & Kenworthy, 2017). Berliner and Kenworthy argue that one cause of this is crowdfunding’s competitive marketplace, where campaigners must be extraordinarily savvy in multiple media domains to succeed, such as in narrative creation, social networking, medical and media literacy, and demonstrating worthiness (Berliner & Kenworthy, 2017). These criteria exclude people who are not familiar with internet platforms, social media, or do not have a wide social support net with sufficient funds to offer monetary assistance. Furthermore, the very ability to appear “deserving” is shaped by historically-rooted social inequalities. Unsurprisingly, during the COVID-19 crisis, an event that has critically impacted people’s fiscal and social stability, US residents have turned to COVID-related crowdfunding (CCF) in droves, with websites like GoFundMe (which, unlike some other crowdfunding websites, did not make receiving funds at the end of a CCF campaign contingent on meeting any specific goals) seeing surges in campaign numbers throughout the course of the ongoing pandemic (Igra, Kenworthy, Luchsinger, & Jung, 1982; Rajwa et al., 2020; Saleh, Ajufo, Lehmann, & Medford, 2020). During March 1 – August 31, 2020, GoFundMe hosted over 150,000 CCF campaigns, making it a prominent figure in the COVID-19 crowdfunding landscape (Kodner, 2020). At the beginning of the COVID-19 era, GoFundMe’s CEO Tim Cadogan stated that there were ‘unprecedented numbers [of CCF campaigns] for GoFundMe, and people [were] making more COVID-19 campaigns every day’ (Kodner, 2020). This trend has persisted as the COVID-19 crisis matures; however, unlike early on in the pandemic when campaigns were largely based around supporting those in emergency situations (Saleh, Ajufo, Lehmann, & Medford, 2020), the majority of campaigns after the first few months of the pandemic were created to solicit support for basic needs, such as rent, food, or car payments (Lerman, 2021).

Researchers have posited that analysing crowdfunding campaigns can be useful for gleaning nuanced information regarding the unmet needs of communities, and that through analysis of crowdfunding campaigns, policymakers can see how a community perceives a problem as well as the acute needs therein (Saleh, Ajufo, Lehmann, & Medford, 2020). However, this claim may be inaccurate and is only useful if crowdfunding use is proportionate to needs in a population; if not, it could potentially further fuel disparities caused by the pandemic as crowdfunding campaigns could overwhelmingly represent the needs of privileged members of society who have technology access, large and wealthy support systems, and in-depth understanding of personal marketing. Consequently, we aimed to test the extent to which crowdfunding campaign creation and earnings aligned with COVID-19’s epidemiological and economic impacts across Washington State in the first seven months of the pandemic.

1.3. Washington state: a poignant example of inequity

Washington has a diverse socio-economic climate, which represents myriad economic, cultural, ecological, and racial spectra across its 39 counties. Western Washington is politically liberal, with sprawling urban metropolises, such as Seattle and Bellevue, that house tech giants like Amazon and Microsoft (Washington - State Energy Profile Analysis - U.S. Energy Information Administration (EIA), 2021). In part due to its proximity to the Puget Sound, it is a hub for international commerce as well as travel (Washington State Department of Commerce). Eastern Washington, on the other hand, is politically conservative, and dependent on agriculture to support its economy (Washington - State Energy Profile Analysis - U.S. Energy Information Administration (EIA), 2021). It is also landlocked, and instead of experiencing the high volume of rain that the west side gets, it is very arid, with long expanses of near-desert conditions that are ideal for the abundance of crops that thrive there (Western Regional Climate Center, 2021) and attract migrant and seasonal farmworkers largely hailing from Central America (National Center for Farmworker Health and Inc, 2012).

Racially and ethnically, Washington is fairly diverse, with the highest proportion of individuals identifying as white, followed by Latinx, Asian American and Pacific Islander, and Black (U.S. Census Bureau QuickFacts, 2019). It is also one of the few states with more than 200,000 Native American residents, who make up nearly 3% of the population (Washington - State Energy Profile Analysis - U.S. Energy Information Administration (EIA), 2021). Due to racial capitalism and histories of extractive settler colonialism, economic inequities between racial and economic groups are stark. Okanogan County, which houses the Confederated Tribes of the Colville Reservation and nearly 10,000 Native American residents (Colville Tribes, 2021), has a mean house-hold income of only a little over $47,000, while mean household income for King County (where Seattle is located) is nearly $95,000 (Income in the Past 12 Months, 2019). Eastern Washington as a whole has a mean income of almost $76,000, while western Washington’s mean income sits slightly over $110,000; with the state-wide mean household income between these two values at $79,000 (Income in the Past 12 Months, 2019).

COVID-19 exacerbated gaps in social determinants of health despite state attempts to dampen impacts (The Bridgespan Group, 2020). Along with its complex socio-economic climate, Washington state was one of the first states to institute stay-at-home orders in the US, the first state to report a death caused by COVID-19, and experienced one of the earliest outbreaks in a long-term care facility (CDC, 2020). Due to these early experiences, Washington legislators reacted swiftly and implemented stay-at-home orders to quell viral spread, developing a ‘Phased Approach’ where each county was given a rating from 1 to 4 depending on COVID-19 severity (Phase 1 most severe, Phase 4 least severe), based on incidence of new cases reported during the prior two weeks, COVID-19 hospitalization trends, and (if available) the reproductive rate of the virus (Safe Start Washington Phased Reopening, 2020). This rating dictated which activities were allowed and by whom and could only change after being granted Governor approval (McNerthney, 2020; Safe Start Washington Phased Reopening, 2020). Phased stay-at-home orders resulted in widespread fiscal impacts; for instance, during April 2020 unemployment was almost 4 times higher than during the same time period in 2019 (Regional and State Employment and Unemployment Bureau of Labor Statistics, 2021). The number of uninsured in Washington state also increased during this time to its highest rate for the general population at 12.6% in May 2020 (Estimated Impact of COVID, 2021). Newly unemployed workers fared even worse, with the uninsured rate for this cohort peaking at 58% in May 2020, compared to 8.1% before COVID-19 (Estimated Impact of COVID, 2021). Finally, GoFundMe listed Washington as the state with the 4th highest number of donations per capita in their annual report for 2021, indicating that Washington has a fairly high existing level of crowdfunding (GoFundMe’s 2021 Giving Report, 2021). All these factors combined make Washington state a compelling place to explore how pandemic inequities and crowdfunding usage intersected during the pandemic.

1.4. Geovisualization and its application to statistical analysis

Geovisualization assists researchers in interactively detecting and revealing distinctive spatial patterns and characteristics of data through mapping the data onto spatial maps, using programs such as ArcGIS (Exploring Geovisualization, 2005). Knigge and Cope (2006) discuss how Exploratory Data Analytics (EDA) can be used with/in geographic visualization because it “employs statistical techniques to reveal hidden characteristics and facilitate seeing what the data ‘tell’ us in order to develop new questions or hypotheses” (p. 2027) (Grounded Visualization, 2006).

A GIS-based Multi-Criteria Evaluation model (MCE) allows researchers to analyze a series of variables and rank them from most impactful to least impactful; the main challenge of MCE application is determining criteria weights and choosing variables that constitute MCE (Jiang & Eastman, 2000). The outcome of GIS-based MCE is generally a map depicting locations fulfilling all conditions set within threshold values (Ruget et al., 2019). The integration of MCE into GIS analysis has gained significant interest over the last couple of decades, and GIS-based MCE has been vital in advancing GIScience in two particular fields: spatial decision support (Chakhar & Mousseau, 2008; Goodchild, 2008; Jiang & Eastman, 2000) and participatory GIS (Zhang, Sherman, Yang, & Wu, 2013).

Along with GIS-based MCE, other methods of analysis exist to analyze COVID-19 demographics in space. These include the Geographical Random Forest approach estimating local non-linear relationships between the COVID-19 death rate and various socioeconomic and health-related factors, (Grekousis, Feng, Marakakis, Lu, & Wang, 2022), the spatial lag by regimes regression model to examine socioeconomic and health determinants of COVID-19 (Grekousis, Lu, & Wang, 2022), spatial error models (Mollalo et al., 2020), local fuzzy geographically-weighted clustering (Grekousis et al., 2021), machine learning (Moosazadeh, Ifaei, Tayerani Charmchi, & Somayeh, 2022), multivariable logistic regression models (Pan et al., 2021), and spatial statistical analysis (Cordes & Castro, 2020). For instance, Cordes and Castro used spatial statistical analysis (e.g., Global Moran’s I) to analyze clusters of COVID-19 and other contextual factors in order to determine areas of highest need; this allowed researchers to determine which areas had the highest proportion of positive tests paired with low testing (Cordes & Castro, 2020). These types of analyses are useful because not only do they often give visual representations of a nuanced, complex situation, they also allow researchers to analyze statistical significance and relationships between various indicators, in particular through data standardization.

2. Materials and methods

2.1. GoFundMe data acquisition and cleaning

We gathered national GoFundMe data using a custom web-scraping tool that searched each U.S. zip-code for all campaigns containing the words ‘COVID’ or ‘coronavirus’ between January 1, 2020 and July 31, 2020. Within this dataset of 176,561 campaigns, we identified and excluded 12,250 small business campaigns started by GoFundMe without the knowledge of, and unclaimed by, business owners. We then extracted all Washington state-based GoFundMe campaigns from the sample for analysis, yielding 4459 campaigns. 626 campaigns less than 13 days old were removed from the sample due to concerns they might skew data on campaign outcomes, leaving 3832 campaigns (Igra, Kenworthy, Luchsinger, & Jung, 1982). Core data from each campaign were catalogued and analyzed, including measures of outcomes: number of donations, amount raised, average donation size per campaign. Since this study relies on publicly available online data, it was not subject to human subjects research regulations at [the author’s institution]. Nevertheless, data was held on password-protected drives and all research team members received certificates in human subjects training.

2.2. Development of social marginalization and COVID-19 impact indices

County-specific data for Washington’s 39 counties were compiled from various sources and used to develop two specific indices: social marginalization (SM) and COVID-19 impact. Data were cleaned in Microsoft Excel (version 2008) for use in ArcMap (ArcGIS Desktop 10.7.1) and Stata (Stata/SE 16.1 for Windows (64-bit x86–64)) software. The SM index was made up of county-level data for each variable, including: mean income (Mean Income in the Past, 2019), percent with one or more type of computing device (Types of Computers and Internet Subscriptions, 2019), percent below poverty level (Poverty Status in the Past, 2019), percent white-identifying (Detailed Race, 2019), percent that did not complete high school (Educational Attainment for the Population, 2019), percent that completed a bachelor’s degree or higher (Educational Attainment for the Population, 2019), percent households receiving food stamps (Food Stamps, 2019), percent uninsured (Selected Characteristics of Health Insurance, 2019), percent unemployed (ESDWAGOV - Monthly employment report, 2020), and county populations (MRSC - Washington County Profiles, 2020).

The COVID-19 impact index consisted of both county-level data regarding COVID-19 impact measured using county COVID-19 case counts over the study time period (COVID-19 data dashboard, 2020), and the time each county spent in the most acute phase (Phase 1) of Washington’s mandated closures to reduce COVID-19 spread (What’s open, 2020). COVID-19 deaths were not used as a measure because we wanted to capture the broadest impacts of COVID-19, which had impacts on time off work, care needs, health bills, and food insecurity even when people were not experiencing deaths.

In order to develop the SM and COVID-19 indices, variable data were compiled from various resources (listed above). We then mapped the geospatial representations of social marginalization and COVID-19 impact variable data in ArcGIS as choropleth maps, using 4 Quantiles to make a MCE model (results shown in column A of Supplemental Data). This created geospatial representations of the composite variables/indicators, illustrating the impacts of both COVID-19 and social marginalization on Washington state counties.

The MCE broke variable data into four categories, allowing us to categorize the indices variable data from 1 to 4 (1 associated with being least socially marginalized or least impacted by COVID-19, 4 associated with being most socially marginalized or most impacted by COVID-19). We assumed either a positive or negative/inverse relationship between each variable and the SM and COVID-19 impact indices due to each variable either contributing to or protecting from social marginalization or COVID-19, allowing us to properly categorize each variable from 1 to 4 when doing the MCE mapping and categorization. For example, “mean income” is inversely related to the SM index (as mean income goes up, social marginalization goes down), so the counties with the highest mean incomes would be ranked as 1s for our Quantile categorization, while the counties with the lowest mean incomes would be ranked as 4s. Whereas “percent below poverty level” is positively related to social marginalization, so counties with the highest level of “percent below poverty” were ranked as 4s, while the counties with the lowest level were ranked as 1s.

The categorization sums of each variable were determined per county (rows 18 and 23), and counties were ranked from 1 to 5 for COVID-19 impact and SM indices (1 least socially marginalized, 5 most socially marginalized; 1 least impact of COVID-19, 5 most impact of COVID-19) (rows 19 and 24). For both the SM index and COVID-19 impact index, no counties had an overall index ranking of 1. These indices were then mapped as MCE models in ArcGIS using choropleth maps with 4 Quantile classification to geovisually relate the SM and COVID-19 indices with the CCF campaign median amount raised per WA county, along with number of campaigns per 100,000 population.

We understand the results in MCE are dependent on the chosen criteria and how they are weighted and the variables chosen (Jiang & Eastman, 2000); nine variables for the SM index and two variables for the COVID-19 impact index were chosen for this study, but no weighting was applied. To account for this, we normalized our data using Quantile classification method, and added the score of each criterion for the benefits of being easier to calculate, understand, and communicate the outcome. However, we acknowledge the importance of applying a weighting scheme that reduces the uncertainty surrounding the weighting of the criteria for both our SM index and COVID-19 impact index.

2.3. Descriptive statistics

Descriptive statistics were assessed using Stata, looking at campaign donation variables, social media interaction, campaign creation month, and campaign category for the WA state campaigns as a whole. Drawing on previous study findings, we wanted to see whether crowdfunding usage was aligned with COVID-19-related impact across WA state, and if social marginalization were enabling or dis-enabling people to crowdfund. Given existing research has shown that crowdfunding generally advantages individuals with privilege and higher socioeconomic standing and likely is increasing disparities in wealth (Duynhoven, Lee, & Michel, 2019; Igra, Kenworthy, Luchsinger, & Jung, 1982; Lukk et al., 2018), we hypothesized that social marginalization would impede both campaign creation and monetary outcomes. We also expected that COVID-19 severity would lead to increased campaign creation within impacted counties, as counties with higher COVID-19 impacts would have higher needs to address, and would lead to more successful campaign outcomes (measured by campaign median amount raised), due to the fact that within these counties there may be an increased willingness for communities to support each other. Our null hypotheses were that there was no relationship between SM index and campaign prevalence or campaign median amount raised, and that there was no relationship between COVID-19 impact index and campaign prevalence or campaign median amount raised.

To explore these hypotheses and search for emergent patterns and relationships between the mentioned variables, we first looked at the campaign median amount raised (CMAR) for the counties with the highest social marginalization and COVID-19 impact, as well as the lowest social marginalization and COVID-19 impact, for examples of an average performing campaign under these circumstances. We then utilized the COVID-19 impact index and SM index, and mapped these along with density of campaigns and CMAR per county. In order to test these hypotheses we ran ANOVA tests to examine for univariate associations between the variables; the ANOVA is specifically testing if there is a difference in the means between groups. We analyzed the relationships at a county level between both indices and campaign density, as well as between both indices and CMAR. 3 counties with less than 3 campaigns were removed from the median raised calculations to decrease outliers affecting the data.

3. Results

3.1. COVID-19 impact vs. prevalence of campaigns

3832 GoFundMe CCF campaigns were created between January 1 and July 31, 2020 in Washington State. Despite asking for nearly $1.6 billion total, Washington State campaigns raised a little more than $8.8 million from over 99,000 donations. Campaigns meeting their monetary goals were rare and variably distributed, seen by the large spread in interquartile ranges shown in Table 1 for amount raised, donation count, and mean donation. The median campaign in our sample raised a mere $170 from 3 donations (averaging $66), towards a goal of $4,500, which is only 3.8% of the campaign goal. While 61.8% of campaigns had at least a single donation, 38.2% remained without any support, and even less (11.4%) met their monetary goal. In rare cases campaigns would do very well and attract significant attention (shown by the max shares being 1787 for a single campaign and the maximum raised $28,125). Generally, though, the median number of shares per campaign was dismally low at 0 shares, with only 3 ‘likes’ per campaign. The majority of campaigns were created in March (27.5%) and April (25.8%), and there was a wide spread of campaign categories present, with the top categories being “Accidents and emergencies” (22.5%) and “Medical, illness, and healing” (17.5%), which makes sense, considering they were in regard to the COVID-19 crisis.

Table 1.

Descriptive statistics of COVID-related GoFundMe campaigns.

Variable	N	n (%) or Sum	Median (IQR)	Min, Max
Campaign donation variables
Amount asked	3832	$1,597,000,000	$4500 ($1033 - $10,000)	$50, $1,000,000
Amount raised (total campaigns)	3832	$8,876,013	$170 ($0 - $1535)	$0, $28,125
Donation count	3832	99,227	3 (0–21)	0, 306
Mean donation	2360	$216,651	$66 ($43 - $100)	$5, $500
Social media variables
Number of shares	3722	488,857	0 (0–59)	0, 1787
Number of favorites	3722	90,502	3 (0–20)	0, 282
Campaign descriptive variables
Month created	3832
January		26 (0.7%)
February		36 (0.9%)
March		1054 (27.5%)
April		987 (25.8%)
May		646 (16.9%)
June		603 (15.7%)
July		480 (12.5%)
Campaign Category	3832
Accidents and emergencies		862 (22.5%)
Animals and pets		196 (5.1%)
Babies, kids, and family		261 (6.8%)
Business and entrepreneurs		229 (6.0%)
Community/volunteer/faith		589 (15.4%)
Dreams, hopes and wishes		154 (4.0%)
Education and learning		179 (4.7%)
Funerals and memorials		157 (4.1%)
Medical, illness, and healing		671 (17.5%)
Other		534 (13.9%)
At least a single donation	3857
Yes		2385 (61.8%)
No		1472 (38.2%)
Met monetary goal	3857
Yes		441 (11.4%)
No		3416 (88.6%)

We first looked at the campaign goals and outcomes of counties that experienced high levels of social marginalization and COVID-19 impact compared to those with low levels (Table 2) to compare both ends of the spectrum, looking at the “best” and “worst” case scenarios present. In both cases, campaigns had lofty goals (mean goal amount $17,7786 and $7,422, respectively) but did not come close to meeting those goals in the majority of cases (median amount raised $50 and $46, respectively). We then mapped the relationship between COVID-19 impact and prevalence of campaigns per 100,000 population within Washington (shown in Fig. 1) to explore what was occurring across Washington geospatially regarding COVID-19’s impact on campaign prevalence. There was not an apparent trend - some counties that were deeply affected by COVID-19, such as Chelan and Franklin, had low rates of campaign prevalence (0–24 per 100,000 population and 24–31 per 100,000 population, respectively), which could indicate that COVID-19 impact did not necessarily align with county-wide GoFundMe usage. However, other counties that were deeply impacted, such as Spokane, Benton, Yakima, and Douglas counties, had 31–43 campaigns per 100,000 population, which aligns more with our hypothesis that CCF usage would be proportional to COVID-19 impact. Occasionally, a county ranked with the lowest COVID-19 impact had the highest rate of campaigns (43–131 per 100,000 population), which was the case in San Juan, Thurston, and Jefferson counties. To specifically test if there is a difference in the means between groups, we ran an ANOVA to test if there is a statistically significant difference in the campaign prevalence means between counties compared to the COVID-19 impact index (F = 0.13, p = 0.941); since this p-value is larger than 0.05, we accept the null hypothesis that there is no statistically significant difference in mean campaign density between counties of different COVID-19 impacts (Table 3A). This was unexpected, as we thought that with increased impact due to COVID-19, there would be an increase in county-wide usage of CCF to support needs. Consequently, we next tested whether perhaps there was county-wide need but due to social marginalization causing barriers to access, crowdfunding was not being turned to for monetary support.

Table 2.

Campaign goals and outcomes for counties with low social marginalization and COVID-19 impact and high social marginalization and COVID-19 impact.

Indicator	Goal amount		Median Amount Raised (Counties with 3 or less campaigns excluded)
	N	Mean (stdev)	N	Mean (stdev)
High social marginalization, high COVID-19 impact	114	17786.449 (94494.11)	114	50 (0)
Low social marginalization, low COVID-19 impact	35	7422.143 (11917.08)	35	45.85714 (24.38918)

Fig. 1.

COVID-19-related GoFundMe campaigns per 100,000 population and COVID-19 impact index in Washington state (2020).

Table 3A.

ANOVA used for relationships between COVID Impact Index, and campaign density.

Indicator	Campaign Density per 100,000 by County
	N	Mean (stdev)	P-value	F-value
COVID-19 Impact Index	36
2		39.68 (18.245)	0.941	0.13
3		39.125 (32.060)
4		36.183 (16.302)
5		33.268 (6.952)

3.2. Social marginalization vs. prevalence of campaigns

To examine this possibility, we explored whether social marginalization was associated with the prevalence of campaigns; Fig. 2 shows this comparison. Generally, the counties ranked highest on our SM index had the lowest prevalence of campaigns (0–24 campaigns per 100,000 population), seen in Ferry, Stevens, Grant, Adams, Mason, and Pacific counties, and the counties with the lowest social marginalization ranking had the highest rates of campaign creation, in King, San Juan, Thurston, and Clark Counties (43–131 campaigns per 100,000 population). There were a few exceptions, where counties with high social marginalization also had high prevalence of campaigns (Grays Harbor and Yakima with 31–43 campaigners per 100,000 population), but for the most part, the rule was that the higher the social marginalization experienced in the county, the lower the prevalence of campaigns present. ANOVA was used specifically to test if there is a statistically significant difference in the campaign prevalence means between counties compared to SM index; since this p-value was less than 0.05, we reject our null hypothesis. Therefore, the result of the ANOVA test (F(3,32) = 3.86, p = 0.018) in Table 3B showed that there is a statistically significant difference in mean campaign density between counties in different social marginalization categories, and therefore a relationship between these variables.

Fig. 2.

COVID-19-related GoFundMe campaigns per 100,000 population and social marginalization index in Washington state (2020).

Table 3B.

ANOVA used for relationships between Social Marginalization Index and campaign density.

Indicator	Campaign Density per 100,000 by County
	N	Mean (stdev)	P-value	F-value
Social Marginalization Index	36
2		48.935 (14.220)	0.018	3.86
3		46.634 (28.056)
4		27.663 (9.391)
5		24.533 (7.079)

Table 3A and 3B ANOVA used for relationships between COVID-19 impact index and campaign density, and social marginalization index and campaign density.

3.3. Social marginalization vs. campaign median amount raised per county

After confirming a relationship between social marginalization and prevalence of campaigns, we explored whether social marginalization impacted campaign outcome, measured by CMAR (Fig. 3). In order to remove outliers, we excluded counties with less than three campaigns. Fig. 3 showed that in many cases, counties that had the highest social marginalization rating had the lowest county CMAR, present in Mason, Pacific, Stevens, and Franklin counties ($0-$50), and some counties with the least social marginalization had higher median amount raised, seen in King and Thurston counties ($201-$360). However, there were many exceptions to the hypothesis that increased social marginalization would lead to less crowdfunding money raised, such as in Grant county, which was ranked both as the most socially marginalized, and also had the highest amount raised ($361-$1639). This could be due to there being very few campaigns in the area, so a single campaign stood out in the community and was able to mobilize lots of support. Other exceptions included counties with the least social marginalization having the lowest median amounts raised, such as in San Juan and Kitsap counties ($0-$50), which may be due to these areas already having strong fiscal stability causing their campaigns not to pick up traction, as campaign success in raising funds and reaching monetary goals is often due to people considering them worthy of donations (Berliner & Kenworthy, 2017). We tested the relationship between social marginalization and county CMAR using ANOVA (F(3,32) = 0.50, p = 0.687), since p > 0.05, we accept the null hypothesis that there was no statistically significant difference in mean county CMAR between counties and social marginalization categories (Table 4).

Fig. 3.

County median amount raised for COVID-19-related GoFundMe campaigns and social marginalization index in Washington state (2020).

Table 4.

ANOVA used for relationships between social marginalization index, COVID-19 impact index, and median amount raised.

Indicator	Median Amount Raised (Counties with 3 or less campaigns excluded)
	N	Mean (stdev)	P-value	F-value
Social Marginalization Index	32
2a		130.000 (131.415)	0.687	0.50
3		230.000 (249.530)
4		222.714 (212.065)
5		355.188 (589.570)
COVID-19 Impact Index	32
2a		228.000 (355.138)	0.107	2.23
3		131.000 (176.664)
4		434.909 (457.968)
5		79.417 (61.734)

^aNo counties were at either index level 1.

3.4. COVID-19 impact vs. campaign median amount raised per county

Finally, we wanted to see if there was a relationship between the county CMAR compared to COVID-19 impact, shown in Fig. 4. These results again showed differences between COVID-19 impact and relative success in raising funds through CCF. For example, the counties with the highest COVID-19 impacts (Chelan, Douglas, Yakima, Benton, Franklin, and Spokane) all had lower county CMAR ($0-$200), while frequently counties that were less impacted by COVID-19 (such as Jefferson, Clallam, Thurston, Grays Harbor) had higher county CMAR ($201-$1639). In some cases, lower COVID-19 impact aligned with the lowest CMAR (such as in Stevens, Asotin, San Juan, Kitsap, Mason, Pacific, Lewis, and Skamania counties, which were $0-$50), and higher COVID-19 impact aligned with the highest CMAR (such as in Okanagan, Klickitat, Whitman, and Grant counties, at $361-$1639). We then used ANOVA to test for a relationship between county CMAR and COVID-19 impacts (F = 2.23, p = 0.107), which showed that we could accept the null hypothesis. It showed that there was no statistically significant difference between the means of county CMAR and the COVID-19 impact index (Table 4).

Fig. 4.

County median amount raised for COVID-19-related GoFundMe campaigns and COVID-19 impact index in Washington state (2020).

4. Discussion

4.1. General findings

Even though crowdfunding was immensely popular at the beginning of the pandemic with 3832 CCF campaigns created in Washington alone, it is clear that it was not a very successful avenue to gather funds for the majority of people participating, with only 11.4% of CCF campaigns meeting their goal and 38.2% not getting a single donation. Campaigns were generally started during the beginning of the pandemic in March and April, likely due to a combination of people feeling fear and uncertainty surrounding the coming months, along with experiencing high rates of unemployment due to stringent lockdowns and facing severe financial insecurity while waiting for benefit programs to disburse funds. The average campaign for counties highly socially marginalized and impacted by COVID-19 raised a median of only $50, while counties with low social marginalization and COVID-19 impact raised $45; these outcomes showed generally very low rates of success in garnering funds. Overwhelmingly, the median campaign experience was not one that met campaign goals.

4.2. Social marginalization has a statistically significant relationship with CCF campaign prevalence

As past research has indicated (Duynhoven, Lee, & Michel, 2019; Igra, Kenworthy, Luchsinger, & Jung, 1982; Lukk et al., 2018), and as we find here, social marginalization has a statistically significant relationship with campaign prevalence. Counties with more privilege and higher socio-economic status (lower social marginalization) generally had more crowdfunding usage than those with higher social marginalization. We saw that areas with high rates of social marginalization had the lowest rates of campaign creation; this is likely due to intersecting aspects of social marginalization within county communities causing barriers to campaign development, such as a lack of access to necessary technology, smaller social networks from which to draw financial support, social support systems with less available capital, the experience of racism, and other barriers.

4.3. COVID-19 impacts have no relationship with CCF campaign prevalence

Despite our expectations that areas with large impacts from COVID-19 would also have a proportionally high prevalence of CCF campaigns, there was no relationship between these variables. This indicates that crowdfunding does not necessarily reflect actual rates of COVID-related need in a population. This may be explained by areas with substantial COVID-19 impacts also experiencing high rates of social marginalization, which could have caused barriers to creating CCF campaigns. This may lead to an increase in inequity and disparities of wealth within these counties, as people who already have fiscal security could be taking advantage of another way to gain financial stability, while those who have deep need may be unable to use this potentially valuable resource. This is emblematic of existing inequities in Washington state, which experiences immense disparities in wealth, privilege, and access to resources among its populace. With its already profound divide between those who are privileged and those who are socially marginalized, crowdfunding is likely further exacerbating these inequities, which may result in increased problems in socioeconomic disparities in the state.

4.4. Both social marginalization and COVID-19 impacts have no relationship with campaign median amount raised per county

We did not see a relationship between social marginalization and campaign outcomes, measured by county CMAR. This finding does not align with existing literature on the relationship between income and campaign outcomes (Igra, Kenworthy, Luchsinger, & Jung, 1982); perhaps with a larger sample size in which outliers may play less of a role, this finding would not still hold. We also did not see any relationship between COVID-19 impact and campaign outcomes, which is striking and troubling, as it means that counties that were more affected by the pandemic were not seeing greater success raising funds with CCF. This means that within counties with high need, there was less use of the resources available, which may enforce disparities of wealth and exacerbate inequity.

5. Conclusion

5.1. Crowdfunding campaigns do not represent need accurately

Past research has suggested that crowdfunding campaigns are a window into the needs of a community during crisis (Saleh, Ajufo, Lehmann, & Medford, 2020). However, our research shows that campaigns are likely showing a distorted perception of need, as seen by those in counties with less social marginalization more frequently taking advantage of these avenues for funding, while those in more socially marginalized counties not making use of them, likely due to systemic barriers. Using crowdfunding campaigns as an indicator of the needs of a community may actually obscure its true needs, highlighting only those of the privileged, which could lead to increased inequities. Therefore, crowdfunding campaigns must be assessed with immense caution when considering them as a way to evaluate community needs.

Study limitations

This study has several limitations. Since we were only looking at relationships, we did not have more information about the directionality of the statistical associations between variables, and how they may have influenced each other and the outcomes measured. Another limitation within our study is that we only looked at the early months of pandemic-related crowdfunding, though by most measures this was the most acute period of activity. Our data was also set during a specific time frame, which we looked at as a whole, but did not look at more discrete time series within that frame. We also did not do a more in-depth analysis that would require a regression model - this could have looked at the different variables and how they influenced each other.

Another key limitation is that our main method of analysis was MCE, which has a number of limitations, possibly the most important being that it does not solve the issue of subjectivity in interpretation; along with this, there are possible issues regarding results generalization, risks of double counting, potential conflicts of interest, potential arbitrariness in evaluation processes, and others (Zozaya González, Oliva Moreno, Hidalgo Vega, & García-Ruiz, 2018).

We also did not look at websites other than GoFundMe, which limits the representation of alternative crowdfunding resources. Although our SM index included 9 indicators, there are other factors that can contribute to social marginalization; our COVID-19 impact index included 2 indicators, but the same can be said here – there are other factors that contribute to COVID-19 impact. We also looked at a county level, but this does not take into account the nuanced differences between smaller areas within counties. This frame of analysis may have affected our results regarding campaign outcomes, given that we assessed county-level outcome statistics. Future studies should look deeper at smaller census tract areas, as well as counties across the US, to assess the relationships between campaign creation and outcomes. Other crowdfunding websites may also be studied to capture a more complete picture of what is occurring.