NONPROFIT ORGANIZATIONS IN GENESEE COUNTY, MICHIGAN By Andrew T. Guhin A THESIS Submitted to Michigan State University i n partial fulfillment of the requirements f or the degree of Psychology Master of Arts 202 0 ABSTRACT NONPROFIT ORGANIZ ATIONS IN GENESEE COUNTY, MICHIGAN By Andrew T. Guhin Background: Nonprofits organizations deliver a variety of crucial goods and services to communities. Given this, it is important to consider if these organizations are located where they are most need ed. Previous research on this topic has primarily considered location as a function of the needs and resources in an area . More recently, there has been interest in agglomeration as an additional facto r influencing location. This thesis contributes to the literature on this topic by examining the relationship between these factors and nonprofit location. Methods: This thesis analyzed National Center for Charitable Statistics (NCCS) and the American Community Survey data from 2013 and 2018 to evaluate nonpro in Genesee County, Michigan as a function of needs, resources, and agglomeration . Service provider location was measured at the census tract level using kernel density estimation. Results: First, an Ordinary Least Squares (O LS) model was estimated. A indicated significant spatial autocorrelation . A Spatial Durbi n model was then estimated, and the percent of renters in a census tract emerged as the only significant predictor of nonprofit locatio n . Discussion: Study findings show that most nonprofits included in the dataset are clustered in Flint, an area with high levels of need . However, the towns and cities surrounding Flint showed comparable levels of need, but far fewer nonprofits . This study also reaffirms the need to for researchers to be sensitive to the spatial nature of this type of data. Copyright by ANDREW T. GUHIN iv TABLE OF CONTENTS LIST OF TABLES ... . .. . v . LITERATURE REVIEW.. . 2 The Importance of Location ... 4 Theories Influencing Nonprofit Locat .. 10 . . 5 7 8 . 9 9 Dependent Variable 2 1 Independent Variable s.. . ... 2 2 Kernel Density Estimation ... 3 6 6 . 3 1 . . . ... 7 Limitations ... . 4 1 Future Directions . ... 4 2 . . ....4 3 REF . . . 4 5 v LIST OF TABLES Table 1 . NTEE Categories . . .. ..... 3 Table 2. Variable Li st 2 3 Table 3. Descriptive Statistics 2 7 Table 4. 2013 Zero - Order Correlations . 2 8 Table 5. 2018 Zero - Order Correlations . .. . ... . 2 9 Table 6 . 2 9 Table 7 . Lagrange Multiplier Diagnostic . ... . 2 T able 8 . Spatial Durbin Model . ... 4 Table 9 . SDM Effects 5 vi LIST OF FIGURES Figure 1. NTEE Codes of Nonprofits .. .. 9 Fi gure 2 . Kernel Density Map of Genesee County Nonprofit Organizations 5 Figure 3 . OLS Residual Plots .. . ... 3 1 Figure 4 . Percent Renting in Genesee County 3 9 Figure 5 . Choropleth Map for Model V ariables .. .. 4 4 1 INTRODUCTION Nonprofits are a vital part of American life. While diverse in scope of activity, arguably the most crucial function of nonprofit organizations ( NPOs ) is in the provision of basic goods and services to c ommunities (Allard, 2009). Compared to n ations with large welfare states, in the United States a more decentralized government with heavy emphasis on voluntarism, localism, and philanthropy nonprofits are the primary providers for many types of service s, with many of them being local, place - based organizations (Lipsky & Smith, 1993). Because of - being, the geographic equity of nonprofits is an important topic of consideration. The present stu dy is an empirical investigation of nonp rofit location in Genesee County , Michigan, and provides unique contributions to the literature on nonprofit organizations , along with useful contextual data. Historically, empirical literature on the location of non profit organizations has ignored conside rations of spatial statistics (Yan, Guo, & Paarlberg, 2014). This study accounted for previous methodological limitations by estimating a spatial Durbin model. Outside of the scholarly community, the information from this study may be of use to : (a) commun ity members in Genesee County who receive services from these organizations , (b) nonprofit organizations who may have interest in , as well as (c) stakeholders in the community, who could utilize the study results to inform decisions re lated to the service location. 2 LITERATURE REVIEW Terms and Definitions (3) tax exempt organizations. Additionally, t he literature revie wed mainly uses three distinct terms - location, sector size, and density to describe nonprofit activity of some kind in a concentrated area. Many studies will use one description while citing literature that uses another and, for most, the unit of analy sis is generally no smaller tha n a census tract. The term used by the researcher shapes the preceding language used in the paper, the form of the research questions and, sometimes, the variables included in models. This thesis use throughout to refer to a n in a community, bearing in mind that the level of analysis location is being d study used the term, and it w as appropriate to describe it in a similar way. The broad nature of the nonprofit sector makes defining exactly what constitutes a nonprofit organization difficult. In the United States, a primary legal distinction between a nonprofit and a for - profit orga nization is that the latter can generate and distribute profits to owners/shareholders ( Hopkins, 1987). This distinction offers a useful starting point for understanding the sector and some of its behavior. Scholars have furt her noted that nonprofits are: (a) formal organizations, (b) private, institutionally separate from government, (c) nonprofit distributing, (d) self - governing, (e) voluntary, and (f) serve some public benefit (Salamon, 1999; Hammack, 2002). Attempts to fu rther define the entirety of no nprofits beyond these broad characteristics become difficult. The sector, like business and government, are highly diverse and carry out 3 many functions. The National Center for Charitable Statistics (NCCS) started the Nationa l Taxonomy of Exempt Entities ( to support the work of researchers and policy analysts ( Jones, 2019 ). Shown by Table 1 below, the NTEE is comprised of ten broad categories and twenty - six major fi elds. Each major field is furth er divided into subfields that total close to a thousand categories (Salamon, 2015). Table 1 . NTEE Categories Broad Category Major Group I. Arts, Culture, and Humanities A II. Education B III. Environment and Animals C, D IV. Health E, F, G, H V. Human Services I, J, K, L, M, N, O, P VI. International, Foreign Affairs Q VII. Public, Societal Benefit R, S, T, U, V, W VIII. Religion Related X IX. Mutual/Membership Benefit Y X. Unknown, Unclassified Z Note (2019). Retrieved from https://nccs.urban.org/project/national - taxonomy - exempt - entities - ntee - codes#overview In many ways, the NTEE has been successful in improvi ng nonprofit research. Nearly all researchers studying nonprofits use it to categorize and breakdown data according to their interests and substantive focus. It has also facilitated easier communication between scholars. But i t is not without limitations. The main limitation identified in the existing literature is that, while beneficial to nonprofit research, NTEE categories may exclude organizations of interest to researchers (Grønbjerg, 1994; Fyall, Moore, and Gugerty, 2018) . For example, if a researcher were interested in studying human service nonprofits , the researcher could simply use the NTEE . entirely clear what distinguishes a human service nonprofit from o ther categories. For example, the line separating health and human services is clear in some instances ( e.g., a medical research 4 organization vs. a religious antipoverty organization ) but is more often blurred ( e.g., should a e classified as a health nonprofit or human service nonprofit?) (Grønbjerg, 2001). Relatedly, some organizations may serve dual purposes and, thus, could fall under multiple categories. Using the previous example, the religious antipoverty nonprofit might be classified as human services, religion - related, or be placed in another category. If a they very well may exclude o rganizations by only using the designa ted NTEE codes (refer to Fyall et al. , 2018 for recent empirical evidence of this). The method section will elaborate more on this and describe how this study plans to address some of these concerns ; however , the impor tant takeaway point is that much of th e literature reviewed uses NTEE codes and is subject to th is limitation. The Importance of Location A clear account of the spatial dimensions of inequality and the relevance of location (200 9 Out of Reach: Place, Pov erty, and the New American Welfare State . book focuses on the social safety net within the United States , which includes entities other than nonprofits. However, as previously stated and as the book notes, much of the responsibility for delivering bas ic goods and services to communities in the United States is shouldered by nonprofits. Researchers who study nonprofits have noted that many of these organizations, especially those providing vital goods and services t o communities, are largely local, plac e - based organizations (Wolpert, 1993; Bielefeld, Murdoch, & Waddell, 1997; Never & Westberg, 2016). Where a nonprofit chooses to locate has important ramifications for communities . 5 Specifically, Allard (2009) identifie d three reasons why the location of an organization impacts service outcomes : connectivity, trust, and geographic accessibility . First, part of the argument for shifting more of the burden for service delivery from the state to nonprofit organization s rest s on the idea that nonprofits are bett er connected with local communities (Trudeau, 2008). When someone knows more about a nonprofit, they are more likely to seek them out and interact with them as opposed to unfamiliar ones (Kissane, 2003) , and a lack of awareness may result in underutilizati on of services (McDougle, 2014). Allard asserted expected people will know more about organizations nearer to them than those further away because knowledge of providers is shared between community members. A dditionally, many and engagement efforts are typically geographically concentrated, with a focus on increasing local awareness. that offer servi ces that are sensitive, stigmatized, o agencies are seen as active and invested members of a community and are connected to local 9 , p. 50). Many place - based nonprofits wor k to establish and maintain ties to th eir neighborhoods and communities for this reason. L iterature on social capital and organizations supports this (Snavely & Tracy, 2002; Bryce, 2007; Schneider, 2009). If a nonprofit is located outside of the area where most of its patrons reside, this coul d damage their ability to form relationships and establish trust. The last reason Allard identified was that as distance to a service location decreases for an individual, the costs of seeking out services and certain access barriers (e.g., problems with t ransportation, time spent traveling, willingness to seek out provider) are likely, though not necessarily, to decrease in kind, thereby increasing the chances that the organization and 6 individual interact. More general ly, geographic accessibility to amenit ies ( e.g., schools, parks, health facilities, etc.) has been studied by researchers in a variety of fields (Dalton, Jones, Ogilvie, Petticrew, White & Cummins, 2013) and focuses on the spatial equity of amenities. Spat ial equity is a concern for this study place matters much more to the success of social programs in a safety net driven by social 9 , p. 14). Theories Influencing Nonprofit Locatio n There are a variety of theories on nonprofit organizations, often approaching the subject from different disciplinary perspectives (Anheier, 2006). Many of the theories described below are attempts to explain the ex istence of nonprofits and do not explic itly theorize why a nonprofit locates where it does. While this review does not exhaust all theoretical perspectives, those included may help explain why a nonprofit would locate in a particular community versus anoth er. The approach of this review is like other studies on this topic, where needs, resources, and agglomeration/prior density are all possible explanations to the question: Why does a nonprofit locate where it does? Heterogeneity and Failure The primary the ories that emphasize the role of commun ity need (which can also be the creation of nonprofit organizations are market/ g overnment failure theory (Weisbrod, 1977; 198 6 , 2009 ) and contract failure theory (Ha nsmann , 1980; 1987). Market/Government failure theory A good is widely considered to be people than to one, and (b) once provided, there is no way to exclude ce rtain peoples from 7 consuming it . For ex ample, air pollution control is a popular example of a public good. Private - (DiMaggio & Anheier, 1990) ; that is, most people woul d not choose to pay for a private servi ce offering air pollution control if they could benefit from the service without paying. The question is then, in the presence of market failure, what entity will provide public goods? Weisbrod (1977; 1986 ; 2009 ) argu ed that, in a democracy, when market fa ilures arise the government will provide at least some public goods to the citizenry, and the decision of which goods to provide is part of an elected officials job. For Weisbrod, and for an entire school of political thought, the - making. Theorists have varying descriptions of who included in this subpopulation , but one way is to think of the median voter is the demands she would make on governme public goods is then reflective of the preferences of the statistically average voter . Under this paradigm, if there is homogeneit y in demand for a good or service and t he government responds accordingly, then the processes will satisfy voters. However , if there is only a small constituency desiring a good or service (i.e., demand heterogeneity), government officials are not likely t o address these concerns or demands. Ye - profit firm would likely not provide it either (Anheier, 2006). Weisbrod argued that the existence of nonprofits in a market economy can be explaine d by this scenario, one where heterogen ous demand creates a situation of unmet needs. As an extension of the critiques of it. As noted by Romer and Rosenthal (1979), these come down to a disagreeme nt about the influence of voters on gov ernment expenditure and whether other factors, like 8 bureaucratic inertia and political economy, are more decisive. While t here are also critiques of the theory on its own terms, some of which will be discussed in deta il further on , what i s most relevant fo r this thesis is nonprofit location and its marked influence on the empirical literature . For example, a recent meta - analysis from Lu (2017) notes that ation primarily tests one of them, the demand heterogeneity hypothesis. The hypothesis predicts a positive relationship between d emand heterogeneity ( e.g., diversity of demand) and nonprofit sector size, and Lu notes that this is seen as the fundamental hy pothesis for the theory. To measure demand heterogeneity, researchers have used demographic characteristics such as race, gender , age, sex, religion, etc., as proxy variables. The meta - analysis contained thirty - seven studies that had one or more measures of heterogeneity as a predictor of nonprofit sector size. asuring any of the individual studies framed their work as investigating location or service access. O verall, the meta - demand heterogeneity, indicating that population heterogeneity had a positive effect in determining nonprofit sector size. Specifically, f ive of the ten measures were significant and had positive effect sizes, indicating support for the heterogeneity hypothesis: age, education, ethnicity, language, and religion. However, t he overall weighted average effect size was 0.034 ( p <.001), and the effect sizes ranged from 0.020 to 0.147, which are conside red small by most research standards. Moreover, as Lu (2017) point ed out, measuring heterogeneity is difficult and these results likely reflect that observation. The author recommend ed using a comprehensive index to measure heterogeneity and focusing on t he measures that showed support for the hypothesis . The 9 relationship between a heterogeneity measure and nonprofit sector size is also highly likely to be moderated by the type of work a nonprofit does , which Lu notes as a limitation of the current existin g literature and a direction for future research. An analysis of this interaction would be of particular use because some of the difficulty in using demographic characteristics as a measure of heterogeneity is the assumption that ascriptive demographic cat egories have unified, shared interests, needs , and demands. This assumption may hold true in some cases. For example, Biel efeld and colleagues (1997) found that racial heterogeneity was related to the location of health, social service, and education nonpr ofits. In their conclusion, Bielefeld and colleagues noted , of diverse preferences because it can be reasonably assumed that each racial group has an equal desire for more of its own type of pr T he authors seem ed to imply that racial segregation of services was desired as t he racial diversity in an area increased; this speculation leaves open the question of whether this desire stems from racism, an attem pt to account for racial disparities in services , or both. Ben - Ner a nd Van Hoomissen (1992) found similar results, and sta I n education, racial diversity enhances [nonprofit] provision, which may be due to resegregation attempts by white parents seeki ng to avoid the busing of their children to findings may reflect the u nified interest of white individuals in an area to racially exclude nonwhite individuals, which could hold across a variety of types o f nonprofits. It could also be the case that some characteristics, such as religious heterogeneity, may influence some act ivity fields (most obviously , religious organizations) more than others (e.g., food, agriculture, and nutrition). 10 Contract Failure Theory that provide goods and se rvices that for - profit organizations also provide (Anheier , 2006). If consumers can: (a) make a reasonably accurate comparison of the quality and price of goods or services offered by different organizations , (b) reach an agreement on the good or service to be provided and its price, and (c) determine if those co nditions were met, a for - profit organization should be able to provide goods or services at an efficient level (Hansmann, 1980, p. 843). For a variety of reasons these conditions are often not met ( e.g., - .) , and the resulting inefficiency is termed a market failure . Hansmann (19 80 ; 1987) focused on a specific type of market failure, informational asymmetry when one party in a transaction has more relevant knowledge than the opposite party as a way to e xplain the existence of nonprofits alongside for - profit businesses. In an in stance of information asymmetry favoring a business, a consumer may not be able to adequately evaluate a good or service because the business knows more than the consumer do es abou t the product. When the business is aware of this discrepancy, they can use this to take advantage of the consumer. However, asserts that, because of information asymmetry, nonprofits have an advantage over for - profit organizations beca use of their perceived trustworthiness. This trustworthiness comes from the , key distinguishing mark of nonprofits that prohibits the organization from distributing net earnings to individuals who have ownership in the busine ss. Thus , nonprofits are thought to have little incentive to take advantage of informational asymmetries and consumers will trust the nonprofit over the for - profit organization . In this way, nonprofits may act as a fiduciary, of sorts. 11 M arket/government f ailure theory has been tested far more than contract failure theory. Perhaps this is because testing the contract theory requires analyzing the market share of nonprofit and for - profit organizations in various sectors along with the population of interests levels of trust in for - profit business (Corbin, 1999). One study conducted by Salamon and Anheier ( 1998) tested a host of theoretical perspectives of nonprofits, including contract failure, at a cross - national, country - wide level (i.e., the unit of analy sector). As a proxy of trust in nonprofi ts, the researchers used survey data from the World Values Survey, which had a measure of trust in various institutions. Salamon and Anheier used the difference between trust in co rporations and average trust in all other institutions to capture trust in n onprofits, and then averaged results across all participants in the survey by country of origin to create one composite score. The authors did not find support for the theory, even after breaking the data down by country and type of nonprofit. While theor etical concerns have greatly influenced the empirical literature, many studies simply include measures of community need and then test its influence on nonprofit location without m uch explicit discussion of theory. For example, Peck (2008) found that antip overty nonprofits in Phoenix , Arizona were locate d where there we re higher levels of need, as measured by the unemployment rate and the number of people under the federal poverty l ine. However, Yan and colleagues (2014) conducted a study of Hartford antipo verty nonprofits by essentially replicat ing and addressing some of its methodological limitations. As discussed further in the Method section, many studies of nonprofits do not always choose the best model fit for their data and fail to accoun t for spatial variability, both of which could seriously bias observed results , as Yan and colleagues observed (2014) . Yet, a fter accounting for these limitations , the Yan and colleagues (2014) 12 had a signi ficant, positive effect on the number of antipoverty nonprofits in a census tract. So, theory driven or strictly empirical, the literature indicates that community needs ar e a factor in where nonprofits locate, whether need is operationalized by demographi c characteristics or material resources. Resources Nonprofits, like for - profits or government entities, require monetary and human resources to carry out their operations (Grønbjerg & Paarlberg, 2001). It is necessary for their survival to maintain reven ue streams along with attracting employees, volunteers, or both. This has been - Ner and Va stakeholder theory is useful for considering why a nonprofit mi ght failure theories , previously discussed , while also addressing supply side c onsiderations. The authors argued that unmet demand for a good or service cannot alo ne explain why nonprofits sufficiently motivated to form an organization for nonmonetary reasons, but those stakeholders must also be able to provide or secure the resources needed to maintain the organization. The stakeholders are then simultaneously supply and demand side actors (Anheier, 2006). While stakeholder theory does not suggest resources alone motivate location decision making, it leaves room for the i dea that the number of nonprofit organizations in an area may be influenced by available stakeholders (i.e. , entrepreneurs willing to start an organization, funders, volunt eers, etc.). A key resource provider for nonprofits is the government. At least that 1995) interdependence theory . One of his criticisms of the failure theories 13 was the assumption that governments and nonprofits have a competitive relationship, where nonprofits serve to fill a void left by insufficient responses from markets and governments. Instead, he argued that there is good reason to believe that the nature of the relationship between governments a nd nonprofits is cooperative and more of a partnership. For unique historical and political reasons, t model; rather than directly providing welfare (e.g., something like cash ass istance, nationalized health care, etc.,) , the government provides and directs funding to third party entities, like nonprofits, as a way to reconcile the desire for public services but general skepticism of government (Salamon, 1987). Salamon further arg ue d weaknesses. Nonprofits may possess loc al knowledge that governments do not have and can help to advocate for various causes (Salamon & Anheier, 1998). But, the weaknesses of the sector terme may showcase its limitations for supplanting government action. ailing of the voluntary system as a provider of collective goods has been its inability to generate resources on a scale that is both adequate enough and reliable enough to cope with the human - 987, p. 39). Philanthropy is also prone to discriminatory and paternalistic giving, conferring the power to decide between the deserving and the undeservi ng to those with the most resources. For these reasons, Salamon argue d that the governments and nonpro fits have an interdependent, complementary relationship. These two theories , stakeholder theory and interdependen ce theory, are not necessarily in conflic t with one another and have both been used to guide empirical research. Interdependence theory is comm only tested by hypothesizing a positive relationship between some measure of 14 government activity and nonprofit activity. The measure used depends on facto unit of analysis, geographic scope, and the availability of data. Variables like soc ial welfare spending (Salamon & Anheier, 1998), county library expenditures (Grønbjerg and Paarlberg, 2001), government grants (Luksetich, 2008), and gove rnment wages per capita (Kim, 2015) have been previously used by researchers as proxies for government activity. Generally, this theory has empirical support (Lecy & Van Slyke, 201 3 ) but results are not always consistent across studies. A recent meta - analy sis (Lu & Xu, 2018) of 30 studies on government size and nonprofit sector size found a significant pos itive relationship between the two variables ; however , the effect was small. The studies included in the meta - analysis differed on important characteristics including year of study, country of origin , measurement of variables , unit of analysis , and type of nonprofit (arts, social services, religious, etc.) . However, even a fter accounting for these moderators, the relationship between government size and nonprofit sector size showed a significant positive but small correlation ( r = .063 ) . As for stakeholder theory and nongovernmental resources on location of nonprofits, the results are mixed. Financial measures of resources are typically used to capture potential revenue streams for nonprofits, which flow from a few sources. Nongovernment sources include thi ngs like investments, service delivery fees, and private donations (Fischer, Wilsker, & Young, 2011). While resources have been operationaliz ed in a variety of ways depending on the study, generally , researchers have used financial measures and/or characte ristics of local populations as proxies. For example, studies have confirmed the relationship between resources and nonprofit location (Peck, 2008; Yan et al., 2014; Never & Westberg, 2016). In an empirical study that tested their stakeholder theory, Ben - N er and Van Hoomissen (1992) found that communities 15 with more educated, wealthier residents had more nonprofits. Bielefeld et al. (1997) and C orbin (1999) also found that income levels were significantly related to nonprofit location . Agglomeration Effects Scholars across a variety of disciplines interested in cities, particularly economists, have long studied agglomeration economies. In the ur ban economics literature, agglomeration eople locate near advantage of agglomeration econo mies is the reduction in transportation costs for goods, people, and ideas (Ellison, Glaeser, & Kerr, 2010). Essent ially, agglomeration economies are generally believed to be what drives clustering of people and firms into specific locations (Chatterjee, 2 process works. The au thors hypothesized two different spatial patterns of hotel locations in Manhattan: differentiation and agglomeration. If a newly formed organ ization was offering a similar product to an already established organization, they may consider differentiating th emselves from competitors and maintain a geographic distance. Alternatively, however, the new organization may perceive benefits, like shared infrastructure, information, and reduced consumer search costs, as reasons to locate near established competition. The authors found that hotels located near one another to benefit from agglomeration economies and differentiated themselves from competitio n through other means, such as size. While researchers have paid considerable attention to agglomeration economies in for - profit industries (for example, see Puga, 2010), a paucity of literature exists exploring agglomeration economies amongst nonprofit or ganizations. 16 The limited studies that have tested for it generally find some positive effects, although the result s are contingent on factors related to the study. In one of the earliest, if not the earliest, studies exploring this topic, Bielefeld and Mu rdoch (2004) examined the location of nonprofit education and human service providers and for - profit counterparts i n six large metropolitan areas in the United States. Mostly, the authors found little evidence of agglomeration economies, but when there wer e positive results, findings differed by metropolitan area and type of service. In Boston, nonprofits tended to clu ster near for - profit firms, nonprofits in Dallas/Fort Worth tended to cluster near other nonprofits of similar sizes, and in Minneapolis, sma ller nonprofits tended to cluster around larger for - profits. Another major study to include a test for the effects of specifically focused on religious, advocacy, professional, and cultural organizations. Except for religious organizations, the results showed support for the ag glomeration hypothesis. Because these studies have found some initial support for this hypothesis and because there is a significant gap in t he literature, the current study may add a much - needed data point to the field by including a test for agglomeratio n. 17 CURRENT STUDY The three explanatory factors presented needs, resources, and agglomeration are not necessarily in conflict with one another. Most research acknowledges that no one factor is likely dominant over another, and it i s likely that a mix of factors inform where a nonprofit locates. Furthermore, modeling limitations, external validity assumptions, and activity specific re sults (i.e., the type of nonprofit), likely contribute to the mixed results and lack of clarity in previous literat ure. To answer the main research question the following hypotheses were put forward: 1. There is a significant, positive relationship between the number of nonprofits in Genesee County census tracts and levels of need. 2. There is a significant, positive rel ationship between the number of nonprofits in Genesee County census tracts and levels of resources. 3. There is a significant, positive relationship between the number of nonprofits in Genesee County census tracts and the previous number of nonprofits in a ce nsus tract. 18 METHOD Study Area This study exclusively focused on nonprofits in Genesee County, Michigan. Contained within the county is the Flint Metropolitan Area (FMA), one of the most racially segregated s a deep historical context (Highsmith, 2015) with intentionally discriminatory and racist policies, along with white flight to surrounding townships being key contributors (S adler & Highsmith, 2016). ned with the automobile crisis in the late 20 th extension a stable tax base, declined, ci ty government revenues and the available workforce fell in tandem (Reckhow, Downe y, & Sapotichne , 2018). The surrounding townships were thus able to provide services for residents, while the city center was left deeply impoverished and sorely underserved ( at less than half of its administrative capacity and represents an extreme case of a nationwide crisis in local government (Reckhow, Downey, & Sapotichne , 2018). Additionally, the Flint water crisis - an egregious instance of environmental racism that pois oned residents of the mostly African American city (Pulido, 2016) - was formally recognized in April 2014. Nearly six years later, the effects of the crisis are still ongoing. Reckhow, Downey, and Sapotichne (2018) noted that the nonprofit sector has histo rically had a prominent role in Flint, stepping in to provide funding for services when the city could not. Their research indicates it has been particularly key in the afterm ath of the water crisis. Nonprofits in the community have collaborated to host me etings, disseminate information, and provide vital services to community members (e.g., bottled water). Further, Flint residents frequently cited nonprofits as important leadi ng 19 organizations in the community. With these facts in mind, Genesee county is a unique setting to investigate what contributes to nonprofit service distribution. Sample Selection In line with most of the prior literature in this area, nonprofits were selected using the National Taxonomy of Exempt Entities (NTEE), the classification system used by the IRS and Figure 1 . NTEE Codes of Nonprofits 20 alth and human services are two of the ten broad categories of nonprofit s in the NTEE system; each of these broad categories are further subdivided by spec ific area activity. For example, the category groups, (1) Health Car e, (2) Mental Health & Crisis Intervention , (3) Voluntary Health Associa tions & Medical Disciplines , and (4) Medical Research. Within these groups, there are subdivisions by activity area and type of organization. This thesis was concerned with nonprofit s like health and human service providers, specifically those that provide direct services to community members . Within the NTEE categories for health and human services, many were immediately excluded from eligibility. For example, the Medical Research subd ivision of Health was excluded, as these organizations do not provide he alth services. But there were also organizations that fall outside the health and human service NTEE categories that were included in the analysis. The NTEE is a useful but imperfect way to categorize nonprofits and relying solely on them for inclusion/ex clusion presents challenges. The previously discussed definitional problem of human services raised by Grønbjerg (2001) is applicable here. While Grønbjerg wrote about the blurry disti nction between health and human services, that blurriness applies elsewh ere too. For example, the research on the social determinants of health (SDH) shows that education, ethnicity and cultural orientation, exposure to crime, and spiritual/religious value 2005). A reasonable argument could be made that, given the SDH, a religious advocacy nonprofit is an important component of the health and well - being of those involved with the organization. It obviously makes little sense to categorize the religious nonprofit as a health non profit for the goals of the NTEE, but for research purposes it is worth considering what 21 an illustration of this, Fyall and colleagues (2018) found, for example , that if a researcher were interested in nonprofits providing housing and shelter, only using the designated NTEE category would exclude many nonprofits that clearly provide housing a nd shelter services based on their mission statements; in their sample o f Washington state nonprofits, it excluded 80% of relevant organizations. Instead of just using NTEE codes, as some studies have done, a more useful strategy is to predefine the relev ant population, service type, or SDH of focus, for example, and then eit her use mission statements to select organizations or include all relevant NTEE codes that match the nonprofits as those in the following NTEE categories: education, health, mental health, justice, food banks/soup kitchens, shelters, legal services, community development, housing, youth development, residential services, foster care and adoption, and homeless services . Joassart - Marcelli and Wolch (2003) as well as Pol son (2017) used similar lists. Likewise, this thesis is concerned with nonprofit organizations that provide direct goods and services to communities. NTEE codes nutritiou s food, clean water, s anitation, health services, education services, housing, electricity, and security services . Figure 1 shows the list of NTEE codes used. Dependent Variable The two most common approaches to measure nonprofit activity are: (a) nonprofi t density , or (b) orga nization expenditures (Never & Westberg, 2016). The former was approach was adopted for this study and is described in more detail in the following section. Part of this siness Master File (BM F). 22 Prior research by McDougle (2015) identified important limitations of the NCCS dataset for nonprofit addresses, including: the use P.O. Box addresses, incorrect addresses, and multiple service locations. The McDougle paper investi gated the accuracy of a ddresses in the NCCS core files dataset, but the issues raised are equally applicable to the BMF dataset because the core files are constructed using the descriptive information from the BMF. In this thesis, if an organization in the dataset listed a P.O. Box as their nonprofit address , the street address was manually identified. Yan and colleagues (2014) were able to successfully do this for roughly half of the organizations in their dataset; t he other half were still included in the ir analysis because P. McDougle also empirically investigated this question and found that many nonprofits had P.O. boxes in the same area as their operating address, supportin g the idea that it is better to keep these addresses in the data set rather than excluding them from the dataset. Independent Variables Following prior research, the following variables from the American Community Survey 5 Year Estimate were included in the first models. 23 Table 2. Variable List Category Variable Description Source & Year Dependent Variable (DV) Nonprofit Density per 1,000 The average kernel density per census tract per 1,000 (weighted by total expenditures) NCCS 201 3 & 2018 BMF & Core Files I ndependen t V ariable (IV) - Needs Poverty Level % of people in census tract below the poverty line 2013 & 2018 ACS 5 Year Unemployment Level % of people in census tract unemployed 2013 & 2018 ACS 5 Year % Renter Occupied % of people in a census tract who rent as a opposed to own their living space 2013 & 2018 ACS 5 Year IV - Resources Educational Attainment % of people in a census degree or higher (including professional degrees) 2013 & 2018 ACS 5 Year Housing Value Median housing v alue in a census tract 2013 & 2018 ACS 5 Year IV - Diversity Simpson Diversity Index for Race Index of racial diversity in a census tract 2013 & 2018 ACS 5 Year IV - Agglomeration Change in Density Change in nonprofit density from 2013 to 2018 2013 NCCS BMF Note. IV = Independent variable. Kernel Density Estimation per en. First, the addresses for all nonprofits active in 2013 ( N = 60) and 2018 ( N = 71) in Genesee County, MI with NTEE codes matching those listed in Table 1 were geoco ded using Texas boxes were included in the analysis. NCCS core files were then used to obtain organization operating expenses 24 Next, kernel density estimation methods were used to create a continuous density surface of nonprofit organizations. By using kernel density metho ds instead of quadrat counts which in this case would entail summing the total number of nonprofits in each census tract the modifiable area unit problem (MAUP) is avoided (Carlos et al., 2010 ; Openshaw, 1984) . The MAUP is particularly an issue in this study, as an arbitrary boundary like a census tract almost certainly does not reflect interactions between Ge nesee county residents and nonprofits (i.e., a nonprofit in one census tract can provide services to people in multiple census tracts). A second i ssue with quadrat counts in this study is that the number of nonprofits in a census tract does not relay any i nformation about the level of expenditures. For example, a tract with five small, low budget nonprofits may spend an amount equivalent to one larg e nonprofit in another tract. To work around this issue, the kernel function was weighted by total organizatio n operating expenditures. Of crucial importance for kernel density estimation is the choice of the bandwidth parameter, as this is the search radi the bandwidth increases, the surface becomes smoo ther and results in less visible variation in point intensity; conversely, as the bandwidth decreases the surface is less smooth and intensity is concentrated near point locations (Anselin et al., 2000) . For this study, the bandwidth was set to the TexMix package in R (Tiefelsdorf et al. average density was extracted, th en adjusted for population per 1,000. The final maps are displayed in Figure 2 . As the map demonstrates, the nonprofits providing basic goods are heavily concentrated in the Flint Metro Area, with little visible change between 2013 and 2018. 25 Figure 2 . Ke rnel Density Map of Genesee County Nonprofit Organizations 26 RESULTS OLS To begin, an ordinary least squares model was fit for 2018 da ta with all planned variables included in the model. The results indicated the need for data transformation and model trimming. First, the change variable to measure agglomeration caused severe i ssues; when this variable was removed, the model substantiall y improved. Because of its removal, however, hypothesis three could not be directly tested. Although not a test for the agglomeration hypothesis, separate regression models for 2013 and 2018 were subsequently estimated to compare changes over five years. In each model, the dependent variable, as Table 3 shows, was highly right skewed. To account for this, a Box - Cox transformation was applied. Unemployment rate was log transformed, and one variabl e - poverty rate - removed to account for multicollinearity. Zero - OLS results are shown in Table 6 . Also included in Table 6 obta in these estimates, a first - order queen contiguity matrix was constructed. The queen matrix defines as a neighbor any spatial unit that shares an edge or vertex. Table 6 mode l was significant and able to explain over half of the varian ce in the dependent variable (average kernel density per census tract per 1,000 (weighted by total expenditures) , with 2018 performing slightly better (2013 model R 2 = . 52 , F ( 5 , 123 ) = 26 . 25 , p < .001 ; 2018 model R 2 = . 59 , F ( 5 , 123 ) = 35 . 21, p < .001 ). T which are plotted in F igure 3 - is significant with each year, indicating residual dependence and that OLS is not an appropriate model choice. With respect to individual covariates, a few things are of interest. 27 First , there is no change in the direction of the relationships over time although beta estimates do change between 2013 and 2018. The coefficient values for the two variables that lost significance in the 2018 model Simpson Diversity Index Score and % Rent each substantially decreased; although, Table 3 indicates little fluctuation in these variables. Median housing value, while significant in each model, had an almost negligible impact. The esti mate for educational attainment substantially increased in 201 significant, and high for both years; this indicates that that these attributes are clustered over space. 28 Table 3 . Descriptive Statistics 2013 (N=129) 2018 (N=129) Total (N=258) NPO Density p er 1 , 000 Mean (SD) 0.151 (0.289) 0.204 (0.376) 0.178 (0.336) Median (Q1, Q3) 0.010 (0.001, 0.134) 0.016 (0.002, 0.214) 0.014 (0.001, 0.168) Min - Max 0.000 - 1.340 0.000 - 1.908 0.000 - 1.908 % Rent Mean (SD) 0.303 (0.196) 0.307 (0.1 96) 0.305 (0.195) Median (Q1, Q3) 0.305 (0.113, 0.435) 0.329 (0.114, 0.426) 0.326 (0.113, 0.434) Min - Max 0.023 - 0.768 0.012 - 0.873 0.012 - 0.873 Median Housing Value Mean (SD) 85 , 639.535 (44 , 248.602) 91 , 424.016 (58 , 439.876) 88 , 531.775 ( 51 , 812.382) Median (Q1, Q3) 87 , 700.000 (45 , 000.000, 116 , 300.000) 90 , 800.000 (33 , 200.000, 134 , 100.000) 88 , 700.000 (41 , 875.000, 121 , 800.000) Min - Max 12 , 700.000 213 , 400.000 9 , 999.000 262 , 300.000 9 , 999.000 262 , 300.000 Simpson Diversity Index S core Mean (SD) 0.232 (0.166) 0.234 (0.173) 0.233 (0.169) Median (Q1, Q3) 0.194 (0.089, 0.356) 0.174 (0.090, 0.395) 0.186 (0.090, 0.367) Min - Max 0.000 - 0.582 0.004 - 0.650 0.000 - 0.650 Unemployment Rate Mean (SD) 0.193 (0.104) 0.1 25 (0.103) 0.159 (0.109) Median (Q1, Q3) 0.165 (0.112, 0.261) 0.092 (0.050, 0.183) 0.132 (0.077, 0.213) Min - Max 0.022 - 0.590 0.008 - 0.509 0.008 - 0.590 or Higher Mean (SD) 0.164 (0.103) 0.177 (0.113) 0.170 (0.108) Median (Q1, Q3) 0.144 (0.097, 0.214) 0.146 (0.094, 0.232) 0.144 (0.094, 0.224) Min - Max 0.012 - 0.502 0.006 - 0.497 0.006 - 0.502 29 Table 4 . 2013 Zero - Order Correlations % Rent Median Housing Value Simpson Diversity Index Unemployment Rate % Degree or Higher % Rent Median Housing Value - .55** Simpson Diversity Index .55** - .45** Unemployment Rate .47** - .72** .32** Degree or Higher - .34** .75** - .19* - .64** NPO Density per 1000 .57** - .62** .52** . 52** - .36** Note: * p<0.05; ** p<0.01 Table 5 . 2018 Zero - Order Correlations % Rent Median Housing Value Simpson Diversity Index Unemployment Rate % Degree or Higher % Rent Median Housing Value - .59** Simpson Diversity Index .5 7 ** - .4 7 ** Unemployment Rate . 55 ** - . 69 ** . 28 ** Degree or Higher - .3 5 ** .7 9 ** - . 14 - . 56 ** NPO Density per 1000 .5 4 ** - . 71 ** . 48 ** .5 8 ** - . 41 ** Note: * p<0.05; ** p<0.01 30 Table 6 . Dependent varia ble: Density per 1000 2013 2018 2013 2018 % Rent 0.333 ** 0.074 0. 454 *** 0. 461 *** (0.127) (0.130) Med. Housing Value < - 0.00 0 *** < - 0.00 0 *** 0. 759 *** 0. 813 *** ( < 0.000 ) ( < 0.000) Si mpson Diversity Index 0.332 ** 0.147 0. 589 *** 0. 493 *** (0.142) (0.141) Log Unemployment Rate 0.605 * 0.641 ** 0. 547 *** 0. 574 *** (0.342) (0.306) Educational Attainment 0.547 * 0.939 *** 0. 603 *** 0. 638 *** (0.294) (0.284) Constant 0.261 ** 0.518 *** (0.119) (0.088) Residuals 0.691*** 0. 6 0 9*** Observations 129 129 R 2 0.516 0.589 Adjusted R 2 0.497 0.572 Residual Std. Error (df = 123) 0.213 0.202 F Statistic (df = 5 ; 123) 26.250 *** 35.205 *** Note: * p<0.1; ** p<0.05; *** p<0.01 31 Figure 3 . OLS Residual Plots Spatial Models 6 (2013: 0.691, p<.001; 2018: 0.609, p<.001) subsequent models attempted to a ccount for the spatial dependency in the data. A series of spatial regression models were estimated, all using the same queen contiguity matrix previously described. Given the natu re of the data, the spatial Durbin model (SDM), both conceptually and empiri cally, was the best fit. The SDM takes the following form: The first term, variable value; the second term, , is the standard matrix of explanatory variables (i.e., a census the third term, , accounts for the influ ence of a this data because previous literature indicates that the presence of nonprofits in one region may influen ce the presence of nonprofits in neighboring re gions (the lagged y component); additionally, 32 Table 7. Lagrange Multiplier Diagnostic characteristics of a region may influence the number of nonprofits in neighboring regions (the lagged x component); for example, high levels of unemployment rate in one tract leading to more nonprofits in a neighboring tract. Empirically, the SDM was shown to be the best fit for the data as well. Lagrange multiplier tests, shown in Table 7 , on both OLS models indicated model misspecification in t he error term as well as the presence of a missing spatially lagged dependent varia ble. With all four df LM p LM Error 2013 2018 1 1 173.21 134.65 <.001 <.001 LM Lag 2013 2018 1 1 186.26 155.49 <.001 <.001 LM Error Robust 2013 2018 1 1 11.33 6.2 5 <.001 .01 LM Lag Robust 2013 2018 1 1 24.38 27.09 <.001 <.001 33 tests significant, this is grounds for defending the choice of the SDM ( Anselin , 2013; Golgher & Voss, 2016) . The results from the SDM are shown in Table 8 , while Table 9 displays the indirect, direct, and total effects. 34 Table 8 . Sp atial Durbin Model Dependent variable Density per 1 , 000 2013 2018 % Rent 0.077 * - 0.002 (0.040) (0.052) Med. Housing Value - 0.00000 - 0.00000 (0.00000) (0.00000) Simpson Diversity Index 0.057 0.019 (0.054) (0.057) Log Unemployme nt Rate 0.190 * - 0.045 (0.102) (0.122) Educational Attainment 0.015 - 0.071 (0.098) (0.122) Lag of % Rent 0.251 *** 0.222 * (0.093) (0.120) Lag of Med. Housing Value 0.00000 - 0.00000 (0.00000) (0.00000) Lag of Simpson Diversity Inde x - 0.114 - 0.151 (0.092) (0.108) Lag of Log Unemployment Rate - 0.369 0.124 (0.236) (0.219) Lag of Educational Attainment - 0.341 * 0.380 * (0.206) (0.229) 0.981*** 0.937*** Constant - 0.049 - 0.008 (0.070) (0.065) Observations 129 129 Log Likelihood 161.011 138.514 sigma 2 0.003 0.005 Akaike Inf. Crit. - 296.023 - 251.027 Wald Test (df = 1) 7,180.304 *** 1,195.225 *** LR Test (df = 1) 244.133 ** * 183.737 *** 35 Table 9. SDM Effects Direct Effects Indirect Effects Total Effects % Rent 2013 2018 .337*** 0.123 7.206 *** 2.865* 7.544 *** 2.988* Med Housing Value 2013 2018 <0.001 < - 0.001 <0.001 < - 0.001 <0.001 < - 0.001 Simpson Diversity Index 2013 2018 .009 - 0.057 - 1.334 - 1.739 - 1.325 - 1.796 Log Unemployment Rate 2013 2018 0.043 0.001 - 4.153 1.065 - 4.110 1.067 Educational Attainment Rate 2013 2018 - 0.245 0.108 - 7.246 4.092 - 7.492 4.199 Note: * p<0.1; ** p<0.05; *** p<0 .01 36 In addition to the Lagrange Multiplier results from Table 7 lower than those from a Spatial Error Model and a Spatial Lag Model (not included here), further suggesting that the SDM is preferred. A few results from Table 8 co mpared to those from Table 6 standout. First, median housing value, while maintaining its negative direction, loses its significance. Educational attainment and unemployment rate lose significance, change direction, and coefficient values are substantially reduce d, suggesting the OLS model overestimated their importance. The only variable significant between both years is the lag variable for the percentage of people renting in a tract. The results from Table 9 , which are necessary to properly interpret the SDM re sults according to LeSage and Pace (2009), support the finding on percent renting from Table 6 ; that is, the model suggests that in each year the percent renting in a census rcent r enting rate in a census tract also influenced the density in the same tract, though this is not as strong of an effect compared to the lag. 37 DISCUSSION This study set out to test the influence of three factors in determining nonprofit locatio n in Ge nesee County, Michigan: needs, resources, and agglomeration. Each factor was hypothesized to have a positive relationship to the density of nonprofits in an area. The results of the study demonstrated the importance of spatial modeling to assess thi s resea rch question and provided useful information about each of these factors relationship to nonprofit density. A number of researchers among transdisciplinary fields conducting research related to social problems influenced by explicit spatial elemen ts, such as, ecolog y (Beale et al., 2010) , demograph y (Voss et al., 2006) , and econometrics (Pace & LeSage, 2004) , have emphasized the importance of accounting for spatial autoco rrelat ion in regression models. Left unaccounted for, spatial autocorrelation can increase type 1 error rates (Beale et al., 2010) and inflate parameter estimates ( Mauricio Bini et al., 2009 ; Lennon, 2000). In comparing the results between OLS and spatial models, it is readily apparent that far fewer variables are statistically significant and coeff icient estimates are substantially smaller in the spatial models, with some also shifting in the opposite direction. These results indicate the potential pitfalls of nonspatial models for this type of data. That is, had this study only used OLS model s, bot h of these issues would have gone unnoticed and inaccurate conclusions would have been drawn. Fortunately, researchers in nonprofit studies do utilize spatial regression methods and have for some time (see Bielefeld and Murdoch, 2004 for an earlier e xample ; for more recent ones, Yan et al., 2014, or Never and Westberg, 2016), although it is hard to gauge their current prevalence. At minimum, testing 38 model residuals for autocorrelation should become a best practice for this line of research, given the spatia l nature of the data. As for the hypotheses, this study was able to statistically test two of the three hypotheses nonprofit density. The results from the f inal spatial model found limited support for the needs hypothesis, and no support for the resources hypothesis. For each year, the only significant variable in Table 8 was the percent renting in a census tract. In 2013, all effects for percent renting were statistically significa nt, while in 2018 only the indirect and total effects were statistically significant. This finding is consistent with prior research (Peck, 2008; Yan et al., 2014; Never and Westberg, 2016) and provides some evidence that the nonpro fits of focus those pr oviding basic goods and services are located in areas where they needed. s nonprofit density than on its own. An examination of Table 9 reveals the indirect effects are larger for all variables. One way to make sense of this finding, along with coefficient values and directions, is by examining choroplet h maps for the variables . The Appendix includes choropleth maps for all the variables, but Figure 4 only shows the percent renting in a tract. As can be seen, in the tract surrounding Flint - Mount Morris and Mount Morris Township, West Burton, Beecher, an d parts of Genesee Towns hip the percentage of residents who rent is fairly high, especially compared to the outskirts of the county. But when examining Figure 1 the Kernel maps 39 density in these areas is not as high as the core metr o area of Flint. This wo uld explain the The nonsignificant results are also themselves interesting and useful to investigate. First, following the theoretical literature on divers ity and nonprofit organizations, a Simpson Diversity Index for race was calculated and included in models as a proxy for need. Its non - significance and small coefficient size in the spatial models imply extremely limited influence, if any. One reasonable e xplanation, hinted at earlier in the literature review, is that heterogeneity is usually not by itself a reliable ind icator of needs and preferences, and in the instance it is, then this is true for a limited class of things. People who are members of some group (e.g., a race, an ethnicity, a religion, etc.), are themselves diverse individuals with membership in many oth er membership in a certain group; if enou gh people in that same group share the same need or preference, and a nonprofit could meet this, then it makes sense to anticipate a relationship between the number of nonprofits in an area and the diversity along some group membership. For example, it wou ld make intuitive sense to expect a relationship between religious diversity Figure 4 . Percent Renting in Genesee County 40 and the number of religious nonprofits i n an area. In this study, there is less of a reason to expect racial diversity to be related to the number of nonprofits providing basic go ods and services, and the results possibly show this. This finding also indicates the potential limitations of using an aggregated diversity index. F uture studies should proceed with caution when using an index for variables like race or perhaps include ca tegories of interest in regression models as separate variables alongside the total index. Additionally, neither of t he proxy variables for resources educational attainment and median housing value - were significant in the spatial models. The coefficien t for the median housing value each year is very small, indicating a negligible impact. The choropleth map of median negative direction; the core Flint area in the center of the map has the lowest housing values and the largest cluster of nonprofits, but the areas surrounding the core metro area - especially to the north - also have lower values compared to the wealthier areas on the edges of the county. Taken together, these results indicate that nonprofits providing basic goods and services are located in areas with higher needs, as indicated by the k ernel and choropleth maps along with the spatial models showing the percent renting as a significant variable. Importantly however, these organizatio ns are clustered in the core metro area of Flint. The surrounding tracts have comparable levels of needs, yet the den sity of nonprofits in these areas is much lower, possibly explaining some of the nonsignificant coefficients in the model. This finding ech oes Yan et al., (2014) who found that nonprofits in the greater Hartford, Connecticut area are primarily located in u rban areas with higher proportions of renters, as opposed to suburban or rural areas. The authors attributed this result to the fact that o ther organizations that these types of nonprofits work with, such as foundations, are located in urban downtown areas , along with potential zoning issues. That interpretation holds well in this study, too. For instance, two major 41 foundations, the Charles S tewart Mott Foundation and the Community Foundation of Greater Flint. , are both located in downtown Flint and are likely a key funder for many of the organizations included in this study. This may also shed light on the non - significance of the resource var iables in the model. In theory, these types of nonprofits are likely aiming to be near those they serve, while at the same time staying close to resources for their continued financial viability. In the case of Genesee county and nonprofits in the Flint me tro area, perhaps access to foundation resources and others downtown are enough to sustain their organizations, ra ther than household (i.e., individual) resources. L imitations There are a number of limitations to this study. It is unfortunate that the aggl omeration hypothesis was not able to be statistically tested due to modeling issues. As discussed above, the patte rn observed in this study indicated evidence for some process of agglomeration, and a formal test would have provided further illuminating inf ormation. Methodologically, there are key aspects of this study that should be kept in mind. Nonprofit organizatio ns were selected based on the authors criteria and NTEE codes; the critique from Fyall et al., ( 2018 ) of NTEE codes is noteworthy, and there were likely some organizations of interest to the st udy incidentally excluded. The dependent variable was constructed using kernel density estimation, which is sensitive to bandwidth selection; if future studies use this method, they could present a series of models with different bandwidth selections or in clude sensitivity analyses. Similarly, this study chose the census tract as the spatial level for analysis, but the results could have shifted , or a new pattern emerged were a different level chosen (e.g., census block group). Finally, the lack of data on 42 conclusions. Were this data to become available, a fuller picture could em erge of areas where services are lacking. Future Directions The results of this study point toward a few differen t lines of work worth pursuing on the topic of nonprofits and service provision. First, a better measure for agglomeration economies in future studies would be of interest to the field, as this study indicates that some process of urban clustering is prese nt. Second, future studies could use more advanced modeling approaches to provide richer details on the nonprofit sector. A spatial panel mode l approach similar to the one employed by Call and Voss (2016) for child poverty using the SPLM package in R (Mill o & Piras, 2012) could account for temporal effects in nonprofit density. The difference between the OLS model results and the spatial models highlight the need to explicitly account for space in this research; a meta - evious studies for the potential presence of spatial autocorrelation in prior studies. Lastly, future work should investigate Morris, Burton, and Swartz Creek, may con tribute to the pattern of urban clustering observed in this study. Suburban poverty is a well - recognized issue by scholars (s ee, for example, Kneebone & Berube, 2013) and if exclusionary zoning is what is prohibiting nonprofit organizations from operating in these areas, this wou ld indicate a gap in services that deserves attention from community members and policy makers. 43 AP PENDIX 44 Figure 5 . Choropleth Map for Model Variables 45 REFERENCES 46 REFERENCES Allard, S. W. (2009). Out of reach: Place, poverty, and the new American welfare state . Yale University Press. Anheier, H. K. (2006). Nonprofit organiz ations: an introduction . Routledge. Anselin, L., Cohen, J., Cook, D., Gorr, W., & Tita, G. (2000). Spat ial analyses of crime. Criminal justice , 4 (2), 213 - 262. Anselin, L. (2013). Spatial econometrics: methods and models (Vol. 4). Springer Science & Busin ess Media. Baum, J. A., & Haveman , H. A. (1997). Love thy neighbor? Differentiation and agglomeration in the Manhattan hotel industry, 1898 - 1990. Administrative Science Quarterly , 304 - 338. Beale, C. M., Lennon, J. J., Yearsley, J. M., Brewer, M. J., & El ston, D. A. (2010). Regression analysis of spatial data. Ecology letters , 13 (2), 246 - 264. Annals of pub lic and cooperative economics , 62 (4), 519 - 550. n Hoomissen, T. ( 1992). An empirical investigation of the joint determination Annals of Public and Cooperative Economics , 63 (3 ), 391 - 415. Bielefeld, W., Murdoch, J. C., & Waddell, P. (1997). The influence of demographics and distance on nonprofit location. Nonprofit and Voluntary Sector Quarterly , 26 (2), 207 - 225. Bielefeld, W., & Murdoch, J. C. (2004). The locations of nonprofi t organizations and their for - profit counterparts: An exploratory analysis. Nonprofit and voluntary sector quarterly , 33 (2), 221 - 246. Bryce, H. J. (2007). The public's trust in nonprofit organizations: The role of relationship marketing and management. Ca lifornia Management Review , 49 (4), 112 - 131. Burton, W. (1988). T he Nonprofit Economy. Call, M. A., & Voss, P. R. (2016). Spatio - temporal dimensions of child poverty in America, 1990 2010. Environment and Planning A , 48 (1), 172 - 191. Carlos, H. A., Shi, X., Sargent, J., Tanski, S., & Berke, E. M. (2010). Density estima tion and adaptive bandwidths: a primer for public health practitioners. International journal of health geographics , 9 (1), 1 - 8. 47 Chatterjee, S. (2003). Agglomeration econ omies: the spark that ignites a city?. Business Review , 4 , 6 - 13. Corbin, J. J. (1999) . A study of factors influencing the growth of nonprofits in social services. Nonprofit and voluntary sector quarterly , 28 (3), 296 - 314. da Costa, M. M. (2016). What infl uences the location of nonprofit organizations? A spatial analysis in Brazil. VOLUNTAS : International Journal of Voluntary and Nonprofit Organizations , 27 (3), 1064 - 1090. Dalton, A. M., Jones, A., Ogilvie, D., Petticrew, M., White, M., & Cummins, S. (2013) . Using spatial equity analysis in the process evaluation of environmental interventio ns to tackle obesity: the healthy towns programme in England. International journal for equity in health , 12 (1), 43. Ellison, G., Glaeser, E. L., & Kerr, W. R. (2010). W hat causes industry agglomeration? Evidence from coagglomeration patterns. American Ec onomic Review , 100 (3), 1195 - 1213. Fischer, R. L., Wilsker , A., & Young, D. R. (2011). Exploring the revenue mix of nonprofit organizations: Does it relate to publicness?. Nonprofit and Voluntary Sector Quarterly , 40 (4), 662 - 681. Fyall, R., Moore, M. K., & Gugerty, M. K. (2018). Beyond NTEE codes: Opportunities to understand nonprofit activity through mission statement content coding. Nonprofit and Voluntary Sector Quarterly , 47 (4), 677 - 701. Glaeser, E. L. (Ed.). (2010). Agglomeration economics . Universit y of Chicago Press. Golgher, A. B., & Voss, P. R. (2016). How to interpret the coefficients of spatial models: Spillovers, direct and indirect effects. Spatial Demography , 4 (3), 175 - 205. Grønbjerg, K. A. (1994). Using NTEE to classify non - profit organisa tions: an assessment of human service and regional applica tions. Voluntas: International Journal of Voluntary and Nonprofit Organizations , 5 (3), 301 - 328. Grønbjerg, K. A. (2001). The US nonprofit human service sector: A creeping revolution. Nonprofit and Voluntary Sector Quarterly , 30 (2), 276 - 297. Grønbjerg, K. A., & Paarlberg, L. (2001). Community variations in the size and scope of the nonprofit sector: Theory and preliminary findings. Nonprofit and voluntary sector quarterly , 30 (4), 684 - 706. Hammack, D. C. (2002). Nonprofit organizations in Amer ican history: Research opportunities and sources. American Behavioral Scientist , 45 (11), 1638 - 1674. 48 Hansmann, H. B. (1980). The role of nonprofit enterprise. The Yale law journal , 89 (5), 835 - 901. Hansmann, H. (1987). The effect of tax exemption and other factors on the market share of nonprofit versus for - profit firms. National Tax Journal , 71 - 82. Highsmith, A. R. (2015). Demolition means progress: Flint, Michigan, and the fate of the American metropolis . Univ ersity of Chicago Press. Hopkins, B. R. (198 7). Partnerships or Joint Ventures as Vehicles To Achieve Charitable Objectives. Cath. Law. , 31 , 123. Joassart - Marcelli, P., & Wolch, J. R. (2003). The intrametropolitan geography of poverty and the nonprofit s ector in Southern California. Nonprofit and v oluntary sector quarterly , 32 (1), 70 - 96. Jones, D. (2019, April 2). National Taxonomy of Exempt Entities (NTEE) Codes. Retrieved from https://nccs.urban.org/projec t/national - taxonomy - exempt - entities - ntee - codes Kim, M. (2015). Socioeconomic diversity, political engagement, and the density of nonprofit organizations in US counties. The American Review of Public Administrat ion , 45 (4), 402 - 416. Kissane, R. J. (2003). What's need got to do with it - barriers to use of nonprofit social services. J. Soc. & Soc. Welfare , 30 , 127. Kneebone, E., & Berube, A. (2013). Confronting suburban poverty in America . Brookings Institution Pre ss. Lecy , J. D., & Van Slyke, D. M. (2013). Nonprofit sector growth and density: Testing theories of government support. Journal of Public Administration Research and Theory , 23 (1), 189 - 214. al ecology. Ecography , 23 (1), 101 - 113. Journal of Regional Science , 48 (5), 941 - 967. LeSage, J. P ., & Pace, R. K. (2010). Spatial econometric models. In Handboo k of applied spatial analysis (pp. 355 - 376). Springer, Berlin, Heidelberg. Lipsky, M., & Smith, S. R. (1993). Nonprofits for hire: The welfare state in the age of contracting . Harvard Universi ty Press. 49 Lu, J. (2017). Does population heterogeneity really matter to nonprofit sector size? Revisiting VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations , 1 - 27. Lu, J., & Xu, C. (2018). Complementary or supplementary? The relationship between gover nment size and nonprofit sector size. VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations , 29 (3), 454 - 469. Luksetich, W. (2008). Government funding and nonprofit organizat ions. Nonprofit and Voluntary Sector Quarterly , 37 (3), 434 - 442. Marmot, M., & Wilkinson, R. (Eds.). (2005). Social determinants of health . OUP Oxford. Albuquerque, F . S., ... & Isabel Bellocq, M. (2009). Coefficient shifts in geographical Ecograp hy , 32 (2), 193 - 204. McDougle , L. (2014). Understanding public awareness of nonprofit organizations: exploring the awareness confidence relationship. International Journal of Nonprofit and Voluntary Sector Marketing , 19 (3), 187 - 199. McDougle, L. M. (2015) . The accuracy of the core fi les for studying nonprofit location: How many nonprofits are there?. Nonprofit and Voluntary Sector Quarterly , 44 (3), 609 - 624. Millo, G., & Piras, G. (2012). splm: Spatial panel data models in R. Journal of Statistical Softwar e , 47 (1), 1 - 38. Never, B., & Westberg, D. (2016, September). Place Matters: The Spatial Effects of Human Service Expenditures. In Nonprofit Policy Forum (Vol. 7, No. 3, pp. 369 - 388). De Gruyter. Openshaw, S. (1984). Ecological fallacies and the analysis of areal census d ata. Environment and planning A , 16 (1), 17 - 31. Peck, L. R. (2008). Do antipoverty nonprofits locate where people need them? Evidence from a spatial analysis of Phoenix. Nonprofit and Voluntary Sector Quarterly , 37 (1), 138 - 151. Polson, E. C. (2017). Relig ious environments and the distribution of anti - poverty nonprofit organizations. Nonprofit and Voluntary Sector Quarterly , 46 (1), 156 - 174. Puga, D. (2010). The magnitude and causes of agglomeration economies. Journal of regional science , 50 (1), 203 - 219. Pu lido, L. (2016). Flint, environmental racism, and racial capitalism. 50 Reckhow, S., Downey, D., & Sapotichne, J. (2018). Governing Without Government: Nonprofit Governance in Detroit and Flint. Urban Affairs Review , 1078087419847531. Reine rt, K. A. (2011). No small hope: The basic goods imperative. Review of Social Economy , 69 (1), 55 - 76. Reinert, K. A. (2015). Food security as basic goods provision. World Medical & Health Policy , 7 (3), 171 - 186. Romer, T., & Rosenthal, H. (1979). The elusi ve median voter. Journal of Public Economics , 12 (2), 143 - 170. Sadler, R. C., & Highsmith, A. R. (2016). Rethinking Tiebout : the contribution of political fragmentation and racial/economic segregation to the Flint water crisis. Environmental Justice , 9 (5), 143 - 151. Salamon, L. M. (1987). Of market failure, voluntary failure, and third - party government: Toward a theory of gove rnment - nonprofit relations in the modern welfare state. Journal of voluntary action research , 16 (1 - 2), 29 - 49. Salamon, L. M., & da C osta Nunez, R. (1995). Partners in public service: Government - nonprofit relations in the modern welfare state (Vol. 19). Ba ltimore, MD: Johns Hopkins University Press. Salamon, L. M., & Anheier, H. K. (1998). Social origins of civil society: Explaining th e nonprofit sector cross - nationally. Voluntas: International journal of voluntary and nonprofit organizations , 9 (3), 213 - 24 8. New York: The Foundation Center . Salamon, L. M. (2015). The resilient sector revisited: The new challenge to nonprofit America . Brookings Institution Press. Schneider, J. A. (2009). Organizational so cial capital and nonprofits. Nonprofit and Voluntary Sector Quarterly , 38 (4), 643 - 662. Snavely, K., & Tracy, M. B. (2002). Developme nt of trust in rural nonprofit collaborations. Nonprofit and Voluntary Sector Quarterly , 31 (1), 62 - 83. Tiefelsdorf, M., Ch un, Y., & Al Aswad, L. (2020) TexMix. R package version 0.5.3. Trudeau, D. (2008). Towards a relational view of the shadow state. Po litical geography , 27 (6), 669 - 690. 50 Voss, P. R., Long, D. D., Hammer, R. B., & Friedman, S. (2006). County child poverty rates in the US: a spatial regression approach. Population Research and Policy Review , 25 (4), 369 - 391. Weisbrod, B. A. (1977). The v oluntary nonprofit sector: An economic analysis. Weisbrod, B. A. (1986). Toward a theory of the voluntary nonprofit sector in a three - sector economy. In The economics of nonprofit institutions . Oxford University Press. Weisbrod, B. A. (2009). The nonprof it economy . Harvard University Press. Wolpert, J. (1993). Decentralization and equity in public and nonprofit sectors. Nonprofit and Voluntary Sector Quarterly , 22 (4), 281 - 296. Yan, J., Guo, C., & Paarlberg, L. E. (2014). Are nonprofit antipoverty organi zations located where they are needed? A spatial analysis of the greater Hartford region. The American Statistician , 68 (4), 243 - 252.