JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000, pp. 649–669
LEVIATHAN VS. LILLIPUTIAN: A DATA ENVELOPMENT
ANALYSIS OF GOVERNMENT EFFICIENCY
Patricia A. Nold Hughes and Mary E. Edwards
Economics Department, St. Cloud State University, St. Cloud, MN 56301, U.S.A.
E-mail: [email protected] and [email protected]
ABSTRACT. In this paper we present a new approach to measuring government
efficiency, based on the theory that communities that allocate resources efficiently in the
local public sector maximize property values. We use Data Envelopment Analysis (DEA)
to identify the counties in Minnesota that are characterized by property-value maximization and hence an efficient public sector. The results indicate that the dominant source of
public sector inefficiency is an inappropriate scale of operations. It appears that some
county jurisdictions are too large to service the population efficiently. The size and
concentration of government power are also responsible in part for observed inefficiencies.
1.
INTRODUCTION
The advantages of government decentralization arise as each jurisdiction
must compete for residents through its use of fiscal policy. Smaller governmental
units are more responsive to consumer demand, tend to be more accountable,
and are less apt to develop into a “Leviathan.” In areas with a diverse population,
disaggregation of government activity may result in smaller, more homogeneous
districts as a result of Tiebout sorting. However, excessive decentralization or
downscaling of decision making bears an additional cost in the provision of goods
and services. The benefits of economies of scale may be lost in small-scale
provision, and administrative costs may increase due to higher overhead. Given
the conflicting nature of economies of scale and Leviathan tendencies, the goal
of this paper is to determine whether local government is efficient, and if not,
what causes the inefficiency.
In Section 2 we provide a review of the literature pertaining to the behavior
of local government. In Section 3 we introduce the method of Data Envelopment
Analysis (DEA). DEA allows us to identify inefficient counties and determine
whether the inefficiency is due to scale economies or government waste independent of the scale of operations. With efficiency scores provided by DEA, in
Section 4 we determine the major predictors of an efficient public sector using
Tobit regression analysis. These results are compared to those obtained using a
Received March 1998; revised November 1999, December 1999, and March 2000; accepted
May 2000.
© Blackwell Publishers 2000.
Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA and 108 Cowley Road, Oxford, OX4 1JF, UK.
649
650
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
more traditional stochastic production-frontier approach: corrected ordinary
least squares. A brief summary and conclusion follow in Section 5.
2.
LITERATURE REVIEW
In the world envisioned by Charles Tiebout (1956) consumers voting with
their feet resolve the problem of efficiency in the public sector. In a perfect
Tiebout world individuals sort themselves into communities based on the
provision of public goods. Each individual migrates to the community offering
the optimal fiscal package, that is, the mix of publicly provided goods, services,
and taxes that is compatible with his or her own tastes and preferences. The
result is a set of homogeneous communities with efficient public sectors. The
diversity of the population determines the size of each community and the level
of decentralization required to achieve efficient fiscal policy.
Sonstelie and Portney (1978, 1980) use Tiebout’s hypothesis to focus on how
communities might act in their role as suppliers of local public goods and
services. They demonstrate that local communities acting as profit maximizers
allocate resources efficiently in the local public sector. In the short run, profit
maximization is equivalent to property value maximization. In the long run it
is equivalent to land-value maximization.
Brueckner (1979, 1982) addresses the question of whether heterogeneous
communities are providing public goods efficiently. The analysis is based on the
effect of changes in the level of government expenditures on property values,
incorporating the budget constraint of the local government into the equation.
The hypothesized influence of local government expenditures on property values
is demonstrated in Figure 1.
According to Brueckner, if local governments are currently under-supplying
the local public good, increases in their expenditures should increase property
values. However, at some point, further increases in government expenditures
FIGURE 1: Total Property Values as a Function of Local Government
Expenditures.
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
651
cause property values to decline, indicating an over-supply of local public goods.
Brueckner tests his hypothesis in two regions: 53 northeastern New Jersey
communities and 54 Massachusetts communities. He concludes that the New
Jersey communities may be over-providing public goods, but he has reservations
about the quality of the data used in the analysis (1979, pp. 244–245). Assuming
that the 1979 study is plausible, the New Jersey communities are providing
amount c of public goods in Figure 1. Communities providing c amount of local
public goods are over-providing goods, and they can increase property values by
decreasing the size of government.
Brueckner’s 1982 study concludes that Massachusetts communities show
no systematic tendency to over- or under-provide public goods. Massachusetts
communities are providing amount b of public goods. Communities providing
amount b are acting as profit maximizers. By maximizing property values they
are also maximizing community profits and providing an efficient amount of
public goods.
To capture interjurisdictional spillover effects, Deller (1990) applies a
Box-Cox procedure to county-level Illinois data. Although educational spending
is efficient, he finds that public goods, except for education, are under-provided
in Illinois. This corresponds to point a in Figure 1. Communities that provide
quantity a of local public goods can increase property values by providing more
local public goods. In addition, Deller loosely implies that although larger
governments may experience economies of scale, they are not as responsive to
the needs of the constituents as are smaller governments (p. 404).
In contrast to the traditional view that government acts in the best interest
of its citizens, Brennan and Buchanan (1980) present an alternative view of
government behavior. Acting as a Leviathan, the government maximizes its own
utility by maximizing its fiscal surplus to provide perquisites of office. Fiscal
surplus is defined as the difference between tax revenue and expenditure. The
ability of the government to exploit its citizens is diminished as the degree of
competition for residents increases between jurisdictions. The Leviathan model
implies that the size of the public sector will vary inversely with the degree of
decentralization.
The search for Leviathan begins with a search for a satisfactory measure
of decentralization. Using the percentage of the state and local public sector
controlled by the state as a measure of centralized decision-making, Oates (1985)
does not find evidence supporting the Leviathan hypothesis. Questioning Oates’s
measure of decentralization, Nelson (1987) estimates a revised version of the
model. He uses the number of local governments within a state normalized by
population as his decentralization measure. With this measure of decentralization, Nelson’s findings are consistent with the hypothesis that competition
between general-purpose local governments restrains government spending at
the state level.
Eberts and Gronberg (1988) provide further statistical support for the
decentralization hypothesis at the metropolitan and county levels. As a more
© Blackwell Publishers 2000.
652
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
complete test of the intrusiveness of government, Joulfaian and Marlow (1991)
include federal with state and local spending to measure the size of government.
Their results also support the Leviathan hypothesis. These studies support the
decentralization hypothesis at the local level as it applies to general-purpose
governments.
Although support of the Leviathan hypothesis is provided by numerous
studies, the empirical evidence is not consistently supportive of such a view.
Oates (1989) provides a comparison of the differences between model specification, variable definitions, and levels of aggregation, which may explain some of
the inconsistencies. Not only are the empirical results inconsistent, implicitly
the searches for the effect of decentralization on the level of spending have
postulated that increases in government spending are inefficient. This is not
necessarily true. Whereas centralized decision-making may create a Leviathan,
fragmentation may not allow small jurisdictions, “Lilliputians,” to provide
beneficial services that exhibit economies of scale. Larger jurisdictions may
provide a greater range of services that are valued by residents and in doing so
are mislabeled Leviathans.
We determine whether inefficient political jurisdictions tend toward overspending with greater government concentration. With DEA we identify specifically which political jurisdictions are inefficient. DEA provides a score that
measures the degree of inefficiency. Using these efficiency scores we differentiate
between scale economies and government waste as the source of inefficiency. In
addition to DEA, in a second stage we use Tobit regression analysis to ascertain
the causes of the inefficiency.
To validate the results obtained using DEA, the model is reestimated using
a more traditional stochastic production-frontier approach, corrected ordinary
least squares (COLS). The comparison of the two approaches is necessary
because of the absence of statistical measures of significance and goodness
of fit in DEA. Although DEA and COLS are different approaches to
estimating production frontiers, a consensus of the two results will validate
the findings produced under the different sets of assumptions inherent in the
two methods.
3.
METHODS
Data Envelopment Analysis
Based on the frontier analysis suggested by Farrell (1957), data envelopment analysis (DEA) was developed by Charnes, Cooper, and Rhodes (1978) to
estimate the level of technical efficiency in production. The applications of DEA
in government, nonprofit, and for-profit institutions are vast. Seiford (1990) has
compiled a list of over 500 works that use DEA as an evaluative tool. Applications
specific to urban and regional issues include studies on industrial production by
township and village enterprises in China (Tong, 1996, 1997), the efficiency of
electricity retail distributors in Sweden (Hjalmarsson and Veiderpass, 1992),
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
653
analyses of the efficiency of local governments in Belgium (De Borger and
Kerstens, 1996) and New South Wales (Carrington et al., 1997), and studies that
analyze urban and highway transit programs (Nolan, 1996; Viton, 1998; Rouse,
Putterill, and Ryan, 1997). Yue (1992), Haag and Raab (1993), and Sherman
(1992) provide some of the more lucid explanations of how DEA works.
Seiford and Thrall succinctly introduce DEA as
a methodology directed to frontiers rather than central tendencies. Instead of
trying to fit a regression plane through the center of the data, one ‘floats’ a
piecewise linear surface to rest on top of the observations. Because of this
unique perspective, DEA proves particularly adept at uncovering relationships
that remain hidden for other methodologies. (1990, p. 8).
DEA uses linear-programming techniques to compare decision-making
units (DMUs) which may produce multiple outputs using multiple inputs.
Without specifying a functional form for the production technology, DEA is able
to estimate a production frontier defining the maximum output for the given
input level. Each production unit’s efficiency is measured relative to the efficiency of all other units. By construction, all units are on or below the frontier.
A single organization is defined as technically efficient if it cannot increase the
amount of one of its outputs without reducing other outputs or increasing inputs.
Thus, the method described by DEA is consistent with the economic theory of
optimization. In addition to being technically efficient an organization must
compare input prices and productivity to demonstrate economic efficiency. For
an organization to attain economic efficiency it must first demonstrate technical
efficiency.
The current study uses IDEAS Version 5.13 to solve the linear-programming
models of DEA. Linear-programming models are solved for each of the DMUs
in the analysis set. For each unit, the model searches for a linear combination
of units in the sample that produces a greater level of output with fewer inputs.
The linear combination represents a hypothetical composite unit that must
satisfy two inequality constraints: (1) all output levels are greater than or equal
to the output levels of the DMU under analysis, and (2) all input levels of the
hypothetical composite unit are less than or equal to the input levels of the unit
under analysis. The model is searching for a comparison that identifies output
slack or excess input usage of the unit under analysis, as defined by the above
inequality constraints.
In solving the linear-programming problem the user must specify three
characteristics of the model: the returns to scale, the evaluation system, and the
orientation system. Returns to scale may be either constant returns or variable
returns, following the standard economic definitions. The evaluation system
refers to weights placed on the inputs and outputs in the objective function,
subject to the inequality constraints. Weights may be desired in situations where
the scale of the inputs or outputs varies.
The orientation system, which defines the objective function, can be designated as input, output, or base. The input orientation system searches for a linear
© Blackwell Publishers 2000.
654
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
combination of DMUs that maximizes the excess input usage of DMUi subject
to the inequality constraints noted above. Under output orientation, the output
slack of DMUi is instead maximized. The base or nonoriented specification
maximizes the combined input excess and output slack.
The most general form of the DEA linear programming model is the base
or nonoriented model with variable returns to scale. The statements for DMUi
of this model are
Min – (ui si +vi ei ) with respect to λi
(1)
Objective Function
subject to
(2)
Y λi – Y i = s i
Output Slack
(3)
Xi – X λi = ei
Excess Input
n
with
∑λ
ij
= 1, λ i ≥ 0, e i ≥ 0, s i ≥ 0
j =1
With m inputs, s outputs, and n DMUs
Y ~ (s × n) matrix of outputs
X ~ (m × n) matrix of inputs
Yi ~ (s × 1) matrix of outputs for unit i
Xi ~ (m × 1) matrix of inputs for unit i
si ~ (s × 1) matrix of output slack
ei ~ (m × 1) matrix of excess input
λi ~ (n × 1) matrix of weights assigned to linear combination of comparison
set
ui, vi ~ (1 × s) and (1 × m) matrix of weights used in the evaluation process
of the objective function.
The base orientation of the model is apparent in the objective function,
Equation (1). The sum of the excess input and output slack is maximized. The
primary objective in the output-oriented model is to maximize the proportional
increase in outputs produced with a given level of inputs, Equation (2). For the
input-oriented model the objective is to maximize the proportional decrease in
inputs necessary to produce a given level of output, Equation (3). The model to
determine government efficiency used in this paper is based on an input
orientation. By using an input orientation, one can determine whether a political
jurisdiction can produce the same level of output with less input.
In defining the model as variable returns-to-scale, the weights determining
the hypothetical composite comparison λij must sum to one. Under constant
returns-to-scale there is no restriction on the sum of the weights. There is no
reason to assume that constant returns exist in community development so both
versions of the DEA model are estimated to determine possible sources of
inefficiency. When ui = 1 and vi = 1 the evaluation system is referred to as being
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
655
standard (or equal bounds). The model is searching for a hypothetical composite
comparison, the linear combination of DMUs, that maximizes the combined
output slack and excess input usage compared to DMUi.
In the case of the units-invariant (or DMU-specific) evaluation system, the
output slack vector and excess input vector are weighted in the objective function
by uir = 1/Yir, r = 1, . . . , s and vit = 1/Xit , t = 1, . . . , m. The units-invariant
specification accounts for variations in the scale of inputs and outputs when
evaluating the distance to the production frontier. For example, we do not want
to weight capital and labor equally when measuring excess input usage, if one
unit of capital represents a one-million dollar machine and one unit of labor
represents one hour’s worth of time. In the present study, which focuses on
property-value maximization, where the scale of the inputs varies substantially,
the appropriate evaluation system is units-invariant.
DEA produces an efficiency score, which estimates the input requirement
if the DMU operates efficiently. An efficient DMU receives a score of one
(100 percent). An efficiency score of one means that the DMU cannot produce
the same output level by using any less than 100 percent of the current inputs.
With an input orientation, inefficient DMUs receive a score of between zero and
one. For example, a score of 0.9 implies that the DMU could produce the same
output using 90 percent of its current inputs.
By designating the production process as constant returns rather than
variable returns, one can determine the cause of inefficiency to be either from
scale inefficiency or purely technical inefficiency. Scale efficiency relates to the
size of operation and whether it is cost efficient. If a unit is operating in a range
of increasing returns to scale, expansion of that unit’s operations will decrease
average production costs. Similarly, under decreasing returns to scale a contraction in the size of operations will reduce average costs. Only if the firm is
operating in a range of constant returns to scale will average costs be minimized.
Pure technical efficiency relates to waste in the production process due to
mismanagement of resources, regardless of the size of the operation.
The efficiency score measures overall technical efficiency by imposing
constant returns to scale in the DEA model. The overall technical efficiency is a
nonadditive combination of pure technical and scale efficiency. When variable
returns are stipulated, the efficiency score measures pure technical efficiency
only. Scale efficiency is not in question because the DMUs are, by construction,
compared to others of roughly equal size. Given that constant returns measures
inefficiencies due to both scale and pure waste, and variable returns captures
pure waste only, the measure of scale efficiency is produced by taking a ratio of
the two sets of efficiency scores. The scale-efficiency score equals one if and only
if technology exhibits constant returns to scale at the current level of operation
for DMUi. A score less than one implies that scale inefficiencies exist; the DMU
in question is either too small and experiencing increasing returns to scale, or
too large and experiencing decreasing returns to scale. Unfortunately, the linear
program used in this paper, IDEAS 5.13, does not allow one to determine whether
© Blackwell Publishers 2000.
656
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
the source of the scale inefficiency is from increasing or decreasing returns to
scale. In either case, the size of the operation is not cost effective, and either
expansion or contraction is in order. See Miller and Noulas (1994) and Färe,
Grosskopf, and Lovell (1994) for a more detailed exposition.
Corrected Ordinary Least Squares
As a validation of the DEA results the model is reestimated using regression
analysis. Aigner, Lovell, and Schmidt (1977) introduce a method of estimating a
production function that captures the frontier of production possibilities using
a modification of traditional regression analysis. Rather than estimating the
average production function, which allows for a symmetric measurement error
in the specification, their composed-error frontier model includes an error term
composed of both a two-sided measurement error and a one-sided inefficiency
error. Banker, Gadh, and Gorr (1993) demonstrate a relatively easy method of
estimating the production frontier using corrected ordinary least squares
(COLS). Once a production frontier is estimated, Jondrow et al. (1982) provide
a formula for estimating the individual inefficiencies, the ωi, for each DMU. Their
method provides efficiency scores comparable to that of DEA.
The general formulation of the COLS model is
Yi = βXi – εj
εi = υi + ωi
where υ i ~ N 0, σ 2υ is the two-sided measurement error term, and ωi ≥ 0 is the
e
j
one-sided measure of technical inefficiency. The one-sided inefficiency term is
generally modeled as half-normal, exponential, or gamma. Based on the assumed distribution of the inefficiency term, maximum likelihood may be used
to estimate the parameters of the production function.
An alternative to the maximum likelihood method, Banker, Gahd, and Gorr
(1993) use corrected ordinary least squares (COLS) as a less demanding way to
estimate the production frontier. Ordinary least squares (OLS) provides unbiased and consistent estimates of all of the production parameters except the
intercept term. COLS corrects for the bias in the intercept term from OLS,
allowing a practical means of estimating composed error frontier models.
The bias of the intercept term is given by the mean of ε
µ = − 2σ 2ω π
e
j
12
The individual inefficiencies are given by the mode of ωi conditional on εi
c hRST
2
2
M ω ε = −ε σ υ σ ε
=0
© Blackwell Publishers 2000.
if ε ≤ 0
if ε > 0
UV
W
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
657
The parameters σ ω2 , σ 2υ , σ 2ε can be consistently estimated using the second and
third moments of the OLS residuals m2 and m3, respectively.
σ ω2 =
{bπ 2g
12
b
g }
π π − 4 m3
b
23
g
σ 2υ = m2 − π − 2 π σ 2ω
σ 2ε
=
σ ω2
+ σ 2υ
The value of ε is the corrected ordinary least squares residual. The efficiency
score is given by
b
g
COLS Efficiency Score = Yi − ω i Yi
Comparison of DEA and COLS
The major advantage of DEA is that it is nonparametric; it does not stipulate
a functional form for the production process and there is no assumption as to
the distribution of the error term. In addition, it is able to handle multiple inputs
and multiple outputs easily. The major disadvantage of DEA is the assumption
that all deviations from the frontier are due to technical inefficiencies, with no
allowance for randomness in the production process and measurement errors
in the variables. In cases involving large measurement errors DEA frontier
estimates are biased outward, overstating the degree of inefficiency of units
within the frontier.
In contrast to DEA, stochastic frontier estimation allows deviations from
the frontier to include both measurement errors and inefficiency. Regression
analysis has the advantage over DEA in cases involving relatively large measurement errors assuming the classical assumptions defining the model are met.
In particular, regression analysis requires the stipulation of both a functional
form for the production process and probability distribution for the error term.
Incorrect specification of either the production function or the error distribution
can result in estimation errors.
4.
THE DATA AND EMPIRICAL RESULTS
Overview
In this section we describe the model of property-value maximization and
the empirical results. Using DEA we identify which political jurisdictions are
efficient or inefficient. Using the efficiency scores from both the constant and
the variable returns models, we differentiate between scale economies and
government waste as the source of inefficiency. Using Tobit analysis we regress
the efficiency scores on the size and concentration measures of government
activity to determine specific causes of the inefficiency. A similar analysis is
performed using COLS in an attempt to validate the results of the DEA analysis.
© Blackwell Publishers 2000.
658
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
We investigate county government efficiency in the State of Minnesota.
Sonstelie and Portney (1978, 1980) point out that if communities act as profitmaximizing firms then resources will be efficiently allocated to the provision of
local public goods. Therefore, public sector efficiency will correspond to maximization of property values with respect to local fiscal policy. Following Deller
(1990), we use county-level data to capture interjurisdictional spillover effects.
Although fiscal policy exerts a significant impact on property values, other
significant input factors include land and water area, residential characteristics, market factors, community characteristics, and employment opportunities. Table 1 lists the variables used as inputs in the analysis. In keeping with
the concept of a technical production function, all input measures theoretically
exert a positive influence on total property value. Hence, we transform variables
TABLE 1: Variables Used in the Frontier Estimation
Output:
Total Property Value ($)
Mean
Median
Standard Deviation
$891,090,000
$539,440,000
$1,220,600,000
91,365
Inputs:
Discretionary
Education ($000)
40,101
16,826
Discretionary
Social Services ($000)
16,041
6,429
47,666
Discretionary
Transportation ($000)
10,033
4,964
22,565
Discretionary
Public Safety ($000)
6,505.9
1,604
22,021
Discretionary
Environment ($000)
10,262
2,176
34,277
Discretionary
Administration ($000)
5,283.2
1,899
14,863
Discretionary
Intergovernmental transfers
(Percent of revenue from
federal and state sources)
0.50169
.51017
0.086917
NonDiscretionary
Land Area (thousands
of squared kilometers)
2370.2
1735.8
2069.5
NonDiscretionary
Discretionary
Water Area (thousands
of squared kilometers)
Total number of homes
218.10
39.080
630.85
21,246
9,675
53,053
Discretionary
Median number of rooms
5.7823
5.8759
0.41121
Discretionary
Discretionary
Median income ($)
(1 – Poverty Rate)
25,052
.877
23,278
.88048
5,758.9
0.042449
Discretionary
(1 – Crime Rate)
.92886
.93274
0.034436
Discretionary
Employment (number of
employees per county)
18,997
4,616
74,476
Discretionary
1/housing density
(land area/homes)
0.28982
0.22894
0.24497
Discretionary
1/average commute time
(minutes)
0.046868
0.046235
0.0095850
Discretionary
Percent who live and work
in the county
0.82600
0.86437
0.10488
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
659
that negatively affect property values into positive effects using 1-poverty rate,
1-crime rate, 1/housing density, and 1/average commute time.
The output measure, total property value, is defined as the market value of
residential, commercial, and industrial property. Fiscal policy is represented by
government expenditures, found by aggregating county, municipality, township,
special district, and school district spending. Government expenditures are
grouped by major categories of spending: education, social services, transportation, public safety, environment, and administration. In addition, interjurisdictional government aid is represented by the percent of revenue received from
federal and state sources. County size is measured by both square kilometers of
land area and, perhaps particularly pertinent in Minnesota, square kilometers
of water area. Determinants of residential property value include the number
and quality of homes with quality proxied by median number of rooms. Median
income will influence housing value following the bid-rent concept. In addition,
median income, poverty rate, crime rate, and housing density proxy community
characteristics. Employment opportunities are represented by the percent of
residents who live and work in the same county, average commute times, and
the number of employees per county. The employment opportunities reflect the
value of commercial and industrial property.
Data pertaining to government composition are obtained from the Census
of Governments, Compendium of Government Finances (1987). Employment
data are drawn from the County Business Patterns (1987). Housing and population data are drawn from the Census of Population and Housing, 1990: Summary
Tape File 1 (Minnesota) and Summary Tape File 3 (Minnesota). The journey to
work data set is provided by the Regional Economic Information System,
(1969–1995). The Minnesota Annual Report on Crime, Missing Children, and
Bureau of Criminal Apprehension Activities (1987) gives crime rates per county.
DEA Empirical Results
Table 2 presents four sets of efficiency scores, three for the DEA analyses
and one from the COLS analysis. The first set of scores is generated by imposing
constant returns to scale (CRS). The second set of scores is generated from
variable returns to scale (VRS). Both models are based on an input orientation
with the inputs weighted to account for disparate units of measurement. In
addition, both land and water area within a county are designated as fixed or
nondiscretionary from a policy standpoint. Nondiscretionary variables are not
subjected to proportional reductions in an input-oriented model. CRS scores
measure overall technical efficiency. Thus, the CRS scores are composed of a
nonadditive combination of pure technical and scale efficiencies. VRS scores
measure pure technical efficiency only. A ratio of the overall efficiency scores to
pure technical efficiency scores provides a scale efficiency measure, the third set
of efficiency scores. The fourth set of efficiency scores provided in Table 2 is for
the COLS method. This last set of scores will be discussed later.
© Blackwell Publishers 2000.
660
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
TABLE 2: Efficiency Scores
FIPS Code — County Name
DEA-CRS
DEA-VRS
Scale (CRS/VRS)
COLS
1 - Aitkin
3 - Anoka
5 - Becker
7 - Beltrami
9 - Benton
11 - Big Stone
13 - Blue Earth
15 - Brown
17 - Carlton
19 - Carver
21 - Cass
23 - Chippewa
25 - Chisago
27 - Clay
29 - Clearwater
31 - Cook
33 - Cottonwood
35 - Crow Wing
37 - Dakota
39 - Dodge
41 - Douglas
43 - Fairbault
45 - Fillmore
47 - Freeborn
49 - Goodhue
51 - Grant
53 - Hennepin
55 - Houston
57 - Hubbard
59 - Isanti
61 - Itasca
63 - Jackson
65 - Kanabec
67 - Kandiyohi
69 - Kittson
71 - Koochiching
73 - Lac Qui Parle
75 - Lake
77 - Lake of the Woods
79 - Le Sueur
81 - Lincoln
83 - Lyon
85 - McLeon
87 - Mahnomen
1.000000
1.000000
0.446740
0.447430
1.000000
0.538060
0.599510
0.659560
0.311200
0.831730
1.000000
0.628880
0.617390
0.487560
0.528240
1.000000
0.771270
1.000000
1.000000
1.000000
0.480710
0.769100
0.691480
0.473060
0.827240
0.827600
0.179950
0.515160
0.795570
0.589540
0.564510
1.000000
0.707400
0.486740
1.000000
0.290400
0.775200
0.439070
0.554210
0.693470
1.000000
0.629140
0.551580
0.622550
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.995520
1.000000
N/A
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.979270
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.973340
1.000000
1.000000
0.980940
1.000000
1.000000
1.000000
1.000000
0.970750
1.000000
1.000000
1.000000
1.000000
0.446740
0.447430
1.000000
0.538060
0.599510
0.659560
0.311200
0.831730
1.000000
0.631710
0.617390
N/A
0.528240
1.000000
0.771270
1.000000
1.000000
1.000000
0.480710
0.769100
0.691480
0.483074
0.827240
0.827600
0.179950
0.515160
0.795570
0.589540
0.564510
1.000000
0.707400
0.500072
1.000000
0.290400
0.790262
0.439070
0.554210
0.693470
1.000000
0.648097
0.551580
0.622550
1.000000
1.000000
1.000000
1.000000
1.000000
0.984710
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.936157
0.962583
0.992473
0.988125
1.000000
1.000000
1.000000
1.000000
0.995036
1.000000
1.000000
1.000000
1.000000
0.992878
1.000000
1.000000
1.000000
1.000000
1.000000
0.985478
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.947258
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
661
FIPS Code — County Name
DEA-CRS
DEA-VRS
Scale (CRS/VRS)
COLS
89 - Marshall
91 - Martin
93 - Meeker
95 - Mille Lacs
97 - Morrison
99 - Mower
101 - Murray
103 - Nicollet
105 - Nobles
107 - Norman
109 - Olmsted
111 - Otter Tail
113 - Pennington
115 - Pine
117 - Pipestone
119 - Polk
121 - Pope
123 - Ramsey
125 - Red Lake
127 - Redwood
129 - Renville
131 - Rice
133 - Rock
135 - Roseau
137 - St. Louis
139 - Scott
141 - Sherburne
143 - Sibley
145 - Stearns
147 - Steele
149 - Stevens
151 - Swift
153 - Todd
155 - Traverse
157 - Wabasha
159 - Wadena
161 - Waseca
163 - Washington
165 - Watonwan
167 - Wilkin
169 - Winona
171 - Wright
173 - Yellow Medicine
Averages
1.000000
1.000000
0.419430
0.420590
0.464850
1.000000
0.903610
0.850000
0.621750
1.000000
1.000000
0.578840
0.407710
0.771300
0.565640
0.543720
0.746160
1.000000
0.744640
1.000000
1.000000
0.528260
1.000000
0.677950
0.360360
0.916030
1.000000
0.764030
0.694600
0.849860
0.657180
0.614760
0.436410
1.000000
0.521580
0.340980
0.685780
1.000000
1.000000
1.000000
0.508580
0.829400
0.737580
0.718308
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
N/A
0.980820
1.000000
1.000000
0.985480
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.990970
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
N/A
0.998299
1.000000
1.000000
0.419430
0.420590
0.464850
1.000000
0.903610
N/A
0.633908
1.000000
1.000000
0.587369
0.407710
0.771300
0.565640
0.543720
0.746160
1.000000
0.744640
1.000000
1.000000
0.528260
1.000000
0.684128
0.360360
0.916030
1.000000
0.764030
0.694600
0.849860
0.657180
0.614760
0.436410
1.000000
0.521580
0.340980
0.685780
1.000000
1.000000
1.000000
0.508580
0.829400
N/A
0.720295
0.999049
1.000000
1.000000
1.000000
1.000000
0.989776
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.936632
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
0.967150
1.000000
1.000000
0.997807
1.000000
1.000000
0.996266
Note: Efficiency scores for three counties are not calculated due to scaling problems in the DEA
programming.
© Blackwell Publishers 2000.
662
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
By imposing CRS in the programming, we identify 25 efficient counties,
leaving 62 inefficient ones. The inefficient counties are either operating at an
inappropriate level or are mismanaged. The inappropriate level could be due to
not taking advantage of scale economies or not offering an efficient quantity of
public goods. Mismanagement refers to pure waste, that is, an inappropriate use
of taxpayer dollars.
Under VRS, 76 counties are identified as efficient, 8 are inefficient, and 3
scores are not reported due to scaling problems with the data in the DEA
program. Under VRS, counties are compared to those of like size, hence, the eight
counties identified as inefficient exhibit some type of mismanagement or waste
given their scale of operation.
Comparing the overall level of inefficiency (CRS scores) to that component
due to pure technical inefficiency (VRS scores), the dominant source of inefficiency is clearly due to scale economies. The average efficiency score under CRS
is equal to 0.718. Including all sources of inefficiency, on average Minnesota
counties could operate at 71.8 percent of their current input levels and maintain
the same total property value. However, the average efficiency score under VRS
is equal to 0.998. Given the scale of operation, a majority of counties are efficient
in managing their resources. Indeed, the magnitude of waste due to inappropriate spending is on average less than 1 percent.
Although DEA identifies inefficient counties in the sample it does not
identify the cause of the inefficiency. The inefficient counties are each given a
reference set which allows for specific recommendations to improve efficiency.
By studying and incorporating the fiscal operations of efficient counties in the
reference set, inefficient counties may realize substantial cost savings or increased property values. As an example, data for Morrison County, data for one
county in the reference set for Morrison County, and data for adjacent counties
are provided in Table 3.
Under constant returns to scale Morrison County receives an efficiency
rating of 0.46485 (see Table 2). For Morrison County, aggregating 0.0025 of the
inputs and outputs of Dakota County and 0.42896 of the inputs and outputs of
Sherburne County creates the hypothetical composite county used in the comparison. For the sake of exposition, consider a roughly similar hypothetical
county based on a scaled-down version of Sherburne County only. The scale
(0.4424569), the ratio of property values of Morrison County to Sherburne
County, is chosen to present a comparison of equal market value. Morrison
County is judged inefficient because we can take half of Sherburne County,
retain the same market value as Morrison County, but use less input to achieve
this value. By definition, this imposes constant returns to scale in the definition
of efficiency.
The data from Table 3 suggest that one major difference between Morrison
and Sherburne County is population density. Sherburne County is more congested but it has more and larger homes in less than half the land (and water)
area of Morrison County. The level of enterprise and employment is greater, as
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
663
TABLE 3: DEA Inputs and Output for Morrison and Adjacent Counties
Morrison
Actual
Data
Outputs
Total Property Value
Inputs
Revenue Transfer (Percent)
Median Number Rooms
Total Number of Homes
Median Income
Employment
1/(Average Commute Time)
1/(Housing Density)
Work in Community
(Percentage)
Education
Social Services
Transportation
Public Safety
Environment
Administration
1 – Poverty Rate
1 – Crime Rate
Land Area
Water Area
Hypothetical
Composite
Data = 0.4425*
Sherburne
Crow
Wing
Sherburne
Benton
Actual
Data
Actual
Data
Actual
Data
702250
702249.9
1587160
611579
1506720
0.57
5.55
12434
22102
5607
0.0478
0.2342
0.20353
2.681289
6620.925
15744.83
2735.269
0.01353918
0.03344974
0.46
6.06
14964
35585
6182
0.0306
0.0756
0.55
5.53
11521
26619
6800
0.044
0.0918
0.6
5.42
29916
22250
11757
0.0475
0.0863
0.87
25591
6538
7238
1691
1836
2116
0.84
0.92
2912.58
74.65
0.256625
12660.9
2395.019
2152.553
1148.618
848.6323
1170.299
0.40706
0.40706
500.2904
16.54346
0.58
28615
5413
4865
2596
1918
2645
0.92
0.92
1130.71
37.39
0.51
15349
4530
2770
1401
676
1967
0.9
0.96
1057.48
12.24
0.88
39758
17928
9292
4713
2715
5041
0.85
0.9
2581.44
414.22
is the overall level of public goods provided, except for transportation and social
services. Economic development seems to be the key difference. The level of
economic development activity may be instigated and compounded by Sherburne County’s closer proximity to the Twin Cities of Minneapolis and St. Paul.
Although Sherburne County is chosen to most clearly identify the inefficiency of Morrison County, some adjacent counties deemed as efficient by DEA
may shed some light on the problems encountered in Morrison County. Benton
County is sandwiched between Sherburne and Morrison Counties. Benton
County is about one-third the size of Morrison with a just slightly smaller total
market value. Again the key difference lies in the level of economic development.
Benton County has nearly the same number of homes and slightly more jobs in
one-third of the land area. Benton County also has a significantly smaller public
sector, particularly in transportation and education. The percent of Benton
residents working outside of the county is 49 percent, higher even than that of
Sherburne County. Although closer proximity to the Twin Cities may again be
responsible for the higher property values, the average commute time of 22.7
minutes, with 82 percent traveling 30 minutes or less to work, does not indicate
© Blackwell Publishers 2000.
664
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
a strong tie to the Twin Cities’ job market. It takes over one hour to commute
from central Benton County to the closest sector of the Twin Cities.
Adjacent and north of Morrison County, Crow Wing County is also judged
efficient by DEA standards. Crow Wing County has slightly less land area but
a significantly greater water area than Morrison. The level of economic development in Crow Wing County is significantly greater than in Morrison County.
Crow Wing County supports nearly twice the employment and number of homes
as Morrison. It is interesting to note that although public sector support
increases in most areas in relation to the increased scale of the private sector,
spending on social services and public safety nearly triples in Crow Wing County.
By location and commuting patterns, proximity to the Twin Cities cannot be the
determining factor affecting development. The greater water area and resort
development may be a key factor in the increased development and property
value of Crow Wing County.
Morrison County is bordered on the north and south by efficient, propertymaximizing counties. Superficially we may claim that Sherburne County’s
success lies in its proximity to the Twin Cities, and Crow Wing’s success lies in
its recreational water resource. However, the cause of Benton County’s efficiency
is not obvious. In all three cases the level of economic development is higher
than in Morrison County, driving up property values. A larger agricultural sector
in Morrison County is a detriment to property-value maximization. Although
Sherburne is highlighted in Morrison’s reference set, collaboration with Benton
County may provide a more useful insight into the area of property-value
maximization.
Although the strategy for realizing the input reductions specified by DEA
is unique to each county, we use Tobit regression analysis to determine if there
are some universal strategies or county characteristics that influence the
efficiency rating. Table 4 presents the regression estimates of the Tobit model
that are used to pinpoint the factors that influence the size of the efficiency score
for both the DEA and COLS models. The Tobit estimation procedure is used
because the efficiency scores have a maximum value of one. In a conventional
two-stage DEA model the technical efficiency scores obtained from the firststage DEA model are regressed on some location-specific fixed inputs in the
second stage. While the DEA inputs are expected to increase property values,
the question remains as to why some counties are able to increase property
values above that of other counties with similar inputs. To shed light on this
question the scale efficiency scores are regressed on factors that reflect the size
of the jurisdiction to capture scale economies, and the number of government
units and level of government spending to capture Leviathan tendencies. The
VRS efficiency scores do not show enough variation to warrant consideration.
The results in Table 4 indicate that an increase in county size, as measured
by land area, has a negative and significant impact on the DEA efficiency score.
Given that the dominant source of inefficiency is attributable to scale economies
these results indicate that diseconomies of scale are present in the expansion of
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
665
county boundaries. The presence of Leviathan tendencies implies that the size
of the public sector will vary inversely with the degree of decentralization. This
implication is supported by the Tobit results. Efficiency significantly increases
as the number of government units per capita increases. An increase in the size
of the public sector paid for by local residents has a negative, although insignificant, impact on efficiency. Overall the results indicate that counties encompass
too large of an area, benefit from decentralization and, to a limited extent, spend
too much per capita. Even though the level of pure waste in government appears
minimal, the above tendencies are consistent with the Leviathan view of
government.
COLS Empirical Results
The estimates of the stochastic production function are presented in Table 5.
The function is log linear with squared and interactive terms included for
government expenditures. The nonlinear functional form, specifically the inclusion of squared terms, allows the return to property values from increasing
government expenditures to decline and eventually become negative if excessive
spending occurs. The interaction terms allow for the public-good nature of
government spending, with larger communities receiving greater benefit from
a given level of spending due to the greater number of consumers.
Theoretically, the coefficients are expected to be positive for the linear terms,
negative for the squared terms, and positive for the interactive terms. However,
the empirical results do not strictly adhere to such a pattern due to the
significant degree of multicollinearity in the data. Although the estimators
remain unbiased, the reliance that can be placed on the value of the estimates
will be small, and interpretation of the coefficients quite difficult. Although
multicollinearity affects the precision of the coefficient estimators, it has little
effect on the overall fit of the equation. The R2 of 0.9574 shows a reasonably good
TABLE 4: Regression Coefficient Estimates of Tobit Model: Dependent
Variable = Efficiency Score: Maximum Value of One (Asymptotic normal
statistics in parentheses)
Independent Variables:
Constant
DEA Scale Efficiency
Scores
0.79709**
(5.58)
Land
–0.3255E-04**
(–2.15)
Total Government Expenditures Per Capita Financed Locally –0.7018E-01
($000)
(–0.54)
Number of Governmental Units Per Capita
56.202**
(2.29)
Number of Observations
84
Squared Correlation Between Observed and Expected
0.12497
**five percent significance level
© Blackwell Publishers 2000.
COLS Efficiency
Scores
1.0257**
(131.0)
–0.2412E-05**
(–2.90)
–0.1914E-01**
(–2.75)
0.22248
(0.18)
87
0.25712
666
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
TABLE 5: Corrected Ordinary Least Squares Stochastic Production
Function: Dependent Variable = Property Value
Estimated Coefficient
Independent Variables:
Intergovernmental Transfers
Land
Water
Homes
Median Rooms
Median Income
Employment
Inverse Commute Time
1-Poverty Rate
1-Crime Rate
Inverse Housing Density
% Work in Community
Education
Social Services
Transportation
Public Safety
Environment
Administration
Education Squared
Social Services Squared
Transportation Squared
Public Safety Squared
Environment Squared
Administration Squared
Education × Homes
Social Services × Homes
Transportation × Homes
Public Safety × Homes
Environment × Homes
Administration × Homes
Constant
Number of Observations
87
.9574
R2
–1.2018**
2888.9
0.0225
–2890
3.8985**
0.4264
–0.14572
0.0142
–0.37257
–1.1405
–2888.6
–0.26303
–4.0743**
2.2323**
–2.4818**
0.44918
0.35154
6.3679**
0.00683
0.0333
0.4332*
–0.25578
0.0964
–0.65808
0.44273
0.31411**
–0.51337
0.37851
–0.2104
0.45459
5.1175
t-ratio
–3.401
0.6537
1.004
–0.654
5.115
0.799
–1.443
0.0571
–0.2321
–1.027
–0.6537
–0.5393
–2.807
3.776
–2.416
0.1858
0.4184
2.878
0.0437
1.45
1.916
–0.9289
1.662
–1.585
1.373
–3.173
–1.249
0.5719
–1.352
0.5364
0.7282
**5 percent significance level
*10 percent significance level
Note: All variables are logged, with the exception of those reported in percentages.
fit of the average production function to the data. Because multicollinearity has
little effect on the overall fit of the equation, it will also have little effect on the
use of the equation for prediction. The main focus of the model is on the fit of
the production function rather than isolating the independent effects on market
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
667
value. The presence of multicollinearity does not affect the consistency of the
resulting efficiency scores.
The efficiency scores derived from the COLS procedure along with the three
DEA efficiency scores are presented in Table 2. The COLS scores identify 14 of
the 87 counties as inefficient. The average efficiency score over the 87 counties
is 0.996, roughly equal to the average of the DEA scores allowing variable
returns to scale. However, the correlation between the COLS efficiency scores
and any of the DEA efficiency scores is virtually zero.
Although the efficiency scores from the two procedures show a surprisingly
low correlation in indicating and ranking county efficiency, the determinants of
the rankings using Tobit regression analysis show much greater consistency in
the two methods. Table 4 shows the results of regressing the efficiency scores
generated from COLS on the size of the jurisdiction and characteristics of
government in that jurisdiction. The signs of the parameters are the same in
both methods. In both models the size of the county has a negative and significant
impact on the efficiency score, again indicating diseconomies of scale. Per capita
expenditures financed locally exert a negative and significant impact on the COLS
efficiency scores, indicating excessive spending. In the DEA model the effect of
increasing expenditures was negative but not significant. Whereas the number
of government units per capita has a positive impact on both sets of efficiency
scores, the impact is not significant in the COLS model. The consistency in the
Tobit regression analysis between the DEA and COLS models provides further
support for the Leviathan view of government behavior.
5.
CONCLUSION
Our paper extends the research on the efficiency of government performance using the method of data envelopment analysis (DEA). With DEA we
identify the counties in the State of Minnesota that are maximizing property
values, and hence the counties that have an efficient public sector. DEA produces
efficiency scores that measure the degree of inefficiency. Using the efficiency
scores we differentiate between scale inefficiency and pure technical inefficiency,
and we conclude that the primary cause of inefficiency is due to the size of the
jurisdiction relative to scale economies. The aspect of production inefficiency or
government waste is a minor problem.
Using Tobit regression analysis we examine the cause of scale inefficiency.
Scale inefficiency may come from a jurisdiction that is too small and therefore
experiencing increasing returns to scale, or one that is too large and experiencing
decreasing returns to scale. The results indicate that jurisdictions with larger
land area tend to be less efficient, that is, they tend to experience diseconomies
of scale. Inefficiencies relating to the size and degree of concentration in
government indicate that greater decentralization and decreased spending by
the public sector increase efficiency. The evidence indicates the presence of
Leviathan tendencies, with inefficient counties typified by increased land area,
increased government concentration, and greater levels of spending. The smaller
© Blackwell Publishers 2000.
668
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 4, 2000
Lilliputian governmental units are more responsive to consumer demand. These
results are supported by an alternative method of estimation, corrected ordinary
least squares which also indicates the negative effect of land area, government
expenditures, and government concentration on public sector efficiency.
Although greater decentralization tends to increase government efficiency,
an area of further research is to determine which level of government is more
efficient. As a measure of decentralization we use the number of government
units in the county. We do not address the question of the level of governmental
unit (municipality, township, school district, or special district) that is responsible for the increased efficiency.
Inconsistencies in the empirical evidence supporting the Leviathan hypothesis are attributable to differences in model specification, variable definitions, and levels of aggregation. Although DEA will suffer the same problems
regarding variable definitions and aggregation levels, the lack of model specification in DEA may provide some consistency to the empirical results.
REFERENCES
Aigner, Dennis, C.A. Knox Lovell, and Peter Schmidt. 1977. “Formulation and Estimation of
Stochastic Frontier Production Function Models,” Journal of Econometrics, 6, 21–37.
Banker, Rajiv, Vandana Gadh, and Wilpen Gorr. 1993. “A Monte Carlo Comparison of Two Production
Frontier Estimation Methods: Corrected Ordinary Least Squares and Data Envelopment Analysis,” European Journal of Operational Research, 67, 332–343.
Brennan, Geoffrey and James Buchanan. 1980. The Power to Tax: Analytical Foundations of a Fiscal
Constitution. New York: Cambridge University Press.
Brueckner, Jan K. 1979. “Property Values, Local Public Expenditure and Economic Efficiency,”
Journal of Public Economics, 11, 223–245.
———. 1982. “A Test for Allocative Efficiency in the Local Public Sector,” Journal of Public Economics,
19, 311–331.
Carrington, Roger, Nara Puthucheary, Deirdre Rose, and Suthathip Yaisawarng. 1997. “Performance
Measurement in Government Service Provision: The Case of Police Services in New South Wales,”
Journal of Productivity Analysis, 8, 415–430.
Charnes, A., W.W. Cooper, and E. Rhodes. 1978. “Measuring the Efficiency of Decision-Making Units,”
European Journal of Operations Research, 2, 429–444.
De Borger, Bruno and Kristiaan Kerstens. 1996. “Cost Efficiency of Belgian Local Governments: A
Comparative Analysis of FDH, DEA, and Econometric Approaches,” Regional Science and Urban
Economics, 26, 145–170.
Deller, Steven. 1990. “An Application of a Test for Allocative Efficiency in the Local Public Sector,”
Regional Science and Urban Economics, 20, 395–406.
Department of Public Safety, Bureau of Criminal Apprehension. 1987. Minnesota Annual Report on
Crime, Missing Children, and Bureau of Criminal Apprehension Activities. St. Paul, MN: Criminal
Justice Information Systems Section.
Eberts, Randal W. and Timothy J. Gronberg. 1988. “Can Competition Among Local Governments
Constrain Government Spending?” Federal Reserve Bank of Cleveland Economic Review, 24, 2–9.
Färe, Rolf, Shawna Grosskopf, and C.A. Knox Lovell. 1994. Production Frontiers. New York: Cambridge University Press.
Farrell, M.J. 1957. “The Measurement of Productive Efficiency,” Journal of the Royal Statistical
Society, A, General, 120, 253–290.
© Blackwell Publishers 2000.
HUGHES & EDWARDS: GOVERNMENT EFFICIENCY
669
Haag, Steven E. and Raymond L. Raab. 1993. “A Current Survey of the Development and Application
of Data Envelopment Analysis and a Comparison to Stochastic Approaches,” Working Paper No.
93-8, Bureau of Business and Economic Research, Center for Economic Development, School of
Business and Economics, University of Minnesota, Duluth.
Hjalmarsson, Lennart, and Ann Veiderpass. 1992. “Efficiency and Ownership in Swedish Electricity
Retail Distribution,” Journal of Productivity Analysis, 3, 7–23.
Jondrow, James, C.A. Knox Lovell, Ivan Materov, and Peter Schmidt. 1982. “On the Estimation of
Technical Inefficiency in the Stochastic Frontier Production Function Model,” Journal of Econometrics, 19, 233–238.
Joulfaian, David and Michael L. Marlow. 1991. “Centralization and Government Competition,”
Applied Economics, 23, 1603–1612.
Miller, Stephen and Athanasios G. Noulas. 1994. “The Technical Efficiency of Large Bank Production,” Working Paper No. 94-203, Economics Department, University of Connecticut.
Nelson, Michael A. 1987. “Searching for Leviathan: Comment and Extension,” American Economic
Review, 77, 198–204.
Nolan, James F. 1996. “Determinants of Productivity Efficiency in Urban Transit,” Logistics and
Transportation Review, 32, 319–342.
Oates, Wallace E. 1985. “Searching for Leviathan: An Empirical Study,” American Economic Review,
75, 748–757.
———. 1989. “Searching for Leviathan: A Reply and Some Further Reflections,” American Economic
Review, 79, 578–583.
Rouse, Paul, Martin Putterill, and David Ryan. 1997. “Towards a General Managerial Framework
for Performance Measurement: A Comprehensive Highway Maintenance Application,” Journal of
Productivity Analysis, 8, 127–149.
Seiford, Lawrence M. and Robert M. Thrall. 1990. “Recent Developments in DEA: The Mathematical
Programming Approach to Frontier Analysis,” Journal of Econometrics, 46, 7–38.
Sherman, H. D. 1992. “Data Envelopment Analysis (DEA): Identifying New Opportunities to Improve
Productivity,” Tijdschrift voor Economie en Management, 37, 153–180.
Sonstelie, Jon C. and Paul R. Portney. 1978. “Profit Maximizing Communities and the Theory of Local
Public Expenditures,” Journal of Urban Economics, 5, 263–277.
———. 1980. “Gross Rents and Market Values: Testing the Implications of Tiebout’s Hypothesis,”
Journal of Urban Economics, 7, 102–118.
Tiebout, Charles. 1956. “A Pure Theory of Local Expenditures,” Journal of Political Economy, 64,
416–424.
Tong, Christopher S. P. 1996. “Industrial Production Efficiency and Its Spatial Disparity Among the
TVEs of China: A DEA Analysis.” Singapore Economic Review, 41, 85–101.
———. 1997. “China’s Spatial Disparity Within the Context of Industrial Production Efficiency: A
Macro Study by the Data Envelopment Analysis (DEA) System,” Asian Economic Journal, 11,
207–217.
United States Department of Commerce, Bureau of the Census. 1987a. Census of Governments,
Compendium of Government Finances, Washington: U.S. Government Printing Office.
———. 1987b. County Business Patterns, Washington: U.S. Government Printing Office.
———. 1991a. Census of Population and Housing, 1990 Summary Tape File 1 (Minnesota), Washington: Economic and Statistics Administration.
———. 1991b. Census of Population and Housing, 1990 Summary Tape File 3 (Minnesota), Washington: Economic and Statistics Administration.
United States Department of Commerce, Bureau of Economic Analysis. 1997. Regional Economic
Information System, 1969–95, Washington, DC: Economic and Statistics Administration.
Viton, Philip A. 1998. “Changes in Multi-mode Bus Transit Efficiency, 1988–1992,” Transportation,
25, 1–21.
Yue, Piyu. 1992. “Data Envelopment Analysis and Commercial Bank Performance: A Primer with
Applications to Missouri Banks,” Federal Reserve Bank of St. Louis Review, 74, 31–45.
© Blackwell Publishers 2000.