The Generous Congregation in Context

Transcription

1 The Generous Congregation in Context Spatial and Social Patterns of Congregational Giving in a Metropolitan County Becky R. Haney Assistant Professor of Economics Calvin College Calvin College, 3201 Burton Street, S.E., Grand Rapids MI, Phone: (616) , brh22@calvin.edu Jeffrey Schiman Research Assistant Calvin College Center for Social Research Calvin College, 3201 Burton Street, S.E., Grand Rapids MI, Phone: (269) , jcs25@calvin.edu Neil Carlson Assistant Director Calvin College Center for Social Research Calvin College, 3201 Burton Street, S.E., Grand Rapids MI, Phone: (616) , neil.carlson@calvin.edu Abstract While most existing research focuses on individual behavior, congregations as units of analysis may be the critical intermediaries in the transfer of donated resources from individuals to recipients. We control for an array of variables and explore the effects of both geographical and social proximity to other congregations that give to local and international recipients. We also explore the role of a congregation s income on the level of giving to local and international recipients and estimate income elasticity curves for each type of giving. We find that congregational giving differs from household giving patterns in the literature in that it is less sensitive to demographics. We also find that congregational giving to local recipients generally declines as income increases but remains a normal or necessary good, while giving to international recipients is a luxury good.

2 Introduction In the spring of 2007, a large group of local clergy met to advise the scholars designing the questionnaire for the Kent County Congregations Study, a comprehensive study of all the religious congregations in the county, which includes the city of Grand Rapids, Michigan. One of those clergy advisors was a minister at an urban, predominantly African- American Baptist church. The minister was understandably frustrated. He saw that local churches gave millions and millions of dollars to international missions, but he perceived that very little of large church budgets came back to benefit the local community. This pastor asked us to find out: how much money leaves the county, and how much stays? And who is giving the money that stays in the community? We found out, and this paper documents some key findings. The main contribution of this paper is to examine in detail the role of congregations as mediators in charitable giving, particularly with respect to their selection of aid recipients. We focus on local and international recipients. 1 In addition, we explore the effects of both geographic and social proximity to other congregations that give to local recipients. Our key findings for Kent County are: international giving dwarfs local giving in total dollars by a factor of about 3 to 1; international giving and local giving are complements (more of one is associated with more of the other); the income elasticity of international giving is concave and elastic, but the income elasticity of local giving is convex and inelastic that is, international giving is a luxury good, while local giving is a necessary good. 1 A third category, giving to domestic U.S. recipients outside the county, is omitted to simplify the analysis. Findings not presented here suggest that non-local domestic giving behaves in a way that falls between local and international.

3 Theory and literature Nearly 60% of all U.S. households donate to religious institutions (Andreoni 2001). Household-level giving is the backbone of charitable giving in the United States. Households provide 77% of all charitable giving in the U.S., far more than foundations (10%), bequests (8%), and corporations (5%) combined. The absolute amount of household charitable giving is also large: in 1998, households gave $138 billion out of a total of $174 billion (Andreoni 2001). Social scientists who study philanthropy have typically focused on the giving behavior of these households. However, few studies have attempted to study what happens to the funds once they have been given to religious congregations; notable exceptions include Rosborough (2008) and Zaleski, Zech, and Hoge (1992). Thus, congregational-level giving behavior is vital to understanding the true nature of charitable giving, as congregations are intermediaries of billions of dollars of donated funds. We begin to examine congregational-level giving behavior by comparing it to what is known about household giving behavior. Sociologists and others have studied the demographic and background characteristics related to giving, such as age, sex, education, religious participation, theological orientation, and income. Studies have shown that older, more educated, married, high income, urban, and theologically conservative (evangelical) households are related to higher levels of giving, although there is wide variance in the strength of the estimated relationships (Hoge 1994; Bekkers and Wiepking 2007; Wiepking 2007). Thus, our estimation models include aggregate background characteristics of the congregation to examine whether congregational giving patterns are similar to those of households. In addition, we examine the relationship of giving to organizational 3

4 characteristics of the congregation, such as leadership style of the pastor and strength or influence of the lay leadership. Economic studies of charitable giving typically examine the price and income effects on household giving, after accounting for these background characteristics. The price effect for giving comes from tax policy which allows households to deduct charitable giving from income and is thus related to the marginal tax rate. The price effect is a common explanation for higher giving levels and proportions among higher income households, as their marginal tax rates are higher. The price of giving is lower because the benefit of deducting charitable giving from taxable income is larger. Giving is also considered a normal good; thus, giving levels are expected to increase as income increases, in addition to the price effect. For congregations, there is no price effect (that is, no tax-reducing benefit from giving); therefore the relationship of giving and income or budget level will be a pure income effect. The focus of our study is on the relationship of a congregation s income to its pattern of giving, both locally and internationally. The general consensus is that household charitable giving has a large income elasticity ranging from 0.60 to 0.80 in Steinberg (1990), 0.40 to 0.87 in Auten, Sieg, and Clotfelter (2002), and 0.66 to 0.97 in McClelland and Brooks (2004); it is quite responsive to changes in income. However, another finding is that giving may follow a U-shaped curve in relationship to income, particularly for contributors to religious institutions, and specifically as a proportion of income, rather than in absolute levels. That is, low income and high income households give a higher proportion of their incomes than do middle income households (Schervish and Havens 1995; Andreoni 2001). 4

5 McClelland and Brooks (2004) examine the determinants of the U-shaped giving curve as well as compute the income elasticities related to household charitable giving at each income level, after accounting for background characteristics. Using a similar multivariate elasticity model for each type of giving, we will estimate a giving curve for congregations and estimate income elasticities of giving at various income levels. Data and methods The data are from the 2007 Kent County Congregations Study (KCCS), which is a comprehensive survey of 583 religious institutions in Kent County, Michigan. 2 Kent County is a large metropolitan county with about 600,000 residents, including the city of Grand Rapids. The study s response rate is 81%, including rates above 80% for predominantly Hispanic and African congregations. To avoid multiple issues of heterogeneity, we exclude 31 non-christian congregations, reducing the sample to 552 cases. The unit of analysis is the congregation. For each congregation, a key informant (usually the senior ordained clergy) gave us a 90 to 120-minute interview on congregational demographics, staffing, training needs, religious education, finances, networking, programs, social services, and vision. A summary report is available online (Carlson and Hernandez 2008), and data is available on the Association of Religion Data Archive (thearda.com 2010). The KCCS asks three questions about charitable giving in the following form: 2 The survey was conducted by the Calvin College Center for Social Research (CSR) and funded by the Douglas and Maria DeVos Foundation. The Center for Marketing and Opinion Research (CMOR) in Akron, Ohio collected telephone-mode responses for 188 congregations. The 395 remaining interviews were conducted face to face by Calvin College students and by well-connected representatives of the African- American and Hispanic religious communities. 5

6 1. In your most recent fiscal year, did your congregation provide financial support to any international projects, individual missionaries or aid workers? This might include outreach, missions, relief or development work. 2. In your most recent fiscal year, did your congregation provide financial support to any outreach, missions, relief or development projects or to individual missionaries or workers in the U.S. but outside of Kent County? 3. In your most recent fiscal year, did your congregation provide financial support to any outreach, missions, relief or development projects or individual missionaries or workers in Kent County? If the respondent answered yes to any one of these questions, he or she was then asked: How much? These six questions were used to calculate international, local, and domestic giving variables. The 583 responding congregations reported $32.3 million in gifts to international recipients; $10.3 million to domestic U.S. but non-local recipients; and $10.8 million to Kent County recipients. The ratio of international giving to local giving is just shy of 3 to 1. In the interest of simplicity, domestic giving has been dropped from this analysis. The remainder of the paper will focus on local and international giving. The selection model Several techniques used to model survey data about charitable giving include: Tobit analysis (Carrol, McCarthy et al. 2005); OLS that excludes those that give nothing (Rooney, Seinberg et al. 2001; McClelland and Brooks 2004); and Heckman Selection models (Bekkers 2006; Bekkers and Wiepking 2006). In a single-equation model, it is not ideal to include or exclude congregations that give nothing, because coefficients become biased. 6

7 Tobit is not appropriate, because the data are not censored (that is, we can observe the lack of giving); rather, the data are subject to selection bias. Selection bias arises from two factors. First, only those congregations that give have measurable outcomes, which can bias the results if giving congregations and non-giving congregations differ in unmeasured ways (Golder 2010). Second, restrictions imposed on the models (such as analyzing only nonzero values, as we do) can impact the probability of selection and bias the results (Heckman 1979). To control for the selection problem, we use a Heckman Selection model, with selection equation: (1) and outcome equation, (2) Where is a dummy variable indicating if the congregation gave any funds or not, is the amount donated divided by the congregation s budget, and are both vectors of congregation characteristics, and and are error terms. If the error terms and are correlated (, there is evidence that the Heckman selection model is needed to estimate this problem without selection bias. Care must be taken when interpreting parameters of variables that are shared between the selection and outcome equation. Because interpretation is not the same as in ordinary least squares, a slight transformation is necessary. The following equation shows the standard method by which one transforms shared variables (Sigelman and Zeng 1999): (3) 7

8 Here, is the parameter estimate from the outcome equation, is the parameter estimate from the selection equation, is a measure of how correlated the selection and outcome equations are, is the inverse Mills ratio, and is the standard error of the residual in the outcome equation. 3 Findings and results Local giving regression results 4 Table 1 presents the results of a Heckman selection model regressing nonzero local giving (as a percentage of congregational budget, following the example of McClelland and Brooks) on a set of explanatory variables, while modeling selection bias on a similar set. In local giving, income is important. The coefficient for income is negative (-1.620), and the coefficient for income squared is positive (0.232, though non-significant), producing a somewhat convex (or U-shaped ) curve as found by McClelland and Brooks, Haven and Shervish. That is, giving to local aid recipients falls as income increases up to an inflection point, and then possibly rises again for congregations with very large incomes. See Figure 1 on page 15 for a visualization of this shape in the raw data. Among household-level predictors of giving, age, sex, education, and race have significant coefficients in much of the published literature; however, at the congregation level, we find no significant relationships between these congregational demography 3 Stata software s marginal effects command ( mfx ) using the y-conditional option accomplishes the process in equation (4); see 4 Variables removed from the local and international giving model include a dummy indicating whether social justice or individual morality is at the heart of the gospel, the percent of the congregation that is theologically liberal, and the percent of college students in the congregation. These variables lacked significance and reduced the sample size. 8

9 variables and local giving, except for the percent of the congregation that is female, which has a negative and significant coefficient (-0.072). We calculate the distance in miles from the nearest spatially-neighboring congregation that gives to local recipients, and also include the log of the amount that congregation gives. Although these variables attained or approached significance in some model specifications, they are not significant in the outcome equation of the best model we are presenting here. The variable, however, is important in the selection equation described below. Community engagement variables such as the natural log of paid community staff and school involvement positively affect local giving as a proportion of budget. The model s value for is with a marginally significant chi-square p-value of 0.081, suggesting that the errors in the outcome and selection equations may be sufficiently correlated for there to be some value in using the Heckman procedure. In the selection equation, we find that congregations are more likely to give anything at all to local recipients, thus entering the outcome equation discussed above, provided they have: larger budgets; suburban location (compared to rural); a pastor or leader who participates in civic affairs; a sense of efficacy (capacity to influence local government); an influential governing board (council, consistory, etc.); a Mainline denominational affiliation; any participation in joint service projects with other congregations; or a nearest spatially neighboring congregation that contributes to local recipients. These results suggest that local giving is sensitive to and declines with rising incomes; that congregations behave differently from households; and that selection bias is a potential issue in analyzing congregational giving to local recipients. Of particular interest is the indication of spatial autocorrelation in the selection equation, given that the more 9

10 their nearest neighbor gives locally, the more likely congregations are to give at least something to local recipients. Table 1 Heckman selection model results for local giving Model dependent variable: *100 Variables Coefficient Std. Err. p-value Sig. Outcome Equation Socio-demographic Income (millions) * Income (millions) squared Flag for imputed income Percent Minority Percent Female * Percent Four-year degree Percent Over Urban Suburban Community Involvement Total programs (log) Social interactions (log) Pastor participation (log) School involvement Staff assigned to comm ty (log) ** Volunteer recruitment is easy Volunteer staff count (log) Other Seminary Has experienced conflict Nearest local giver (dollars) Nearest local giver (miles) Constant *** Selection Equation Socio-demographic Income *** Flag for imputed income Urban Suburban * Other Pastor participation (log) ** Has any volunteer staff Governing board influence ** Congregational efficacy * Mainline denomination ** Joint service projects ** Local Interest Nearest local giver (dollars) *** Nearest local giver (miles) Constant Sample Size 384 Censored Observations 119 Uncensored Observations 265 Rho Lambda Sigma Wald test of independent equations chi2=3.05 Prob>chi2 =

11 International giving regression results Table 2 presents the results of a Heckman selection model regressing nonzero international giving on a set of explanatory variables. In the international model, income appears to have a concave relationship, with an income beta of and an income squared beta of , significant at the p < 0.10 and p < 0.05 levels, respectively. See Figure 1 on page 15 for a visualization of this relationship. Of particular interest in the international giving model are the coefficients for percent minority, percent with a four year college degree, and the percent over age sixty. All are significant, and all are negative. This indicates that, all else equal, greater minority populations, greater college educated populations, and greater elderly populations are associated with less international giving as a proportion of budget. None of these coefficients were significant in the local giving model. In terms of leadership characteristics, the number of hours that the pastor or key leader spends engaged in civic affairs reduces the amount given internationally as a proportion of budget. Seminary educated leaders are found in congregations with lower international giving, all else equal. The model s value for is with a very significant chi-square p-value of , meaning the error terms from the selection and outcome equation are correlated and there is good reason to use the Heckman procedure. In the selection equation, variables that significantly and positively affect selection include the congregation s income, the nearest congregation (measured in miles) that gives internationally, and whether or not the governing board has decisive influence within the congregation. 11

12 Table 2 Heckman selection results for international giving Model dependent variable: *100 Variables Coefficient Std. Err. p-value Sig. Outcome Equation Socio-demographic Income (millions) Income (millions) squared * Flag for imputed income Percent minority * Percent female Percent four-year degree * Percent over Urban Suburban Community Involvement Social interactions (log) Pastor participation (log) Paid staff assigned to comm ty (log) Provides social services Volunteer recruitment is easy Volunteer staff count (log) Other Seminary Has experienced conflict Nearest international giver (dollars) Nearest international giver (miles) Constant ** Selection Equation Socio-demographic Income ** Flag for imputed income Percent four-year degree Percent over Urban ** Suburban Nearest international giver (dollars) Nearest international giver (miles) * Has any volunteers Governing board influence ** Constant * Sample Size 378 Censored Observations 69 Uncensored Observations 309 Rho Lambda Sigma Wald test of independent equations chi2=8.48 Prob>chi2=

13 Marginal effects Below in Table 3 are the marginal effects results for both the local and international giving models. The marginal effects calculate the combined effects of variables included in both the outcome and selection equation, conditional on the dependent variable being observed or selected. After accounting for selection, income and income squared maintain their signs and significance in both equations, and the magnitude of their effect increases. The rest of the variables appear to maintain the same overall effect implied in the separate equations. In the international giving equation, the percent of the congregation with four-year degrees and the percent of the congregation over sixty years of age are shared between the selection and outcome equation and experience no real change in their coefficients after the marginal effect transformation. Again, the negative coefficients on education (-0.026) and age (-0.023) differ from common findings of much of the household giving literature; dealing with a larger corporate unit of analysis changes the role of these demographic variables. 13

14 Table 3 Marginal effects of local giving and international giving models Model dependent variable Variables dy/dx p-value dy/dx p-value Income (millions) * Income (millions) squared * Flag for imputed income Percent minority * Percent female * Percent four-year degree * Percent over Urban Suburban Social interactions (log) Total programs (log) Pastor participation (log) Staff assigned to comm ty (log) ** Provides social services School involvement Volunteer recruitment is easy Has volunteer staff Volunteer staff count (log) Seminary Has experienced conflict Close giver (amount) Close giver (distance) Governing board influence * Congregational efficacy Mainline denomination Joint service projects Local interest Estimator MFX y(conditional) MFX y(conditional) Significance: + p < 0.10; * p < 0.05; ** p < 0.01; *** p <

15 Income elasticity Here we split congregations into income deciles and examine both the raw and modeled data. Figure 1 shows the raw local and international giving data by income decile. In the raw data, the international giving curve wobbles up and down but remains relatively insensitive on average to changes in income. However, the local giving curve exhibits the U-shape that has been found in household giving data by Shervish and Havens (1995). Figure 1 Percent given by income decile Actual Local International 10% 9% 8% 7% 6% 5% 4% 3% 2% 1% 0% 8.8% 7.7% 6.5% 3.5% Income Decile Now, using the mean predicted values from the Heckman models, we draw the curve while controlling for all the model covariates (Figure 2). Just as McClelland and Brooks found using household data, after controlling for model covariates, the underlying U-shape mostly disappears. International giving still exceeds local giving as a proportion of budget by an increasing ratio, ending at more than 2 to 1 in the highest decile. Instead of the U-shape, we see a generally declining percentage of budget allocated to local giving, given the local giving model s negative coefficient for income and non-significant coefficient for income squared. We also see a somewhat upward-trending international giving curve, given the 15

16 positive coefficient for income. These results imply that congregations with higher incomes will donate a smaller portion of their budget locally and a larger portion of their budget internationally. Figure 2 Predicted percent of budget given by income decile with 95% CI bands Local International 8% 7% 6% 6.1% 6.4% 6.4% 5% 4% 3% 3.9% 3.5% 2.6% 2% 1% 0% Income decile Estimating the income elasticity using the same method as McClelland and Brooks (2004), we use the following equation: (4) where and are income and income squared, is a vector of the means of the covariates and is a vector of their parameters. Unlike McClelland and Brooks, we include covariates in our elasticity model to control for possible confounding variables. We hold all the other variables constant at their means and vary income in increments of $100,000. Figure 3 displays the resulting elasticity curves from both models. 16

17 100, , , , , , , , ,000 1,000,000 1,100,000 1,200,000 1,300,000 1,400,000 1,500,000 1,600,000 1,700,000 1,800,000 1,900,000 2,000,000 Income elasticity Figure 3 Income elasticity by income category International Local > 1 = "Luxury good" > 0 and < 1 = "Normal good" < 0 = "Inferior good" Income in dollars Our local giving elasticity results exhibit a different pattern than the household analysis of McClelland and Brooks, which finds a downward sloping curve indicating that high income households tend to react less strongly to income increases than low-income households. We find that local giving begins on the inelastic portion of the curve and then becomes increasingly more inelastic, or increasingly less responsive to increases in income, until it reaches an inflection point near the $1.5 million income level. Beyond $1.5 million, the income elasticity begins to move back towards one (however, the model includes only 15 congregations in this high-income bracket, so the finding is potentially sensitive to sample size). For the entire extent of the curve, local giving is a normal or necessary good; as income increases by 1%, consumption as a proportion of budget increases by less than 1%. The income elasticity for international giving exhibits a much different curve. It begins on the elastic portion of the curve and remains there for the duration of the curve. It, too, experiences an inflection point at an income level near $1.1 million. At this point, the 17

18 elasticity slowly begins to return to one. This means that international giving is a luxury good; as income increases by 1%, consumption of international giving as a proportion of budget increases by more than 1%. This discussion indicates that congregations at different income levels behave differently toward local and international giving. Conclusion This paper contributes to the literature of charitable religious giving by focusing on congregations, instead of households, as the unit of analysis. The results suggest that congregations and households behave differently when it comes to what drives giving; demographic variables are much less important for congregations. We also explore how congregations allocate funds between local and international giving and determine that these two types of giving do in fact behave differently: local giving appears to be a normal or necessary good, while international giving is a luxury good. Further analysis and research is necessary to fine tune the models and explore in greater detail the positive and normative implications of these findings. The minister who inspired this analysis can now rightly expect the study s scholars and sponsors to address the question: What can we do about this? How can we increase congregational support for local agencies and charities? 18

19 Appendix A: Codebook of variables used in models and raw means Variable Definition Mean Standard Deviation Local giving/budget (Local giving amount divided by the budget)* International (International giving amount divided by the budget)*100 giving/budget Income (millions) Income of congregation divided by one million Income (millions) Income (divided by one million) squared squared Flag for imputed 1 if income was imputed, 0 otherwise income Percent minority Percent of congregation that is non-white Percent female Percent of congregation that is female Percent four-year Percent of congregation with a four-year degree degree Percent over 60 Percent of congregation over 60 years old Urban 1 if urban location, 0 if suburban or rural Suburban 1 if suburban location, 0 if urban or rural Social interactions (log) Natural log of the number of congregation interactions Total programs (log) Natural log of the number of social programs Pastor participation Natural log pastor hours spent in community activity (log) Seminary 1 if pastor has seminary education, 0 otherwise Staff assigned to Natural log of paid community service staff members comm ty (log) Provides social services 1 if congregation provides social services, 0 otherwise School involvement 1 if congregation involved with a school, 0 otherwise Volunteer recruitment 1 if very easy to find volunteers, 0 otherwise is easy Has volunteer staff 1 if congregation has volunteer staff, 0 otherwise Volunteer staff count Natural log of the number of volunteer staff (log) Has experienced 1 if significant internal conflict in church in past two conflict years, 0 otherwise Nearest local giver Amount given locally by nearest congregation that gives (dollars) locally Nearest local giver Distance in miles of nearest congregation that gives (miles) locally Nearest international Amount given internationally by nearest congregation giver (dollars) that gives internationally Nearest international Distance in miles of nearest congregation that gives giver (miles) internationally Governing board 1 if governing board holds influence on major decisions, influence 0 if other lay leader or pastor hold influence Congregational efficacy 1 if congregation feels it can influence local government decisions, 0 otherwise Mainline denomination 1 if Mainline denomination, 0 otherwise Joint service projects 1 if participates in joint service projects, 0 otherwise Local Interest 1 if congregation has a local interest, 0 otherwise

20 Appendix B: Modeling strategy We tested two different model specifications. First, we tested standard ordinary least squares excluding the instances where the dependent variable was zero. Second, we tested a Heckman selection model. Ultimately, the authors chose the Heckman selection model as a more effective way to model charitable giving, because it takes into account the decision to give and then how much to give, thereby incorporating those that give nothing. Another decision made by the authors was to exclude observations in which the giving amount divided the budget was greater than 20 percent. The reason for this restriction is that 95 percent of the cases have a value below this level, indicating that values above 0.20 are likely outliers. This restriction forced us to drop 10 observations in the local giving model and 26 observations in the international giving model. To compensate for potential heteroskedasticity, we used Stata s robust command for all the Heckman selection models. We used imputed income variables. We regressed natural log of regular participants over 18 years old, natural log of total giving, natural log of staff members, percent of the congregation that is minority, and a dummy indicating if the congregation is rurally located on the natural log of income (R2 =.7877) and then generated the predicted values. We then generated a new variable that is equal to income if income is not missing and the predicted value otherwise. In the final analysis we gained only three observations, but we believe the results warranted it. If a congregation responded affirmatively to the question type In your most recent fiscal year, did your congregation provide financial support to any international projects, individual missionaries or aid workers? This might include outreach, missions, relief or development work and gave no information for How much, the congregation was coded as missing. 20

21 Acknowledgements: The authors thank Kurt Schaefer and Steve McMullen for helpful comments throughout the research and writing of this paper, and Kathryn Bardolph for proofreading the paper. We d also like to thank the Douglas and Maria DeVos Foundation for its generous funding of the Kent County Congregations Study. References Andreoni, J., Ed. (2001). The Economics of Philanthropy. International Encyclopedia of the Social and Behavioral Sciences. London, Elsevier. Auten, G. E., H. Sieg, et al. (2002). "Charitable Giving, Income, and Taxes: An Analysis of Panel Data." American Economic Review 92(1): Bekkers, R. (2006). "Traditional and Health-Related Philanthropy: The Role of Resources and Personality." Social Psychology Quarterly 69(4): Bekkers, R. and P. Wiepking (2006). "To Give or Not to Give, That Is the Question: How Methodology Is Destiny in Dutch Giving Data." Nonprofit and Voluntary Sector Quarterly 35(3): Bekkers, R. and P. Wiepking (2007). "Generosity and Philanthropy: A Literature Review." Social Science Research Network. Carlson, N. and E. L. Hernandez (2008). "Gatherings of Hope: How Religious Congregations Contribute to the Quality of Life in Kent County." Carrol, J., S. McCarthy, et al. (2005). "An Econometric Analysis of Charitable Donations in the Republic of Ireland." Economic and Social Review 36(3): Golder, M. (2010). "Selection Models." Retrieved 4 June, 2010, from Heckman, J. J. (1979). "Sample Selection Bias as a Specification Error." Econometrica 47(1): Hoge, R. (1994). "Introduction: The Problem of Understanding Church Giving." Review of Religious Research 36(2): McClelland, R. and A. C. Brooks (2004). "What Is the Real Relationship between Income and Charitable Giving?" Public Finance Review 32(5): Rooney, P. M., K. S. Seinberg, et al. (2001). "A Methodological Comparison of Giving Surveys: Indiana as a Test Case." Nonprofit and Voluntary Sector Quarterly 30(3): Rosborough, J. (2008). "A Theory of Congregational Giving." Schervish, P. G. and J. J. Havens (1995). "Explaining the curve in the U-shaped curve." Voluntas: International Journal of Voluntary and Nonprofit Organizations 6(2): Sigelman, L. and L. Zeng (1999). "Analyzing Censored and Sample-Selected Data with Tobit and Heckit Models." Political Analysis 8(2): Steinberg, R. (1990). "Taxes and giving: new findings." Voluntas: International Journal of Voluntary and Nonprofit Organizations 1(2): thearda.com. (2010). Retrieved 4 June, 2010, from Wiepking, P. (2007). "The Philanthropic Poor: In Search of Explanations for the Relative Generosity of Lower Income Households." Voluntas: International Journal of Voluntary and Nonprofit Organizations 18(4):

22 Zaleski, P. A., C. E. Zech, et al. (1992). "Determinants of Contributions to Religious Organizations: Free Riding and Other Factors." American Journal of Economics and Sociology 51(4):