Revealed Preference for Nonprofit Organizations: A Hedonic Price Analysis

2.1 The Benefits of Nonprofit Organizations

Nonprofits contribute to the well-being of their communities in a variety of ways. Economic theories posit that nonprofits enrich the lives of residents by satisfying excess demands, such as those not met by government or private sector firms, and are positioned to tailor services to their locale (Lipsky 2010; Steinberg 2003; Wolpert 1993). Nonprofits are also strong community actors and play an essential role in the generation of social conditions (Mayer 2023c; McQuarrie and Marwell 2009). Nonprofits can fulfill spiritual needs for those that prefer collective settings, and residents that participate in nonprofit activities may be more inclined to volunteer and improve community conditions (Cnaan and Curtis 2013; Marwell 2004; Yeung 2018). Strong institutional ties can facilitate and maintain social networks (Sampson 2004), and organizational participation, in particular, can grant residents greater control over their neighborhood through increased collective efficacy (Mayer 2023a; Sampson and Groves 1989). Fundamentally, nonprofits can convene residents and facilitate goal alignment, assisting in reducing collective action problems. Indeed, the presence of nonprofits can enhance social capital, parental involvement, as well as reduce crime and child maltreatment (Marwell 2004; Mayer 2023a, 2023c; Ressler 2020; Wo et al. 2016).

2.2 Services for Stigmatized Groups

As discussed above, organizations benefit greatly from maintaining fit with their surroundings and procuring support from constituents. However, nonprofits are a heterogeneous group, and acceptance of nonprofits may vary over space. A hostile environment can result in difficulty achieving social recognition, acceptance, or obtaining support, and ultimately lower the viability of the organization (Fernandez 2008; Hager et al. 2004). Young (1983) emphasizes this when he suggests “the legitimacy of a nonprofit agency itself depends on it maintaining a stance consistent with the community interests” (p. 113). Consequently, a hostile neighborhood setting may make existence and service delivery more challenging, with implications for groups that rely on such services. Such hostility is widely associated with services for groups with stigmatized conditions or marginalized identities, encapsulated by the term “not in my back yard” (NIMBY) which refers to “the protectionist attitudes of and oppositional tactics adopted by community groups facing an unwelcome development in their neighborhood” (Dear 1992, p. 288). When a nonprofit is focused on services to such groups, residents may associate the behaviors of clients (or employees) with the values of organizations, effectively transferring stigma to the organization (Devers et al. 2009). In extreme cases these attitudes manifest as land-use conflicts, where residents mobilize in a formal attempt to prevent an organization from locating nearby or continuing to deliver services. Land use conflicts are common in the placement of facilities in several areas of human services including services for populations with developmental disabilities, substance use, citizen re-entry, homelessness, and AIDS (Dear 1992; Piat 2000; Solomon and Davis 1984).

Studies seeking to understand the complexities in NIMBY attitudes have often relied on case studies such as document reviews, which make differentiating between opposition to the organization and other environmental conditions difficult (Davidson and Howe 2014; Dear 1992; Dear and Gleeson 1991; Lyon-Callo 2001). For example, the arguments presented opposing a needle exchange centered around the idea that the new location was in a “different neighborhood” (although, it was a block away), concern over the safety of children or elderly residents, or that the location change placed the neighborhood “under siege” due to an influx of those seeking services (Davidson and Howe 2014). A similar set of concerns exist surrounding organizations focused on homelessness, or those seeking migrant or mental health services, including the impact on the neighborhood and businesses (Abraham and Maney 2012; Dear and Gleeson 1991; Piat 2000). Further, opposition to homeless shelters has been documented in communities considered progressive, including participation from local officials (Lyon-Callo 2001). Overall, several common features are found across occurrences, including references to the impact on property value, personal safety, or other neighborhood amenities (Dear 1992). It is important to note that the extant research suggests residents opposing the location of services often claimed they were not in opposition to the existence of the organization at a different location, or of increasing resources to the target group, were it not for the negative effects on their neighborhood. For this reason, quantitatively differentiating between existing neighborhood conditions and the contribution of nonprofits is of paramount importance.

3.1 The Hedonic Model

The study of revealed preference and implicit pricing has often centered on the hedonic model, which has been widely used in economics since its first application to automobile pricing in 1939 (Sopranzetti 2015). The hedonic model is an example of an approach for studying revealed preference; it leverages market information to infer the value of features that are otherwise difficult to value. The appeal of the hedonic model comes from its ability to decompose an asset’s price into the implicit price of its components (Sopranzetti 2015). The hedonic model for revealed preference is particularly well-suited to understand the implicit value consumers and residents place on environmental features, which allow inference for preference. For example, it has been applied to investigate consumer preferences and willingness to pay for open space, urban design, and transit access, as well as to investigate opposition to affordable housing (Bartholomew and Ewing 2011; Hamidi et al. 2020; Nguyen 2005). In this study, the hedonic model is selected to understand the role nonprofits play in the price paid, and correspondingly, revealed preference for, nearby nonprofit organizations by allowing the estimation of a premium, or discount, for home near nonprofits after accounting for a variety of features demonstrated in the existing literature.

A detailed description of the underlying model of behavior that gives rise to the hedonic model of equilibrium can be found in several texts, so it is not reproduced here (see, for example, Baranzini et al. 2008; Bishop et al. 2020; Rosen 1974). Important assumptions of the model include perfect information of consumers and the measurement of covariates at a level which is perceived by market participants (Taylor 2008). In this framework housing prices reflect demand for amenities, materials, bidding, and the allocation of households across neighborhoods (Rosen 1974). The theory provides little guidance with respect to an empirical specification, however, for home sales the model may be written as s P i = α + ∑ j β j z j + ϵ i . Where P is the price of home i, a function of J characteristics, while α and the J β′s are estimated from the data (ordinary least squares or maximum likelihood are common choices), with ϵ uncorrelated with z _j for all J. The function s may include a log or other transformation, and any number of the z’s may also be transformed, although the “semi-log” specification is most common (Taylor 2008).

4.1 Dependent Variable

The outcome of interest in this study is the price paid for residential sales in Cuyahoga County from 2014 to 2016. Although proximity to nonprofit organizations may benefit commercial entities as well, the sample is limited to residential sales as the focus of this paper is the benefits nonprofit organizations provide to neighborhood residents. The price paid for a home provides a better measure of revealed preference than alternatives such as the median in the neighborhood or a valuation provided by another service. In contrast to the former, home sales can be mapped to characteristics of the home, which allow a detailed decomposition of price. In contrast to the latter, this variable provides direct information on preference, and market valuations often fail to account for important neighborhood features and other amenities (Sirmans et al. 2005). For example, in these data, the estimated market value and the realized sale value have a correlation of 0.88. Sales for less than 10,000 dollars are excluded from the sample as they are unlikely to be occurring at market rate, however, maintaining variation in both tails of the distribution of sale prices is important to include a range of households in the equilibrium (Freeman et al. 2014). Additionally, all sale prices are inflation adjusted to represent 2016 dollars.

4.2 Key Independent Variable: Nonprofit Density

The key independent variable in this study is nonprofit density, measured by the number of active public charities present in the census tract each year, a measure consistent with the extant literature on nonprofit density (Joassart-Marcelli and Wolch 2003; Wo 2018; Yan et al. 2014). Information about the number of nonprofit organizations is obtained from the National Center for Charitable Statistics Business Master File (BMF), a longitudinal data source based on a file maintained by the Internal Revenue Service containing basic information on nonprofits that submit any variety of tax documents, which has become a widely used source of data for nonprofit scholars working with smaller geographies seeking accurate counts (Mayer 2023c; Wo 2018; Yan et al. 2014). The BMF provides accurate information on the number of organizations that submit tax forms, however, may also contain inactive nonprofits. Accordingly, organizations that have not filed tax forms in the past two years are removed from the sample as they are unlikely to be active. Private foundations and supporting organizations are excluded as they often maintain a distinct purpose and may have drastically different relationships to their surroundings. The remaining organizations were geo-coded and placed in census tracts using 2010 census geographies. In the event that an organization listed a post office box as their main address, the post office was used (Yan et al. 2014). Using this process, just over 99 % of active organizations were successfully geocoded, while the remainder were excluded.

4.3 Control Variables

Control variables must be carefully chosen in applications of the hedonic model. Taylor (2008) suggests selecting covariates that fall into three categories, 1) characteristics of the home or lot, 2) features of the neighborhood, and 3) the property’s locational characteristics. Features related to consumers, or “demand variables”, must be avoided as they can substantially alter the interpretation of the model, resulting in a regression for bid-function (see discussion of this in Yinger and Nguyen-Hoang 2016).

4.4 Empirical Model

There are several complications to consider when approaching an empirical application of the hedonic model. Several authors caution against purely additive and linear specifications, as they do not allow the flexibility required to reflect sorting behaviors or utility functions (Bishop et al. 2020; Taylor 2008; Yinger and Nguyen-Hoang 2016). For this reason, the empirical model includes squared terms for the number of bedrooms, bathrooms, and property size. Squared terms may allow, for example, decreasing marginal returns for higher numbers of rooms in very large homes. The spatial dynamics of housing markets make the contextualization of parameters of paramount importance. Yet, the use of fixed effects for a geography of interest shares many of the same concerns as covariates: although geographic fixed effects may reduce omitted variable bias, they also shift the model toward bid-function regression to the degree that they are correlated with demand variables (Yinger and Nguyen-Hoang 2016). The multilevel approach first introduced by (Orford 2000) is particularly advantageous in this study with variables measured on three distinct geographic units, the parcel, tract, and city.^[3] Accordingly, the model includes Gaussian random effects (i.e., random “intercepts”) varying over tracts and cities.

The tract level variables contain all neighborhood features discussed in Section 4.3.2, such as the vacancy rate and number of gun-involved events, as well as nonprofit density. However, the preference for a home may reflect the nearby surroundings. To account for the possibility that sale price reflects the features of nearby neighborhoods, spatially lagged neighborhood variables are employed.^[4] After spatially weighting, neighboring nonprofit density is strictly positive and has a heavy right tail, consequently it is log-transformed. All models include fixed effects for the year of the sale, and the land use code.

Two models are examined to address the research questions. Each model is estimated with a Gaussian likelihood and includes three variance components, one from the parameterization of the Gaussian (e.g. a residual), in addition to non-nested tract and city random effects. The first model includes the total number of nonprofits in the census tract, as well as the number of nonprofits in nearby tracts through the spatial weight. The second model decomposes nonprofit density into nonprofits focused on populations with stigmatized conditions or marginalized identities and those that do not. The second model addresses the second research question, allowing heterogeneous response to nonprofits based on primary purpose. A direct comparison of the resulting parameter estimates is done by using “statistical simulation” drawing 10,000 samples from the variance-covariance matrix assuming the distribution of the parameters is multivariate normal with mean vector equal to point estimates, then taking a 95 % interval of the difference (linear combination) of the samples (King et al. 2000). This method is justified by the central limit theorem and allows posterior inference by taking the differences between samples, obtaining an interval estimate for the difference in preference based on the focus of the organization.

5.1 Description of Data

Descriptive statistics for the sample are shown in Table 1. Table 1 shows the average residential sale occurred in a tract with between 12 and 13 nonprofits, less than one of which serves stigmatized groups. The average residential sale also occurs in tracts that average less than one gun involved event a year and a residential vacancy rate of nearly 6 %. The average sale price is just over 130,000 dollars, less than half the median sale price of homes in the US in 2016 (U.S. Census Bureau and U.S. Department of Housing and Urban Development 2016).

Spatial dynamics are at play in Cuyahoga County’s housing market, panel B in Figure 1 shows the residential sale price averaged over sales and years within each census tract. The figure shows a clear pattern with higher average sale prices in suburban areas, with low average sale prices in the northeast section of Cuyahoga County. The pattern of nonprofit density, shown discretized in panel A of Figure 1, is less distinct. The northeast appears to have lower average density over the study period, with higher density found in the suburbs near the outer edges of the map.

Figure 1:

Spatial dynamics in average nonprofit density (A) and average residential price (B) in Cuyahoga county Ohio from 2014 to 2016. Panel A shows the average nonprofit density in each tract from 2014 to 2016. Panel B shows the residential sale price, averaged over all sales in the census tract and years on the log scale.

As discussed in Section 2, nonprofits tend to co-locate to borrow legitimacy, as well as share information and resources (Mayer 2023b). This presents a complication when seeking to disaggregate preference for different types of nonprofits. Table 2 shows the product-moment correlation for the main variables included in the study (omitting parcel and home characteristics). Several correlations are worth noting, including the strong negative correlation between parks and school performance and strong positive correlation between parks and residential vacancy. However, of paramount importance is the strong positive correlation between the locations of nonprofits that provide services to groups with stigmatized conditions or marginalized identities and those that do not, a result of strong agglomerative forces.

5.2 Model Results

Convergence was unproblematic for all models, the full results of which are shown in Table 3. The results show consideration of the multilevel structure is an important part of the phenomenon. Akaike Information Criterion (AIC) shows substantial improvements in estimated prediction error when the random effects are included. With all variables and nonprofits aggregated, the linear model estimated by OLS produced an AIC of 88160.32, including the tract-level random effect reduced this to 72010.62, and the city-level random effect reduced it further to 71773.86, with similar improvements when nonprofits are disaggregated.^[5] Additionally, the model and covariates have conditional R² of just over 0.75, suggesting the fixed and random effects explain over 75 % of the variation in log residential sale price.

Note that Table 3 omits the year fixed effects, which are not of interest in this study. Model 1 shows the estimate for nonprofit density at the tract level is positive and statistically significant. This suggests that an increase of one standard deviation in nonprofit density predicts an increase between 4 and 10 % in the expected residential sale price of a home. Further, the estimate for nonprofit density in neighboring census tracts is positive as well: a 1 % increase predicts an increase in the expected residential sale price of between 5 and 12 %. Model two allows different responses to nonprofit density. The results show the estimate for nonprofits focused on those with stigmatized conditions or marginalized identities is consistent with zero and other negligible values, suggesting these nonprofits have little to do with the determination of sale prices.

Following the method described in Section 4.4, the difference in marginal effects between non-stigmatized nonprofits and stigmatized nonprofits is between 0.025 and 0.099. This suggests a one standard deviation increase in the number of nonprofits predicts an increase in residential sale price between 2.5 and 10 % beyond that which is expected from a one standard deviation increase in stigmatized nonprofits. Although not the focus of this study, the estimates for the school performance index show an increase of one standard deviation predicts an increase of 1 and 9 % in the expected sale price in the city. The results also suggest the vacancy rate is important, as the residential sale price is sensitive to the rate of vacancy in the tract and in nearby areas. The estimates for the random effects also capture substantial heterogeneity, with a standard deviation between tracts of 0.32 and 0.29 between cities.