Institutional features of Italy

In order to understand the relationship between accountability and corruption in Italy in the 1980s–1990s, it is important to describe some key institutional features. During the period in question in Italy, at the national level, an open list proportional (OLP) electoral rule was in place. The party system of the time was also exceedingly stable. The same party, the DC (in coalition with smaller parties), was in government from 1947 to 1994. This resulted in a democracy that was “‘blocked’, noncompetitive and [with] a governing class [that] continued to be unaccountable” (d’Alimonte Reference d’Alimonte2005, p. 253). OLP systems provide incentives for politicians to cultivate the personal vote (Carey and Shugart Reference Carey and Shugart1995) as voters have influence over the order in which candidates are elected. Furthermore, OLP electoral systems are associated with higher levels of corruption (Chang and Golden Reference Chang and Golden2007), precisely due to their incentives to foster the personal vote.

With regard to subnational government, there are three levels in Italy: regions, provinces, and municipalities. These levels are directly linked to national politics in that local politicians frequently run under the flag of the national parties. For example, in the 1990 municipal elections in Milan there were candidates running under the name of all the major national parties of the time: Democrazia Cristiana, Partito Comunista Italiano, Partito Socialista Italiano, Lega Nord, and Partito Repubblicano Italiano. The national committee of the party in question vets and authorises the local candidate to run on their platform, thus giving local voters a heuristic to assess what a given candidate stands for and who she is associated with on the national political stage. There are different areas of competence that dictate which governmental level administers which services. More information about the functioning of subnational government in Italy is available in the Appendix (Section 1.1).

Prior to “Clean Hands,” Italian deputies engaged in extensive patronage practices, appointing public officials on a nonmeritocratic basis (Zuckerman Reference Zuckerman1979; Golden Reference Golden2003). Public offices during this period were granted an unusual amount of discretion despite (or perhaps due to) excessively detailed and copious legislation (Predieri Reference Predieri1963; Golden Reference Golden2003). This manifested itself in myriad ways of working the system such that the Italian bureaucracy became an immense reservoir of patronage (Chubb Reference Chubb1982; Golden Reference Golden2003). There are too many nefarious hiring techniques described in the qualitative literature to be able to cover them all here, so I shall provide a selection of compelling examples. The first is known as “titularisation” and is described concisely in the following steps:

Another similar technique, “svolgimento di mansioni superiori,” ^{Footnote 3} circumvents the requirement for a public competition by advertising nonadministrative positions (e.g., clerks, typists, ushers, janitors, labourers) that can be hired on a direct personal basis (Chubb Reference Chubb1981, p. 113). In her case study of Palermo, Chubb finds that “[o]f 3,056 positions in the city government in 1976, 2,514 were nonadministrative; only 542 required public competitions” (Chubb Reference Chubb1981, p. 113).

It has been estimated that between 1973 and 1990, 350,000 individuals were recruited to the public administration without entrance exams – subsequently having their posts made permanent through special laws – thereby bypassing regular procedures, as compared to 250,000 individuals who were recruited regularly (Cassese Reference Cassese1993, p. 325). This intensive use of patronage hiring also extended to public holding companies such as IRI, ENI, and EFMI ^{Footnote 4} where figures close to the DC held prominent leadership positions. Moreover, high-level politicians were directly engaged in managing and directing local administrative activities:

This translated into significant numbers of public officials being de facto political appointees, even if the de jure system would see them as appointed meritocratically. I discuss these extensive patronage practices and party infiltration of almost every facet of the state’s activities in the Appendix (Section 2).

These practices resulted in an inflated and inefficient bureaucracy. To this day, the quality of Italy’s public administration is widely pilloried and satisfaction with public services as well as trust in governing bodies are consistently low (Transparency International 2013; Eurofound 2016). Golden (Reference Golden2003) argues that Italy’s “bad bureaucracy” has been purposefully fostered by deputies in order to offer constituent services to navigate the dysfunctional system and boost their personal vote. ^{Footnote 5}

Mechanisms and hypotheses

We have seen, then, that there were a series of conditions in place in Italy in the 1980–1990s that would undermine the correct functioning of the accountability mechanisms that should characterise an established democracy (for more details see the Appendix, Section 2). Combined with theory and evidence that elections can be ineffective as devices to hold politicians to account for their actions and that corruption displacement is a real risk in contexts where systemic corruption exists, which mechanisms might be at work behind an accountability-driven corruption displacement effect?

I limit myself to describing two channels, which seem most likely given the context. ^{Footnote 6} They both pertain to the political vacuum created at the national level by the “Clean Hands” investigations whereby the accused deputies were (temporarily or otherwise) distanced from office due to their being investigated for corruption. The first channel is that bureaucrats become freer to consolidate their own autonomous sphere of influence in the system of corruption whilst national deputies are in crisis. ^{Footnote 7} Public officials have multiple incentives to increase their role in corrupt exchanges: first, a remunerative benefit in that they receive side payments for their work on top of their salary. Second, the opportunity to administer favours for oneself for professional or political (if one is in local political office or is interested in entering politics) gain. Scholars of corruption in Italy have recognised (in a descriptive spirit, not focusing on accountability and displacement) precisely such a “new,” more decentralised form of corruption following “Clean Hands” (more on this in ‘Discussion’ section), whereby public officials stepped into the political vacuum left by the investigations, increasing their bargaining power with respect to politicians and, subsequently, their ability to extract resources through corrupt exchanges (Della Porta and Vannucci Reference Della Porta and Vannucci2007, Reference Della Porta, Vannucci, Tänzler and Maras2016b; Asquer Reference Asquer2013).

The second channel works through what I term “surrogacy” behaviours, whereby bureaucrats effectively stand in for investigated deputies. Public officials who wish to stay in the good graces of their political principals (the widespread patronage practices of the time meant that significant numbers of bureaucrats were personally indebted to national deputies for their positions) continue to distribute electoral favours to maintain key constituencies. This is widely documented in the qualitative literature: “[t]he raccomandazione by which the public employee accedes to the object of his aspirations is only the first link in a chain of reciprocal obligation that will progressively bind him ever more irrevocably to his patron and protector” (Chubb Reference Chubb1982, p. 98). ^{Footnote 8} Thus, the public official has a strong incentive to ensure that favours continue to be administered to constituents. As Oliveros (Reference Oliveros2021) argues, patronage contracts are “self-sustaining over time” [p. 382] because public officials believe that their jobs are dependent on the success of their (incumbent) patrons. ^{Footnote 9}

This channel links to the literature on brokers or mediators in clientelistic relationships between politicians and voters. Public officials, here, are hypothesised to act as agents of the politician, collecting and responding to the requests of voters. Recent literature has enriched our understanding of clientelistic exchanges (Gingerich and Medina Reference Gingerich and Medina2013; Stokes et al. Reference Stokes, Dunning, Nazareno and Brusco2013; Weitz-Shapiro Reference Weitz-Shapiro2012; Rodrigo Zarazaga Reference Rodrigo Zarazaga2014) by focusing on political brokers or “ground-level intermediaries between parties and voters […] who can identify the needs of clients, distribute goods, and monitor behavior” (Holland and Palmer-Rubin Reference Holland and Palmer-Rubin2015, p. 1187). Such figures are hypothesised to become especially important in times of scandals when a political vacuum is created “by virtue of their capacity to cushion the uncertainties associated with the illegal nature of [corrupt] exchanges” (Della Porta and Vannucci Reference Della Porta and Vannucci2016a, p. 171).

An increase in local-level corruption would be expected via this “surrogacy” channel, as prior to “Clean Hands” deputies would look to personally manage as many electoral favours possible to claim credit directly and obtain preference votes. Qualitative evidence shows that “every public official from the national deputy to the municipal assessor maintains copious and detailed files recording every favour ever performed. These files […] are then activated during every electoral campaign” (Chubb Reference Chubb1982, p. 174). There is evidence that national deputies were able to influence even the highest ranked public officials (see the Appendix, Sections 1.3 and 2), who are the figures who would be best placed to “take over” from deputies during a period of incapacitation. It is important to recognise, however, that the impact of patronage certainly varied according to both the level of government and the type of public body under consideration. More high-ranking public officials are more difficult to appoint through patronage and control than low-ranking officials, and public holding companies were infamously under political control in the 1980–1990s in Italy, whereas this was less true of other administrative bodies such as the judiciary.

These two channels would also have been self-reinforcing in that individuals who accept a public position through patronage and understand “the rules of the game” are likely to spontaneously start making their own deals, to advance professionally. This mechanism is succinctly expressed by the ex-DC mayor of Reggio Calabria: “[o]ne of the people I initiated into taking mazzette [bribes] afterwards started to make small deals of his own, gli “assolo” [solos] as they’re called in the jargon” (Licandro and Varano Reference Licandro and Varano1993, p. 48). As such, corrupt exchanges take on a life of their own as officials appointed through patronage and organised into multilevel hierarchies (Appendix, Section 2) seek to administer favours and collect votes or money.

This leads to my first hypothesis that:

This implies that I expect to see an increase in local-level corruption in those electoral districts affected by “Clean Hands.” My second, competing, hypothesis is instead that:

My findings overwhelmingly support the first hypothesis. I find a substantial and robust increase in local-level corruption in the electoral districts affected by the investigations.

Data

Thus, I require data on both national deputies and local public officials over time, including accusations as part of “Clean Hands” and a local-level corruption measure. First, focusing on the national level, I collect demographic data (age, gender, district of election etc.) from Gagliarducci et al. (Reference Gagliarducci, Nannicini and Naticchioni2011) for Italian Members of Parliament from 1987 to 2008. These data are supplemented with the same information for the 2013 election. ^{Footnote 10} Thus, I have detailed individual-level data on the universe of elected politicians from all levels of government from 1984–2013 in Italy.

Second, still at the national level, I bring in Golden (Reference Golden2014)’s dataset of Italian parliamentary malfeasance. This dataset contains information on all requests by the Italian judiciary to remove parliamentary immunity (richieste di autorizzazioni a procedere, RAP) from deputies elected in 1948 through 1992, including the nature of the judicial charges, the partisan affiliation of the deputy charged, the deputy’s electoral district and legislature, etc.. The Italian constitution (Article 68) required that the Ministry of Justice transmit requests to remove immunity to parliament in order to investigate a legislator for suspected criminal wrongdoing or in order to proceed with an arrest warrant. A majority vote by the floor of the relevant chamber was required to lift immunity. The kinds of corrupt practices that the deputies in my sample undertook fall snugly into the category of government allocations to special interests. The most frequent allegation was “violation of the regulations on public financing of political parties,” ^{Footnote 11} with “abuse of office” and “corruption for an act contrary to official duties” running close behind (Chang et al. Reference Chang, Golden and Hill2010, p. 182).

Third, focusing on the subnational level, these datasets are supplemented with local administrative data held by the Italian Institute of Statistics (ISTAT) and manually collected by the author from the ISTAT archive in Rome. These data include the number of “corruption crimes” indicted and regard public officials (not citizens). ^{Footnote 12} These “corruption crimes” are expressly classified as “crimes against the public administration” (Codice Penale 1930a). Public officials (pubblici ufficiali) are defined by the Italian Penal Code as those who “exercise a public function, be it legislative, judicial or administrative” (Codice Penale 1930b), ^{Footnote 13} for example judges, technical consultants, health inspectors, policemen, employees of municipal governments (Codice Penale 1930b). ^{Footnote 14} Local politicians (such as mayors or councillors) are also included in this category. Given that there is so much variation in the types of roles covered in my sample, I consequently adopt a broad term to refer to them: specifically, public officials, which also closely reflects the notion laid out in the Italian penal code. The Corruption Indictments variable that is used as the dependent variable in my main specifications captures illegal activities such as embezzlement, misappropriation of yield to the detriment of the government, extortion, and bribery agreements. ^{Footnote 15}

I match these data sources at the electoral district level, to create a corruption score ^{Footnote 16} for all the provinces within a district (see Figure 3 for a treatment map and the Appendix, Section 3 for more details). To be able to do so, I geocode the municipalities and provinces across the datasets, matching them to the electoral districts. Thus, a deputy can be linked to the corruption performance of the bureaucracies in the provinces within her electoral district. Descriptive statistics are presented in Table 1.

Table 1.

Summary statistics

Note: Electoral District refers to the electoral districts for Italy’s national legislative elections. In the estimations, the baseline region for comparisons for Macro Regions of Italy is the North. All variables are described in ‘Dependent variable’. ‘Theoretical covariates’ and ‘Additional covariates’ sections.

Empirical strategy

My identification strategy, a DiD design, exploits the timing of the 1987 and 1992 elections and the fact that the composition of both governments is very similar. I run the following regression model:

(1) $${Y_{dt}} = \alpha + \gamma trea{t_d} + \lambda pos{t_t} + {\delta _{DiD}}(trea{t_d} \times pos{t_t}) + \beta X_{dt}^\prime + {e_{ct}}$$

where ${Y_{dt}}$ is the number of individuals with Corruption Indictments in the local bureaucracies of a given district d and year t, $trea{t_d}$ is a dummy variable equal to 1 for a district that experienced an increase in the indictment rate of national deputies following “Clean Hands” and 0 for a district that experienced no meaningful increase, $pos{t_t}$ is a dummy variable equal to 1 if the deputy was elected after the scandal (1992) and 0 if elected before the scandal, $trea{t_d} \times pos{t_t}$ indicates treated electoral districts for postscandal years. ${\delta _{DiD}}$ is the DiD estimator that captures the effect of the investigations on the levels of corruption in the local public administrations. $X_{dt}^\prime$ is a vector of district-level controls (discussed in ‘Additional covariates’ section) measured for each district in each year.

With respect to the “parallel trends” assumption, the most critical of the DiD assumptions in terms of internal validity: here, the assumption means that the districts with no increase in the indictment rate of national deputies (control) and those with an increased indictment rate (treated) were moving in parallel in terms of local public official corruption crimes ahead of the 1992 scandal. A visual inspection of the data, as presented in Figure 2, as well as analyses introducing lag years to the main model to test for pre-trends (Figure 4) strongly support parallel trends in the treated and control groups. The other assumptions required to justify inference from the DiD model are discussed in the Appendix (Section 4).

Figure 2.

Parallel trends.

Dependent variable

The dependent variable in my models, Corruption Indictments (mean: 66.80, S.D.: 82.55), captures the number of local public officials who were indicted for corruption in an electoral district. I stress that these are indicted crimes and, as such, it must be recognised that the measure captures not only corruption in the province concerned, but also the effort and efficiency of the national and local authorities in combating corruption. I am able to account for efficiency through the Length of Judicial Proceedings (see ‘Additional covariates’ section), but capturing the effort of state authorities is, unfortunately, more problematic. Although I have no way of empirically capturing the effort that individual prosecutors put into fighting corruption, I hope that some contextual information (Appendix, Section 1.2) about the judiciary will assuage concerns about a differential amount of effort across prosecuting authorities in Italy. To the degree that one might be concerned about systematic differences in the amount of information available to judicial authorities across the country about corruption crimes, I seek to account for this through district and regional-level fixed effects (Table 2) and an analysis factoring in historical organised crime presence (Appendix, Table A2).

Table 2.

The effect of “Clean Hands” on local-level corruption, difference-in-differences estimations

Note: DiD and control variable coefficients from OLS regressions of the effect of “Clean Hands” on the number of local public officials indicted for corruption crimes. Descriptive statistics for all variables are presented in Table 1. *p  <  0.1; **p  <  0.05; ***p  <  0.01.

Judicial statistics on indictments are an imperfect measure of the underlying phenomenon of corruption. They capture only when a corruption crime is discovered rather than realised and they likely only capture a small percentage of the actual corruption crimes being undertaken. However, if one assumes, ungenerously, that the judiciary is even only slightly effective at identifying and investigating corruption, one can expect such statistics to mark a lower bound of the amount of corruption that is really occurring.

I argue that corruption crime indictments are one of the best measures that we have at our disposition, particularly for noncontemporary periods. Data on verdicts of political or administrative corruption are not very informative in the sense that corrupt activities are notoriously difficult to prove and punish adequately, in the Italian context. Davigo (one of the principal investigators leading “Clean Hands”) in fact noted that Italy makes fewer corruption convictions than Finland and that the province of Reggio Calabria (an area notorious for corruption) has had only two convictions for corruption in twenty years. ^{Footnote 17} Indeed, as Decarolis et al. (Reference Decarolis, Fisman, Pinotti and Vannutelli2020) put it, “corruption cases are generally complex, and convictions relatively rare. This is particularly true in Italy […] [f]or these various reasons, official data on (convicted) offenders may greatly understate the extent of corruption” [p. 12].

Indictments also allow for the measurement of specific types and perpetrators of corruption, which would not be possible with, for example, survey data capturing perceptions of corruption (Heywood and Rose Reference Heywood and Rose2014). Moreover, survey-based perception measures of corruption have a series of weaknesses (see Golden and Picci Reference Golden and Picci2005, pp. 39–41) that make them suboptimal for addressing the research question I pose. Whilst there are excellent new ways to capture corruption in the Italian context (e.g., corruption risk indicators (Decarolis and Giorgiantonio Reference Decarolis and Giorgiantonio2020) or measures of how discretionary mechanisms affect corruption (Decarolis et al. Reference Decarolis, Fisman, Pinotti and Vannutelli2020)), unfortunately they do not exist for the time period under analysis here.

Theoretical covariates

Assessing whether local public officials respond to national political scandals involves a comparison of districts that were and were not exposed to the accountability shock, before and after the scandal. Thus, I include a dummy variable, Treated, to indicate those districts that were “treated” with an increase in the indictment rate of national deputies following the investigations (1) and those that were not (0). This measure, created using the RAP data, captures the increase in horizontal (judicial scrutiny) accountability that “Clean Hands” involved, given that diagonal (media or civil society scrutiny) accountability is difficult to capture empirically in this setting. Figure 3 presents the treated and control districts geographically (for details on how the indictment rate is calculated see the Appendix, Section 3 and Table A3). This Treated variable is interacted with a dummy indicating whether the time period is before or after the outbreak of the scandal in 1992, Post. Thus, the interaction term Treated*Post captures the effect of an increased indictment rate at the national level – the accountability effect – in the post-“Clean Hands” period on local-level Corruption Indictments.

Figure 3.

Map of treated and control districts.

Dark grey districts: treated, light grey: control. Circles represent treatment intensity.

A potential confounding factor for my treatment would be if the “Clean Hands” investigations were originally driven by the same underlying interest in eradicating corruption at both the local and national levels. To speak to this, I have to lean on qualitative evidence that the “Clean Hands” investigators were initially interested in targeting illegal party financing in the upper echelons of the party administrations. The investigations were not a nationwide effort to root out corruption across the board. They were the work of a specific and small ^{Footnote 18} “pool” of investigators and were shaped by the zeal and specific interests of individual investigators. At the beginning of “Clean Hands” (when my treatment is defined), the investigators thought that they were dealing with a phenomenon of illegal party financing that only affected certain political parties’ leaderships and certain businesses with extensive ties to the public administration, only to later discover the reach and intensity of the corrupt networks (Davigo Reference Davigo2021). Indeed, “Clean Hands” has been called “the biggest missed opportunity in Italian history” (Di Pietro cited in Amato (Reference Amato2013), translation by the author ^{Footnote 19} ) precisely because the investigations failed to target the root of corruption. By focusing on politicians’ grand corruption, the investigations left the petty corruption that pervaded all levels of government to flourish.

In fact, experts argue that following “Clean Hands,” corruption actually worsened due to a legal system that made it very hard to punish corrupt individuals and the involvement of organised crime in local politics and public administration (La Voce 2015; Il Fatto Quotidiano 2015). Silvio Berlusconi later made judicial reform a top priority in order to rein in magistrates such as those involved in “Clean Hands” who he saw as “overweening and politically biased” (Dickie Reference Dickie2004). He also rendered judges’ work more difficult by introducing ‘several ad personam’ laws, aimed at shielding Berlusconi from certain charges, but ultimately generating an increase of the potential impunity for all white-collar criminals” (Vannucci Reference Vannucci2016, pp. 65–66). Moreover, it is argued that the investigations and their ineffectiveness in terms of translating indictments into guilty verdicts led to an increased “sense of impunity” amongst the leading actors in corrupt activities (Della Porta and Vannucci Reference Della Porta, Vannucci, Tänzler and Maras2016b, p. 145).

Additional covariates

My main statistical models include three control variables. My model already includes fixed effects at the district level (capturing time-invariant features of each district), so these additional control variables are intended to account for time-varying factors that are unrelated to the investigations, but might affect local-level corruption.

The first is a linear year trend, Year, to help rule out the possibility that the treatment and control districts were already on different growth trajectories in terms of Corruption Indictments before “Clean Hands.”

The second measures the average length of penal trials, that is to say the “ratio of the number of pending judicial proceedings, at the beginning and the end of each year, to the number of judicial proceedings started and completed in the same year” (Acconcia and Cantabene Reference Acconcia and Cantabene2008). I use this measure, Length of Judicial Proceedings (mean: 152.66, S.D.: 60.50), to capture the efficiency of local public administrations.

The third variable controls for the Macro Regions of Italy (1 = North, 2 = Centre, 3 = South, 4 = Islands), ^{Footnote 20} given the varying levels of economic development and intensity of corruption that are associated with the country’s different regions, with the South of the country being associated with higher corruption than the North (Banfield Reference Banfield1958; Golden and Picci Reference Golden and Picci2005).