1
Kent (1998) has documented the use of privateers during the 16
th
century reign of Elizabeth I as even earlier evidence of
contracting out overseas.
2
As Boyne (1998) notes, population size also has been used in many studies as a measure of scale economy to test the
hypothesis that cities with smaller populations are more likely to contract out than larger ones. The results, however, have
been mixed and inconsistent with the scale hypothesis.
3
Where exactly do the savings from contracting lie? According to Rehfuss, private contractors are able to lower their labor
costs by offering lower wages and fringe benefits, by hiring employees for fewer hours, and by hiring younger employees.
Kettl (1993), after observing patterns in state and local government contracting during the 1980s and 1990s, concluded that
contractors were able to reduce costs through more flexible use of the workforce, lower wages, and lower fringe benefits.
“With so much of state and local spending concentrated on personnel costs, any argument to reduce that spending inevitably
is an argument to reduce the number of employees as well” (1993, pp. 161-162).
4
This finding is not surprising, given the assertion made by advocates of privatization that privatized service delivery is
nearly always more efficient (Savas 1987, 2000), even though the empirical evidence to support this claim is somewhat
mixed (see Hodge, 2000; Sclar, 2000).
5
Interestingly, Greene (2002) also suggests that a sharp decline in population can cause an increase in the use of contracting
out due to fiscal strain caused by a shrinking economy and tax base.
6
Missing values are replaced with the average value.
7
The examples are property tax levy limit, general revenue or expenditure increases limit, overall and specific property tax
rate limitations, and limits on assessment increases.
8
The Level I submodel is specified as follows (Singer 1998; Heinrich 2000, 84-85; Hox 1995, 10-23; Littell, Milliken,
Stroup, and Wolfinger 1996, 135-169, 253-266):
Y
ij
=
β
0j
+
β
1j
X
1ij
+ … +
β
nj
X
nij
+ r
ij
(1)
Y
ij
is a measure of contracting out by local governments, where i denotes each local government responding to the 2002
ICMA survey and j denotes the state to which each local government belongs. At level I, we express the measure of
contracting out as a function of an intercept (the estimated state level measure of contracting out,
β
0j
), Level I variables (X
1ij
to X
nij
), and errors associated with the i
th
local government in the j
th
state (r
ij
). The subscript j denotes that each state has
unique intercepts and slopes. It is notable here that our dependent variable is a count variable, which counts the number of
total services provided through for-profit and non-profit providers for each local government. Counts have typically been
assumed to follow a Poisson distribution. For this reason, we have run HLM models with the Poisson error distribution
(Littell, Milliken, Stroup, and Wolfinger 1996, 423-460). Our preliminary HLM regression analyses indicate that the
models with unique slopes (i.e., randomly varying slopes across states) do not perform well in terms of the fit statistics or
are unable to solve. For this reason, we assume that only intercepts vary across states. Thus, our Level II submodel is the
following:
β
0j
=
γ
00
+
γ
01
W
1j
+ … +
γ
0n
W
nj
+ u
0j
β
1j
=
γ
10
…
β
nj
=
γ
n0
(2)
Equation (2) indicates that the intercept in the Level I submodel (
β
0j
)
is a function of the intercept (the state level measure of
contracting out,
γ
00
), Level II predictors (W
1j
… W
nj
), and random variance across states (u
0j
). By combining Equation (1)
and Equation (2), we have the following random intercept and fixed slope model:
Y
ij
=
γ
00
+
γ
01
W
1j
+ … +
γ
0n
W
nj
+
γ
10
X
1ij
+ … +
γ
n0
X
nij
+ u
0j
+ r
ij
(3)
9
Y
ij
=
γ
00
+ u
0j
+ r
ij
.
(4)
This model was run without the assumption of the Poisson error distribution. The results obtained from the Poisson model
showed almost similar findings for this unconditional means model when converted into raw numbers.
10
The variance in the state level measure in Equation (4) is 1.8644 (p = 0.0673). The variance associated with local
governments in Equation (4) is 110.01 (p < 0.0001).
11
We should note that local governments in the West and Northeast are somewhat overrepresented and underrepresented,
respectively, among ICMA survey respondents. Similarly, urban local governments were slightly overrepresented among
ICMA survey respondents compared to independent local governments in rural areas.
12
Kloha, Weissert, and Kleine (2005) point out that simple average spending and taxation measures may not accurately or
objectively measure fiscal need. This may help to explain these null findings.
13
Brown and Potoski (2003a) examined the role of isomorphic pressures on local government contracting in 1997 and found
that the council-manager form of government was more likely to engage in contracting out. Data on the specific type of
local government, however, was not available for 2002.
14
The coefficient is negative and approaches statistical significance (p < 0.12). This may offer evidence that local TELs
actually put localities in a better fiscal condition, thereby discouraging localities from trying to reduce costs through
privatization. Local TELs may be generating more revenues in the form of state grant funding for local governments than
the shortfall they cause in local revenues.
Convention