Taxes and agglomeration economies: how are they related to nonprofit firm location?
Nonprofits are an ever increasing component of the U.S. economy. From 1997-2001, employment growth in the nonprofit sector averaged 2.5%, outpacing both the business (1.8%) and the government sectors (1.6%) (Moore 2004). Much of this growth stems from new nonprofits, yet few studies have investigated possible determinants of nonprofit entry and location decisions. Most of these studies focus on demand-side factors (Wolch and Geiger 1983; Downes and Greenstein 1996; Bielefeld, Murdoch, and Waddell 1997). For example, Downes and Greenstein (1996) focus on demographic and religious characteristics of the surrounding community and its relation to private school location. In this paper, I use firm-level data and focus on potentially important supply side variables: tax rates and agglomeration economies.
Most nonprofits are exempt from corporate, sales, and property taxes. Hansmann (1987) and Gulley and Santerre (1993) find evidence that higher corporate, property, and sales taxes are positively related to nonprofit market share. However, market share measures obfuscate the dynamics of the industry in terms of entry and exit rates. If market share for nonprofits increases in response to higher tax rates, this could be because new nonprofits are entering or because for-profits are leaving the industry. Identifying these two cases is impossible with market share data. Using nonprofit tax return data, I examine the relation between these tax exemptions and state/county location decisions of
new nonprofits. To my knowledge, this is the first paper to study these interactions at the individual firm level.
Nonprofits also receive tax-favored status due to the individual tax deduction on charitable contributions (Gentry and Penrod 1998). Employees of and donors to the nonprofits will be affected by the individual tax rates. This in turn alters the costs to the nonprofit of location in a high or low tax area. Unlike previous studies, I therefore investigate the relation between individual marginal tax rates and nonprofit location.
Another potential supply-side determinant of firm location decisions is the existing firm concentration in the area. The potential economies of scale generated by proximity to other firms have been extensively examined in for-profit industries. (1) Previous studies have generally found a positive relation between agglomeration and location (Carlton 1983; Bartik 1985; Woodward 1992; Gius and Frese 2002; Rosenthal and Strange 2003). To my knowledge, Bielefeld et al. (2004) is the only study that examines nonprofit agglomeration, finding evidence of agglomeration in Dallas, Texas. This warrants additional examination because nonprofit output is generally service-driven and less mobile. Thus, agglomeration implies greater direct competition, making market saturation a larger issue in the nonprofit market. Using a nationwide dataset, I examine how location and existing firm concentration are related and explore the potential presence of market saturation. I also investigate possible differences between inter- and intra-industry agglomeration.
My results provide new insight into nonprofit firm behavior. In most cases, higher tax rates are associated with new firm location, but the results for corporate and property taxes vary across the sample years. I find that individual tax rates are an important consideration for nonprofits and are positively correlated to location. My analysis also contributes to the literature by examining how the sensitivity to tax rates varies according to the sources of revenues for the institution. Firms that depend highly on individual donations are more likely to locate in high individual tax states while nonprofits with large mission-related revenues are more sensitive to the property tax rate. I find some evidence that nonprofit agglomeration outside an entrant's industry is negatively associated with new firm location. Moreover, the total elasticity for agglomeration within a nonprofit's industry is positive, robust, and appears to be more strongly associated with new firm location than tax rates.
The next section discusses the theory, specific hypotheses, and empirical model. Section 3 describes the data. Section 4 presents the results and discusses possible implications of the findings. The final section concludes.
2. Theory and Empirical Specification
Each new firm i chooses to locate in a state/county j from all possible choices in the set J. For clarity, I frame the initial discussion in terms of state choice. I then discuss alterations to the model for county-level location decisions. Each state has (i) an individual income tax rate, [I.sub.j]; (ii) corporate, sales, and property tax rates, [C.sub.j]; and (iii) agglomeration of existing nonprofits in the state, [A.sub.j]. Other characteristics of the state, particularly demand-side factors, that are related to the location decision are also included ([X.sub.j]). Firm i derives utility [U.sub.ij] from locating in a state j and chooses the state that maximizes its utility such that:
[U.sub.ij]([C.sub.ij], [I.sub.ij], [A.sub.ij], [X.sub.ij]) > [U.sub.ik]([C.sub.ik], [I.sub.ik], [A.sub.ik], [X.sub.ik]) [for all]j, k [member of] J and j [not equal to] k, (1)
where utility is a function of the tax rates, agglomeration, and other characteristics of each state.
Although most nonprofits are tax-exempt, previous theory suggests that tax rates still influence their behavior. In industries where nonprofits and for-profits compete, the marginal tax rate paid by the for-profit creates a higher cost of capital relative to the tax-exempt nonprofit firm, suggesting that nonprofits should locate in states with higher corporate tax rates (Rose-Ackerman 1986; Steinberg 1991). However, changes in tax rates are also directly related to other attributes potentially important in the location decision. For example, increased government provisions due to an increase in the tax rate could trigger a decline in entry by nonprofits, particularly if the provisions are substitutes for nonprofit services. In this scenario, the relation between location and [C.sub.ij] would instead be negative.
Individual taxes are a common regressor in the for-profit literature on location decisions (Bartik 1985; Coughlin, Terza, and Arromdee 1991; Woodward 1992; Devereux and Griffith 1998; and Gius and Frese 2002). Although the nonprofit firms are tax exempt, the workers are not. So, due to the tax burden on their employees, nonprofits may be motivated to locate in states with lower individual income tax rates. A negative relation between individual taxes and location could also occur if higher tax rates decrease individual migration and thus, population growth. Lower population growth would not only affect the size of the potential employee pool but also may decrease the demand for nonprofit services.
Alternatively, individual tax rates and nonprofit location could be positively related since individual taxes affect the price of the tax deduction on charitable contributions. Many studies have found a negative price elasticity on donations, indicating that higher individual tax rates increase charitable giving. (2) If the nonprofit believes that higher tax rates positively impact their donations, then they will be more likely to locate in higher tax states. (3)
Thus, it is not clear a priori whether higher individual tax rates increase or decrease the probability of location by a nonprofit firm. As I discuss in the data section, I attempt to remove the indirect effects that taxes have on the location decision through my controls. However, there may be other omitted factors correlated with the tax rates that could bias the estimates. (4) Therefore, I will not be able to identify the direct cause of my observed relations.
A firm's response to changes in tax rates will also vary by industry due to variation in the extent of for-profit competition and the degree of financial dependence on donations. Nonprofits that do not compete with for-profit institutions should not be directly influenced by the corporate tax rate; an increased cost of capital to a for-profit competitor is a moot point. Nonprofits with more mission-related revenues are more likely to have for-profit competitors since for-profits in such an industry could also receive service-related revenues. Similarly, nonprofits with donations as the primary source of revenue should be more influenced by the individual tax effect on donations. Thus, I include measures of the percentage of revenues obtained through mission-related revenues and donations to examine the relation between revenue sources and nonprofit location decisions.
Labor market pooling, knowledge spillovers, and input sharing are generally the three primary theories used to explain the (generally for-profit) firm benefits of agglomeration. (5) However, agglomeration economies can equally exist for nonprofit firms. Nonprofits clearly employ workers, so they could benefit from higher quality job matches due to labor market pooling. A larger workforce can also produce internal economies of scale in production. Similarly, knowledge spillovers could take the form of learning from other nonprofits about popular fundraising events or successful methods to connect with the target population.
In my paper, I distinguish between within- and outside-industry agglomeration. The first measure is generally thought to capture localization economies in which it is the shared knowledge and inputs within an industry that spur growth. Increased diversity or improvement in employer/employee matching due to labor pooling is generally captured by the outside-industry agglomeration and are referred to as urbanization economies (Rosenthal and Strange 2003; Henderson 2003). Rosenthal and Strange (2003) find a positive relation for within-industry agglomeration but a negative relation for higher firm concentration outside the industry. Although I expect a similar relation for within-industry nonprofit agglomeration, the effect of firm concentration outside the industry is unclear. Existing nonprofits, even those outside the industry, may provide a signal about whether the region is hospitable to other nonprofits. Thus, new nonprofits may engage in herding behavior, creating a positive correlation (Banerjee 1992).
Within nonprofit industries, agglomeration economies, if they exist, could be smaller. Nonprofit production is generally more labor-intensive and thought to have relatively low fixed costs. Thus, the gains from knowledge spillovers or sharing indivisible inputs may be lower. Moreover, unlike manufacturing firms that can easily export their products to other areas, most nonprofits are engaged in service activities that directly benefit the local community (Wolpert 1993). In this case, a nonprofit might achieve external economies of scale by locating close to other firms but face increased local competition for their services precisely due to the proximity. In this case, crowding and market saturation within an area become a larger issue. (6) I, therefore, include both a linear and a squared agglomeration term. In the case where market saturation diminishes the agglomeration benefits, I would observe a concave relation. I examine the potential nonlinearity for within and outside agglomeration and hypothesize that saturation should be more of an issue for within-industry agglomeration.
Empirical Analysis for State Location
From Equation 1, it is natural to use discrete choice estimation techniques. Let
[U.sub.ij]([C.sub.ij], [I.sub.ij], [A.sub.ij], [X.sub.ij]) = [theta] + [alpha][C.sub.ij] + [beta][I.sub.ij] + [delta][A.sub.ij] + [gamma][X.sub.ij] + [for all]j, k [member of] J, (2)
where [theta], [alpha], [beta], [delta], and [gamma] are coefficients to be estimated. Then, the probability that firm i locates in statej is:
Prob([Y.sub.i] = j) = Prob([U.sub.ij] > [U.sub.ik]) (3)
= Prob([theta] + [alpha][C.sub.ij] + [beta][I.sub.ij] + [delta][A.sub.ij] + [gamma][X.sub.ij] + [[epsilon].sub.ij] > [theta] + [alpha][C.sub.ij] + [beta][I.sub.ik] + [delta][A.sub.ik] + [gamma][X.sub.ik] + [[epsilon].sub.ik]) [for all]j, k [member of] J and j [not equal to] k (4)
= Prob([[epsilon].sub.ij] - [[epsilon].sub.ik] > [alpha]([C.sub.ik] - [C.sub.ij]) + [beta]([I.sub.ik] - [I.sub.ij]) + [delta]([A.sub.ik] - [A.sub.ij])
+ [gamma]([X.sub.ik] - [X.sub.ij])) [for all]j, k [member of] J and j [not equal to] k. (5)
If each error term is distributed Weibull, then [[epsilon].sub.ij] - [[epsilon].sub.ik] is a logistic distribution, and I arrive at McFadden's conditional logit model. Let [d.sub.ij] = 1 if state j is chosen by firm i, otherwise [d.sub.ij] = 0 for all other states in the choice set. Thus, the log-likelihood function is:
LLF = [N.summation over (i = 1)] [J.summation over (j = 1)] [d.sub.ij]Prob([Y.sub.i] = j). (6)
I maximize Equation 6 with respect to [alpha], [beta], [delta], and [gamma]. A positive (negative) coefficient implies that the attribute increases (decreases) the likelihood of a nonprofit locating in that state. Note that in a conditional logit model, any variable that does not vary by choice cannot be identified.
The independence of irrelevant alternatives (IIA) property arising from a conditional logit model is well documented in the econometric literature (Hausman and Wise 1978; Hausman and McFadden 1984). To correct for this issue, I control for unobserved heterogeneity between states through state demographic regressors. In addition, I account for general business migratory trends through the addition of regional dummies. As Bartik (1985) shows, this method is equivalent to a nested logit model where the firms first choose the region of the country and then choose the state within that region. (7)
Empirical Analysis for County Location
I also examine location decisions at the county level. Unfortunately county-level tax data for 1997 and 2002 are unavailable. (8) Although tax rates are not included, I control for taxes by including county-level tax revenues and expenditures. The exact measures and interpretative limitations are discussed in the data section.
Due to this limitation, my analysis at the county level will focus on the potential agglomeration effects. I perform this more detailed analysis since theory and empirical evidence suggest that agglomeration economies function at a local level. For most theories of agglomeration, particularly localization effects, interaction between the firms is necessary, and therefore, some degree of geographic proximity is required in order to acquire the benefits. Indeed, Rosenthal and Strange (2003) find that agglomeration attenuates rapidly as distance increases.
Now, all regressors are measured at the county level and do not vary by firm. Guimaraes, Figueiredo, and Woodward (2003) show that in this circumstance a count data model is equivalent to the conditional logit. Thus, I aggregate the choice of county for each nonprofit to obtain [N.sub.j], the number of nonprofits in county j. Let [N.sub.j] be Poisson distributed, then
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (7)
The parameter [[lambda].sub.j] is modeled to incorporate agglomeration, tax revenues, and other demographic characteristics of the county such that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].
The Poisson count data model has the unfortunate property that the E([N.sub.j]) = Var([N.sub.j]). In cases such as mine, where many of the counts are zero, the data generally are overdispersed (E([N.sub.j]) < Vat([N.sub.j])). The negative binomial specification allows for overdispersion by adding a random error component, [u.sub.j], to the probability density function. Now,
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (8)
where [u.sub.j] is distributed as a Gamma distribution. I then integrate over [u.sub.j] to get:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (9)
[r.sub.j] = [[lambda].sub.j]/[[lambda].sub.j] + [theta]
and [GAMMA](*) is the gamma distribution. Under this specification, E([N.sub.j]) = [[lambda].sub.j] and Var([N.sub.j]) = [[lambda].sub.j](1 + (1/[theta])[[lambda].sub.j]). The Poisson specification is nested within the negative binomial when 1/[theta] = 0. Thus, I perform a likelihood ratio test to test whether the Poisson regression is rejected in favor of the negative binomial.
Firm-level data on the nonprofit institutions are from the 1997 and 2002 tax return files compiled by the National Center on Charitable Statistics (NCCS) at The Urban Institute. For the state-level analysis, I use both years to investigate possible changes in behavior across time. The county location estimates only use 2002 due to data availability for the control variables. The variable RULEDATE reports the year the IRS granted 501(c)3 status to the nonprofit. For 1997 (2002), I limit the sample to nonprofits classified as operating, public charity organizations, whose RULEDATE was 1995 (2000) or later. (9) Inclusion of existing nonprofits would obfuscate studying the location decision directly because only nonprofits that survived in previous years would be observed in the data. By focusing on new nonprofits, I avoid this issue. (10) Recall that the state analysis is at the individual firm level, while the county-level analysis uses the number of new nonprofits (NUMNEWNP) as the dependent variable.
The data set also includes selected financial information as well as information on the type of charity and the state/county of residence. Total revenues for the fiscal year are subtracted from total expenses to obtain net income (NETINC). Revenues received from donations and government grants are divided by total revenues to get the percentage of revenues from public and donative sources (PERDON). (11) Revenues obtained from fulfilling the mission of the organization (program service revenues) divided by total revenues produces PERPROG. (12) The type of charity is defined by the National Taxonomy of Exempt Entities (NTEE), which is analogous to the North American Industry Classification System (NAICS) codes for for-profit firms. The NTEE is a four-digit classification system for nonprofit institutions, where the first digit of the NTEE divides charities into 26 category types ranging from arts and culture to religious organizations. (13)
Information on corporate, sales, property, and individual marginal tax rates is obtained from the 1996 and 2002 All States Tax Handbook. I calculate the corporate marginal tax rate (CORMTR) the nonprofit would pay on net income if it was not exempt from taxation. The corporate marginal tax rate, therefore, varies by state and also by firm. Sales tax (SALES) is the state portion of the tax rate. Property taxes are reported as the minimum and maximum tax rate across all counties within a state. To capture this range, AVGPROP is the average between the minimum and maximum. The individual marginal tax (INDMTR) is calculated as the marginal tax rate based on the per capita income (INCOME) for each state.
For the agglomeration effects in 1997 (2002), I measure the total number of existing nonprofits in a state/county as the number of nonprofits in the dataset whose RULEDATE was prior to 1995 (2000). (14) EXISTNTEE are those nonprofits with the same one-digit NTEE code. All other existing nonprofits are included in EXISTNP. The agglomeration variables therefore vary by industry and geographic area (state or county). Since the number of existing nonprofits is highly correlated with the population, I divide EXISTNP and EXISTNTEE by the state/ county population. Each measure is also squared to investigate whether the benefits of agglomeration are nonlinear.
Two common econometric issues with identifying agglomeration are the omission of important site characteristics and the potential endogeneity of the regressors (Hanson 2001). I address these concerns in several ways. I include information on demographics of each state/ county to control for demand side factors that simultaneously affect location decisions. The data are obtained from the State and Metropolitan Area Data Book for 1997-1998 (U.S. Bureau of the Census 1998a), the Statistical Abstract of the United States (U.S. Bureau of the Census 1997, 1998b, 2002, and 2003) and the Bureau of Economic Analysis Regional Economic Accounts (U.S. Bureau of Economic Analysis 2005). I control for population, population change, elderly population, educational attainment, poverty level, the percent of uninsured, infant mortality, and per capita income in the state (POP, POPCH, POPO65, POPEDHS, PERCBPOV, UNINS, INFMORT, INCOME). I also include measures for the level of government-provided services for welfare, retirement, and veteran benefits, Medicare, Medicaid, and unemployment insurance (INCMAIN, PAYRET, VETBEN, MEDICARE, MEDICAID, UNEMPINS). These services may be substitutes for nonprofit services. I also control for total government expenditures (GEXP) since, as discussed earlier, higher tax rates and government services are correlated. I also include a proxy for the political and cultural characteristics within the state (REPPRES), and regional fixed effects. These dummies are admittedly broad, but similar to Rosenthal and Strange (2003), they should control for some of the omitted site characteristics.
For the county-level analysis, I include analogous county-level variables for population, population change, demographics and income, derived from the 2000 County and City Data Book (U.S. Bureau of the Census 2000). In addition, I control for local government finances by including total tax revenues (TAXREV), property tax revenues (PROPREV), and total expenditures (GEXP) within the county. Unfortunately, increases in property tax revenues, even after controlling for total tax revenues and expenditures, could be due to an increase in (i) the tax/mills rate, (ii) the assessment rate, or (iii) the total amount of assessed property (i.e., new construction). Thus, I cannot examine the direct relation between property taxes and location decisions. Following Mofidi and Stone (1990), I exclude total revenues as a regressor in order to avoid near-perfect multicollinearity. Since I could not obtain information on government level social service provisions at the county level, I include state-level fixed effects to identify unobserved site characteristics.
Use of the NCCS Core Files is advantageous because they contain the universe of nonprofits filing a tax return. Unfortunately, due to the number of nonprofit tax returns and the fact that no tax revenue is collected from these tax returns, the IRS does not verify the financial information reported on all of the tax returns. In order to mitigate this problem, I delete observations reporting negative revenues or expenses and observations where the percentage of donations or program service revenue divided by total revenues (PERDON and PERPROG) are either negative or greater than 1. After removing firms with implausible or missing data, the dataset contains 16,541 new nonprofits in 1997 and 29,370 in 2002, representing the 48 contiguous states. (15) For the county-level analysis, many counties within the entire United States will not observe nonprofit firm births in a given year. I, therefore, L51 was unable to obtain government expenditure data for the District of Columbia. investigate county location decisions for the 10 largest states (16) providing a total of 10,706 counties.
Table 1 summarizes the variables used in this analysis, while Tables 2 and 3 report the descriptive statistics for 1997 and 2002, respectively. I report the tax rates across all firms as well as across states since the former incorporate the location decision of each nonprofit. For example, in 1997, the average sales tax rate by firm is 5.4% but by state is 4.8%. This implies that nonprofit entrants are locating in higher sales tax states. Of course, this conclusion does not account for correlation between sales tax rates and other state characteristics, such as population or income. Overall, the average corporate, sales, and individual marginal tax rate are around 5%. The density of nonprofits, both within and outside an industry, has increased since 1997, consistent with a growing nonprofit sector. At the county level, there is an average of 1.5 new nonprofits. In addition, the dispersion of the number of new nonprofits is quite high across counties, suggesting that the negative binomial specification may fit the data better.
Tables 4 and 5 present the conditional logit results for the state-level analysis for 1997 and 2002 separately. (17) I present several different specifications. For Tables 4 and 5, the first column presents estimates excluding (i) the interactions between tax rates and the type of revenue source (PERDON/PERPROG) and (ii) state level government spending. Column 2 includes the interaction terms. I then add state-level government spending that may be a substitute for nonprofit services and also include total state expenditures (column 3). (18) Column 4 provides estimates without the squared agglomeration terms in order to investigate whether introduction of the nonlinearities biases other coefficients. Finally, given the measurement error in property taxes, estimates excluding this regressor are reported in column 5.
Table 6 provides the county-level estimates using count data techniques. I present results, including elasticities, with and without government taxes and expenditures. The estimates for both regressions use a negative binomial specification since the likelihood ratio test rejects the Poisson specification. The chi-squared statistic for this test is provided.
Comparing coefficients across specifications for Tables 4 and 5, I note that, overall, the tax and agglomeration estimates are robust. I perform likelihood-ratio tests to assess the best specification. The chi-squared statistic comparing column 3 to all other specifications suggests that column 3 is superior. Therefore, based on these coefficient estimates, I provide average own-elasticities in Table 7 in order to compare the magnitudes of the tax and agglomeration effects. I provide both the individual elasticities as well as the total elasticities, taking into account the interaction and squared terms. (19)
I also observe variation in the tax coefficients across years. These differences highlight the complexity of identifying the direct effects of the location decision. To simplify exposition, I first present the results for taxes and agglomeration in the next two sections. In the final results section, I then discuss the implications of the findings, focusing on the most robust results.
For 1997, corporate tax rates and location decisions are negatively related in four of the five specifications but only significant when government service levels are excluded and interaction terms are included (column 2). The inverse relation in 1997 between property taxes and location is more robust. Conversely, increases in sales and individual tax rates are positively related to nonprofit location. This positive location effect for sales and individual taxes also continues to hold for 2002; although, the average elasticity for sales taxes is lower.
However, the 2002 estimates for corporate and property taxes are quite different. Now, these tax rates have a positive relation to nonprofit location, even after including government expenditures. The elasticity estimates for the first-order effect suggest that a 1% increase in corporate/property taxes increases the likelihood of new nonprofit location by .05% and .07%, respectively.
For the interaction terms in 1997 and 2002, nonprofits with higher program service revenues are more likely to locate in higher property tax states. This interaction effect counteracts the first-order effect in 1997 and magnifies it for 2002. States with larger corporate income taxes also attract relatively more nonprofits with mission-related revenues, but only for 1997. Conversely, for sales taxes, the coefficient is significant for 2002 and negative, indicating that service-driven nonprofits are less likely to locate in high sales tax states. For individual taxes, results indicate that firms who receive a larger proportion of revenues from donations and government grants are more likely to locate in high individual tax states.
Examining within-industry agglomeration, I find strong evidence of localization economies both at the state and county level. The coefficient on the first-order effect is positive and significant in Tables 4-6 across specifications. The elasticity estimates are also relatively large. A 1% increase in within-industry firm concentration increases the probability of state location by 1.94% in 1997 and 1.20% in 2002. Similarly, the county-level estimates suggest a .6% increase in the number of new nonprofits. (20) However, the squared term for within-industry concentration is negative and significant. The magnitude for the second-order effect is also relatively large for both years. This suggests that new nonprofits are initially more likely to locate in states with more existing nonprofits and then the probability of location declines. When the second-order effect is omitted in column 4 for both years, I find that the overall relation is positive. The total average elasticity is also positive.
Urbanization economies measured as agglomeration outside the nonprofit's industry seem to have the opposite relation, particularly for the 2002 state-level analysis and the county-level results. In 2002, nonprofit state location exhibits a convex relation to the number of existing nonprofits outside their industry; the first-order effect is negative while the squared term is positive. The 1997 state-level coefficient on the linear term is positive while the squared term is negative. For both years however, the overall relation is negative for column 4 of Tables 4 and 5 but insignificant for 1997. Moreover, when I examine the county-level estimates, I find that both the linear and squared terms are negative. Taken together, my results suggest that an increased presence of nonprofits outside the firm's industry decreases the likelihood of new firm location.
Analysis and Discussion
In terms of the tax rates affecting businesses, I have found evidence to suggest that, in most cases and particularly in 2002, tax rates and nonprofit location are positively correlated. These results are consistent with a higher nonprofit/for-profit market share found in Gulley and Santerre (1993) and Hansmann (1987). However, even when I find a positive relationship, the total elasticities are relatively small. For a 1% increase in taxes, the increase in the probability of location ranges between .01% and .36%. Moreover, property and, to a lesser extent, corporate taxes are negatively related to location in 1997. This empirical finding could suggest changes in nonprofit behavior over time but could also be due to other reasons. For example, changes in the macroeconomic conditions between 1997 and 2002 could have affected the types of nonprofits entering the industry between the two years. (21) Additionally, differences across specifications confirm the previous discussions that tax rates are correlated to the level of government services, and both, in turn, could affect a nonprofit's location choice. I strive to control for these factors by including regressors for economic conditions and government provisions, but disentangling the independent effect of the tax rates is difficult.
My results interacting tax rates with the source of revenue provide stronger evidence. I find that nonprofits with greater dependence on mission-related revenues are more likely to locate in states with high property and corporate taxes. The results for property taxes are particularly robust across specifications and years. This finding lends credence to the arguments that nonprofits competing with for-profits take advantage of the tax exemption by locating in higher tax states. The negative coefficient on the interaction between revenue source and sales taxes could be contradictory. This difference perhaps stems from the fact that the exemption affects goods that are purchased but not goods sold. Nonprofits are still expected to pay sales taxes on revenues from services rendered as the tax burden is assumed to be on the consumer rather than the firm. If the tax burden is on the nonprofit, this would work against any positive tax advantage from goods purchased. Furthermore, I assume that all states and municipalities exempt nonprofits from the sales tax. However, if nonprofits are not exempt from the sales tax, then the coefficient is as expected.
New to the literature on taxes and nonprofit location, I find a positive relation between individual tax rates and nonprofit location, highlighting the importance of controlling for individual taxes. The elasticity for individual tax rates is larger than other tax variables and is consistent in magnitude across years. This finding indicates that individual tax rates are perhaps more important in the location decision than business-related taxes. As discussed earlier, this positive correlation may be due to many factors. However, the positive coefficient on the interaction between revenue source and individual taxes indicates that nonprofits with more dependence on donations are more sensitive to the individual tax rates. Thus, the effect of taxes on the price of charitable contributions may contribute to firm location choice.
I also find evidence of nonprofit agglomeration and the relation differs between own- and outside-industry concentration. Within-industry agglomeration is positively associated with new nonprofit location while a higher density of nonprofits outside the firm's industry decreases the probability of location in 2002. These effects are similar to those found in previous for-profit studies (Rosenthal and Strange 2003; Henderson 2003). However, my results for a potential nonlinear relationship between outside-industry concentration and new firm location are not particularly robust, and I therefore hesitate to draw conclusions.
For own-industry concentration, agglomeration increases the likelihood of location but at a decreasing rate. This finding is robust across specifications, years, and geographic definition. Compared to the outside-industry elasticity, the average magnitude is larger in absolute value in 1997 but smaller in 2002. The negative second-order effect contributes to the smaller overall magnitude. I attribute the observed concavity to the fact that nonprofits tend to serve a smaller locality and do not generally export goods and services. Thus, these results might suggest that competition and saturation issues begin to dominate the economies of scale benefits associated with agglomeration (i.e., labor pooling, knowledge spillovers, input sharing). Is this relation specific to nonprofits or does this nonlinear agglomeration relation exist for for-profits? To my knowledge, much of the for-profit agglomeration literature focuses on industries where the produced output is easily transportable (e.g., manufacturing, software). If the rationale for the observed concavity is correct, similar relations in for-profit service industries, such as restaurants and banks should be observed.
This paper is the first, to my knowledge, to simultaneously examine the relation of taxes and agglomeration to nonprofit entry and location decisions. The dynamics of the industry are an important consideration and lead to a more detailed understanding of nonprofit behavior. Previous studies found a positive relation between tax rates and nonprofit/for-profit market share, but this result could be due to entry or exit. My paper demonstrates an increase in market share due to taxes may be the result of greater entry rather than decreased exits. I also show that larger individual tax rates are positively associated with new nonprofit location. Moreover, the revenue source of nonprofits affects their sensitivity to tax rates. Nonprofits dependent on service-related revenues are more likely to locate in states with high property taxes. Alternatively, donation-driven nonprofits locate in states with high individual taxes. Both of these results are consistent with theories of nonprofit location behavior.
My results perhaps speak even more strongly to the agglomeration literature. Like for profit studies, I find evidence of within-industry agglomeration economies and, to a lesser extent, a negative relation between outside-industry concentration and location. However, the concave relation for within-industry agglomeration is a new finding and may suggest the benefits to agglomeration are more limited in service-oriented industries with high transport costs. Much of the theoretical agglomeration literature discusses the diseconomies of agglomeration due to congestion and crowding, but I am unaware of much empirical evidence.
My results perhaps suggest a direction for additional for-profit agglomeration studies.
I would like to thank the Urban Institute for providing data and Hongfang Zhang for excellent research assistance. Seminar participants at the 2004 Eastern Economic Association Conference and at Drexel University provided helpful comments. I am grateful to John Pepper and two anonymous referees for valuable comments and suggestions. Any errors are my responsibility.
Received June 2006; accepted June 2007.
Ai, Chunrong, and Edward C. Norton. 2003. Interaction terms in logit and probit models. Economics Letters 80:123-9. Auten, Gerald E., Holger Sieg, and Charles T. Clotfelter. 2002. Charitable giving, income, and taxes: An analysis of panel data. American Economic Review 92:371-82.
Auten, Gerald E., James M. Cilke, and William C. Randolph. 1992. The effects of tax reform on charitable contributions. National Tax Journal 45:267 90.
Banerjee, Abhijit V. 1992. A simple model of herd behavior. Quarterly Journal of Economics 107:797 817.
Bartik, Timothy J. 1985. Business location decisions in the United States: Estimates of the effects of unionization, taxes, and other characteristics of states. Journal of Business and Economies Statistics 3:14 22.
Bielefeld, Wolfgang, and James C. Murdoch. 2004. The locations of nonprofit organizations and their for-profit counterparts: An exploratory analysis. Nonprofit and Voluntary Sector Quarterly 33:221-46.
Bielefeld, Wolfgang, James C. Murdoch, and Paul Waddell. 1997. The influence of demographics and distance on nonprofit location. Nonprofit and Voluntary Sector Quarterly 26:207 25.
Carlton, Dennis W. 1983. The location and employment choices and new firms: An econometric model with discrete and continuous endogenous variables. The Review of Economics and Statistics 65:440-9.
Center on Philanthropy. 2005. Indiana Gives 2004." Charitable Giving in Indiana. Indianapolis, IN: Indiana University, Purdue University.
Clotfelter, Charles T. 1985. Federal Tax Policy and Charitable Giving. Chicago, IL: University of Chicago Press.
Coughlin, Cletus C., Joseph V. Terza, and Vachira Arromdee. 1991. State characteristics and the location of foreign direct investment within the United States. The Review of Economics and Statistics 73:675-83.
dePalma, A., V. Ginsburgh, Y. Papageorgiou, and J.-F. Thisse. 1985. The principle of minimum differentiation holds under sufficient heterogeneity. Econometrica 53:767 81.
Devereux, Michael P., and Rachel Griffith. 1998. Taxes and the location of production: evidence from a panel of U.S. multinationals. Journal of Public Economics 68:335 67.
Downes, Thomas, and Shane M. Greenstein. 1996. Understanding the supply decisions of nonprofits: Modelling the location of private schools. RAND Journal of Economics 27:365-90.
Duranton, Giles, and Diego Puga. 2004. Micro-foundations of urban agglomeration economies. In Handbook of regional and urban economies, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: Elsevier, pp. 2063-I 18.
Fujita, Masahisa, and Jacques-Francois Thisse. 1996. Economics of agglomeration. Journal of the Japanese and International Economies 27:399-442.
Fujita, Masahisa, and Jacques-Francois Thisse. 2002. Economics of agglomeration: Cities', industrial location, and regional growth. New York: Cambridge University Press.
Gentry, William M., and John R. Penrod. 1998. The tax benefits of not-for-profit hospitals. NBER Working Paper No. 6435.
Gius, Mark P., and Phillip Frese. 2002. The impact of state personal and corporate tax rates on firm location. Applied Economics Letters 9:47 9.
Greene, Pamela, and Robert McClelland. 2001. Taxes and charitable giving. National Tax Journal 54:433 53.
Guimaraes, Paulo, Octavio Figueiredo, and Douglas Woodward. 2003. A tractable approach to the firm location decision problem. Review of Economics and Statistics 85:201 4.
Gulley, O. David, and Rexford E. Santerre. 1993. The effect of tax exemption on the market share of nonprofit hospitals. National Tax Journal 46:477 86.
Hansmann, Henry. 1987. The effect of tax exemption and other factors on the market share of nonprofit versus for-profit firms. National Tax Journal 40:71 82.
Hanson, Gordon H. 2001. Scale economies and the geographic concentration of industry. Journal of Economic Geography 1:255 76.
Hausman, Jerry A., and Daniel McFadden. 1984. Specification tests for the multinomial logit model. Econometrics 52:1219 -40.
Hausman, Jerry A., and David A. Wise. 1978. A conditional probit model for qualitative choice: discrete decisions recognizing interdependence and heterogeneous preferences. Econometrics 46:403 26.
Henderson, J. Vernon. 2003. Marshall's scale economies. Journal of Urban Economies 53:1 28.
Mofidi, Alaeddin, and Joe A. Stone. 1990. Do state and local taxes affect economic growth? The Review of Economics and Statistics 72:686-91.
Moore, Cassie J. 2004. Nonprofit organizations are hiring workers at a faster pace than government, businesses. The Chronicle of Philanthropy, 10 June, 37.
Ottaviano, Gianmarco, and Jacques-Francois Thisse. 2004. Agglomeration and economic geography. In Handbook of regional and urban economies, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: Elsevier, pp. 2564- 608.
Quigley, John M. 1998. Urban diversity and economic growth. Journal of Economic Perspectives 12:127 38.
Research Institute of America. 1996. 1996 All States Tax Handbook. New York: Research Institute of America.
Research Institute of America. 2002. 2002 All States Tax Handbook. New York: Research Institute of America.
Rose-Ackerman, Susan. 1986. Unfair competition and corporate income taxation. In The economics of nonprofit institutions: Studies in structure and policy, edited by Susan Rose-Ackerman. Oxford: Oxford University Press, pp. 394-414.
Rosenthal, Stuart S., and William C. Strange. 2003. Geography, industrial organization, and agglomeration. The Review of Economies and Statistics 85:377-93.
Rosenthal, Stuart S., and William C. Strange. 2004. Evidence on the nature and sources of agglomeration economies. In Handbook of regional and urban economics, edited by J. Vernon Henderson and Jacques-Francois Thisse.
Amsterdam: Elsevier, pp. 2119-72.
Steinberg, Richard. 1991. 'Unfair' competition by nonprofits and tax policy. National Tax Journal 44:351-64.
U.S. Bureau of the Census. 1997. Statistical Abstract of the United States: 1997. Washington, DC: Government Printing Office.
U.S. Bureau of the Census. 1998a. State and Metropolitan Area Data Book 199a98. 5th edition. Washington, DC: Government Printing Office.
U.S. Bureau of the Census. 1998b. Statistical Abstract of the United States: 1998. Washington, DC: Government Printing Office.
U.S. Bureau of the Census. 2000. County and City Data Book, 2000. Washington, DC: Government Printing Office.
U.S. Bureau of the Census. 2002. Statistical Abstract of the United States: 2002. Washington, DC: Government Printing Office.
U.S. Bureau of the Census. 2003. Statistical Abstract of the United States: 2003. Washington, DC: Government Printing Office.
U.S. Bureau of Economic Analysis. 2005. Regional Economic Accounts, Accessed 15 November 2005. Available at http:// www.bea.doc.gov/bea/regional/reis/default.cfm#a.
Wolch, Jennifer R., and R. K. Geiger. 1983. The urban distribution of voluntary resources: An exploratory analysis.
Environment and Planning 15:1067-82.
Wolpert, Julian. 1993. Decentralization and equity in public and nonprofit sectors. Nonprofit and Voluntary Sector Quarterly 22:281 96.
Woodward, Douglas P. 1992. Locational determinants of Japanese manufacturing start-ups in the United States. Southern Economic Journal 58:690 708.
(1) The agglomeration literature discusses the effect on location as well as production, growth, wages, and profits. For this paper, I focus on the location studies but refer an interested reader to Rosenthal and Strange (2004) for a complete survey of the empirical findings.
(2) See Clotfelter (1985); Auten, Cilke, and Randolph (1992); Randolph (1995); Greene and McClelland (2001); and Auten, Sieg, and Clotfelter (2002), for examples.
(3) As noted by a referee, in the extreme where all giving is interstate, an observed positive relation between state individual tax rates and nonprofit location should not be due to the charitable deduction. However, a study of philanthropy in Indiana reports that 79% of giving is to organizations located within the state (Center on Philanthropy 20O5).
(4) Perhaps one of the most important factors would be the presence of within-industry for-profit concentration. Including for-profit agglomeration is difficult because the NTEE/NAICS mapping is not one-to-one, particularly at the one-digit level. I therefore performed a county-level analysis on just National Taxonomy of Exempt Entities (NTEE) "A6," which maps to North American Industry Classification System (NAICS) "7111." I did not directly examine tax rates, but my agglomeration results were consistent with those presented here.
(5) See Fujita and Thisse (1996; 2002); Quigley (1998); and Duranton and Puga (2004) for a thorough discussion of the microfoundations of agglomeration.
(6) Ottaviano and Thisse (2004) discuss these issues as "the interplay between a dispersion force and an agglomeration force" (p. 2574). Transportation costs and increased competition are the dispersion forces that when high enough can deter agglomeration (de Palma et al. 1985).
(7) I also attempted to run a nested and random coefficients model and was unable to achieve convergence.
(8) County-level tax rate data for 1997 and 2002 (or prior) are not available from a common repository, even at the state level (i.e., Commerce Clearing House [CCH] Tax Research Network, Department of Revenue, etc.). Such repositories only report the most current tax rates and even then, the data is available for only some states. Use of current tax data would lead to biased estimates since changes to tax rates after a nonprofit's entry decision may very well be influenced by that entry decision. Moreover, there are additional issues with consistency of the reporting. Some states report average or effective tax rates, while others report mills rates. In addition, the unit of observation for some states is at the city/township/borough level. Analysis of tax rates and location decisions at this Finer geographic level is, perhaps, an area for further research.
(9) Nonprofits are classified as operating, supporting, or mutual benefit organizations, and as public charities, private foundations, or other nonprofits.
(10) The exit rates of nonprofits are an important component of nonprofit firm dynamics. However, identifying exits using the 990 tax return is less precise due to noncompliance on filing. In order to avoid these measurement error issues, I only look at entry rates.
(11) Unfortunately, individual donations cannot be separately identified in this data.
(12) As an example of program service revenue, hospitals would include revenues from treating patients under this category.
(13) See http://nccs.urban.org for further details on the classification system.
(14) Existing firm concentration has been measured both as size of the workforce (Carlton 1983; Bartik 1985) and number of existing firms (Woodward 1992; Gius and Frese 2002; Rosenthal and Strange 2003). Ideally, I would employ both measures as in Rosenthal and Strange (2003) and Henderson (2003). Unfortunately, my data do not contain employment information. However, Henderson's (2003) finding that number of firms is a better measure suggests that my omission of employment agglomeration may not be severe.
(16) California, Florida, Georgia, Illinois, Michigan, New Jersey, New York, Ohio, Pennsylvania, and Texas.
(17) I run the regressions separately for each year because variables, such as year dummies, that do not vary by choice cannot be identified with the conditional logit specification. Differences across time could be captured by interacting the year dummy with particular covariates. However, since it is not clear a priori which coefficients might vary across time, I investigate differences for all covariates. Indeed, I find that the tax rate coefficients vary across time so pooling the years would just mask these year differences.
(18) Although demographic and government service provisions are controlled for in this regression and the count data model, I do not report the individual coefficients due to large variations in results across specifications. (Results for these coefficients provided upon request.) I, therefore, simply report the likelihood ratio test for joint significance of the demographic and government provision variables and note that these regressors perform well as controls.
(19) For every state j, recall that regressor m varies by state ([x.sub.j,m]) and the probability of locating in a state is a function of every state's characteristics. Thus, given a total of J choices and M regressors, there are J * J * M own and crosselasticities (i.e., [partial derivative]log(Pj)/[partial derivative]log([x.sub.k,m), [for all]m, j, and k). To simplify interpretation, I only present the average ownelasticities for the tax and agglomeration variables. For the interaction terms x * y, where x is the tax variable, the elasticity for firm i and state j is [[epsilon].sub.ij] = ([[beta].sub.1] + [beta].sub.2] x [y.sub.i,j]) x (1 - [P.sub.i,j]) x [x.sub.i,j], where [[beta].sub.1] and [[beta].sub.2] are the coefficients on x and x * y respectively. Similarly for agglomeration, [[epsilon].sub.ij] = ([[beta].sub.1] + 2[[beta].sub.2] x [x.sub.ij]) x (1 [P.sub.ij])x [x.sub.ij], where [[beta].sub.1] and [beta].sub.2] are the coefficients on x and [chi square] respectively. Note that these elasticities take into account the inherent nonlinearity in a conditional logit model (Ai and Norton 2003).
(20) The smaller county level magnitudes, relative to the state elasticities, do not necessarily contradict Rosenthal and Strange's (2003) finding of strong local agglomeration effects. Unlike Rosenthal and Strange's specification, my model does not measure agglomeration attenuation within the same regression.
(21) Ideally, I would also include industry dummies, but recall that regressors must vary across choices in order to be included in the conditional logit.
Teresa D. Harrison, Department of Economics, Drexel University, 3141 Chestnut Street, Philadelphia, PA 19104, USA; E-mail email@example.com.
Table 1. Variable Descriptions Variable Description TOTREV Total revenues TOTEXPS Total expenses NETINC TOTREV--TOTEXPS CONT Total public support PROG Program service revenues PERDON CONT/TOTREV PERPROG PROG/TOTREV CORMTR Corporate marginal tax rate INDMTR Individual marginal tax rate SALES Sales tax rate AVGPROP Average property tax rate EXISTNP Outside industry agglomeration/POP EXISTNTEE Inside industry agglomeration/POP NUMNEWNP Number of new nonprofits by industry and county POP State/county population INCOME Per capita income POPCH Population change POP065 Percent of population over 65 POPEDHS Percent of population with at least high school diploma PERCBPOV Percent of population below poverty level REPPRES Percent voting Republican in 1996 Presidential election UNINS Percent uninsured INFMORT Infant mortality rate INCMAIN Welfare expenditures MEDICARE Medicare expenditures MEDICAID Medicaid expenditures VETBEN Veterans' benefits UNEMPINS Unemployment insurance benefits PAYRET Retirement benefits TAXREV Total county tax revenues PROPREV Total property tax revenues GEXP Total government expenditures Table 2. Descriptive Statistics for 1997 Standard Variable N Mean Deviation By firm CORMTR 16,541 4.66 3.97 SALES 16,541 5.37 1.59 AVGPROP 16,541 2.95 3.88 INDMTR 16,541 4.38 2.50 EXISTNP 16,541 0.57 0.14 EXISTNTEE 16,541 0.06 0.04 PERPROG 16,541 0.28 0.39 PERDON 16,541 0.52 0.42 By state CORMTR 48 4.61 3.88 SALES 48 4.82 1.74 AVGPROP 48 2.63 3.11 INDMTR 48 4.87 2.57 POP (000) 48 5425.63 5806.25 INCOME 48 22,067.31 3305.31 POPCH 48 8.25 7.40 POPO65 48 12.86 1.75 POPEDHS 48 75.99 5.51 PERCBPOV 48 13.12 4.10 REPPRES 48 42.02 6.90 UNINS 48 14.23 4.12 INFMORT 48 7.58 1.33 INCMAIN 48 2,067,751.00 3,111,375.00 MEDICARE 48 3,707,159.00 4,250,484.00 MEDICAID 48 3,194,396.00 4,402,345.00 VETBEN 48 420,497.20 387,308.10 UNEMPINS 48 446,699.60 607,933.40 PAYRET 48 7,252,985.00 7,335,888.00 GEXP 48 17,193.25 20,010.49 Variable Minimum Maximum By firm CORMTR 0.00 14.00 SALES 0.00 7.25 AVGPROP 0.31 13.19 INDMTR 0.00 9.00 EXISTNP 0.28 1.40 EXISTNTEE 0.00 0.27 PERPROG 0.00 1.00 PERDON 0.00 1.00 By state CORMTR 0.00 14.00 SALES 0.00 7.25 AVGPROP 0.31 13.19 INDMTR 0.00 9.00 POP (000) 479.00 31,558.00 INCOME 16,690.00 31,814.00 POPCH -1.60 39.50 POPO65 8.80 18.50 POPEDHS 64.30 85.10 PERCBPOV 5.30 25.30 REPPRES 26.80 54.40 UNINS 7.30 25.60 INFMORT 5.20 10.50 INCMAIN 115,239.00 18,200,000.00 MEDICARE 226,666.00 20,400,000.00 MEDICAID 173,787.00 26,000,000.00 VETBEN 40,315.00 1,706,891.00 UNEMPINS 14,580.00 3,200,073.00 PAYRET 663,487.00 34,700,000.00 GEXP 1880.00 109,231.00 Table 3. Descriptive Statistics for 2002 Standard Variable N Mean Deviation By firm CORMTR 29,370 4.41 3.77 SALES 29,370 5.35 1.47 AVGPROP 29,370 3.78 4.50 INDMTR 29,370 4.16 2.45 EXISTNP 29,370 0.66 0.18 EXISTNTEE 29,370 0.07 0.05 PERPROG 29,370 0.24 0.37 PERDON 29,370 0.60 0.42 By state CORMTR 48 4.24 3.48 SALES 48 4.84 1.74 AVGPROP 48 3.26 3.77 INDMTR 48 4.78 2.54 POP (000) 48 5882.58 6338.75 INCOME 48 28,787.27 4462.00 POPCH 48 1.02 1.11 POPO65 48 12.66 1.66 POPEDHS 48 85.34 3.97 PERCBPOV 48 11.52 2.96 REPPRES 48 50.54 8.61 UNINS 48 13.04 3.79 INFMORT 48 7.03 1.49 INCMAIN 48 2,253,063.00 3,205,262.00 MEDICARE 48 5,037,328.00 5,719,273.00 MEDICAID 48 4,842,161.00 6,284,315.00 VETBEN 48 544,660.50 504,937.80 UNEMPINS 48 662,983.90 751,835.90 PAYRET 48 9,322,486.00 9,335,835.00 GEXP 48 33,412.31 40,049.25 By county NUMNEWNP 10,706 1.49 5.34 EXISTNP 10,706 6.16E-04 3.11E-04 EXISTNTEE 10,706 4.96E-05 6.38E-05 POP 10,706 309,480.90 688,198.40 INCOME 10,706 24,260.65 6632.57 POPCH 10,706 13.78 16.25 POPO65 10,706 13.87 4.16 POPEDHS 10,706 73.17 8.20 PERCBPOV 10,706 13.62 5.70 TAXREV 10,706 537,638.60 2,166,466.00 PROPREV 10,706 326,236.60 946,964.40 GEXP 10,706 1,257,841.00 4,683,111.00 Variable Minimum Maximum By firm CORMTR 0.00 13.60 SALES 0.00 7.25 AVGPROP 0.00 14.39 INDMTR 0.00 9.33 EXISTNP 0.31 1.80 EXISTNTEE 0.00 0.34 PERPROG 0.00 1.00 PERDON 0.00 1.00 By state CORMTR 0.00 10.50 SALES 0.00 7.25 AVGPROP 0.00 14.39 INDMTR 0.00 9.33 POP (000) 494.00 34,501.00 INCOME 21,643.00 41,930.00 POPCH -1.20 5.40 POPO65 8.50 17.60 POPEDHS 77.10 91.80 PERCBPOV 7.30 19.30 REPPRES 31.90 67.80 UNINS 7.50 23.50 INFMORT 3.80 10.70 INCMAIN 113,506.00 18,800,000.00 MEDICARE 326,241.00 27,400,000.00 MEDICAID 257,402.00 33,400,000.00 VETBEN 55,138.00 2,239,865.00 UNEMPINS 22,759.00 3,466,374.00 PAYRET 884,308.00 45,300,000.00 GEXP 3352.00 217,970.00 By county NUMNEWNP 0.00 172.00 EXISTNP 0.00 2.59E-03 EXISTNTEE 0.00 1.18E-03 POP 849.00 9,637,494.00 INCOME 8,225.00 72,194.00 POPCH -26.30 123.20 POPO65 4.00 34.70 POPEDHS 31.60 91.90 PERCBPOV 3.10 46.70 TAXREV 826.00 19,400,000.00 PROPREV 795.00 7,387,996.00 GEXP 2437.00 40,400,000.00 Table 4. Conditional Logit Estimates for 1997 Variable (1) (2) CORMTR -0.0063 (0.0043) -0.0100 ** (0.0048) SALES 0.0325 *** (0.0088) 0.0367 *** (0.0096) AVGPROP -0.0010 (0.0033) -0.0067 * (0.0037) INDMTR 0.0143 ** (0.0060) 0.0061 (0.0073) PERPROG * CORMTR 0.0138 * (0.0082) PERPROG * SALES -0.0145 (0.0129) PERPROG * AVGPROP 0.0189 *** (0.0053) PERDON * INDMTR 0.0162 ** (0.0082) EXISTNP 3.1178 *** (0.4347) 3.1296 *** (0.4348) EXISTNTEE 35.2854 *** (1.4249) 35.1562 *** (1.4268) EXISTNP (2) -2.7379 *** (0.3180) -2.7429 *** (0.3180) EXISTNTEE (2) 149.5685 *** (6.8680) -149.1140 *** (6.8737) LR test demographic variables 3203.93 *** 3202.15 *** LR test government provisions N/A N/A LLF Value -56,678.324 -56,669.464 LR test specifications 644.052 626.332 Variable (3) (4) CORMTR -0.0012 (0.0049) 0.0008 (0.0049) SALES 0.0818 *** (0.0135) 0.0423 *** (0.0127) AVGPROP -0.0327 *** (0.0062) -0.0160 *** (0.0060) INDMTR 0.0647 *** (0.0086) 0.0789 *** (0.0087) PERPROG * CORMTR 0.0124 (0.0082) 0.0150 * (0.0082) PERPROG * SALES -0.0177 (0.0134) -0.0182 (0.0133) PERPROG * AVGPROP 0.0182 *** (0.0053) 0.0204 *** (0.0053) PERDON * INDMTR 0.0159 * (0.0082) 0.0149 * (0.0083) EXISTNP 2.8951 *** (0.5639) -0.2459 (0.1616) EXISTNTEE 33.9806 *** (1.4535) 5.2871 *** (0.4595) EXISTNP (2) -2.0334 *** (0.3654) EXISTNTEE (2) -135.8838 *** (6.9113) LR test demographic variables 306.23 *** 297.67 *** LR test government provisions 579.4 *** 786.49 *** LLF Value -56,356.298 -56,611.791 LR test specifications -- 510.986 Variable (5) CORMTR -0.0030 (0.0049) SALES 0.0677 *** (0.0129) AVGPROP INDMTR 0.0676 *** (0.0086) PERPROG * CORMTR 0.0139 * (0.0082) PERPROG * SALES -0.0236 * (0.0131) PERPROG * AVGPROP PERDON * INDMTR 0.0108 (0.0081) EXISTNP 2.2465 *** (0.5471) EXISTNTEE 33.5033 *** (1.4448) EXISTNP (2) -1.6686 *** (0.3592) EXISTNTEE (2) -133.5289 *** (6.8731) LR test demographic variables 289.72 *** LR test government provisions 560.67 *** LLF Value -56,372.985 LR test specifications 33.374 Note: Standard errors are in parentheses. Coefficients with *** **, and * are significant at [alpha] = .01, .05, and .10, respectively. All regressions include regional dummies and demographic variables (POP, POPCH, INCOME, POPO65, POPEDHS, PERCBPOV, REPPRES, UNINS, and INFMORT). Government provisions (INCMAIN, MEDICARE, MEDICAID, VETBEN, UNEMPINS, PAYRET, and GEXP) are included in columns (3)-(5). We only report the likelihood ratio (LR) tests for the joint significance of the demographic and government provision variables. Individual coefficients for these variables provided upon request. We also report the LR statistic comparing the specification in column (3) to all other specifications. Table 5. Conditional Logit Estimates for 2002 Variable (1) CORMTR 0.0034 (0.0033) SALES 0.0281 *** (0.0065) AVGPROP 0.0110 *** (0.0023) INDMTR -0.0068 (0.0043) PERPROG * CORMTR PERPROG * SALES PERPROG * AVGPROP PERDON * INDMTR EXISTNP -0.4978 ** (0.2359) EXISTNTEE 19.5054 *** (0.7890) EXISTNP (2) -0.1625 (0.1223) EXISTNTEE (2) -58.4762 *** (2.9397) LR test-demographic variables 5942.6 *** LR test-government provisions N/A LLF Value -102,604.27 LR test of specifications 1400.92 Variable (2) CORMTR 0.0032 (0.0036) SALES 0.0343 *** (0.0071) AVGPROP 0.0090 *** (0.0025) INDMTR -0.0209 *** (0.0056) PERPROG * CORMTR -0.0011 (0.0070) PERPROG * SALES -0.0251 ** (0.0107) PERPROG * AVGPROP 0.0081 ** (0.0036) PERDON * INDMTR 0.0241 *** (0.0063) EXISTNP -0.4953 ** (0.2359) EXISTNTEE 19.5212 *** (0.7903) EXISTNP (2) -0.1638 (0.1224) EXISTNTEE (2) -58.5301 *** (2.9430) LR test-demographic variables 5945.45 *** LR test-government provisions N/A LLF Value -102,593.78 LR test of specifications 1379.94 Variable (3) CORMTR 0.0119 *** (0.0039) SALES 0.0101 (0.0090) AVGPROP 0.0213 *** (0.0037) INDMTR 0.0495 *** (0.0064) PERPROG * CORMTR -0.0021 (0.0070) PERPROG * SALES -0.0320 *** (0.0112) PERPROG * AVGPROP 0.0080 ** (0.0036) PERDON * INDMTR 0.0240 *** (0.0064) EXISTNP -3.8442 *** (0.3355) EXISTNTEE 16.9142 *** (0.8170) EXISTNP (2) 1.5466 *** (0.1558) EXISTNTEE (2) -45.9229 *** (2.9900) LR test-demographic variables 626.34 *** LR test-government provisions 1317.14* ** LLF Value -101,903.81 LR test of specifications -- Variable (4) CORMTR 0.0116 *** (0.0039) SALES 0.0196 ** (0.0086) AVGPROP 0.0162 *** (0.0034) INDMTR 0.0510 *** (0.0064) PERPROG * CORMTR -0.0005 (0.0070) PERPROG * SALES -0.0305 *** (0.0112) PERPROG * AVGPROP 0.0093 ** (0.0036) PERDON * INDMTR 0.0230 *** (0.0064) EXISTNP -0.9108 *** (0.0948) EXISTNTEE 4.3187 *** (0.2960) EXISTNP (2) EXISTNTEE (2) LR test-demographic variables 620.62 *** LR test-government provisions 1421.81 *** LLF Value -102,113.3 LR test of specifications 418.98 Variable (5) CORMTR 0.0128 *** (0.0039) SALES 0.0144 (0.0090) AVGPROP INDMTR 0.0500 *** (0.0064) PERPROG * CORMTR -0.0002 (0.0070) PERPROG * SALES -0.0339 *** (0.0111) PERPROG * AVGPROP PERDON * I NDMTR 0.0227 *** (0.0063) EXISTNP -2.8240 *** (0.2941) EXISTNTEE 17.7554 *** (0.8065) EXISTNP (2) 1.1445 *** (0.1419) EXISTNTEE (2) -48.6916 *** (2.9596) LR test-demographic variables 593.28 *** LR test-government provisions 1311.09 *** LLF Value -101,926.66 LR test of specifications 45.7 Standard errors are in parentheses. Coefficients with ***, **, and * are significant at [alpha] = .01, .05, and .10, respectively. All regressions include regional dummies and demographic variables (POP, POPCH, INCOME, POPO65, POPEDHS, PERCBPOV, REPPRES, UNINS, and INFMORT). Government provisions (INCMAIN, MEDICARE, MEDICAID, VETBEN, UNEMPINS, PAYRET, and GEXP) are included in columns 3-5. I only report the likelihood ratio (LR) tests for the joint significance of the demographic and government provision variables. Individual coefficients for these variables provided upon request. I also report the LR statistic comparing the specification in column 3 to all other specifications. Table 6. Count Data for 2002 County Level Variable (1) Elasticity EXISTNP -219.5666 (159.1900) -0.1353 EXISTNTEE 12,098.3400 *** (409.2985) 0.5997 EXISTNP (2) -1.98E+05 ** (7.90E+04) -0.0942 EXISTNTEE (2) -1.84E+07 *** (9.41E+05) -0.1203 CONSTANT -5.2920 *** (0.3619) LR test-demographic 3459.30 *** variables LR test-government N/A provisions LLF value -13,609.612 LR test-Poisson 7990.51 Variable (2) Elasticity EXISTNP -300.8638 * (166.8603) -0.1854 EXISTNTEE 12,034.5500 *** (411.6021) 0.5965 EXISTNP (2) -1.77E+05 ** (8.28E+04) -0.0843 EXISTNTEE (2) -1.83E+07 *** (9.57E+05) -0.1194 CONSTANT -5.1681 *** (0.3662) LR test-demographic 1163.12 *** variables LR test-government 13.01 *** provisions LLF value -13,603.114 LR test-Poisson 6602.09 Note: Standard errors are in parentheses. Coefficients with *** **, and * are significant at [alpha] = .01, .05, and .10, respectively. All regressions include regional dummies and demographic variables (POP, POPCH, INCOME, POPO65, POPEDHS, and PERCBPOV). Government provisions (TAXREV, PROPREV, and GEXP) are included in column (2). We only report the likelihood ratio (LR) tests for the joint significance of the demographic and government provision variables. Individual coefficients provided upon request. Table 7. Average Own-Elasticities for Conditional Logit Variable 1997 2002 CORMTR -0.0050 0.0484 SALES 0.3848 0.0477 AVGPROP -0.0842 0.0677 INDMTR 0.3090 0.2327 PERPROG * CORMTR 0.0035 -0.0005 PERPROG * SALES -0.0049 -0.0075 PERPROG * AVGPROP 0.0051 0.0019 PERDON * INDMTR 0.0081 0.0140 EXISTNP 1.7114 -2.7252 EXISTNTEE 1.9358 1.1975 EXISTNP (2) -0.8037 0.8885 EXISTNTEE (2) -0.7628 -0.3959 Total effects CORMTR 0.0077 0.0465 SALES 0.3611 0.0115 AVGPROP -0.0709 0.0738 INDMTR 0.3487 0.2998 EXISTNP 0.1040 -0.9483 EXISTNTEE 0.4101 0.4058
|Printer friendly Cite/link Email Feedback|
|Comment:||Taxes and agglomeration economies: how are they related to nonprofit firm location?|
|Author:||Harrison, Teresa D.|
|Publication:||Southern Economic Journal|
|Date:||Oct 1, 2008|
|Previous Article:||Corruption, centralization, and the shadow economy.|
|Next Article:||Third degree price discrimination in linear-demand markets: effects on number of markets served and social welfare.|