Is transparency spatially determined? An empirical test for Italian municipalities

ABSTRACT

In this article, we aim to assess whether transparency is characterized by a spatial dimension. To this end, we use a new composite indicator (CTI) for a large sample of Italian municipalities and control for several factors (socio-economic, fiscal, and politico-institutional), which, according to the literature, affect transparency. Our results suggest that there is a statistically significant transparency clustering across the Italian municipalities which follows a dichotomic pattern, i.e. either very low or very high. Moreover, the empirical analysis shows that spatial dependence matters and ‘transparency mimicking’ takes place among neighbouring municipalities. This behaviour mainly occurs among small municipalities where citizens’ political participation is likely to be greater and the single-ballot electoral system strengthens the incentives for government accountability.

KEYWORDS:

• Transparency
• local governments
• spatial dimension
• determinants

JEL CLASSIFICATION:

• K2
• K4
• H3
• H7

Disclosure statement

No potential conflict of interest was reported by the authors.

Data available on request from the authors

The data that support the findings of this study are available from the corresponding author, E.G., upon request.

Notes

¹ For a recent and extensive review of transparency, see Cucciniello, Porumbescu, and Grimmelikhuijsen (2016).

² For further details, see Section 2 below.

³ The ‘Compass of Transparency’ is a web-portal managed by the Italian government, which only assesses whether in the public administrations’ websites the ‘Transparent Administration’ section exists and if its structure formally complies with the standard legal requirements (see Section 2 below) without verifying the data and information content.

⁴ Legislative Decree n. 150/2009 containing provisions on ‘optimization of the productivity of public employees and efficiency and transparency of public administrations’.

⁵ Law no. 190/2012, containing ‘Provisions for the prevention and repression of corruption and illegality in Public Administration’.

⁶ Legislative decree n. 33/2013 containing ‘Rules about publicity, transparency and information provision of public administrations’.

⁷ Moreover, access to information has been made easier (civic access). In the same direction, the generalized dissemination of information upon request was also introduced in 2016.

⁸ The legislative decree n. 97/2016 has entitled the ANAC to diversify obligations across administrations depending on the type and the size and has enlarged the ANAC’s sanction powers.

⁹ The mean and median population of all the Italian municipalities in 2013 were about 7,500 and 2,500 inhabitants, respectively. Therefore, a municipality with more than 15,000 people is considered a medium-large city.

¹⁰ By law, all institutions are requested to publish the OIVs certifications on their websites under the section Amministrazione Trasparente.

¹¹ Table A1 in the Appendix reports all the transparency items listed in the above-mentioned ANAC resolution and the selected items used to construct the CTI. Moreover, Table A2 shows the values of the CTI, with selected items (Galli, Rizzo, and Scaglioni 2017) and a CTI including total items at regional level. The two indexes are highly correlated both at regional level (0.92) and at municipal level (0.87). Data for municipalities are available upon request.

¹² For further details, see Galli, Rizzo, and Scaglioni (2017).

¹³ Specifically, in small municipalities citizens vote once, i.e. there is only a single round while in large municipalities citizens vote twice, i.e. a dual ballot mechanism is used to select a winner. For the investigation of the effect of different electoral rules on the selection of politicians, see Galasso and Nannicini (2011) and Bordignon, Nannicini, and Tabellini (2016).

¹⁴ In the Appendix, Table A3 reports variables’ descriptions and sources.

¹⁵ For a discussion on alternative spatial autocorrelation indicators, see, among others, Griffith (1987).

¹⁶ We use a matrix determined by applying an algorithm based on the closest neighbours, since in our sample there are two islands (Sicilia and Sardegna). See Table A4 in the Appendix.

¹⁷ Randomization was tested on 999 permutations.

¹⁸ We recall here that the ‘so-called spatial clusters shown on the LISA cluster map only refer to the core of the cluster. The cluster is classified as such when the value at a location (either high or low) is more similar to its neighbours (as summarized by the weighted average of the neighbouring values, the spatial lag) than would be the case under spatial randomness. Any location for which this is the case is labelled on the cluster map. However, the cluster itself likely extends to the neighbours of this location as well’ (Anselin 2005). See also Ord and Getis (1995).

¹⁹ W_CTI has been selected among different types of weight matrices based on the geographical distance across the municipalities (see Table A4 in the Appendix).

²⁰ As well-known, both are special cases of a so-called general nesting spatial model (GNS), i.e. the model that includes a spatially lagged dependent variable, spatially lagged independent variables, and a spatially autocorrelated error term simultaneously (Manski 1993; Elhorst 2010; Golgher and Voss 2016; Burridge, Elhorst, and Zigova 2017).

²¹ We compared Spatial autoregressive model (SAR); Spatial error model (SEM); Spatial lag of X model (SLX); Spatial autoregressive combined model (SAC-SARAR); Spatial Durbin error model (SDEM). We re-estimated the OLS model including spatially lagged independent variables (WX); the hypothesis H₀: Θ = 0 was rejected. To test the additional hypothesis H₀: $\rho$= 0, we estimated the different models with spatially lagged independent variables: SLX, SDM and SDEM (Elhorst (2010, 17). Results are available upon requests.

²² It is worth mentioning that SDM approach turned to be the most appropriate in other studies conducted on Italian municipalities (Santolini 2020; Bocci, Ferretti, and Lattarulo 2019).

²³ See Tables A5 in the Appendix.

²⁴ The average direct effect is the marginal effect of a one-per cent variation in the independent variable on the dependent variable of the municipality; the average indirect effect is the marginal effect of a one-per cent variation in the independent variable on the dependent variable of the neighbouring municipalities. The total effect is the sum of both effects.