## Abstract

This paper examines voluntary contributions to public education via charitable school foundations, booster clubs, parent teacher associations, and parent teacher organizations. We use panel data on school-supporting charities with national coverage from 1995 to 2010, which we geocode and match to school districts. We document the meteoric rise of school-supporting nonprofits during this panel, and then estimate a series of regression models to examine the distributional consequences of voluntary contributions. We find relatively large districts have higher probabilities of receiving revenues from a school-supporting nonprofit but the level of per-pupil voluntary contributions declines with student enrollment. In addition, we find school districts with higher endowments have higher probabilities of being served by at least one school-supporting nonprofit and higher levels of per-pupil contributions. Finally, we find no evidence that impressive recent growth in the number and financial size of these school-supporting charities relates to reductions in the public financing of schools.

## 5.  Distributional Consequences of Nonprofit Contributions

In this section, we examine the distributional consequences of voluntary contributions to public schools. The presence of school-supporting nonprofits and the level of voluntary contributions to public schools are not uniform. The uneven distribution of charitable resources enables us to examine demand heterogeneity across school districts and associated equity implications, including whether the distribution of nonprofit revenues results in funding inequalities. To answer these questions, we use panel data on school-supporting nonprofits linked to a complete panel of unified school district data for the years 1999 through 2009, in which we observe school district data obtained from the NCES CCD, the Local Education Agency (School District) Finance Survey (F-33), the Decennial Census, and the ACS. Excluding observations from 2010 in our sample due to changes in 501(c)(3) IRS filing requirements, our final analysis sample includes 13,058 unique school-supporting nonprofits that meet the criteria for sample inclusion described in section 3.

### Empirical Strategy

We first model the probability that a unified school district receives revenues from a school-supporting 501(c)(3) charity as a function of time-varying school district characteristics: the log of state revenues per pupil, the log of property tax revenues per pupil, the log of federal revenues per pupil, the proportion of English learner students, the proportion of students enrolled in free and reduced-price meal programs, the log of district enrollment, the proportion of minority residents, the proportion of non-U.S. citizens, the proportion of residents above age 65, the proportion of female-headed households, the proportion of households below the poverty threshold, the unemployment rate, the proportion of the population with a bachelor's degree or higher, the homeownership rate, and the log of median household income.

We begin with the following reduced form model:
1
The dependent variable is a dichotomous variable indicating whether district d receives any revenues from a school-supporting 501(c)(3) charity in year t (i.e., whether any organization files an IRS Form 990 in the district in a given year), X is a set of time-varying district characteristics, η denotes year fixed effects and captures systematic differences in IRS filings over time, and ϵ denotes the random error term. The coefficient β is the average effect of that district characteristic on the probability of the district having at least one school-supporting nonprofit in a given year. The model coefficients β are descriptive and not causal, because unobserved district characteristics are likely to be correlated with both the observed district characteristics (X) and with whether any organization files an IRS Form 990 in the district (Y). Thus, omitting these variables is likely to bias estimates of the effects of district characteristics on our dependent variable of interest. We estimate a linear probability model using ordinary least squares (OLS) and robust standard errors to account for heteroskedasticity resulting from the application of OLS techniques to limited dependent variables. We also cluster standard errors by district in all models to correct for within-district serial correlation of the errors; we assume independence of between-district errors.
A major benefit of using panel data for our analysis is that it enables us to reduce bias in our estimated coefficients by including a rich set of fixed effects to control for time-invariant observed and unobserved variation nested within units. Our data have broad national coverage and span eleven years, so we are able to modify equation 1 to include additional fixed effects:
2

In equation 2, γ denotes a set of fixed effects, which may be specified at either the state level s or at the school district level d. The fixed effects model reduces bias associated with time-invariant observed and unobserved differences in states (districts) that may be correlated with both the time-varying district characteristics and our dependent variable. Although this model reduces bias due to systematic differences in district characteristics across states (and in district characteristics over time), the estimated β coefficients may still be biased due to time-varying unobservables at the state (district) level. The inclusion of school district fixed effects in equation 2 requires that the β coefficients are estimated using within-district variation in district characteristics, and net of time trends. Such a parameterization reduces bias but is likely to yield relatively imprecise estimates of β. Alternatively, a less restrictive parameterization with fixed effects at the state level will yield β estimates that are more precise but relatively biased.12

### Results: Probability of Receiving Revenues from a School-Supporting Nonprofit

Table 6 reports results from our linear probability regressions modeling on whether a unified school district receives revenues from a school-supporting 501(c)(3) charity. Model I displays the reduced form estimates, model II displays results including state fixed effects, and model III displays results including school district (rather than state) fixed effects.

Table 6.
The Distribution of School-Supporting Nonprofits and Revenue
DV: Any OrganizationDV: Log of Per-Pupil Voluntary Contributions
IIIIIIIVVVI
ReducedStateDistrictReducedStateDistrict
FormFEFEFormFEFE
Log of state revenues −0.010 0.025* −0.009 −0.398*** −0.018 0.005
per pupil (in $100) (0.011) (0.012) (0.009) (0.050) (0.058) (0.040) Log of property tax revenues 0.019** 0.048*** 0.024** −0.230*** 0.115* 0.051 per pupil (in$100) (0.007) (0.009) (0.008) (0.043) (0.056) (0.055)
Log of federal revenues −0.015 −0.007 −0.003 0.128* 0.126* 0.050
per pupil (in $100) (0.009) (0.009) (0.006) (0.053) (0.049) (0.027) Proportion of English learner 0.333*** 0.267*** 0.094* 0.322 −0.054 −0.013 students (0.056) (0.062) (0.040) (0.429) (0.439) (0.230) Proportion of free/reduced −0.042 −0.088* −0.037 0.466 −0.035 −0.252 lunch students (0.038) (0.040) (0.029) (0.278) (0.287) (0.164) Log of enrollment 0.181*** 0.183*** 0.167*** −0.283*** −0.307*** −0.618* (0.004) (0.005) (0.030) (0.025) (0.027) (0.250) Proportion minority −0.001** −0.001 0.002 −0.008*** −0.012*** 0.006 (0.000) (0.000) (0.001) (0.003) (0.003) (0.007) Proportion aged 65 and over 0.010*** 0.009*** 0.006 0.053*** 0.053*** −0.005 (0.001) (0.001) (0.004) (0.009) (0.009) (0.022) Proportion female head −0.003** −0.001 −0.002 −0.003 0.019* 0.003 of households (0.001) (0.001) (0.002) (0.008) (0.008) (0.009) Proportion below poverty line 0.001 0.002 0.003* 0.019* 0.023** −0.007 (0.001) (0.001) (0.001) (0.009) (0.009) (0.012) Proportion unemployed −0.001 −0.003** 0.000 −0.029*** −0.025*** 0.010 (0.001) (0.001) (0.001) (0.006) (0.006) (0.009) Proportion non-U.S. citizens −0.002 −0.003** −0.003 0.009 0.017 0.010 (0.001) (0.001) (0.002) (0.009) (0.009) (0.015) Proportion with bachelor's 0.006*** 0.007*** 0.005*** 0.024*** 0.022*** 0.007 degree or higher (0.001) (0.001) (0.001) (0.003) (0.003) (0.007) Proportion homeowners −0.004*** −0.002*** −0.000 −0.008* −0.004 −0.009 (0.001) (0.001) (0.001) (0.003) (0.004) (0.009) Log of median 0.240*** 0.200*** 0.079 0.837*** 0.902*** 0.844* household income (0.040) (0.045) (0.058) (0.235) (0.256) (0.406) 2000 −0.011 0.001 0.004 0.023 0.048 0.083** (0.006) (0.005) (0.005) (0.035) (0.031) (0.030) 2001 −0.019** −0.008 0.005 0.048 0.066 0.143*** (0.007) (0.007) (0.007) (0.039) (0.038) (0.038) 2002 −0.003 −0.000 0.019* 0.042 0.093* 0.200*** (0.009) (0.009) (0.009) (0.047) (0.046) (0.053) 2003 −0.003 −0.001 0.030** 0.085 0.137* 0.272*** (0.010) (0.011) (0.011) (0.055) (0.056) (0.069) 2004 0.001 −0.000 0.038** 0.178** 0.205** 0.360*** (0.012) (0.013) (0.013) (0.065) (0.067) (0.086) 2005 −0.001 −0.003 0.042** 0.236** 0.230** 0.430*** (0.014) (0.015) (0.016) (0.077) (0.080) (0.104) 2006 0.000 −0.001 0.053** 0.301*** 0.300** 0.513*** (0.016) (0.017) (0.019) (0.087) (0.091) (0.121) 2007 0.009 0.010 0.073*** 0.344*** 0.342*** 0.594*** (0.018) (0.020) (0.021) (0.099) (0.104) (0.139) 2008 0.009 0.016 0.088*** 0.327** 0.325** 0.623*** (0.021) (0.023) (0.024) (0.114) (0.120) (0.157) 2009 −0.041 −0.037 0.034 0.131 0.184 0.460** (0.024) (0.026) (0.027) (0.131) (0.136) (0.171) Constant −3.483*** −3.736*** −2.162** 0.468 −6.273* −1.981 (0.431) (0.503) (0.701) (2.407) (2.692) (4.575) Observations 37,886 37,886 37,886 13,938 13,938 13,938 Pseudo-R2 0.445 0.463 0.079 0.272 0.331 0.213 R2 0.444 0.463 0.079 0.271 0.328 0.211 DV: Any OrganizationDV: Log of Per-Pupil Voluntary Contributions IIIIIIIVVVI ReducedStateDistrictReducedStateDistrict FormFEFEFormFEFE Log of state revenues −0.010 0.025* −0.009 −0.398*** −0.018 0.005 per pupil (in$100) (0.011) (0.012) (0.009) (0.050) (0.058) (0.040)
Log of property tax revenues 0.019** 0.048*** 0.024** −0.230*** 0.115* 0.051
per pupil (in $100) (0.007) (0.009) (0.008) (0.043) (0.056) (0.055) Log of federal revenues −0.015 −0.007 −0.003 0.128* 0.126* 0.050 per pupil (in$100) (0.009) (0.009) (0.006) (0.053) (0.049) (0.027)
Proportion of English learner 0.333*** 0.267*** 0.094* 0.322 −0.054 −0.013
students (0.056) (0.062) (0.040) (0.429) (0.439) (0.230)
Proportion of free/reduced −0.042 −0.088* −0.037 0.466 −0.035 −0.252
lunch students (0.038) (0.040) (0.029) (0.278) (0.287) (0.164)
Log of enrollment 0.181*** 0.183*** 0.167*** −0.283*** −0.307*** −0.618*
(0.004) (0.005) (0.030) (0.025) (0.027) (0.250)
Proportion minority −0.001** −0.001 0.002 −0.008*** −0.012*** 0.006
(0.000) (0.000) (0.001) (0.003) (0.003) (0.007)
Proportion aged 65 and over 0.010*** 0.009*** 0.006 0.053*** 0.053*** −0.005
(0.001) (0.001) (0.004) (0.009) (0.009) (0.022)
Proportion female head −0.003** −0.001 −0.002 −0.003 0.019* 0.003
of households (0.001) (0.001) (0.002) (0.008) (0.008) (0.009)
Proportion below poverty line 0.001 0.002 0.003* 0.019* 0.023** −0.007
(0.001) (0.001) (0.001) (0.009) (0.009) (0.012)
Proportion unemployed −0.001 −0.003** 0.000 −0.029*** −0.025*** 0.010
(0.001) (0.001) (0.001) (0.006) (0.006) (0.009)
Proportion non-U.S. citizens −0.002 −0.003** −0.003 0.009 0.017 0.010
(0.001) (0.001) (0.002) (0.009) (0.009) (0.015)
Proportion with bachelor's 0.006*** 0.007*** 0.005*** 0.024*** 0.022*** 0.007
degree or higher (0.001) (0.001) (0.001) (0.003) (0.003) (0.007)
Proportion homeowners −0.004*** −0.002*** −0.000 −0.008* −0.004 −0.009
(0.001) (0.001) (0.001) (0.003) (0.004) (0.009)
Log of median 0.240*** 0.200*** 0.079 0.837*** 0.902*** 0.844*
household income (0.040) (0.045) (0.058) (0.235) (0.256) (0.406)
2000 −0.011 0.001 0.004 0.023 0.048 0.083**
(0.006) (0.005) (0.005) (0.035) (0.031) (0.030)
2001 −0.019** −0.008 0.005 0.048 0.066 0.143***
(0.007) (0.007) (0.007) (0.039) (0.038) (0.038)
2002 −0.003 −0.000 0.019* 0.042 0.093* 0.200***
(0.009) (0.009) (0.009) (0.047) (0.046) (0.053)
2003 −0.003 −0.001 0.030** 0.085 0.137* 0.272***
(0.010) (0.011) (0.011) (0.055) (0.056) (0.069)
2004 0.001 −0.000 0.038** 0.178** 0.205** 0.360***
(0.012) (0.013) (0.013) (0.065) (0.067) (0.086)
2005 −0.001 −0.003 0.042** 0.236** 0.230** 0.430***
(0.014) (0.015) (0.016) (0.077) (0.080) (0.104)
2006 0.000 −0.001 0.053** 0.301*** 0.300** 0.513***
(0.016) (0.017) (0.019) (0.087) (0.091) (0.121)
2007 0.009 0.010 0.073*** 0.344*** 0.342*** 0.594***
(0.018) (0.020) (0.021) (0.099) (0.104) (0.139)
2008 0.009 0.016 0.088*** 0.327** 0.325** 0.623***
(0.021) (0.023) (0.024) (0.114) (0.120) (0.157)
2009 −0.041 −0.037 0.034 0.131 0.184 0.460**
(0.024) (0.026) (0.027) (0.131) (0.136) (0.171)
Constant −3.483*** −3.736*** −2.162** 0.468 −6.273* −1.981
(0.431) (0.503) (0.701) (2.407) (2.692) (4.575)
Observations 37,886 37,886 37,886 13,938 13,938 13,938
Pseudo-R2 0.445 0.463 0.079 0.272 0.331 0.213
R2 0.444 0.463 0.079 0.271 0.328 0.211

Note: Standard errors in parentheses. DV = dependent variable; FE = fixed effect.

***Statistically significant at the 1% level;   **statistically significant at the 5% level;   *statistically significant at the 10% level.

Coefficients on the year fixed effects confirm the descriptive trends in table 5. In the district fixed effects model, the probability that a school district receives revenues from any school-supporting nonprofit increases monotonically from 1999 through 2008, before dropping off in 2009, and the probability is 0.088 higher in 2008 as compared to the 1999 baseline year.13

In the reduced form, state fixed effect, and district fixed effect models, the probability that any school-supporting nonprofit files an IRS Form 990 is significantly higher in districts with higher property tax revenues per pupil, higher enrollments, higher proportions of English learner students, and higher shares of residents with at least a bachelor's degree. The probabilities generally are less variable within districts over time, as we observe in the more restrictive district fixed effects model; estimated coefficients remain positive and statistically significant, however.

Coefficients on other school district characteristics are less consistent across model specifications. For example, the share of senior residents and district median household income levels are associated with significantly higher probabilities of receiving revenues from a school-supporting nonprofit, but only in the reduced form and state fixed effects models; within-district variation may not be sufficiently large over time to identify the effect precisely in the district fixed effects model. State revenues per pupil are associated positively and significantly with the probability of receiving revenues from a school-supporting nonprofit in the state fixed effects model. In the district fixed effects model, higher shares of residents living below the poverty line also are associated positively and significantly with the probability of a district being served by a nonprofit organization. The proportion of female-headed households is associated with significantly lower probabilities of receiving school-supporting nonprofit revenues in the reduced form model but the effect is not statistically significant in the state or district fixed effects models. Finally, the share of students in poverty (as measured by enrollment in free- and reduced-price meal programs), the proportion of non-U.S. citizens, the unemployment rate, and the homeownership rate are associated with significantly lower probabilities of receiving school-supporting nonprofit revenues in the state fixed effect models, but the effects are not statistically significant in the reduced form or district fixed effects models.14

Taken together, the results indicate that relatively large districts with higher property tax (local, own-source) revenues per pupil, higher state revenues per pupil, relatively educated and wealthy residents, lower unemployment, and lower shares of students in poverty, female-headed households, and non-U.S. citizens, have higher probabilities of receiving revenues from a school-supporting nonprofit. The coefficient estimate signs, magnitudes, and statistical significance are consistent with the theory that communities with larger endowments exhibit a greater capacity to co-produce public services and address government failure.15

### Results: Modeling the Level of Per-Pupil Voluntary Contributions

We next turn to modeling the level of per-pupil voluntary contributions at the district level as a function of school district characteristics. We adapt the reduced form and fixed effects specifications in equations 1 and 2, changing the dependent variable Y from a dichotomous outcome to a continuously defined measure of the log of nonprofit revenues per pupil, calculated using total revenues across all nonprofits in the school district and dividing by the total number of enrolled students. We estimate our models using OLS. Because our dependent variable is no longer dichotomous, we no longer encounter problems in interpreting unbounded predicted coefficients. In addition, heteroskedasticity is of less concern with a continuously defined dependent variable (though we continue to use robust standard errors in our models and to cluster the errors by district to correct for serial autocorrelation).

We run our district-level models of per-pupil revenue conditional on the sample of districts that received revenues from a school-supporting organization (i.e., the sample of districts for which the dependent variable was equal to one in our first set of regressions in table 6). This specification raises two issues. First, our data are left-censored because we do not observe revenues from school-supporting nonprofits that do not file with the IRS, and so (1) our analysis sample is likely to be truncated and (2) we are likely to underestimate total per-pupil nonprofit revenues (Figlio and Kenny 2009). We are unable to use a Tobit specification to account for this censoring, however, because the likelihood estimator for fixed effects is biased and inconsistent. Second, the selection of school districts into our sample is nonrandom and will lead to biased coefficient estimates, but we are unable to use the results from our first set of regressions in table 6 in a standard Heckman selection model because we do not observe variables in our data set that form a valid exclusion restriction. Therefore, coefficient estimates should be interpreted with caution as descriptive rather than causal.

Table 6 also reports our results modeling the log of per-pupil voluntary contributions. Model IV displays the reduced form estimates, model V displays results including state fixed effects, and model VI displays results including school district (rather than state) fixed effects. Coefficients on the year fixed effects confirm the time trends reported in table 4. Per-pupil voluntary contributions increase monotonically in each year from 1999 through 2008 (and decline slightly in 2009). Notably, the time trends are the most sizable and statistically significant in the district fixed effects regression, suggesting that time trends within districts explain large differences in voluntary contributions. The district fixed effect regression suggests that within districts, average per-pupil voluntary contributions increased by 62.3 percent from 1999 through 2009.

Our results also provide support for Brunner and Sontelie's (2003) model of partial cooperation, which posits that the marginal price of voluntary contributions increases with the number of students enrolled in the district. Using California data, Brunner and Sonstelie (2003) find a negative cross-sectional relationship between per-pupil donations and enrollment, and Brunner and Imazeki (2005) confirm this relationship using an updated panel data set but without additional controls. In our data set, which includes national panel data from 1999 through 2009, we find a sizable and statistically significant negative relationship between per-pupil voluntary contributions—as measured by the revenues reported by nonprofits filing IRS Form 990—and the log of district enrollment in our reduced form and fixed effects models.

In our models of per-pupil voluntary contributions, coefficients on the remaining time-varying district characteristics shed additional light on the distributional consequences of nonprofit contributions. In all specifications, we find a positive and statistically significant relationship between median household income and the level of per-pupil voluntary contributions; we also find the share of residents with at least a bachelor's degree positively and significantly predicts per-pupil voluntary contributions in both the reduced form and state fixed effects specifications. In the reduced form and state fixed effects specifications, we also find the unemployment rate and the share of minority residents negatively and significantly predict per-pupil voluntary contributions. These coefficient estimates also are consistent with the theory that communities with larger endowments exhibit a greater capacity to coproduce public services and address government failure.16

Interestingly, the reduced form model finds state revenues per pupil and property tax revenues per pupil negatively and significantly predict per-pupil voluntary contributions, and federal revenues per pupil positively and significantly predict per-pupil voluntary contributions. The coefficient on state revenues per pupil is insignificant in both the state and district fixed effects models, however, whereas the coefficient on property tax revenues per pupil becomes positive and statistically significant in the state fixed effects model, and the coefficient on federal revenues per pupil remains positive and statistically significant in the state fixed effects model. In other words—after accounting for time-invariant, cross-state differences in local, state, and federal contributions to public schools—we find school districts with higher local own-source revenues and higher federal funding on a per-pupil basis generate larger average voluntary contributions. Within districts, however, increases in property tax revenues and federal revenues over time do not predict variation in per-pupil voluntary contributions.

### Summary of Main Results

Taken together, the results indicate substantial growth in both the presence of school-supporting nonprofits and in the level of per-pupil voluntary contributions from 1999 through 2009. We also find that school districts with higher endowments—as measured by property tax revenues per pupil, the share of individuals with a bachelor's degree or more, median household income, and relatively low unemployment rates—have higher probabilities of being served by at least one school-supporting nonprofit and higher levels of per-pupil contributions. We also find that while school districts with relatively high student enrollments have higher probabilities of being served by a school-supporting nonprofit, the level of per-pupil voluntary contributions declines with enrollment.

We also find no evidence that voluntary contributions offset reductions in the public financing of public schools. In fact, we find the opposite in our state and district fixed effects models: School districts with higher levels of per-pupil state and property tax revenues have a higher probability of being served by a school-supporting nonprofit, and school districts with higher per-pupil revenues from property tax and federal sources generate higher average levels of per-pupil voluntary contributions. Collectively, these findings imply voluntary contributions do not constitute an efficient or stable substitute for the financing of K–12 public education. Rather, they serve to enhance spending in school districts that already receive significantly larger per-pupil revenues.

## Notes

1.

Zimmer, Krop, and Brewer (2003) use slightly different categorizations to characterize the voluntary contributions of nongovernmental organizations to California public schools.

2.

The filing threshold increased to $50,000 in gross revenues in 2010. 3. Figlio and Kenny's (2009) survey data allow them to observe total voluntary contributions, including those from organizations that do not file a Form 990 (i.e., who have gross revenues less than$25,000).

4.

Although the NCCS created the National Taxonomy of Exempt Entities (NTEE) coding protocols to provide a standardized means of identifying charitable organizations by their purpose, Guidestar assigns different codes and Gazley (2011) found that relying solely on select NTEE codes to identify these government-supporting entities was insufficient to capture all cases. In the absence of consistent identifiers, a keyword search was performed on the approximately 65,000 entities categorized under the NTEE code B for Educational Institutions (this category also includes libraries). Separately and in combination, keywords such as “friends,” “education,” “foundation,” “school,” “school district,” “booster,” “parent,” “PTO,” and “trust” were used to identify the school-supporting nonprofits.

5.

Our tabulation of school-supporting nonprofits differs from the NCCS tabulations. For example, we report 2,116 foundations in our data, compared with the NCCS estimate of 9,093 registered foundations in the education category. This discrepancy is due to definitional differences. The NCCS figures include education foundations spanning 29 categories of support, including management and technical assistance, single organization support, fundraising and fund distribution, libraries, parent teacher groups, and scholarships and student financial aid. A good example of how the NCCS data tabulations differ from ours is found within the B11 single organization support subcategory, which lists over 1,500 organizations and is composed mainly of athletic and music booster clubs. Our data tabulates these organizations separately and we do not include them within the count of school foundations. Further, the scholarships and student financial aid subcategory includes over 5,000 organizations. We exclude many of these organizations from our data set because they do not provide direct support for school or district operations (e.g., their stated missions are to provide college scholarships to graduating high school seniors).

6.

Hansen (2008) notes that beginning in 2006 the U.S. Census Bureau's survey of school system finances requires school districts to report contributions and donations from private sources. Unfortunately, these data are not available for the full panel in our study and are not disaggregated by source.

7.

Charter schools are considered school districts in some states. We exclude charter school districts from our analysis due to differences in funding charter versus traditional public schools.

8.

Though F-33 school finance data are available for the entire panel from 1995 to 2010, we only merge in school finance data for the years in which we also observe school district demographic data obtained from the NCES Common Core of Data (1999–2010).

9.

As a robustness check, we also assigned the 2006–10 five-year ACS estimates to the year 2010 and re-ran all reported analyses. Results are qualitatively similar and available from the authors upon request.

10.

Table 2 reports that 2.3 percent of charities in our data set support private schools. We exclude these charities from our analysis sample because they do not support public schools or districts.

11.

In 2008 the IRS began to adjust, over three years, the threshold reporting requirements for charities to file a Form 990 or 990EZ. The most significant change for our analysis is that registered charities with annual gross receipts of $25,000 or more and assets of$1,250,000 or more were required to file a 990 or 990EZ in 2009. That threshold changed to $50,000 in receipts or$500,000 in assets in 2010. All else equal, a higher reporting threshold will deflate the growth in revenues and in numbers of school-supporting charitable organizations reported in our trend data.

12.

Linear probability models are prone to problems arising from unbounded predicted values. As we observe in our models, estimated coefficients may take on values that are greater than one and less than negative one. However, we prefer the OLS specification to the standard logit or probit specification because it is not prone to the incidental parameters problem encountered when using maximum likelihood techniques in combination with fixed effects. In untabulated results, we run probit specifications without fixed effects to test the robustness of our reduced form estimates to nonlinear transformations. We obtain qualitatively similar results.

13.

As we discuss in section 4, part of the decline in 2009 may be due to delays in IRS tax filings.

14.

In untabulated results, we also model separately the probabilities that school districts receive revenues from any PTA/PTO, booster club, or school foundation. Results from the reduced form, state fixed effect, and district fixed effect specifications are qualitatively similar to those reported in table 6, especially for the set of regressions modeling the probability that school districts receive revenues from any PTA/PTO (the most ubiquitous form of school-supporting nonprofit organization in our data set). Results are available from the authors upon request.

15.

It is more difficult to interpret the positive coefficient on the proportion of English learners in all model specifications, the positive coefficient on the share of senior residents in the reduced form and state fixed effects models, the positive coefficient on the share of residents below the poverty line in the district fixed effects regression, and the negative coefficient on homeownership rate in the reduced form and state fixed effects regressions. These variable relationships generally are not associated with greater capacity, and additional research on demand heterogeneity is needed to explain these results.

16.

It is more difficult to interpret the positive coefficients on the share of senior residents and on the share of residents below the poverty line in the reduced form and state fixed effects models, the positive coefficient on the share of female-headed households in the district fixed effects regression, and the negative coefficient on homeownership rate in the reduced form model. These variable relationships generally are not associated with greater capacity, and additional research on demand heterogeneity is needed to explain these results.

17.

We report all dollar figures using inflation-adjusted 2000 dollars.

18.

Figlio and Kenny (2009) discuss the limitations of IRS Form 990 data in producing accurate estimates of voluntary contributions to public schools.

19.

Low average levels of per-pupil funding also may belie the influential effect of school-supporting nonprofit activity across districts that may only be observed in a less parametric context (e.g., in quintile regressions). We leave this possibility to future research.

## Acknowledgments

The authors thank the University of Texas’ RGK Center for Philanthropy and Community Service and the Association for Research on Nonprofit Organization and Voluntary Action (ARNOVA) for financial support awarded via the RGK Center-ARNOVA Presidents Award for 2012. The authors also thank David Warren, Ed Gerrish, Thomas Sugimoto, Mir Usman Ali, Quentin Ball, Lynn Nguyen, and Megan Welch for excellent research assistance.

## REFERENCES

Alm
,
James
, and
David L.
Sjoquist
.
2009
.
The response of local school systems in Georgia to fiscal and economic conditions
.
Journal of Education Finance
35
(
1
):
60
84
. doi:10.1353/jef.0.0000
Andreoni
,
James
.
1988
.
Privately provided public goods in a large economy: The limits of altruism
.
Journal of Public Economics
35
(
1
):
57
73
. doi:10.1016/0047-2727(88)90061-8
Andreoni
,
James
.
1989
.
Giving with impure altruism: Applications to charity and Ricardian equivalence
.
Journal of Political Economy
97
(
6
):
1447
1458
. doi:10.1086/261662
Andreoni
,
James
.
1990
.
Impure altruism and donations to public goods: A theory of warm-glow giving
.
Economic Journal
100
(
401
):
464
477
. doi:10.2307/2234133
Bergstrom
,
Ted
,
Larry
Blume
, and
Hal
Varian
.
1986
.
On the private provision of public goods
.
American Economic Review
63
(
3
):
280
296
.
Brandsen
,
Taco
, and
Philip
Marcel
Karré
.
2011
.
Hybrid organizations: No cause for concern?
34
(
13
):
827
836
. doi:10.1080/01900692.2011.605090
Brudney
,
Jeffrey L.
1987
.
Coproduction and privatization: Exploring the relationship and its implications
.
Nonprofit and Voluntary Sector Quarterly
16
(
3
):
11
21
. doi:10.1177/089976408701600303
Brudney
,
Jeffrey L.
, and
Robert E.
England
.
1983
.
Toward a definition of the coproduction concept
.
43
(
1
):
59
65
. doi:10.2307/975300
Brunner
,
Eric
, and
Jennifer
Imazeki
.
2005
.
Fiscal stress and voluntary contributions to public schools
. In
Developments in school finance
:
2004
,
edited by
W. J.
Fowler
, pp.
39
54
.
Washington, DC
:
NCES
.
Brunner
,
Eric
, and
Sonstelie
John
.
2003
.
School finance reform and voluntary fiscal federalism
.
Journal of Public Economics
87
(
9–10
):
2157
2185
. doi:10.1016/S0047-2727(02)00040-3
Cohen
,
Rick
.
2012
.
Philanthropy funding government work? There's a foundation for that—several, actually
.
Nonprofit Quarterly
,
13 April
. Available https://nonprofitquarterly.org/policysocial-context/20145-philanthropy-funding-government-work-theres-a-foundation-for-thatseveral-actually.html. Accessed 21 July 2014.
Cornes
,
Richard
, and
Todd
Sandler
.
1984
.
Easy riders, joint production, and public goods
.
Economic Journal
94
(
275
):
580
598
. doi:10.2307/2232704
Dixon
,
Mark
.
2012
.
Public education finances 2010: Governments division reports. United States Census Bureau
. Available http://www2.census.gov/govs/school/10f33pub.pdf. Accessed 14 May 2014.
Dye
,
Richard F.
, and
Andrew
Reschovsky
.
2008
.
Property tax responses to state aid cuts in the recent fiscal crisis
.
Public Budgeting & Finance
28
(
2
):
87
111
. doi:10.1111/j.1540-5850.2008.00906.x
Figlio
,
David N.
, and
Kenny
Lawrence W.
.
2009
.
Public sector performance management and stakeholder support
.
Journal of Public Economics
93
(
9–10
):
1069
1077
. doi:10.1016/j.jpubeco.2009.07.003
Gazley
,
Beth
.
2011
.
Cooptation, coproduction, or competition? A preliminary empirical look at government-supporting NPOs
. Paper presented at the International Research Society for Public Management,
Dublin, Ireland
, April.
Glazer
,
Amihai
, and
Kai
.
1996
.
A signaling explanation for charity
.
American Economic Review
86
(
4
):
1019
1028
.
Greene
,
Jay
.
2005
.
Buckets into the sea: Why philanthropy isn't changing schools, and how it could
. In
With the best of intentions
, edited by
Frederick
M.
Hess
, pp.
49
76
.
Cambridge, MA
:
Harvard Education Press
.
Hansen
,
Janet S
.
2008
.
The role of nongovernmental organizations in financing public schools
.
In Handbook of research in education finance and policy
, edited by
Helen
F.
and
Edward B.
Fiske
, pp.
314
331
.
New York
:
Routledge
.
Harbaugh
,
William T.
1998
.
What do donations buy? A model of philanthropy based on prestige and warm glow
.
Journal of Public Economics
67
(
2
):
269
284
. doi:10.1016/S0047-2727(97)00062-5
Irvin
,
Renee A.
, and
Patrick
Carr
.
2005
.
The role of philanthropy in local government finance
.
Public Budgeting & Finance
25
(
3
):
33
47
. doi:10.1111/j.1540-5850.2005.00366.x
Jetté
,
Christian
, and
Yves
Vaillancourt
.
2011
.
Social economy and home care services in Quebec: Co-production or co-construction?
Voluntas
22
(
1
):
48
69
. doi:10.1007/s11266-010-9134-8
Johnson
,
Nicholas
, and
Michael
Leachman
.
2013
.
Four big threats to state finances could undermine future U.S. prosperity
.
Washington, DC
:
Center for Budget and Policy Priorities
.
Lampkin
,
Linda
, and
David
Stern
.
2003
.
Who helps public schools: A portrait of local education funds
,
1991
2001
.
Washington, DC
:
The Urban Institute
. Available www.urban.org/UploadedPDF/410915_PENReport.pdf. Accessed 28 July 2014.
Lecy
,
Jesse D.
, and
David M.
Van Slyke
.
2013
.
Nonprofit sector growth and density: Testing theories of government support
.
Journal of Public Administration: Research and Theory
23
(
1
):
189
214
. doi:10.1093/jopart/mus010
Ledyard
,
John O
.
1995
.
Public goods: A survey of experimental research
. In
Handbook of experimental economics
, edited by
John H.
Kagel
and
Alvin E.
Roth
, pp.
111
194
.
Princeton, NJ
:
Princeton University Press
.
Matsunaga
,
Yoshiho
, and
Yamauchi
Naoto
.
2004
.
Is the government failure theory still relevant? A panel analysis using U.S. state level data
.
Annals of Public and Cooperative Economics
75
(
2
):
227
263
. doi:10.1111/j.1467-8292.2004.00251.x
Mizrahi
,
Shlomo
.
2011
.
Self-provision of public services: Its evolution and impact
.
71
(
2
):
285
291
.
National Center for Education Statistics (NCES)
.
2013
.
National public education financial survey data
. Available http://nces.ed.gov/ccd/stfis.asp. Accessed 14 May 2014.
Oliff
,
Phil
,
Chris
Mai
, and
Vincent
Palacios
.
2012
.
States continue to feel recession's impact
.
Washington, DC
:
Center on Budget and Policy Priorities
.
Olson
,
Mancur
, and
Richard
Zeckhauser
.
1966
.
An economic theory of alliances
.
Review of Economics and Statistics
48
(
3
):
266
279
. doi:10.2307/1927082
O’Toole
,
Laurence J.
, and
Kenneth J.
Meier
.
2004
.
Desperately seeking Selznick: Cooptation and the dark side of public management in networks
.
64
(
6
):
681
693
. doi:10.1111/j.1540-6210.2004.00415.x
Parks
,
Roger B.
,
Paula D.
Baker
,
Larry
Kiser
,
Ronald
Oakerson
,
Elinor
Ostrom
,
Vincent
Ostrom
,
Stephen L.
Percy
,
Martha B.
Vandivort
,
Gordon P.
Whitaker
, and
Rick
Wilson
.
1981
.
Consumers as coproducers of public services: Some economic and institutional considerations
.
Policy Studies Journal
9
(
7
):
1001
1011
. doi:10.1111/j.1541-0072.1981.tb01208.x
Reschovsky
,
Andrew
.
2014
.
The future role of the property tax in the funding of K–12 education in the United States
. In
Local Education, Land, and Location
, edited by
Gregory
Ingram
and
Daphne
Kenyon
.
Cambridge, MA
:
Lincoln Institute of Land Policy
.
Roberts
,
Russell D
.
1984
.
A positive model of private charity and wealth transfers
.
Journal of Political Economy
92
(
1
):
136
148
. doi:10.1086/261212
Salamon
,
Lester M
.
1987
.
Of market failure, voluntary failure, and third-party government: Toward a theory of government-nonprofit relations in the modern welfare state
.
Nonprofit and Voluntary Sector Quarterly
16
(
1–2
):
29
49
. doi:10.1177/089976408701600104
Steinberg
,
Richard
.
1987
.
Voluntary donations and public expenditures in a federalist system
.
American Economic Review
77
(
1
):
24
36
.
Su
,
Eleanor Yang
.
2012a
.
Public schools increasingly rely on private donations
.
California Watch Daily Report
, 19 January.
Su
,
Eleanor Yang
.
2012b
.
Foundations funding more basic school needs: California K–12 foundations, PTAs and booster clubs raised about 1.3 billion
.
Berkeley, CA
: Center for Investigative Reporting.
Sugden
,
Robert
.
1982
.
On the economics of philanthropy
.
Economic Journal
92
(
366
):
341
350
. doi:10.2307/2232444
Warr
,
Peter G
.
1983
.
The private provision of a public good is independent of the distribution of income
.
Economics Letters
13
(
3
):
207
211
. doi:10.1016/0165-1765(83)90087-3
Warren
,
Robert
.
1987
.
Coproduction, volunteerism, privatization, and the public interest
.
Nonprofit and Voluntary Sector Quarterly
16
(
3
):
5
10
. doi:10.1177/089976408701600302
Whitaker
,
Gordon P
.
1980
.
Coproduction: Citizen participation in service delivery
.
40
(
3
):
240
246
. doi:10.2307/975377
Young
,
Dennis R
.
2000
.
Alternative models of government-nonprofit sector relations: Theoretical and international perspectives
.
Nonprofit and Voluntary Sector Quarterly
29
(
1
):
149
172
. doi:10.1177/0899764000291009
Zimmer
,
Ron
,
Krop
Cathy
, and
Brewer
Dominic
.
2003
.
Private resources in public schools: Evidence from a pilot study
.
Journal of Education Finance
28
(
4
):
485
521
.