# A new cost efficiency measure for not-for-profit firms: evidence of a link between inefficiency and large endowments.

IntroductionResearch on the performance of not-for-profit organizations has typically focused on their delivery of program services. The National Center for Chantable Statistics, a primary source of data on these organizations, collects information reported on the IRS Form 990 filed by these organizations. While these organizations do not generally pay taxes, most of them must nevertheless file the Form 990 annually with the IRS. It contains information derived from the income statement and balance sheet and divides all expenditures into three broad types: program, management, and fundraising. Research using the Form 990 data to evaluate the performance of not-for-profit firms often employs the ratio of program expenses to total expenses to gauge the effectiveness of organizations in delivering services. For example, Core et al. (2006) employ the program expense ratio to consider whether large endowments insulate not-for-profits from donor and market discipline and reduce their effectiveness at delivering program services. Desai and Yetman (2006) use the program expense ratio to examine how differences in the strictness of the regulation of not-for-profits by states influence their delivery of services.

A higher ratio of program expenses to total expenses suggests a not-for-profit organization delivers its program with fewer total expenses devoted to management and fundraising. In short, it is more cost efficient than those with a lower program expense ratio. While a higher program expense ratio indicates higher cost efficiency, it does not indicate the best-practice level of costs and the divergence of any particular organization's cost from best practice. Such divergence can gauge the extent of agency problems.

Many studies have directly estimated cost functions and cost efficiency for a variety of organizations, both for profit as well as not for profit. These studies typically focus on firms in industries with detailed data on production and cost, such as banking and health care. (1) The data on not-for-profits obtained from the IRS Form 990 lack these details and, consequently, encourage substitution of the program expense ratio for the direct estimation of cost efficiency.

While the IRS Form 990 data are not ideal for estimating cost efficiency, they nevertheless allow the estimation of a simple cost function and cost efficiency by adapting some of their variables to proxy for more standard outputs and inputs. The focus of most studies on program expense suggests that program expense can be used as a proxy for the primary output of a not-for-profit organization in the Form 990 data. The indirect cost of delivering any amount of program services is the sum of the management and fundraising expenses. The fundraising components of this sum of administrative costs point to another potential output, total contributions raised, which occasions some of these administrative costs. Thus, the Form 990 administrative cost function consists of the sum of management and fundraising expenses as a function of the two outputs, program expense and total contributions raised. In addition, since labor and physical capital costs differ geographically, it is important to control for these differences in the cost function. In a particular not-for-profit sector, the best-practice organizations minimize the administrative expense of any given level of program service expense and contributed funds and input prices.

When most studies use the ratio of program expense to total expenses to gauge and compare the effectiveness of charities in delivering services, they implicitly assume that the quality of each dollar of program expense across charities is homogeneous. In using program expense as a proxy for the output of program services in the administrative cost function, the same assumption of homogeneity must be made. In addition, if the administrative cost function takes on a sufficiently flexible functional form such that the marginal administrative cost of an additional dollar of contributions raised is not necessary constant, the efficiency of fundraising can be investigated. If the specification includes an interaction between the two outputs, contributions raised and program services expenditures, the size of program services can influence funds raised so that the optimal level of fundraising would depend on the size of the organization.

The administrative cost function can be estimated for any particular not-for-profit sector (industry). The standard estimation fits the cost function as an average of the data while the challenge to identify the best-practice cost (the lowest costs observed in the sample) requires fitting the cost function as a lower frontier of the data and gauging the efficiency of each organization by the distance of its cost from this best practice cost frontier. Several techniques have been used in the literature to estimate the best-practice frontier. Since panel data are available, this study estimates the distribution-free frontier, which was developed by Berger (1993) from a panel data approach proposed by Schmidt and Sickles (1984). For an example of this frontier applied to banking data, see Altunbas et al. (2001), Berger and Hannan (1998), Berger and Mester (1997), and Stiroh (2000).

In this study, the plausibility of the cost efficiency estimates are investigated by re-estimating the model of Core et al. (2006), which found that charities whose endowment exceeds an estimated benchmark have on average lower program expense ratios and higher executive compensation--evidence that the reduction in managers' need to compete in the market for donors and services results in a shift of expenses from program to administration. This study confirms the plausibility of its frontier measure of administrative cost efficiency by finding evidence consistent with Core et al. that endowments in excess of the benchmark are associated with lower administrative cost efficiency. Several different characterizations of "excess endowment" all lead to the same conclusion: wealthier charities tend to be less administratively efficient than their poorer cousins. Moreover, higher administrative cost efficiency is associated with a higher predicted CEO compensation. The estimates obtained using administrative cost efficiency as the dependent variable are much more precise than those obtained using the program expense ratio.

While the study by Core et al. used data from the entire charitable sector covering many very different industries, this study focuses on the largest sector in the Form 990 data, health care, and, to narrow the data to a more homogenous industry, general hospitals (National Taxonomy of Exempt Entities Core Code E22). The data consist of a balanced panel of 1,028 distinct not-for-profit hospitals in the United States over the period 1998-2003. Observations with implausible data, such as negative assets and liabilities, program expenses, contributions, and so on were eliminated. In addition, hospitals which reported "payments to affiliates" (IRS Form 990 line 16) were eliminated since their total expenses did not equal the sum of management, fundraising, and program expenses. Table 1 provides summary statistics of these data.

In contrast to the median total revenue, $23,149,532, the mean, $77,332,098, reflects the influence of very large hospitals, the largest generating $1,698, 971,684 in revenue. The mean administrative cost, $10,138,015 exceeds the median, $3,130,709 and, not surprisingly, consists largely of management rather than fundraising expense. Program services expense, 83.3 % of total expenses, constitutes the largest expense category: on average, $64,586,493, which is much larger than the median $18,310,289. Fundraising, 0.63 % of total expenses, generates on average 8.95 % of total revenue. The median proportion of contributions to total revenue is 0.55 %. Liquid holdings, the sum of cash, savings, and investment securities, termed "endowment" by Core et al. (2006), amounts to $31,180,048 on average. The largest holding is $34,650,000,000. The mean level of endowment expressed as a percentage of total expenses is 123.59 % while the median is 20.07 %. Core et al. (2006) focus on holdings of liquid assets to investigate agency problems related to excess cash discussed by Jensen (1986) in his paper on "free cash flow." Fund balance, the difference between assets and liabilities, gives a more common measure of "endowment." However, it does not necessarily capture holdings of liquid assets. This investigation explores the existence of agency problems related to relatively large holdings of liquid assets using both the program expense ratio and administrative cost efficiency as performance metrics. The sections that follow detail the specification and estimation of the administrative cost function, the calculation of the best-practice efficiency metric, and a test of its plausibility by using it to retrace the investigation of Core et al. to determine if the efficiency metric yields similar results at the same or better precision.

The Administrative Cost Function, Scale Economies, and Efficiency

The specification of the cost function must accomplish at least two goals. First, it should capture the essential outputs of the charity to the extent allowed by the Form 990 data so that convincing peer groups can be defined for comparing efficiency--that is, groups producing the same output, doing essentially the same thing. Second, the specification should allow sufficient flexibility that it does not attempt to impose, say, linear relationships on nonlinear data. The administrative cost function for firm i in year t is given by:

ln [C.sub.it] = ln [C.sub.it] ([Y.sub.it], [w.sub.it]) + ln[[mu].sub.i] + ln[[mu].sub.i] + [v.sub.it] (1)

where In is the natural logarithm; C is administrative cost, the sum of management and fundraising expenses; y is a vector of two outputs, program services measured by program expenses and total contributions raised; w, a vector of input prices, proxied by state and year dummy variables; [mu], a time-invariant efficiency factor; and v, a time-varying random error.

Equation (1) is estimated by using a translog functional form. The log form presents a problem for observations that report no contributions. To avoid losing these observations, Core et al. (2006), who also use a log specification in a different context, create a dummy variable that takes the value one when total contributions equal zero, and zero when contributions is positive. The log of contributions equals the log of contributions when contributions is positive and zero when contributions equals zero. Their procedure is followed here. Equation (1) is estimated over the six years in the period 1998-2003.

The error term in the estimated cost function captures both inefficiency and random error. When the six random error terms obtained for each organization are averaged, the random error components tend to offset each other so that the average, ln[[mu].sub.i], yields an estimate of the efficiency factor for the i-th firm. This technique assumes that an organization's degree of efficiency persists so it can be captured in the averaging. Of course, over a sufficiently long period, organizational governance and market discipline change and so does efficiency. DeYoung (1997) considers the number of years required to minimize the influence of random error in the averaging and recommends six.

The efficiency factor, [[mu].sub.i], is a multiplicative cost index: the index exceeds one when, on average, the achieved cost is greater than predicted and less than one when smaller than predicted. The smallest value of the cost index, [[mu].sub.MIN], can be used to construct the efficiency ratio for the i-th organization,

[E.sub.i] = [[mu].sub.MIN]/[[mu].sub.i] (2)

When the i-th organization defines the minimum cost index, [[mu].sub.i] = [[mu].sub.MIN], so that [E.sub.i] = 1, and for all other organizations whose cost index exceeds the minimum, [[mu].sub.j] > [[mu].sub.i], 0 < [E.sub.j] < I, so their efficiency registers less than one. Thus, higher values of the efficiency ratio correspond to lower values of the multiplicative cost index and indicate that any given amount of program services and contributions raised are achieved at lower administrative cost. Following Berger and Hannan (1998), the efficiency factor, ln[[mu].sub.i], is winsorized at the 5 % and 95 % levels to reduce the influence of outliers.

Panel A of Table 2 gives the details of the estimation of the administrative cost function. The translog specification allows for nonlinear administrative cost effects in logs and for interaction between program services and contributions. The estimated effect on administrative cost of a proportional increase in program services is given by the cost elasticity of program services:

[partial derivative]ln C/[partial derivative]ln program

= 0.72292 + (2)(0.00799)(ln program) + (-0.00373)(ln contributions),

and of a proportional increase in total contributions, by the cost elasticity,

[partial derivative]ln C/[partial derivative]ln contributions = 0.15703 + (2)(-0.00308)(ln contributions) + (-0.00373) (ln program). (4)

The overall cost elasticity, or the effect on administrative cost of a proportional increase in both program and contributions, is given by the sum of the elasticities. Table 3 reports that, evaluated at the median values of program services and contributions, the overall cost elasticity is 0.968. The inverse of the cost elasticity gives the degree of scale economies:

degree of scale economies = 1, /[partial derivative]C/[partial derivative]ln program + [partial derivative]lnC/[partial derivative]ln contributions (5)

which, evaluated at the median values of program and contributions, is 1.030 and significantly different from one. It ranges from a minimum of 0.886 to a maximum of 1.179. Values greater than one indicate scale economies. The value 1.030 implies that a 10 % increase in program services and contributions entails a 9.86 % increase in administrative cost.

Table 4 reports the administrative cost efficiency estimates: mean efficiency, 30.41%, and median, 23.29 %, in a range of a minimum of 7.50 % to a maximum of 100 %. The commonly used performance metric, the ratio of program services expenses to total expenses, exhibits a mean of 83.30 %, a median of 85.82 %, and a range from 0.833 % to 100.00 %. The correlation between the performance metrics of administrative cost efficiency and the program expense ratio is 0.574 (Table 5).

As a first step in examining the credibility of the cost efficiency metric, the sample is divided at the median value of efficiency into the more and less efficient halves to compare means of key variables that should be related to efficiency. Table 6 reports these comparisons. The more efficient group contains larger hospitals on average. They devote a higher proportion of total expenses to program services, 89.94 % versus 76.66 %, and operate with lower ratios of management expenses, 9.75 versus 22.40 %, and fundraising expenses, 0.32 versus 0.94 %.

While there is no statistically significant difference in contributions as a percentage of total revenue, the more efficient hospitals spend 22.60 cents on average to raise a dollar of contributions while the less efficient spend 62.57 cents on average. There is no significant difference in their holdings of liquid assets as a proportion of total expenses. However, compensation is significantly higher at the more efficient hospitals. Since some hospitals do not report compensation and others report zero values, the values reported in the tables are computed only for hospitals that report positive values for the respective components of compensation. CEO compensation averages $285,532 at the more efficient hospitals and $244,636 at the less efficient. In turn, officers and directors compensation averages $669,872 at more efficient hospitals and $576,122 at less efficient ones, while average non-officer salaries are respectively $40,214 versus $36,436. These differences between the more and less administratively cost efficient appear intuitive and suggest that the efficiency metric accounts for a variety of factors that explain differences in the performance of charities.

Administrative Cost Efficiency, Excess Cash Holdings, and Organizational Size

As an additional test of the plausibility of the cost efficiency metric, the investigation of Core et al. (2006) into the relationship of the program expense ratio and excess cash holdings is recast to examine the relationship of administrative cost efficiency to excess cash holdings. A finding similar to Core et al. is that as higher levels of excess cash are associated with lower ratios of program services, higher levels of excess cash are associated with lower administrative cost efficiency. This would add to the credibility of the efficiency metric as a measure of not-for-profit organizational performance. In addition, it would shed light on a different aspect of charities'

administrative performance relative to best practice. The efficiency metric does not simply substitute an administrative cost ratio for the program expense ratio, it considers how close a charity's administrative cost is to best practice cost. The finding that higher levels of excess cash are associated with lower administrative cost efficiency describes how the distance of such charities from the best practice frontier depends on their excess cash.

In this section, the data on 1,028 hospitals are used to estimate a benchmark cash-holding equation and to measure excess cash, either positive or negative, from the residual of the benchmark equation. In turn, the measure of liquid assets in excess of the benchmark is used to investigate how such large cash holdings are related to the provision of program services, to administrative cost efficiency, and to the size of managerial compensation. The agreement of the findings for program services and managerial compensation with the findings for administrative cost efficiency and the overall agreement with the evidence of Core et al. suggest the plausibility of the administrative cost efficiency metric.

The benchmark endowment (holdings of liquid assets) follows from a model developed by Fisman and Hubbard (2005), which emphasizes a precautionary motive for holding an endowment to maintain program services against variations in revenue from year to year. Core et al. specify a benchmark regression to capture the precautionary motive:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)

where endowment is stated as a percentage of total expense. Endowment, following Core et al. (2006), consists of liquid assets at year-end: cash--non-interest-bearing balances (line 45 of IRS Form 990), savings and cash investments (line 46), and investment securities (line 54). The benchmark endowment is estimated by two-stage least squares. The log of total revenue (line 12) is treated as endogenous and, in addition to the other independent variables, lagged values of the log of assets and log of liabilities are used as instruments. Revenue risk is the standard deviation of revenue divided by its mean over five years. Hence, revenue risk has values only for 2002 and 2003. Debt is a dummy variable that equals one when the value of bond liabilities (line 64a) or mortgages and other notes (line 64b) is positive. The revenue risk variables limit the sample to two years, 2002 and 2003. Several zero values of assets and liabilities further reduce the number of observations to 1912. State and year dummies are also included.

The estimation reported in Table 7 shows that higher revenue risk is associated with higher benchmark holdings of liquid assets. For 96 % of the organizations, access to debt reduces the benchmark need to hold liquid assets through the negative coefficient on the interaction of access to debt and revenue risk. The fitted benchmark equation is used to predict the holdings of liquid assets for all 1,028 hospitals in the years 2002 and 2003 and to obtain their residuals. The residual defines excess liquid holdings as a percentage of total expenses, which can be negative or positive.

Jensen (1986) hypothesizes that excess holdings of cash reduce performance pressure on managers. In the case of not-for-profits, Core et al. look for evidence that organizations with excess endowment use the extra funds for program growth and investment in plants and equipment. They find little evidence that growth opportunities explain the higher holdings of liquid assets. Instead, they find that the proportion of expenses allocated to program services is negatively related to excess endowment, which implies that excess cash is related to higher administrative expenses. Moreover, executive compensation is positively related to excess endowment. They conclude that these relationships provide evidence that excess liquid holdings are related to agency problems.

As part of the investigation of the relationship of administrative cost efficiency to excess liquid holdings, the findings of agency problems related to excess endowment by Core et al. are reproduced using these newer data. Table 8 reports these relationships and compares the evidence obtained from the two performance metrics. Table 8 contains several panels of results, each of which focuses on a different characterization of the residual from the benchmark endowment equation. Core et al. define the excess endowment ratio in several ways. First, they use the residual, a continuous variable without transformation, in a regression of the program expense ratio on its lagged value and on lagged control variables. Second, they allow the coefficient on the lagged excess endowment ratio to differ for positive values and for negative values by creating a variable, lagged positive excess endowment, which is the residual value of the excess endowment ratio for positive values and zero for negative residual values, and lagged negative excess endowment, which is the value of the residual when it is negative and zero when positive. Third, values of the residual endowment ratio in the largest quartile are indicated by the value one of a dummy variable, Q4 excess endowment. Fourth, when an organization's excess endowment residual appears in the fourth quartile for several continuous years (in the case of the current data, the 2 years the variable can be estimated, 2002 and 2003), a dummy, persistent excess endowment, takes the value one. When the residual appears in the fourth quartile for one but not both years, a dummy, transitory excess endowment, takes the value one.

The regressions reported in Table 8 each employ 1,028 observations in the last year of the panel (2003). The shortened sample results from two factors: first, the 5 years of data required to define revenue risk in the benchmark endowment regression leaves revenue risk defined only for 2002 and 2003; and, second, the 2 years of data required to measure persistent and transitory excess endowment leave these two variables measured only for 2003. Thus, the performance regressions are cross-sections.

In Panel A of Table 8, the program expense ratio and administrative cost efficiency are regressed on the lagged continuous value of the excess endowment residual, the lagged natural log of total expenses, and the squared natural log of total expenses, two control variables for size. Two additional control variables Core et al. used to test for accounting strategies organizations might use which would bias the results, the lagged ratio of contributions to total revenue and the lagged ratio of liabilities to total assets, are not used in these regressions. Their omission does not appreciably affect the results. In the program expense ratio regression, the coefficient on the continuous excess endowment residual is negative and statistically significant at 12 % but not at conventional levels. However, the corresponding coefficient in the administrative cost efficiency regression is negative and statistically significant at better than 1%. To evaluate its economic significance, consider an increase in the excess endowment ratio from its minimum to its maximum value: administrative cost efficiency decreases 52.3%.

In Panel B, the coefficient on the excess endowment ratio is broken down into coefficients for positive-valued and negative-valued ratios. These two coefficients provide evidence that the program expense ratio is not significantly related to either formulation of the excess endowment ratio while administrative cost efficiency is significantly negatively related to lagged positive excess endowment and not significantly related to the lagged negative excess endowment. Thus, the negative relationship of cost efficiency and excess endowment reported in Panel A appears to be driven by organizations whose holdings of liquid assets exceed the benchmark--that is, those with positive (residuals) excess endowments--rather than those with deficit holdings. The administrative cost efficiency ratio decreases 43.68 % as the lagged positive excess endowment ratio increases from its minimum positive value to its maximum value.

In Panel C, the coefficients on the dummy variable that indicates holdings of liquid assets in the largest quartile of organizations provide evidence that, on average, these organizations have a statistically significant 2.65 % lower program expense ratio and a 7.35 % lower administrative cost efficiency ratio. In Panel D, membership in the fourth quartile is broken down into those organizations that belong to the fourth quartile for both years where excess endowment can be computed as indicated by a dummy variable, persistent excess endowment, equal to one, and those that belong to the fourth quartile in either 2002 or 2003 but not both, as indicated by a dummy variable, transitory excess endowment, equal to one. Neither the program expense ratio nor administrative cost efficiency is significantly related to transitory excess endowment; however, both performance metrics are significantly negatively related to persistent excess endowment. Organizations with liquid asset holdings in the highest quartile in both 2002 and 2003 have, on average, a statistically significant 3.49 % lower program expense ratio and a 7.43 % lower administrative cost efficiency ratio. Thus, weaker performance is not related to relatively large, temporary holdings of liquid assets, which may reflect growth opportunities in program and physical assets. Instead, as Jensen (1986) hypothesized, relatively large holdings of liquid assets that persist over time reduce performance pressure on management and often lead to poorer organizational performance.

In each of the panels, the relationship of the program expense ratio to size is given by the coefficients on the lagged log and squared lagged log of total expenses. These coefficients are quantitatively similar in each panel so the size effect is examined on in the first panel. In Panel A, the derivative of the program expense ratio with respect to organizational size, measured by the lagged log of the total expenses, is negative for smaller values of total expenses and positive for larger values. The derivative,

[partial derivative] program expense ratio %/[partial derivative]In [expenses.sub.t-1] = -3.49236 + (2)(0.14583)(1n [expenses.sub.t-1]), (7)

reverses sign at In [expenses.sub.t-1] = 11.97. Of the 1,028 organizations in the regression in fiscal year 2003, 67 have total expenses under the value 11.97, while 961 exceed it. Thus, for larger hospitals, which constitute 93.5 % of the sample, the program expense ratio is positively related to size: larger expenses are associated with a higher program expense ratio and, hence, a lower administrative cost ratio.

A similar calculation in Panel A for administrative cost efficiency shows that the derivative,

[partial derivative]administrative cost efficiency/[partial derivative]1n [expenses.sub.t-1] = -24.77615 + (2)(0.73645)(In expenses, i), (8)

reverses sign at In [expenses.sub.t-1] = 16.82. For 469 smaller hospitals, administrative cost efficiency is negatively related to size. However, as these hospitals' size increases, the negative correlation approaches zero. And, for the remaining 559 hospitals, 54.4 % of the sample, cost efficiency increases with size.

The relationship of both performance metrics to organizational size (measured by total expenses), is qualitatively similar. For most of the sample, the program expense as a percentage of total expenses increases with total expenses, which suggests that, on average, there are scale economies in administrative costs which permits the increase in program expense over administrative expense as institutions grow larger. For slightly more than half the sample, the administrative efficiency with which the hospitals produce program services increases with total expenses.

Table 9 divides the sample into quartiles by the amount of total expenses. Both the mean and median program expense ratio are highest in the largest quartile. The mean program expense ratio increases from the smallest to the largest quartile. Administrative cost efficiency is higher in the smallest and largest quartiles; apparently, mid-sized hospitals are at a disadvantage compared to small and large hospitals. This pattern is also found in commercial banking, where small community banks and large money center banks are more efficient than mid-sized banks, which often become acquisition targets of large banks. The IRS data do not provide characteristics of hospitals like the number of beds and the case mix. Thus, small hospitals measured by revenue may be specialized or may treat simple cases, while medium and large hospitals measured by revenue may treat a variety of complicated cases requiring sophisticated technologies. The medium sized hospitals may not have the quantity of patients to compete in a cost effective manner with the larger hospitals. (2)

Administrative Cost Efficiency and CEO Compensation

Two distinctive characteristics of not-for-profit organizations, the absence of owners and the constraint preventing the distribution of profits and assets to employees, limit the ability of these organizations to tie pay to performance. (3) 0LS regressions reported in Table 10 investigate the relationship of CEO compensation to the two performance measures. CEO compensation sums salary, benefits, and expense allowance (IRS Form 990, Part V, amounts in columns c + d + e). Many firms in the IRS data do not report compensation, and some report zero values. The sample used in the compensation regressions is restricted to firms that report positive CEO compensation. Robust standard errors are clustered at the firm level.

In Panel A, the coefficient on administrative cost efficiency, 7191.1, indicates that CEO compensation is positively related to administrative cost efficiency. However, the coefficient on the interaction of efficiency and size, -390.6, shows that the magnitude of the positive relationship is reduced as the organization's size increases. The estimated derivative of CEO compensation with respect to administrative cost efficiency is given by

[partial derivative] CEO Compensation/[partial derivative] Administrative Cost Efficiency = 7171.1-390.6(1n [expenses.sub.t-1]), (9)

which is positive for values of In [expenses.sub.t-1] <18.41. Thus, for the smaller 68 % of the sample CEO compensation is positively related to administrative cost efficiency, and the derivative is statistically significant at better than 1%. On the other hand, the derivative is negative for the larger 32 % of the sample, but the derivative is not significantly different from zero over this range. At the median value of lagged expenses, the value of the derivative is 265; hence, a 1% increase in cost efficiency is associated with a $265 increase in total CEO compensation. For the smallest hospital, the increase is $2,986. Thus, at larger hospitals, the pay-performance relationship is rather weak given the magnitude of the median CEO compensation.

In Panel B, the coefficient on the program expense ratio, -9844.44, indicates that CEO compensation is negatively related to this measure of performance. However, the coefficient on the interaction of the program expense ratio and size, 581.66, complicates this relationship. The derivative of CEO compensation with respect to the program expense ratio (%) is given by

[partial derivative] CEO Compensation/[partial derivative]Program Expense [Ratio.sub.t-1] = -9844.44 + 581.66(1n [expenses.sub.t-1]), (10)

which is positive for values In [expenses.sub.t-1]> 16.92. Thus, for 70 % of the sample, CEO compensation is positively related to the program expense ratio. At the median value of lagged expenses, the value of the derivative is $469. Hence, an increase in the program expense ratio of 1% is associated with an increase of $469 in total compensation. Again, the pay-performance relationship is weak given the magnitude of the median size of CEO compensation. In both regressions, CEO compensation and organizational size are positively related. While the interaction term between size and cost efficiency is negative in the first regression, its magnitude is too small to switch the sign of the compensation-size correlation--that is, the value of cost efficiency at which the sign changes is outside the bound of the definition of cost efficiency.

Conclusions

The IRS Form 990 offers much useful information about not-for-profit organizations in the U. S. In particular, its functional breakdown of expenses into program, management, and fundraising permits a simple calculation of organizational performance in terms of the proportion of its expenses accounted for by program services. The program expense ratio points to the remaining proportion of expenses--administration--the sum of management and fundraising expenses. A larger program expense ratio implies that the organization operates more efficiently. For a given amount of total expenses, a larger program expense ratio implies that less is spent on administration. For a given amount of expenditure on program services, a larger program expense ratio entails smaller spending on administration.

While a larger program expense ratio captures better administrative cost efficiency, it does not gauge best-practice cost and the extent to which an organization's administrative costs exceed best practice. Based on the estimation of administrative cost as a function of program service expenditure and total contributions raised, this study uses the distribution free technique to identify organizations that produce program services at lowest administrative cost--at best practice--and to gauge other organizations' administrative cost in terms of this best-practice cost. Thus, administrative cost efficiency supplements the program expense ratio as a performance metric by providing additional insight into organizational performance on the cost side.

The empirical evidence on performance obtained from the administrative cost efficiency metric complements that of the program expense ratio and reinforces its credibility. The division of the sample into the more and less efficient organizations shows that the more efficient are larger, devote a higher proportion of total expenses to program services, operate with lower ratios of management expenses and fundraising expenses, spend less to raise a dollar of contributions, and pay higher salaries.

The regressions of the program expense ratio and administrative cost efficiency on various characterizations of excess holdings of liquid assets agree: holdings of liquid assets in excess of the benchmark are associated with a lower program expense ratio and lower administrative cost efficiency. Administrative cost efficiency decreases 52 % as the excess endowment ratio increases from its minimum to its maximum value. This relationship is driven by organizations whose holdings exceed the benchmark--those with positive residual endowment rather than deficit holdings. Organizations holding excess liquid assets in the largest quartile have a statistically significant 2.65 % lower program expense ratio and a 7.35 % lower administrative cost efficiency ratio. This result is driven by those organizations which persistently belong to the fourth quartile. Neither the program expense ratio nor administrative cost efficiency is significantly related to transitory excess endowment. Thus, weaker performance is not related to relatively large, temporary holdings of liquid assets, which may reflect growth opportunities in program and physical assets. Instead, as Jensen (1986) hypothesized, relatively large holdings of liquid assets that persist over time reduce performance pressure on management and often lead to poorer organizational performance. The agreement of the evidence on agency problems related to excess holdings of liquid assets from the program expense ratio and administrative cost efficiency reinforce the credibility of the latter as a measure of the performance of not-for-profit organizations.

DOI 10.1007/s11293-013-9374-2

Published online: 14 June 2013

Submitted to the Atlantic Economic Journal for the John Virgo Memorial Issue

The author is grateful to the Research Council at Rutgers University for their financial support of this project, Vivian Valdmanis for her insightful comments, and to John Bowblis and Nicholas Galunic for their research assistance and advice while graduate students at Rutgers.

References

Altunbas, Y., Evans, L., & Molyneux, P. (2001). Bank ownership and efficiency. Journal of Money, Credit, and Banking. 33, 926-954.

Berger, A. N. (1993). 'Distribution-free' estimates of efficiency in the U.S. banking system and tests of the standard distributional assumptions, Journal of Productivity Analysis, 4, 261-292.

Berger, A. N., & Hannan, T. H. (1998). The efficiency cost of market power in the banking industry: a test of the 'quiet life' and related hypotheses. Review of Economics and Statistics, 80, 454-465.

Berger, A. N., & Mester, L. J. (1997). Inside the black box: what explains differences in the efficiencies of financial institutions. Journal of Banking and Finance, 21, 895-947.

Core, J. E., Guay, W. R., & Verdi, R. (2006). Agency problems of excess endowment holdings in not-for-profit firms. Journal of Accounting and Economics, 41, 307-333.

Desai, M., & Yetman, R. J. (2006). Constraining managers without owners: Governance of the not-for-profit enterprise. Working Paper, NBER.

DeYoung, R. (1997). A diagnostic test for the distribution-free efficiency estimator: an example using U. S. commercial bank data. European Journal of Operational Research. 98, 243-249.

Fisman, R., & Hubbard, G. (2005). Precautionary savings and the governance of nonprofit organizations. Journal of Public Economics, 89, 2231-2243.

Hartzell, J. C., Parsons, C. A., & Yermack, D. L. (2010). Is a higher calling enough? Incentive compensation in the church. Journal of Labor Economics, 28, 509-539.

Hughes, J. P., & Mester, L. J. (2009). Efficiency in banking: Theory, practice, and evidence. Chapter 19. In A. N. Berger, P. Molyneux, & J. Wilson (Eds.), The Oxford Handbook of Banking. Oxford, UK: Oxford University Press.

Jensen, M. C. (1986). Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review, 76, 323-329.

Schmidt, P., & Sickles, R. C. (1984). Production frontiers and panel data. Journal of Business and Economic Statistics, 2, 367-374.

Stiroh, K. J. (2000). How did bank holding companies prosper in the 1990s? Journal of Banking and Finance, 24, 1703-1745.

J. P. Hughes ([mail])

Department of Economics, Rutgers University, New Brunswick, NJ 08901, USA

e-mail: jphughes@rci.rutgers.edu

(1) For a review of the literature on bank production and efficiency, see Hughes and Mester (2009).

(2) The author is grateful to Vivian Valdmanis for this explanation of the relative disadvantage of medium sized hospitals.

(3) Hartzell et al. (2010) find evidence that the Methodist Church appears to tic ministers' pay to performance.

Table 1 Summary statistics for the full sample Variable N Mean Panel A Total revenue 6,168 77,332,098 Total expense 6,168 74,724,507 Program expense 6,168 64,586,493 Administrative cost 6,168 10,138,015 Management expense 6,168 10,076,798 Fundraising expense 6,168 61,217 Total contributions 6,168 1,870,932 Endowment (liquid holdings) 6,168 31,180,048 Panel B Program expense/total expenses % 6,168 83.30 Administrative cost/total expenses % 6,168 16.70 Management expense/total expenses % 6,168 16.07 Fundraising expense/total expenses % 6,168 0.63 Total contributions/total revenue % 6,168 8.95 Endowment/total expenses % 6,168 123.59 Variable Median Std Dev Panel A Total revenue 23,149,532 149,135,726 Total expense 22,494,743 144,425,725 Program expense 18,310,289 128,634,214 Administrative cost 3,130,709 18,883,783 Management expense 3,109,150 18,818,765 Fundraising expense 0 557,303 Total contributions 123,952 13,398,351 Endowment (liquid holdings) 2,576,138 615,335,474 Panel B Program expense/total expenses % 85.82 11.97 Administrative cost/total expenses % 14.18 11.97 Management expense/total expenses % 13.71 11.45 Fundraising expense/total expenses % 0.00 3.93 Total contributions/total revenue % 0.55 97.47 Endowment/total expenses % 20.07 811.17 The data consist of a balanced panel of 1,028 not-for-profit hospitals over a 6-year period 1998-2003. They were obtained from the IRS Form 990 filed annually by not-for-profit organizations and published by the National Center for Charitable Statistics. Total revenue is given on Form 990 at line 12; total expense at line 17, which sums program expense, line 13, management expense, line 14, and fundraising expense, line 15. (Hospitals which report payments to affiliates, line 16, are dropped.) Administrative cost sums management and fundraising expenses. Total contributions is given at line Id. Endowment, following Core et al. (2006), consists of liquid assets at year-end: cash--non-interest-bearing (line 45), savings and cash investments (line 46), and investment securities (fine 54), which focuses on the potential of agency problems due to excess cash holdings Table 2 Estimation of the administrative cost function Variable Parameter Robust standard estimate error Dependent variable: administrative cost Intercept -0.12187 0.80282 Log program expenses 0.72292 0.12028 Squared log program expenses 0.00799 0.00407 Log total contribution 0.15703 0.05091 Squared log total contributions -0.00308 0.00198 Zero contributions indicator 0.59470 0.25739 Log program expenses x log total -0.00373 0.00186 contributions Variable t value Pr> [absolute value of t] Dependent variable: administrative cost Intercept -0.15 0.8793 Log program expenses 6.01 <.0001 Squared log program expenses 1.96 0.0495 Log total contribution 3.08 0.0020 Squared log total contributions -1.55 0.1211 Zero contributions indicator 2.31 0.0209 Log program expenses x log total -2.01 0.0448 contributions To estimate the cost function, ordinary least squares with robust standard errors is applied to the data, which consist of a balanced panel of 1,028 not-for-profit hospitals over a 6-year period 1998- 2003. The data were obtained from the IRS Form 990 filed annually by not-for-profit organizations and published by the National Center for Charitable Statistics. The dependent variable is administrative cost, which sums management (line 14) and fundraising expenses (line 15). The two outputs are program expenses (line 13) and total contributions (line Id). In order not to eliminate observations where total contributions equal zero, the log of total contributions is set equal to zero and an indicator variable for zero contributions is set equal to one. To account for differences across states in the cost of inputs and in the regulatory environment, state indicator variables are included as well as time indicator variables. They are not reported in the table State and Time Dummies--Yes Number of Observations = 6,168 Adjusted R-square = 0.853 Table 3 Efficiency and scale economics estimates Elasticities evaluated at median values of program services and total contributions Adminstrative cost elasticity of program services 0.946 * Adminstrative cost elasticity of contributions 0.022 ** Adminstrative cost elasticity of program services and 0.968 * contributions Degree of scale economies 1.032 * The administrative cost elasticity of program services is [partial derivative]ln[C.sub.t]/[partial derivative]ln(program expenses) and the administrative cost elasticity of contributions is [partial derivative]ln[C.sub.t]/[partial derivative]ln(contributions). The estimate of scale economies is the inverse of the sum of output cost elasticities: scale economics= 1/[[partial derivative]ln[C.sub.t]/ [partial derivative]ln(program expenses)+[partial derivative]ln[C.sub.t]/[partial derivative]ln(contributions)] * Significantly different from one at better than p=.01 ** Significantly different from zero at better than p=.01 Table 4 Efficiency estimates Variable N Mean Median Std Dev Total revenue 6,168 77,332,098 23,149,532 149,135,726 Admin cost efficiency % 6,168 30.41 23.29 23.38 Program expense ratio % 6,168 83.30 85.82 11.97 Variable Minimum Maximum Total revenue 161 1,698,971,684 Admin cost efficiency % 7.50 100.00 Program expense ratio % 0.83 100.00 Using the Distribution Free technique of estimating efficiency, the six random error terms obtained for each organization are averaged: since the random error components tend to offset each other, the average yields an estimate of the multiplicative efficiency factor, In [[PHI].sub.i], which is a cost index. The smallest value of the cost index, [[PHI].sub.MIN], is used to construct the efficiency ratio for the i/th organization, [E.sub.i] = [[PHI].sub.MIN]/ [[PHI].sub.i] Table 5 Correlations Total Administrative Program revenue cost efficiency expense ratio Administrative 0.0264 1.0000 0.5745 Cost efficiency 0.038 <.0001 Program expense 0.1353 0.5745 1.0000 Ratio <.0001 <.0001 The correlation of the two performance metrics, administrative cost efficiency and the program expense ratio, are examined in this table Table 6 Summary statistics for less and more efficient hospitals Variable N Mean Panel A: Less cost efficient hospitals Total revenue 3,084 62,711,261# Admin cost efficiency % 3,084 14.61# Program expenses/expenses % 3,084 76.66# Management expenses/expenses % 3,084 22.40# Fundraising expenses/expenses % 3,084 0.94# Contributions/revenue % 3,084 7.09 Fundraising expenses/contributions % 2,596 62.57 * Endowment/expenses % 3,084 119.15 CEO Total compensation (a) 1,148 244,636# Officers and directors compensation (a) 1,957 576,122# Average non-officer salary (a) 2,440 36,436 ** Panel B: More cost efficient hospitals Total revenue 3,084 91,952,935# Admin cost efficiency % 3,084 46.21# Program expenses/expenses % 3,084 89.94# Management expenses/expenses % 3,084 9.75# Fundraising expenses/expenses % 3,084 0.32# Contributions/revenue % 3,084 10.80 Fundraising expenses/contributions % 2,659 22.50 * Endowment/expenses % 3,084 128.02 CEO Total compensation (a) 1,008 285,532# Officers and directors compensation (a) 1,697 669,872# Average non-officer salary (a) 2,240 40,214 ** Variable Median Std Dev Panel A: Less cost efficient hospitals Total revenue 24,772,562 111,432,257 Admin cost efficiency % 14.59 4.75 Program expenses/expenses % 79.46 12.19 Management expenses/expenses % 20.05 11.70 Fundraising expenses/expenses % 0.00 4.94 Contributions/revenue % 0.47 20.35 Fundraising expenses/contributions % 0.00 1192.68 Endowment/expenses % 19.72 420.45 CEO Total compensation (a) 195,850 189,064 Officers and directors compensation (a) 323,274 1,132,372 Average non-officer salary (a) 35,061 16,108 Panel B: More cost efficient hospitals Total revenue 21,091,085 177,891,313 Admin cost efficiency % 36.39 23.91 Program expenses/expenses % 90.59 7.05 Management expenses/expenses % 9.22 6.72 Fundraising expenses/expenses % 0.00 2.52 Contributions/revenue % 0.62 136.31 Fundraising expenses/contributions % 0.00 347.86 Endowment/expenses % 20.56 1067.43 CEO Total compensation (a) 220,754 264,214 Officers and directors compensation (a) 341,452 1,149,180 Average non-officer salary (a) 36,244 74,437 Note: Values of a variable indicate that they are significantly different from each other at the 1% level are indicated with #. The data, which consist of a balanced panel of 1,028 not-for-profit hospitals over a 6-ycar period 1998 2003, were obtained from the IRS Form 990 tiled annually by not-for-profit organizations and published by the National Center for Charitable Statistics. The charities arc restricted to NTEEC category E22, hospitals The sample is divided at the median value of administrative cost efficiency. Administrative cost efficiency is estimated by the Distribution Free technique. Administrative cost is the sum of management (line 14) and fundraising expenses (line 15) Total expenses sums administrative cost and expenses due to the provision of program services (line 13). The ratio (stated as a percentage) of contributions to total revenue is given by the amounts on line Id and line 12. Endowment, following Core ct al. (2006), consists of liquid assets at year-end: cash non--interest-bearing (line 45), savings and cash investments (line 46), and investment securities (line 54). CEO total compensation sums salary, benefits, and expense allowance (Part V, amounts in columns c + d + c). Total compensation for officers and directors is given by line 25a. Average non-officer salary is the ratio of other salaries and wages (line 26a) divided by total full time equivalent employees (line 90b) Between the more and less efficient samples, emboldened values of a variable indicate that they arc significantly different from each other at the 1 % level. ** indicates a significant difference at the 5 % level while * indicates significance at the 10 % level (a) The summary statistics arc computed only for observations that report positive compensation Many observations failed to report compensation, and some reported values of zero Table 7 Estimation of the benchmark endowment equation Variable Parameter Standard t value Pr > estimate error [absolute value of t] Dependent variable: Endowment/expenses x 100 Intercept -30.8466 63.66390 -0.48 0.6281 Log revenue -0.35799 3.433147 -0.10 0.9170 Debt 30.16410 21.17694 1.42 0.1545 Revenue risk 1001.438 50.74173 19.74 <.0001 Debt x revenue risk -828.453 76.15512 -10.88 <.0001 The data, a balanced panel of 1,028 not-for-profit hospitals (NTEEC category E22) over a 6-year period 1998-2003, were obtained from the IRS Form 990 filed annually by not-for-profit organizations and published by the National Center for Charitable Statistics. The ratio (stated as a percentage) of endowment to total expenses (line 17) is the dependent variable. Endowment, following Core et al. (2006), consists of liquid assets at year-end: cash-non-interest-bearing balances (line 45), savings and cash investments (line 46), and investment securities (line 54). The benchmark endowment is estimated by two-stage (cast squares. The log of total revenue (line 12) is treated as endogenous and, in addition to the other independent variables, lagged values of the log of assets and log of liabilities are used as instruments. Revenue risk is the standard deviation of revenue divided by its mean over 5 years. Hence, revenue risk has values only for 2002 and 2003. Debt is a dummy variable that equals one when the value of bond liabilities (line 64a) or mortgages and other notes (line 646) is positive. The revenue risk variables limit the sample to 2 years: 2002 and 2003. Several zero values of assets and liabilities further reduce the number of observations to 1912. State and year dummies are also included State and Year Dummies--Yes [R.sup.2] = 0.28720 Number of observations = 1,912 Table 8 Relationship of performance to various measures of the lagged excess endowment residual Variable Parameter Standard estimate error Panel A: Relationship of performance to the continuous lagged excess endowment residual Dependent variable: Program expense/expense Intercept 94.11309 17.42688 Log [expense.sub.t-1] -3.49236 2.12293 Squared log [expense.sub.t-1] 0.14583 0.06418 Excess [endowment.sub.t-1] -0.00232 0.00149 State dummies-Yes Number of observations = 1,028 R-square = 0.1338 Dependent variable: Administrative cost efficiency Intercept 234.31124 33.19272 Log [expense.sub.t-1] -24.77615 4.13331 Squared log [expense.sub.t-1] 0.73645 0.12753 Excess [endowment.sub.t-1] -0.00871 0.00238 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0836 Panel B: relationship of performance to lagged positive and negative excess endowment residual Dependent variable: Program expense/expense Intercept 94.47088 17.96911 Log [expense.sub.t-1] -3.52481 2.16586 Squared log [expense.sub.t-1] 0.14658 0.06512 Positive excess [endow.sub.t-1] -0.00245 0.00173 Negative excess [endow.sub.t-1] -0.00179 0.00373 State dummies--Yes Number of observations= 1,028 [R.sup.2] = 0.1338 Dependent variable: Administrative cost efficiency Intercept 239.81683 34.31092 Log [expense.sub.t-1] -25.27551 4.22934 Squared log [expense.sub.t-1] 0.74799 0.12969 Positive excess [endow.sub.t-1] -0.01057 0.00276 Negative excess [endow.sub.t-1] -0.00051 0.00626 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0853 Panel C: Relationship of performance to fourth quartile lagged excess endowment residual Dependent variable: Program expense/expense Intercept 93.98147 16.75550 Log [expense.sub.t-1] -3.42809 2.05268 Squared log [expense.sub.t-1] 0.14368 0.06235 Q4 excess [endowment.sub.t-1] -2.65038 1.06189 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.1375 Dependent variable: Administrative cost efficiency Intercept 228.07709 32.40687 Log [expense.sub.t-1] -23.90538 4.05334 Squared log [expense.sub.t-1] 0.71054 0.12555 Q4 excess [endowment.sub.t-1] -7.34831 1.91513 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.852 Panel D: Relationship of performance to persistent and transitory lagged excess endowment Dependent variable: Program expense/expense Intercept 94.16443 16.82080 Log [expense.sub.t-1] -3.46586 2.06599 Squared log [expense.sub.t-1] 0.14500 0.06271 Persistent excess [endow.sub.t-1] -3.48774 1.13153 Transitory excess [endow.sub.t-1] -0.06238 1.61806 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.1414 Dependent variable: Administrative cost efficiency Intercept 227.22584 32.91539 Log [expense.sub.t-1] -23.82198 4.09671 Squared log [expense.sub.t-1] 0.70839 0.12661 Persistent excess [endow.sub.t-1] -7.74279 2.00877 Transitory excess [endow.sub.t-1] -1.77761 3.00752 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0844 Variable t Pr > value [absolute value Panel A: Relationship of performance to the continuous lagged excess endowment residual Dependent variable: Program expense/expense Intercept 5.40 <.0001 Log [expense.sub.t-1] -1.65 0.1003 Squared log [expense.sub.t-1] 2.27 0.0233 Excess [endowment.sub.t-1] -1.56 0.1203 State dummies-Yes Number of observations = 1,028 R-square = 0.1338 Dependent variable: Administrative cost efficiency Intercept 7.06 <.0001 Log [expense.sub.t-1] -5.99 <,p001 Squared log [expense.sub.t-1] 5.77 <,0001 Excess [endowment.sub.t-1] -3.66 0.0003 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0836 Panel B: relationship of performance to lagged positive and negative excess endowment residual Dependent variable: Program expense/expense Intercept 5.26 <.0001 Log [expense.sub.t-1] -1.63 0.1040 Squared log [expense.sub.t-1] 2.25 0.0246 Positive excess [endow.sub.t-1] -1.42 0.1573 Negative excess [endow.sub.t-1] -0.48 0.6312 State dummies--Yes Number of observations= 1,028 [R.sup.2] = 0.1338 Dependent variable: Administrative cost efficiency Intercept 6.99 <.0001 Log [expense.sub.t-1] -5.98 <.0001 Squared log [expense.sub.t-1] 5.77 <.0001 Positive excess [endow.sub.t-1] -3.83 0.0001 Negative excess [endow.sub.t-1] -0.08 0.9351 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0853 Panel C: Relationship of performance to fourth quartile lagged excess endowment residual Dependent variable: Program expense/expense Intercept 5.61 <.0001 Log [expense.sub.t-1] -1.67 0.0952 Squared log [expense.sub.t-1] 2.30 0.0214 Q4 excess [endowment.sub.t-1] -2.50 0.0127 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.1375 Dependent variable: Administrative cost efficiency Intercept 7.04 <.0001 Log [expense.sub.t-1] -5.90 <,p001 Squared log [expense.sub.t-1] 5.66 <,p001 Q4 excess [endowment.sub.t-1] -3.84 0.0001 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.852 Panel D: Relationship of performance to persistent and transitory lagged excess endowment Dependent variable: Program expense/expense Intercept 5.60 <.0001 Log [expense.sub.t-1] -1.68 0.0938 Squared log [expense.sub.t-1] 2.31 0.0210 Persistent excess [endow.sub.t-1] -3.08 0.0021 Transitory excess [endow.sub.t-1] -0.04 0.9693 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.1414 Dependent variable: Administrative cost efficiency Intercept 6.90 <.0001 Log [expense.sub.t-1] -5.81 <.0001 Squared log [expense.sub.t-1] 5.60 <.0001 Persistent excess [endow.sub.t-1] -3.85 0.0001 Transitory excess [endow.sub.t-1] -0.59 0.5546 State dummies--Yes Number of observations = 1,028 [R.sup.2] = 0.0844 The data, a balanced panel of 1,028 not-for-profit hospitals (NTEEC category E22) over a 6-year period 1998-2003, were obtained from the IRS Form 990 filed annually by not- for-profit organizations and published by the National Center for Charitable Statistics. These performance regressions use only the 1,028 observations in 2003. The shortened sample results from two factors: first, the 5 years of data required to define revenue risk in the benchmark endowment regression leaves revenue risk defined only for 2002 and 2003; and, second, the 2 years of data, 2002 and 2003, required to measure persistent and transitory excess endowment leave these two variables measured only for 2003 The percentage of total expenses accounted for by program expense consists of program expense (line 13) divided by total expense at line 17, multiplied by 100. The administrative cost efficiency is based on the percentage of total expenses accounted for by administrative costs, the sum of management and fundraising expenses-lines 14 and 15. It is estimated by the Distribution Free method Endowment, following Core et al. (2006), consists of liquid assets at year-end: cash-non-interest-bearing (line 45), savings and cash investments (line 46), and investment securities (line 54), which focuses on the potential of agency problems due to excess cash holdings. Excess endowment (a ratio of endowment to total expenses) is the residual of the benchmark endowment regression. The residual is also defined by a variable, positive excess endowment, which equals the value of the residual when it is positive and zero when it is negative. Negative excess endowment equals the value of the residual when it is negative and zero when positive. Q4 excess endowment equals one when the organization's excess endowment is in the largest quartile of the data. Persistent excess endowment equals one when the organization's excess endowment places in the largest quartile in 2002 and 2003. Transitory excess endowment equals one when the organization's excess endowment places in the largest quartile in 2002 or 2003 but not both years White robust standard errors are reported in all panels Table 9 Relationship of performance to size Variable Mean Median Largest quartile measured by total expenses Log [expense.sub.t-1] 19.17 19.02 Program expense/expense % 85.88 88.07 Admin cost efficiency % 31.49 24.42 Second largest quartile measured by total expenses Log [expense.sub.t-1] 17.67 17.70 Program expense/expense % 83.96 85.10 Admin cost efficiency % 26.66 20.86 Second smallest quartile measured by total expenses Log [expense.sub.t-1] 16.27 16.31 Program expense/expense % 82.43 84.63 Admin cost efficiency % 25.41 21.13 Smallest quartile measured by total expenses Log [expense.sub.t-1] 13.04 13.17 Program expense/expense % 79.98 84.97 Admin cost efficiency % 38.08 26.33 Variable Minimum Maximum Largest quartile measured by total expenses Log [expense.sub.t-1] 18.31 21.09 Program expense/expense % 55.16 99.44 Admin cost efficiency % 7.50 100.00 Second largest quartile measured by total expenses Log [expense.sub.t-1] 17.05 18.31 Program expense/expense % 53.60 99.77 Admin cost efficiency % 7.50 100.00 Second smallest quartile measured by total expenses Log [expense.sub.t-1] 15.35 17.05 Program expense/expense % 48.03 99.60 Admin cost efficiency % 7.50 100.00 Smallest quartile measured by total expenses Log [expense.sub.t-1] 9.09 15.33 Program expense/expense % 1.39 99.97 Admin cost efficiency % 7.50 100.00 The data, a balanced panel of 1,028 not-for-profit hospitals (NTEEC category E22) over a 6-year period 1998-2003, were obtained from the IRS Form 990 filed annually by not-for-profit organizations and published by the National Center for Charitable Statistics. The percentage of total expenses accounted for by program expense consists of program expense (line 13) divided by total expense at line 17, multiplied by 100. The administrative cost efficiency is based on the percentage of total expenses accounted for by administrative costs, the sum of management and fundraising expenses- lines 14 and 15. It is estimated by the Distribution Free method Table 10 Relationship of CEO compensation to performance Parameter Standard Error t value Pr>|t| estimate Panel A: CEO compensation and administrative cost efficiency (%) Dependent variable: CEO compensation Intercept -1426058.9 109813.203 -12.99 <.0001 Log [expense.sub.t-1] 95707.5 6294.837 15.20 <.0001 Admin cost efficiency % 7191.1 1963.941 3.66 0.0003 Log [expense.sub.t-1] -390.6 113.995 -3.43 0.0006 x admin cost efficiency % State dummies--Yes Number of observations = 1,814 [R.sup.2] = 0.3911 Panel B: CEO compensation and the program expense ratio (%) Dependent variable: CEO compensation Intercept -395358.79 243456.359 -1.62 0.1046 Log [expense.sub.t-1] 34931.23 15382.229 2.27 0.0233 Program expense/ -9844.44 3219.208 -3.06 0.0023 expense % Log [expense.sub.t-1] 581.66 197.691 2.94 0.0033 x program expense/ expense % State dummies--Yes Number of observations = 1,814 [R.sup.2] = 0.3893 The data, a balanced panel of 1,028 not-for-profit hospitals (NTEEC category E22) over a 6-year period 1998-2003, were obtained from the IRS Form 990 filed annually by not-for-profit organizations and published by the National Center for Charitable Statistics. The sample of organizations used in these OLS regressions is restricted to those that report (positive) compensation for the CEO. Robust standard errors are clustered by firm. CEO compensation sums salary, benefits, and expense allowance (Part V, amounts in columns c + d + c). The percentage of total expenses accounted for by program expense consists of program expense (line 13) divided by total expense at line 17, multiplied by 100. The administrative cost efficiency is based on the percentage of total expenses accounted for by administrative costs, the sum of management and fundraising expenses-lines 14 and 15. It is estimated by the Distribution Free method

Printer friendly Cite/link Email Feedback | |

Author: | Hughes, Joseph P. |
---|---|

Publication: | Atlantic Economic Journal |

Geographic Code: | 1USA |

Date: | Sep 1, 2013 |

Words: | 10064 |

Previous Article: | Mother's gender preferences and child schooling in Ethiopia. |

Next Article: | Bringing culture to macroeconomics. |

Topics: |