Electoral accountability and local government efficiency: quasi-experimental evidence from the Italian health care sector reforms

Abstract

This paper evaluates the effect of two policy changes on the efficiency of Italian regional governments in the provision of health care services: first a change in the electoral system, second a process of fiscal decentralisation. The paper provides two main contributions: (1) a comprehensive analysis of the two main reforms that involved Italian regional governments and the health care sector during the 1990s, (2) the evaluation of the impact of the electoral reform in a quasi-experimental setting. Final results are in line with recent theoretical predictions that show a positive relationship between government efficiency and the electoral accountability enhanced by institutions such as electoral rules and fiscal decentralisation.

Appendices

Appendix 1

1.1 Data envelopment analysis

In case of input approach, our index of efficiency $e_{it}^{DEAin} = \phi _{it}$, where $\phi _{it}$ is the solution of the following linear program which provides the efficiency score for the region $i$ in period $t$:

$$\begin{aligned} \underset{\phi \in \mathfrak {R}, {\varvec{\lambda }}\in \mathfrak {R}^{NT}}{min} \phi \quad s.t.\quad \phi \mathbf{x}_\mathbf{it}\geqslant \mathbf{X}\varvec{\lambda },\mathbf{Y}\varvec{\lambda }\geqslant \mathbf{y}_\mathbf{it}; {\varvec{\lambda }} \geqslant 0; {\varvec{\iota }}^{\prime } {\varvec{\lambda }}=1 \end{aligned}$$

(5)

where $\mathbf {x_{it}}$ is the vector of inputs of region $i$ at time $t,\,\mathbf {X}$ is $l\times NT$ the matrix of inputs of all $N\$ regions over all $T$ years, $\mathbf {Y}$ is the $q\times NT\$matrix of outputs of $N\$ regions over all $T$ years, $\mathbf {\lambda }$ is a $NT\times 1\$ vector of optimal weights that identify the benchmark regions on the frontier for each inefficient region, $\mathbf {y_{it}}$ is the vector of outputs of region $i$ at time $t$, and $\mathbf {\iota }^{\prime }$ is a $1\times NT\$vector of $(1,\ldots 1)$. The three first constraints are necessary in order to generate the frontier and the last constraint is important for imposing variable returns to scale.

In the case of the output approach, instead, $e_{it}^{DEAout} = \frac{1}{\phi _{it}}$, where $\phi _{it}$ is the solution of the following linear program which provides the efficiency score for the region $i$ in period $t$:

$$\begin{aligned} \underset{\phi \in \mathfrak {R}, {\varvec{\lambda }}\in \mathfrak {R}^{NT}}{max} \phi \quad s.t.\quad \mathbf{x}_\mathbf{it}\geqslant \mathbf{X}\varvec{\lambda } ,\mathbf{Y}\varvec{\lambda }\geqslant \phi \mathbf{y}_\mathbf{it}; {\varvec{\lambda }} \geqslant 0; {\varvec{\iota }}^{\prime } {\varvec{\lambda }}=1 \end{aligned}$$

(6)

Finally it is important to note that the linear program in (5) and (6) is solved by using a pooled approach where only one production frontier is estimated and each region is compared also with itself in another year. In this way it is possible to use all the $N\times T$ observations in order to minimise the upward small-sample bias that affects this non-parametric estimator of $\frac{y_{it}}{f(\mathbf {x} _{it};\mathbf {\beta })}$. As argued by Kneip et al. (1998) this bias produces a small measurement error in the estimated indices of efficiency $e_{it}^{DEA}$ that vanishes as the number of observations increases.

Appendix 2

See Tables 6, 7 and 8.

Table 6 General statistics and variable definitions

Table 7 Point estimates of the parameters related to the control variables, output of the health care sector measured in terms of infant mortality

Table 8 Point estimates of the parameters related to the control variables, output of the health care sector measured in terms of neonatal mortality at 1 day