Three Models of US State-Level Charity Regulation

1 State-Level Charity Regulation

Interdisciplinary social science literatures in public administration, international studies, and nonprofit management have addressed issues related to the regulation of charitable nonprofits and NGOs, although there are relatively few studies that have specifically addressed state-level charity regulation in the United States outside of the legal literature.

Given the many natural affinities between public administration and nonprofit management (Mitchell and Schmitz 2019; Smith 2010), the ongoing phenomenon of sectoral blurring (Bromley and Meyer 2015), and attention to the “big questions” of nonprofit management (Lohmann 2007) in public administration, it is not surprising that a considerable amount of scholarship concerning the relationships between government and nonprofits and NGOs can be found within the public administration field. Prior such research has examined topics including government-nonprofit contracting (Smith and Lipsky 2009), the effects of government funding of nonprofits (Brooks 2000; Guo 2007; Rushton and Brooks 2007), cross-sectoral collaboration (Boyer, Kolpakov, and Schmitz 2018; Bryson, Crosby, and Stone 2006; Curley et al. 2021; Marwell and Calabrese 2015; Mitchell 2014a; Mitchell, O’Leary, and Gerard 2015; Selden, Sowa, and Sandfort 2006), and accountability (Grizzle and Sloan 2016; Leazes 1997; LeRoux 2009; Robinson and Billingsley 2016). Public administration research specifically relating to federal regulation has addressed topics such as tax policy (Brooks 2004), lobbying restrictions (Berry 2005), and financial disclosure (Keating and Peter 2003), usually referencing the role of the Internal Revenue Service (IRS) and the significance of the IRS Form 990 as a disclosure-based, de facto regulatory tool. Other scholarship in this field has considered the regulation of charitable solicitation by US municipalities (Mead 2019), national charity regulatory regimes across countries (Cordery and Deguchi 2018), and the global phenomenon of voluntary self-regulation through so-called accountability clubs (Tremblay-Boire, Prakash, and Gugerty 2016).

The international studies literature has also addressed the theme of voluntary regulation (Crack 2018), particularly given the absence of a global legal regime governing NGOs. Interest in regulatory patterns across countries has led to research attempting to understand the ebb and flow of so-called regulatory waves over time (Breen, Dunn, and Sidel 2017, 2019) with particular attention paid to the possibility of shifting, closing, or shrinking civic spaces resulting from increasing regulation (Bloodgood and Tremblay-Boire 2011; DeMattee 2019). Other research has explored how domestic charity regulations and norms can affect the transnational behavior of NGOs (Mitchell and Stroup 2020; Mitchell, Schmitz, and Vijfeijken 2020; Schmitz and Mitchell 2022).

The nonprofit management literature has also examined national patterns of regulation across countries (Bloodgood, Tremblay-Boire, and Prakash 2014) and the phenomenon of voluntary self-regulation (Gugerty, Sidel, and Bies 2010) but has tended to focus more on regulatory issues specific to the United States. For example, prior research has identified areas for federal regulatory reform (Anheier and Toepler 2019; Barber and Smith 2020), explored topics related to the Sarbanes-Oxley Act (Benzing et al. 2011; Nezhina and Brudney 2010), and examined issues concerning fundraising regulation (Barber 2012; Barber and Farwell 2016; Barber, Farwell, and Galle 2022; Breen 2012; Dietz et al. 2017). Much of this work acknowledges the role of information disclosure as a means of regulating fundraising activity in the absence of formal regulations. The literature on nonprofit accountability particularly emphasizes the significance of public disclosure in the US context (Chisolm 1995; Lee 2004). So-called charity watchdogs or information intermediaries use this information to assess charities and provide guidance to donors who, in turn, can play an informal regulatory role by predicating their giving decisions on the charity assessments (Mitchell 2014b; Mitchell and Calabrese 2019; Szper and Prakash 2011). This literature has also addressed a variety of regulatory issues related to taxation (Breen and Cordery 2022; Grønbjerg and McGiverin-Bohan 2016; Hager and McArdle 2015; Kearns 2015), lobbying (Prentice 2018), political activities (Carroll, Myser, and An 2022), and the possibility of overregulation (Irvin 2005).

There have been a few important efforts to provide overviews of charity regulation, internationally and domestically. McGregor-Lowndes and Wyatt and others (2017) provide insider accounts of charity regulation across several countries, while Fremont-Smith (2004) provides what is probably the most comprehensive overview of charity regulation in the United States, including state-level regulation. Many other studies have analyzed charity regulation in non-US countries, including Australia (Murray 2018), Ireland (Breen and Smith 2019), and the United Kingdom (Morgan 2015), among others. Overall, beyond the relevant legal literature, there is relatively little recent social science scholarship focused on US state-level charity regulation in the major nonprofit studies journals. The most notable exception is Lott et al.’s (2022) research on regulatory breadth at the state level.

In general, state-level charity regulation in the United States has received only modest attention from social scientists and nonprofit management scholars, despite a considerable body of research making extensive use of IRS Form 990 data in the context of the federal regulatory environment. This is problematic because state-level charity regulation may be a significant factor influencing charitable activities and organizations in the United States. This limited attention is likely related to the complexity and variety of state charity regulations across states, which can be especially daunting for researchers without substantial legal expertise. The derivation of a simple typology that addresses this problem may facilitate the incorporation of state-level charity regulation information into mainstream social science and nonprofit management scholarship. The remainder of this article identifies, interprets, and discusses models of state-level charity regulation.

2 Data and Method

Data are obtained from the Legal Compendium, a dataset originally produced in 2016 containing information on state-level charity regulation compiled by researchers associated with the Regulation of Nonprofits and Philanthropy Project at the Urban Institute and the Charities Regulation and Oversight Project at Columbia Law School (Lott et al. 2016). An updated 2019 version of the Legal Compendium added data on fundraising contributed by Deitz et al. (2017).

The Regulatory Breadth Index (RBI) provides a quantitative measure of state-level charity regulation based on 34 indicator variables selected from the 2019 Legal Compendium. These indicators, which were developed by a team of legal scholars in consultation with state charity regulators to reflect the most important areas of state-level charity regulation, are organized into five main categories. Five corresponding subindexes measure (1) registration and exemptions from registration; (2) fundraising regulations; (3) requirements for notice, filing, or review of transactions; (4) audits, and; (5) requirements for notice or action for dissolution.^[1] Lott et al. (2022) provide an in-depth qualitative and quantitative description of the RBI research, detailing the five subindexes and the indicator variables upon which each is derived. In most cases, the underlying indicator variables measure the presence or absence of specific regulations. For example, the subindex measuring requirements for notice or action for dissolution is equal to the sum of six dichotomous indicators measuring whether a state’s attorney general’s office and/or another state office must review voluntary, judicial, or voluntary dissolutions.^[2] Each subindex is scaled to range from zero to one, with zero indicating the lowest possible amount of regulation and one indicating the highest possible amount of regulation. The RBI is the average of the five subindexes, multiplied by 100 for convivence of display. Table 1 provides summary statistics for the RBI and the five constitutive subindexes.

The RBI provides a scalar measure of the breadth of state-level charity regulation across all 50 states and the District of Columbia. However, the RBI does not take account of the distinct regulatory patterns that exist across states. Latent class analysis (LCA) is performed to assess this possibility. LCA is a model-based clustering technique (Ahlquist and Breunig 2012; Magidson and Vermunt 2003; Magidson, Vermunt, and Madura 2020; McCutcheon 1987; Mitchell 2014c; Mitchell and Schmitz 2021; Vermunt and Jay 2002) that is comparable to (non-model-based) traditional k-means and hierarchical clustering approaches (Magidson and Vermunt 2002; Mitchell 2010). The basic structure of a latent class model expresses the probability of obtaining a specific indicator pattern as a weighted average of the cluster specific probabilities (Vermunt and Jay 2004). Unlike factor analysis, which is typically used to reduce a large number of variables into a smaller number of factors, the typical goal of cluster analysis is to classify cases into clusters such that cases within each cluster are very similar to each other and cases in different clusters are very different from each other. In this respect, LCA could be regarded as a case-oriented method, whereas factor analysis could be regarded as a variable oriented method. As a model-based technique, LCA posits the existence of a categorical latent variable conditional upon which states adopt a specific pattern of charity regulation.

Latent class model estimation proceeds by incrementing the number of posited latent classes until a model fit criterion is satisfied, typically indicated by the Bayesian Information Criterion (BIC). The model with the lowest value indicates the preferred model. A well-performing latent class model also exhibits a condition known as local independence. This is achieved when the residual statistical association among the indicator variables conditional upon the latent variable is insignificant. This condition can be assessed by examining bivariate residuals following model estimation. The local independence assumption can be relaxed by incorporating direct effects between highly correlated indicators (Vermunt and Jay 2016). After a preferred model is selected, a taxonomy is created by assigning cases (states) to the cluster to which they most likely belong based on their posterior membership probabilities. This procedure is called modal assignment.

LCA generates a descriptive profile for each cluster. The profile describes both the estimated size of each cluster and the estimated means of each indicator variable conditional on latent class membership. These profiles are interpreted to derive a qualitative understanding of the latent variable and to label the clusters (Mitchell and Schmitz 2021). The profile information is estimated prior to modal assignment. After classification, an assignment solution may contain classification error that can be statistically accounted for in subsequent analyses if necessary (Bakk, Oberski, and Vermunt 2014; Bakk, Tekle, and Vermunt 2013; Bolck, Croon, and Hagenaars 2004; Vermunt 2010).

The final step in LCA usually involves further use of the categorical latent variable in subsequent analyses. Typically, this is done to facilitate description and interpretation of the latent variable and to assess whether the latent variable is statistically associated with other variables of interest. This is sometimes referred to as step-3 analysis, the three steps being estimate, classify, and analyze.

3 Latent Class Analysis: Estimation and Classification

The indicators employed include the five subindexes that constitute the RBI. These subindexes measure the extent of regulatory requirements for transactions, dissolutions, registration, audits, and fundraising, as summarized in Table 1.

Table 2 displays model fit statistics for latent class models with 1–5 clusters. The 3-class model is preferred because it has the lowest Bayesian Information Criterion (BIC). However, the bivariate residuals table indicates a local dependency between the indicators for transactions and dissolutions, probably because these two indicators both concern the status of charitable assets. A 3-class model (Model 6) accounting for this local dependency with a direct effect between the indicators results in a lower BIC but does not represent a statistically significant improvement in model fit according to a bootstrap log-likelihood difference test. The profiles and assignment solutions for the two 3-class models are identical; however, Model6 is retained as the preferred model because it resolves the local dependency. This model has virtually no classification error (0.00), indicating that each state is correctly assigned to the appropriate cluster.^[3] Accounting for classification error in step-3 analysis is therefore not necessary.

Table 3 displays the profiles for the three latent classes. As shown in the first row of the table, about 54 percent (28) of states fall into Cluster 1, 31 percent (16) fall into Cluster 2, and 14 percent (7) fall into Cluster 3. The remaining values in the table describe the predicted means for each indicator conditional on the respective latent class. Interpreting these means involves identifying substantively large differences both across the clusters and relative to the unconditional (population) means. Substantial variation across a row indicates that the indicator has interpretive importance as large differences in values differentiate the clusters from each other. Comparing the conditional means to the respective unconditional means indicates how the states in a particular cluster compare to the overall population. Substantively large differences similarly suggest that the indicator has interpretive importance. The bolded numbers in Table 3 represent values greater than the unconditional means. Tables 4, 5 and 6 display the states assigned to Clusters 1, 2 and 3, respectively, and provide each state’s values for the regulatory breadth index as well as for the five subindexes.

The profile for Cluster 1 indicates above average scores for all five subindexes. The states assigned to Cluster 1 are shown in Table 4.

Cluster 2 states score highly only for the registration subindex, while scoring a zero for the audits subindex. The states assigned to this cluster are shown in Table 5.

Cluster 3 states score highly for transactions and dissolutions, while scoring a zero for both audits and registration requirements. The states in this cluster are shown in Table 6.

4 Latent Class Analysis: Profile Interpretation

Profile interpretation is an exploratory process. To facilitate this step, Figure 1 displays a map of the United States shaded by cluster (with light grey indicating Cluster 1, medium grey indicating Cluster 2, and dark grey indicating Cluster 3). Additionally, Table 7 displays descriptive means for various state-level covariates by latent class. The rationales for selecting these covariates are as follows. First, it is probable that states with more legal restrictions overall might also impose more regulations on charities, reflecting a consistency in regulatory propensity. Second, states with higher tax rates, tax revenues, GDP, and populations should, in principle, have more resources available to support more comprehensive charity regulation models, such as the model reflected in Cluster 1. Third, political liberalism is typically associated with more regulation and thus may also be associated with more extensive charity regulation. Fourth, the number, size, and density of nonprofit originations in a state might be associated with the charity regulation model because greater number, size, and density may provide more opportunities for activities to take place in the nonprofit sector that may lead to more or different regulation. Fifth, the regulatory breadth index itself may be correlated with the regulatory model because both are constructed from the same underlying data. Finally, states with bifurcated jurisdictions, in which more than one authority has responsibility for charity regulation, might regulate charities differently compared to states without bifurcated jurisdictions because differently situated regulators may have different interests, mandates, incentives, or resources for regulating charities. Additionally, requirements to report to more than one authority in bifurcated jurisdictions might mechanically result in more overall requirements. The following three subsections describe each latent class with reference to these covariates.^[4]

Figure 1:

Cluster map.

5 Latent Class Analysis: Step-3

The derivation of the categorical latent variable enables new research seeking to explain the emergence of state level charity regulation models using quantitative methods. Prior comparative research in the so-called social origins tradition has proposed that economic, social, and political factors influence the development of a country’s charitable sector (Salamon and Anheier 1998; Salamon, Sokolowski, and Haddock 2017). The preceding descriptive analysis similarly suggests that economic, social, and political characteristics may help to explain features of states’ charitable sectors within the United States. To explore this further, Table 8 displays the results of a multinomial logistic regression analysis predicting the regulatory model (cluster) based on a concise set of economic, social, and political variables. The economic importance of the nonprofit sector in a state is indicated by the ratio of nonprofit gross receipts to GDP. The social significance of the nonprofit sector in a state is indicated by density, or the count of nonprofits divided by the population. The political character of a state is indicated by state ideology. The results shown in Table 8 show effects relative to Cluster 1 in units of multinomial log odds.

To facilitate a more intuitive interpretation of the multinomial logistic regression results, Table 9 displays the average marginal effects of the indicators on the probabilities of cluster membership. States in which the charitable sector is more economically important, less socially important, and that exhibit greater political liberalism have an increased probability of broad regulation. States in which the charitable sector is less economically important and that are more politically conservative have an increased probability of adopting a limited regulation model. Finally, states with a more socially important charitable sector are more likely to have the asset oversight regulatory model.

6 Discussion

Why do states adopt a particular charity regulation model? Analysis suggests that regulatory model adoption may be influenced by the economic, social, and political conditions prevailing in each state. However, larger historical forces, coupled with state-level idiosyncrasies, are likely to play a large role in any explanation. The roots of contemporary US charity law have been traced through English law to at least Roman antiquity (Fremont-Smith 2004). Within the United States, the process of westward expansion resulted in different cultural traditions and legal frameworks taking hold in different places at different times. Complex historical and cultural factors likely affect contemporary patterns of state-level charity regulation in ways that may be difficult or impossible to capture using statistical social science methods alone.

For example, in one prior study of state-level charity regulation, Irvin (2005) singled out six of the seven states from Cluster 3 for the absence of certain registration and reporting requirements, in close alignment with the LCA result based on the RBI data.^[5] According to conversations with deputy and assistant attorneys general from those states, this lower level of regulation did not appear to be associated with unusually high levels of charity fraud as might be expected. Irvin (2005, 171, 174) pointed out that both high and low regulation states tend to rely on “the citizen complaint method … to identify potentially fraudulent activity” and proposed that:

the lack of state regulatory action could be due to the small states’ ability to spot the few cases of nonprofit misdeeds when they occur or because fraudulent nonprofits find low-population states to be inefficient places to carry out fraudulent activity.

This also allows the possibility that greater overall levels of political conservatism in these states may mean that residents are simply less likely to complain to state regulators when fraud is suspected. Even if fraud rates are broadly similar across high and low regulation states, a lower propensity to seek government intervention could help explain lower perceived rates of fraud and therefore potentially explain a lesser need for extensive charity regulation. When suspicions about fraud do arise, it may be that concerned stakeholders prefer to address matters through other means.

These complexities notwithstanding, it is informative that economic, social, and political factors relate to a state’s model of charity regulation. Economic importance is associated with a greater likelihood of broad regulation and a lesser likelihood of limited regulation, suggesting that charity regulation is likely to be more comprehensive when the contribution of the charitable sector to a state’s economy is greater. By contrast, states with a more socially important charitable sector are more likely to adopt the narrowest regulatory model focused on asset oversight and are less likely to adopt a model of broad regulation. It may be that the types of charities that predominate in the more sparsely populated Cluster 3 states are relatively smaller and more numerous than those found elsewhere, perhaps reflecting different cultural and geographic characteristics that favor larger numbers of geographically more dispersed organizations (over the greater consolidation that might occur in more densely populated, urbanized areas). Finally, regulatory model adoption is correlated with political ideology, with more liberal states favoring broad regulation and more conservative states favoring limited regulation or asset oversight. This result agrees with the expectation that liberal politics are likely to result in more, not less regulation, and this tendency appears to extend to the charitable sector.

One further possibility is that state-level charity regulation depends not only upon the overall size but also the composition of each state’s nonprofit sector. Table 10 reports statistics describing states’ charitable subsectors by cluster.^[6] The relative sizes of five charitable subsectors as a proportion of the total does not vary significantly by cluster, but a few differences emerge when charitable subsectors are measured in terms of assets. Educational charities account for a larger share of total assets in Cluster 1 (broad regulation), while human services charities account for larger shares in Clusters 2 (limited regulation) and 3 (asset oversight). This suggests that qualitative factors such as the types of charities that predominate in a state may affect the nature of the regulatory models that emerge over time.

Prior research has asserted that the need to increase public trust and confidence in charitable organizations is an important consideration for charity regulators (Cordery and Deguchi 2018; Yang and Northcott 2021). Additionally, so-called contract failure theory—the canonical theory of the nonprofit institutional form—posits that nonprofit organizations are specifically designed to be trustworthy and that this trustworthiness is achieved essentially through regulation (in the form of the prohibition on private inurement referred to as the non-distribution constraint) (Hansmann 1980). One potentially fruitful area for future research would be to rigorously examine whether a state’s charity regulation model is associated with public trust in charitable organizations. To explore this possibility preliminarily, Table 11 summarizes the level of public trust in charitable organizations by cluster for various time periods. Generally, public trust appears to be slightly lower in the cluster of states adopting the asset oversight model that offers the least comprehensive regulation. However, these results should be interpreted with caution due to sparse data for less populated states. A longer panel with annul trust data stratified by state and with regulatory breadth data for each year may be needed to support more conclusive findings in future research.

The emergence of three qualitatively distinct charity regulation models in the United States does not imply that these models were deliberately selected by policymakers or were derived from any explicit political preferences or theories of governance. Rather, it is more likely that these models are the result of myriad idiosyncrasies and path dependencies, including differences in “state nonprofit culture” (Pettijohn and Boris 2018, 1), regulatory responses to specific cases of fraud or abuse, the ongoing need to monitor tax exemptions, the work of particularly entrepreneurial lawmakers, and the uneven influence of model acts, among many other potential factors.^[7] More research, particularly research of a legal-historical nature, is needed to better understand the qualitative processes and characteristics that influence state-level charity regulation.

Although each state may have traveled a unique path to its current regulatory model, analysis shows that state-level charity regulation can be efficiently characterized with respect to three general models: broad regulation characterized by a relatively comprehensive array of requirements; limited regulation emphasizing an interest in primarily knowing what organizations exist in a state; and asset oversight more narrowly focused on actions that may affect the purpose and status of charitable assets. This typology helps to make sense out of an otherwise unwieldy number of state-level regulatory regimes that have heretofore defied inductive, data-driven classification.

This analysis has several important limitations. First, the cross-sectional, observational nature of the data precludes rigorous causal analysis, and it is likely that numerous features of states related to charity regulation coevolved together and exerted mutual influence. Second, as data about enforcement are not available, it is possible that the de jure regulatory breadth data do not accurately describe the de facto level of regulation and enforcement in each state, particularly after taking reporting propensities and regulator discretion into account. Third, the underlying research does not account for common law or additional statutes that may exist at the local level, such as in municipalities. Perhaps most importantly, statistical analysis alone cannot possibly incorporate the myriad relevant qualitative-historical factors that inevitably influence the adoption and non-adoption of state-level charity regulations.

7 Conclusions

Amid a perplexing array of charity regulation regimes across all 50 US states and the District of Columbia, this article identifies three general models of state-level charity regulation in the United States: broad regulation, limited regulation, and asset oversight. Although a detailed tracing of the historical emergence of these regulatory models over time is beyond the scope of this research, analysis suggests that the economic, political, and social characteristics of states may help to explain patterns of charity regulation in the United States.

Future research could examine regulatory model emergence quantitatively using longitudinal data. It could also evaluate additional covariates that may have important connections to charity regulation, such as variables related to religious organizations and land ownership, or explore other interesting regulatory patterns, such as the absence of bifurcated jurisdictions in Cluster 3. Attempts could also be made to estimate enforcement rates, perhaps with creative proxy indicators or potentially based on representative surveys, given the low likelihood of obtaining enforcement data directly from state regulators. Additionally, the underlying RBI data could be expanded to include additional statutory variables and possibly also information concerning common law to produce a richer dataset for analysis. Moreover, in-depth qualitative research based on historical data could provide much more detail about the development of charity law over time, especially at the intersections of important economic, political, and social developments within an across states. Finally, research in international studies and related fields could identify and compare regulatory models across countries to derive further insights about the relationships between regulatory models and key outcome variables such as indexes measuring civil society and democracy, levels of public trust in NGOs, and levels of foreign funding of NGOs. Such work could potentially contribute to existing efforts to study and typologize civil society regulatory regimes internationally (DeMattee 2018). Relatedly, research in the social origins tradition (e.g. Salamon and Anheier 1998; Salamon, Sokolowski, and Haddock 2017) could seek to examine the relationship between charity regulation models and patterns of civil society development internationally.

This research has useful implications for researchers, regulators, practitioners, information intermediaries, and donors. Researchers in nonprofit studies and related fields can employ the typology to gain new insights about the role of state-level charity regulation in their studies. The categorical variable generated by the analysis can be used to contextualize the regulatory environments particular to a given state, to assess the significance of state-level regulatory models in statistical research, or simply to control for differing regulatory models across states in studies of other phenomena. For example, a state’s regulatory model might predict rates of organizational entry and exit, mergers and acquisitors, malfeasance, levels of public trust in nonprofit organizations, and other important outcomes of interest. Regulators may find it useful to have a better understanding of their peer group to facilitate convening and benchmarking. Additionally, proponents of the so-called single portal initiative for fulfilling state registration and reporting requirements could consider developing a triple portal, either instead of, or as an intermediate step toward, a single portal. State charity regulators may have an easier time achieving alignment within each of the three clusters than across all states simultaneously. Finally, practitioners and donors may benefit from a simpler and more convenient understanding of the general regulatory environments that prevail in different states. For example, depending on the regulatory model, charity practitioners may choose to incorporate or operate (or not) in certain states, while donors and information intermediaries may wish to incorporate the regulatory context into their assessments of charities. In short, the novel typology can improve our understanding of the charitable sector and facilitate efforts to pursue a variety of desirable outcomes for the sector and for society.