INTRODUCTION
As conflict, poverty, and climate change continue to put people on the move, the question of how to integrate an expanding population of migrants in advanced economies has become of primary importance. While there is a debate about the determinants of successful legal migrant integration (Dancygier and Laitin Reference Dancygier and Laitin2014), a sizeable and growing fraction of immigrants remains undocumented and vulnerable to exploitative labor practices. According to the International Labour Organization.Footnote 1 27 million people worldwide are victims of forced labor, most of whom are employed in Asia and Europe.
Migrant labor exploitation is not only problematic from a global human rights perspective, but it is also concerning for the economy and the rule of law of destination countries. Illegal labor can change the structure of labor markets, giving a competitive edge to firms willing to hire illegal workers for lower wages. It implies a loss of fiscal revenue for the state. Finally, when migrants are smuggled and controlled by criminal intermediaries, this phenomenon also reinforces criminal organizations. However, policy initiatives against labor exploitation have remained limited, in part because cracking down on illegal labor is difficult if migrants do not report on their racketeers.
In this article, I study the effects of an intervention designed to fight labor exploitation by explicitly targeting informal migrants and seeking to empower them to report on their exploitative employers. The intervention took place in Italy, one of the main hotspots of migrant arrival in Europe and a setting where illegal labor is often controlled by criminal organizations (Parliamentary Commission 2017, 52).
Starting in 2007, one of the main Italian unions of agricultural workers launched a campaign of in-person canvassing to provide migrants employed in agriculture with information about their rights, legal assistance to denounce their racketeers, and incentives to do so. Under Italian law, as in other countries, undocumented migrants can obtain a humanitarian residence permit if they are recognized as victims of exploitation. I use the staggered roll-out of this intervention to study its effects on (i) news reporting about labor racketeering, a measure of whether the intervention empowered migrants to denounce their exploitation, (ii) prosecution of criminal groups involved in labor racketeering, and (iii) public opinion.
Theoretically, it is unclear whether empowering a marginalized community that is trying to hide from the state can be successful. Undocumented workers who speak up against labor exploitation might risk unemployment, retaliation by criminal groups, and deportation. On the other hand, remaining in exploitative labor conditions is also costly, and migrants might decide to act if the risks they would incur when reporting were mitigated. I argue that a combination of information highlighting the mismatch between what migrants experience and what they could attain, and incentives to report tied to the hope of obtaining legal status can empower undocumented migrants to take the costly action necessary to denounce their exploitative employers and improve their socioeconomic condition.
I leverage the staggered roll-out of the intervention across municipalities to study its effects in a difference-in-differences (DiD) framework. I provide evidence consistent with the validity of the DiD assumptions by showing that, first, treatment timing is uncorrelated with reports on the severity of exploitation produced by the union itself, a finding confirmed by the union leader in interviews I conducted; second, trends in outcomes before treatment are parallel; third, results are consistent across the full sample, a matched sample based on propensity score matching, a synthetic DiD approach, and using alternative estimators accounting for treatment effects heterogeneity bias.
One challenge with studying migrant labor exploitation relates to data limitations. The phenomenon is illegal, hard to detect, and its detection might correlate with the capacity of racketeers to hide their activity. I circumvent these problems by scraping news related to labor racketeering from seven news outlets of varying political orientation and by validating this measure with province-level data on unannounced audits to agricultural firms by the Labor Inspectorate. The measure is robust against reporting biases inherent in news-based indicators. Analyses also account for time-invariant differences in news coverage within locations, addressing factors such as population size, xenophobia, and political activity that may influence news coverage.
I present four main findings. The first directly relates to migrants’ political activation: the intervention increased reporting about labor racketeering in the news. The increase is driven by articles discussing cases of migrants whistleblowing and mobilizing in public protests. Evidence indicates that the migrants themselves, rather than the union, are driving these effects. First, news did not cover the intervention itself, and mentions of unions are balanced across the treated and control group. Second, in places where unions limited themselves to collecting information on exploitation without conducting interventions with the migrants, there was no effect on labor racketeering news. This indicates that the campaign was effective at empowering migrants to whistleblow their condition of exploitation.
The second finding relates to the role of the state: the intervention led to a substantial increase in the number of properties seized from organized crime. This finding ties the role of migrants in denouncing exploitation to the role of criminal groups in controlling the system of labor exploitation. When workers denounce their employers, the state has leverage to identify and prosecute criminals and can crack down on their properties.
Third, the intervention increased state and civil society mobilization against labor racketeering: not only do news items reveal the emergence of civil society initiatives in treated municipalities, but public administrations also redistribute mafia-seized properties for public use at higher rates. The increase is not driven by the higher rates at which goods are seized, and is driven by assignment of previously mafia-owned properties to become agricultural cooperatives destined to social use, which are the main method to offer nonexploitative working conditions to migrants employed in agriculture.
Finally, I consider the effects of this intervention on public opinion and on voting behavior of natives leveraging both panel surveys and election results at the municipal level. I document that the same respondent became 9% less likely to express sentiments of distrust toward migrants after the intervention took place in their municipality. Consistently, I show that the intervention increased the vote share for pro-integration far-left parties by 2 percentage points among survey respondents, and by 1 percentage point considering voting in the entire municipality. This gain is driven by a drop in votes for the main center-left party, which in the years of the study had adopted strong actions against the influx of new migrants, including pushing asylum seekers back to Libya before they could enter Italian waters (The Guardian 2017).
This study makes three theoretical contributions. First, while understanding the determinants of migrant integration is increasingly important, most of our knowledge is based on legal migrants, refugees, and asylum seekers (Dancygier and Laitin Reference Dancygier and Laitin2014). This article instead focuses mainly on undocumented migrants. The successful political and economic integration of undocumented migrants is likely determined by different factors than those applying to legal migrants. For economic integration, learning the language and high education increase legal migrants’ chances of finding employment, but they might be insufficient for migrants who cannot access labor markets legally. Factors important for migrants’ political integration, such as socioeconomic resources and networks, might not contribute to the activation of a population lacking political franchise and risking deportation if they make themselves visible to the state. Studies of undocumented migration are rare and often limited to the United States, where scholars have studied the effect of sanctuary policies on undocumented migrants’ health outcomes (Hainmueller et al. Reference Hainmueller, Lawrence, Martén, Black, Figueroa, Hotard and Jiménez2017; Swartz et al. Reference Swartz, Hainmueller, Lawrence and Rodriguez2019) and crime rates (Hausman Reference Hausman2020).Footnote 2 A small set of studies has focused on undocumented migrants’ activation using ethnographic methods to investigate migrants’ mobilization initiatives (Cremaschi Reference Cremaschi2020; Delgado Reference Delgado1993; Omizzolo Reference Omizzolo2019). Most questions related to undocumented migrant workers remain unaddressed. This article starts filling this gap by studying the conditions under which undocumented migrants can gain voice and adopt the costly actions necessary to emancipate from exploitative regimes.
Second, this study considers a case in which migrants are active political agents of change. Previous studies have either considered the impact of interventions and events, such as migrants arrival or economic shocks, on the opinions of natives (Adida, Lo, and Platas Reference Adida, Lo and Platas2018; Dancygier and Donnelly Reference Dancygier and Donnelly2013; Emeriau Reference Emeriau2024; Zhou, Grossman, and Ge Reference Zhou, Grossman and Ge2023) or have considered how migrants are impacted by policies granting or revoking rights, such as the right to access the country, citizenship, and clothing regulations (Abdelgadir and Fouka Reference Abdelgadir and Fouka2020; Hainmueller, Hangartner, and Pietrantuono, Reference Hainmueller, Hangartner and Pietrantuono2017; Masterson and Yasenov Reference Masterson and Yasenov2021).
Here, I consider a case in which, rather than receiving a concession from the state, migrants take costly actions to denounce racketeers to the authorities, seek state protection, and apply for the permission to work legally. This distinction likely matters for public opinion: while interventions favoring migrants’ integration tend to backlash (Zonszein and Grossman Reference Zonszein and Grossman2022), an intervention highlighting migrants’ vulnerability might increase the sympathy of natives toward migrants and trigger more acceptance of immigration, as shown in Bansak, Hainmueller, and Hangartner (Reference Bansak, Hainmueller and Hangartner2016), Bonilla and Mo (Reference Bonilla and Mo2018), and Alrababa’h et al. (Reference Alrababa’h, Dillon, Williamson, Hainmueller, Hangartner and Weinstein2021). In this respect, this study represents a novel contribution to the literature on the determinants of support for migrant integration by showing that, not only natives’ self-reported attitudes and willingness to accept migrants, but also their voting behavior is responsive to migrants’ vulnerability.
Finally, this article contributes to the study of the unintended effects of immigration policies, particularly studies considering spillovers from immigration to organized crime. Previous work has documented how large migration influxes combined with restrictive immigration regimes and tight labor markets can generate profit opportunities for criminal organizations (Dipoppa Reference Dipoppa2024a; Luca and Proietti Reference Luca and Proietti2022). In line with the idea that criminal groups gain strength by exploiting migrants, this article shows that when migrants are empowered and defy exploitation, organized crime is undermined.
THEORETICAL FRAMEWORK
Migrant Political and Economic Integration
Political and economic participation are two fundamental dimensions of migrants’ successful integration in host societies (Harder et al. Reference Harder, Figueroa, Gillum, Hangartner, Laitin and Hainmueller2018). Previous studies identified factors important for migrants’ political integration, ranging from socioeconomic resources and socialization (Cho Reference Cho1999), networks with co-ethnics (Bratsberg et al. Reference Bratsberg, Ferwerda, Finseraas and Kotsadam2021), and political opportunities created by party organization (Dancygier et al. Reference Dancygier, Lindgren, Oskarsson and Vernby2015; Pons and Liegey Reference Pons and Liegey2019). Networks play an important role also for economic integration, together with language training and job matching technologies (Bansak et al. Reference Bansak, Ferwerda, Hainmueller, Dillon, Hangartner, Lawrence and Weinstein2018).
This knowledge was built focusing on legal migrants and refugees, who eventually gain rights to participate in the labor market and politics. Less is known about undocumented migrants, whose networks, opportunities, and socioeconomic resources may not suffice for political and economic integration. In fact, a consistent result in this literature is that obtaining legal status significantly boosts migrants’ political participation (Ferwerda, Finseraas, and Bergh Reference Ferwerda, Finseraas and Bergh2020; Hainmueller, Hangartner, and Pietrantuono Reference Hainmueller, Hangartner and Pietrantuono2015), and employment (Hainmueller and Hangartner Reference Hainmueller and Hangartner2019), and even deferring deportations improves undocumented workers’ labor market participation (Amuedo-Dorantes and Antman Reference Amuedo-Dorantes and Antman2022; Borjas and Cassidy Reference Borjas and Cassidy2019). Evidence underscoring the role of legal status for integration suggests that our knowledge on regular migrants’ integration might not travel to those who are undocumented.
A handful of ethnographic studies have focused specifically on the political organization of undocumented migrants. Of particular relevance to this study is the work by Omizzolo (Reference Omizzolo2019), who documented the extreme exploitation faced by Punjabi agricultural workers in the Agropontino plain in Italy, became an activist for their rights and participated in coordinating their efforts to mobilize to call attention on their situation.Footnote 3 Similarly pertinent to this case is the ethnography by Delgado (Reference Delgado1993), examining a successful union campaign to organize undocumented Mexican workers in Los Angeles in 1986. Delgado argues that workers did not fear deportation: their previous lack of activism was due to insufficient union efforts to involve them. However, the conditions during Delgado’s study period, marked by rare, non-salient, and ineffective deportations resulting in swift returns across the border, may differ significantly from the current context, in which immigration is a salient political topic, deportations are enforced, and return entails expensive and dangerous journeys.Footnote 4
The conditions for the political mobilization of undocumented migrants, who lack franchise and face the threat of deportation, remain still unexplored. However, these migrants, who are employed in low-skill sectors, endure a large wage penalty, and are frequently victims of exploitation, have often strong reasons to mobilize. For a large fraction of the migrant population, therefore, the question of integration is rather a question of how to emancipate from exploitative working regimes. Under what conditions do undocumented migrants become empowered and manage to set themselves free from labor exploitation?
The Determinants of Undocumented Migrants’ Empowerment
Undocumented migrants face distinctive risks and challenges when considering whistleblowing and engaging in political action that citizens and legal migrants do not experience. First, the heightened visibility coming with political activism entails the risk of being identified and deported, with consequences that might outweigh the benefits of mobilization. Second, successful political action hinges on garnering consensus among civil society, which can prove challenging if some natives deem undocumented migrants unworthy of making demands to the state. Third, whistleblowing and protesting can lead to job loss, especially since informal workers lack contracts safeguarding their right to strike without retaliation. Losing employment can be particularly costly for migrants, often devoid of the local support networks that can assist others during unemployment. Fourth, this same lack of networks can also mean that workers possess less information about their rights, and fewer resources to solve collective action problems, two fundamental conditions for reporting and mobilization. Finally, when smuggled and controlled by criminal groups, migrants face the risk of retaliation from organized crime, both against themselves and against their networks back home.
In contrast, legal migrants and native workers enjoy more favorable conditions when engaging in activism. For instance, the struggle of Mexican and Filipino agricultural workers in the US, led by Cesar Chavez, involved legal migrants and U.S. citizens with regular work contracts. They faced no risk of deportation, enjoyed support from local networks of friends and family in case of income loss, accessed resources from a local network offering information and organizational skills, and received backing from civil society, which actively participated in their boycott efforts (Garcia Reference Garcia2016).
Undoubtedly, the costs of denunciation and political activation for undocumented migrants are significant. However, the risks of inaction and remaining in exploitative working regimes are also elevated. Undocumented migrants face high likelihood of work-related and transportation accidents, exposure to diseases while working, and the constant threat of organized crime retaliation. Given these dangerous conditions, it is possible that some may opt for confronting the elevated costs of political activation and whistleblowing, rather than enduring exploitation. This choice becomes especially viable when considering that migrants have incentives to denounce their condition: several states grant migrants a temporary residence permit when they are found to be victim of exploitation:Footnote 5 the hope of attaining a legal (albeit temporary) status might be sufficient to tip the scales in favor of taking action.
I propose that migrants working in exploitative regimes, upon realizing the extent to which their rights are violated and recognizing the disparity between their current condition and the one they could achieve outside of exploitation, may be motivated to take on the risks associated with exiting labor exploitation and exposing their exploiters. This hypothesis aligns with the concept, dating back to Tocqueville (Reference Tocqueville1856), that individuals experiencing a mismatch between their perceived status and the status they could potentially attain may experience dissatisfaction leading them to mobilize.Footnote 6 A similar dynamic might drive the mobilization of undocumented migrant workers: the shared experience of learning about their rights, recognizing the discrepancy between their current situation and the status they could achieve, and being presented with a tangible incentive to report their exploiters could trigger political activation within this marginalized community.
Undocumented migrant activation to exit labor exploitation might manifest in various forms: reporting exploitation to authorities, a political act asserting rights through state institutions, or collective mobilization with other migrant victims. Group action aims to draw public attention to migrant labor exploitation, facilitating institutional mobilization and potential pathways for change.
The Impact of Migrants’ Empowerment
If migrants endeavor to liberate themselves from labor exploitation, what will be the repercussions on state capacity and politics? Past studies have explored the consequences of granting or revoking rights to migrants (Abdelgadir and Fouka Reference Abdelgadir and Fouka2020; Hainmueller, Hangartner, and Pietrantuono Reference Hainmueller, Hangartner and Pietrantuono2017). However, I examine a scenario where migrants are not passive recipients but rather active protagonists in their pursuit of social mobility. Achieving political voice is likely to produce different effects compared to just receiving rights, particularly on two outcomes: state capacity to curb crime, and natives’ public opinion on immigration.
As for crime, studies show that regularization lowers crimes by allowing migrants to accept legal over illegal jobs (Hausman Reference Hausman2020; Pinotti Reference Pinotti2017). Beyond diminishing incentives for crime, exiting exploitation could also deprive criminals of a lucrative source of profits obtained through exploitation. Moreover, if migrants are willing to report on their exploiters, the state stands to gain valuable information about criminals, enabling their prosecution. Hence, I hypothesize that denouncing labor racketeering could lead to increased prosecution of organized crime members responsible for exploiting migrants.
Regarding public opinion on immigration, recent studies suggest that support for the integration of migrants is higher when they are perceived as vulnerable (Alrababa’h et al. Reference Alrababa’h, Dillon, Williamson, Hainmueller, Hangartner and Weinstein2021; Bansak, Hainmueller, and Hangartner Reference Bansak, Hainmueller and Hangartner2016). Notably, Bonilla and Mo (Reference Bonilla and Mo2018) show that emphasizing migrants’ susceptibility to human trafficking influences public opinion toward more pro-immigrant survey responses. Based on these insights, I hypothesize that a political act emphasizing migrants’ victimization through exploitation may shift public perceptions toward more pro-immigration stances and, sometimes, alter their partisan preferences.
CONTEXT AND QUALITATIVE EVIDENCE
Migrant Labor Racketeering
A universally agreed-upon legal definition of labor exploitation is still lacking (Global Migration Group 2013). The ILO proposed a legal definition of forced labor as “work or service exacted under the threat of penalty.”Footnote 7 The term encompasses slavery and other coercive practices, such as debt bondage or retention of identity documents. In this article, I focus on a practice used to accomplish labor exploitation: labor racketeering. This is a form of illegal recruitment and control of the labor force that relies on intermediaries to hire and control workers. Often intermediaries are members of criminal organizations, which contribute to smuggle migrant workforce and control their behavior once they are working, preventing them from reporting their exploitation to the police (UNDOC 2018, 52). This practice is common in seasonal and unskilled sectors, such as agriculture, constructions, and the service industry, and it is present across countries (Augère-Granier Reference Augère-Granier2021; Open Society 2020).
Labor racketeering is widespread in Italy, one of the first hot-spots of immigration into Europe and where criminal organizations are a long-standing presence. All four main Italian criminal organizations are involved in labor racketeering: mafias profit from this business by collaborating with foreign criminal organizations, as documented for the Sicilian mafia (Ministro dell’Interno 2021, 54), the ‘Ndrangheta, the Sacra Corona Unita (Parliamentary Commission 2022, 167 and 205), and the Camorra (La Repubblica 2021). This practice is used across the south, center, and north of Italy.Footnote 8 Reports by the Placido Rizzotto Observatory provide a recent picture of this phenomenon with reference to the agricultural sector: about 400,000 workers are involved in labor racketeering in agriculture—about one-third of the agricultural workforce. Of these, 49% are undocumented migrants and 40% in a situation of severe exploitation. In gross amounts, workers’ daily pay can reach 50 Euro, but this sum is forcefully curtailed: workers pay racketeers for transportation, food and water, such that their net wage is 20–30 Euros—about half the pay of legal workers. While Italians can also be employed through racketeering, the majority of workers in conditions of exploitation are migrants. Migrants are paid less than average—about 1 Euro per hour—they work longer hours, 62% of them live in ghettos without access to the most basic services, including water and sanitation,Footnote 9 and 76% develop diseases they did not have at the start of employment. This phenomenon has attracted periodic attention from media and institutions, usually in correspondence with with woeful news stories.Footnote 10 Mobilized migrants denouncing their condition and seeking change have faced threats, violence, and even homicide.Footnote 11
The Intervention
In 1914, farmworkers from Bari were working in Cerignola, a town in Apulia a few kilometers to the north, accepting lower pays than local farmworkers, resulting in conflicts and occasional violence (Perrotta Reference Perrotta2014). Union leader Giuseppe Di Vittorio, a historical figure of Italian unionism, resolved the conflict by uniting local and foreign farmworkers to demand a uniform salary from landowners. In 2007, following Di Vittorio’s legacy, the union of agricultural workers (Federazione Lavoratori Agro-Industria, or FLAI) organized a campaign to help workers exploited through racketeering.
The campaign had three key components: first, it provided workers with information about their rights. Second, it offered legal assistance to those who chose to report their exploitation to the authorities. Third, they offered an information impacting their incentives to denounce their employers: Italy, like other countries, offers a working permit to victims of exploitation, and the union assisted them in applying for this permit if their exploitation situation was assessed judicially. This permit allows non-E.U. citizens to legally work in Italy for 6 months, renewable for 12, and can be converted into a longer employment permit after expiration (Art 18, D.Lgsl. July 25 1998, N. 286).
The first two components are direct solutions to the most basic obstacles to whistleblowing faced by migrants: the lack of awareness about their rights, unfamiliarity with bureaucratic procedures, and lack of legal resources necessary to navigate a trial successfully. The presence of the union is also likely to facilitate mobilization initiatives, by creating awareness and reflection on migrants’ collective condition. The third component, instead, is likely to address the deeper fundamental challenges to whistleblowing: legal status removes the threat of deportation, and opens the doors to legal employment, resolving issues of job loss and risks associated with criminal intermediaries in illegal labor. Taken together, the three components of the intervention hold the potential to fundamentally change migrants’ decision-making regarding mobilization and reporting about labor exploitation.
The campaign was based on a standard plan delivered by the national union, which provided indications and resources to territorial units on how to implement the intervention. An ex post survey confirmed that local units followed these standard procedures (FLAI CGIL 2016, Section 7). The unions used camper vans with a median of 4.5 unionists to reach migrants in the fields on a weekly basis, employing linguistic mediators for communication. The intervention offered the same information, assistance, and materials to migrants.
Timing of the Intervention
The union-led intervention, initially promoted at the national level, was voluntarily adopted by local territorial units. It began in Apulia, where Di Vittorio originated, and expanded to other locations under the name “Street-Union.” Over 8 years, the intervention reached 49 municipalities across Italy (Figure A.1b in the Supplementary Material).Footnote 12 The decision to adopt the campaign was based on need (the existence of exploitation in that area) and will (the presence of union members willing to lead the intervention).Footnote 13 While data on union membership are not available below the regional level, the union itself collected data on the level of exploitation in different areas. I digitize these data, and test whether there is a relation between exploitation and the order of treatment adoption. Table A.1 in the Supplementary Material reveals an insignificant correlation between the order of treatment and levels of exploitation. This suggests that the highest or lowest exploitation areas were not specifically targeted first or later, even when adding region fixed effects (additional details in Section A.1 of the Supplementary Material). High-exploitation areas were also not more likely to receive treatment in absolute terms—many high-exploitation areas remain untreated. While other unobserved factors could explain why certain locations are treated earlier or later, robustness checks show that findings are not driven by the early-treated locations having specific characteristics different from the other municipalities treated later. After 2016, the union shifted focus toward advocating for institutional change, resulting in Law N.199/2016, which extends punishment to business owners and allocates more resources to support victims of exploitation.
Motives of the Participants
Arguably, the union had economic incentives for embarking in this campaign, to reduce the downward pressure on salaries caused by informal labor. They, however, also faced considerable costs: the possibility of organized crime retaliation. Interviews with unionists suggest that some of them were indeed targets of threats. How could the campaign continue when unionists were victims of attacks? I asked this question to one of the leaders, who replied that they received protection from “not being alone”: being embedded in networks of politically active individuals, with the resources to demand protection from the police and the capacity to increase their visibility and mobilize public opinion, helped unionists escape organized crime threats. Migrants in contact with union members might similarly be less “invisible.”Footnote 14
A second obstacle is that the intervention could not address the root cause of the phenomenon of labor racketeering—the existence of a vulnerable, undocumented population lacking alternative job opportunities and facing deportation risks if they reported exploitation. The union intervention could partially address this obstacle by providing migrants with information about alternative job opportunities and about the humanitarian permit. While information and connections can be helpful, it is still migrants who had to take the costly action of denouncing their racketeers and the risk of unemployment, retaliation, and deportation in case the humanitarian permit did not go through.
DATA
Media Coverage of Labor Racketeering
In Italy, there is no data on labor racketeering and its denunciation provided by the police at levels below the regional. I overcome the lack of data by scraping news items including the word labor racketeering (“caporalato”) from the seven main Italian national newspapers of different political slantFootnote 15 and validating this measure with province-level indicators of racketeering from government audits. The sample also includes the local editions of Repubblica and Corriere. While I do not scrape local-only newspapers, the local media landscape varies considerably across areas and in online archive availability. This non-homogenous sample might introduce nonrandom noise in news coverage by location. The news-based measure takes value different from zero when (i) labor racketeering takes place, and (ii) is uncovered, either through migrants denouncing their exploitation or through independent investigations. Newspapers can also cover the story if, after (i) and (ii), civil society initiatives against labor racketeering take place. This measure should thus be conceptualized not as an indicator of labor racketeering presence only, but as a measure of presence and prosecution of this phenomenon. Another aspect to notice to correctly conceptualize this measure is that newspapers tend to report news when extreme cases take place, such as severe exploitation, migrants’ killings, or revolts. As such, although the term “caporalato” might also refer to Italians employed through labor intermediation, the phenomenon I capture largely refers to extremely exploited migrants.
Figure 1a plots the temporal pattern in reporting about labor racketeering, indicating a similar trajectory across newspapers in time. For all newspapers, coverage of labor racketeering increases in 2014–15. In these years, Italy received two large waves of migrants, those escaping the aftermath of the Arab Spring (2011–12) and refugees from Syria (2014–15). The increase in news related to labor racketeering in those years is consistent with an increase in the phenomenon itself, driven by larger availability of migrant workforce. The presence of labor racketeering news is not systematically related to municipality size, a common concern with news-based measures (Figure A.1a in the Supplementary Material). It is unlikely that changes in local politics (the election of certain mayor) will affect national newspapers coverage, but I test for this possibility in Table A.3 in the Supplementary Material, showing that larger or smaller changes in voting do not affect reporting on labor racketeering. All fixed characteristics of the municipalities which might impact migrant labor exploitation and its reporting (e.g., the level of xenophobia, trust in institutions and state capacity, …) are partialled out by municipality fixed effects.Footnote 16
Trends in Outcomes
Labor Racketeering Data Validation
I validate my news-based measure of labor racketeering by comparing it with information on audits conducted in agricultural businesses throughout Italy by the Labor Inspectorate, the institution inspecting irregularities in the workplace under the Ministry of Interior. I obtained province-level data from the Labor Inspectorate on the number of agricultural firms found hiring workers informally. I normalize this and the news-based measure by the province population.
There is a strong correlation between news-based and audit-based measures of irregular labor in agriculture: the correlation is 0.55 and the distributions map each other well spatially (Figure A.2 in the Supplementary Material). In Table A.2 in the Supplementary Material, I regress the audit-based on the news-based measure controlling for year fixed effects (these account, e.g., for changes in government which might have triggered a different approach to investigations by the Inspectorate), region fixed effects, and a set of controls.Footnote 17 Results indicate a strongly significant correlation even in the most restrictive model, indicating that one additional news in one thousand residents translates into four more agricultural firms with irregularly hired workers over one thousand residents. In the results section, I also regress province-level treatment on the audit-based measure, confirming the results obtained with the news-based measure.
Goods Seized from Mafias and Destined to Public Use
Since the approval of Law n. 646/1982, Italy has had a judicial tool to seize goods, properties, and firms owned by criminal organizations. Since 1996 (L. 109/96), the law mandates that seized mafia assets be used for social purposes. These assets can either become part of the state patrimony (often used as offices) or be assigned to local administrations for redistribution to cooperatives, NGOs, and associations. Information on seized and destined goods is publicly available through the National Agency for Seized Goods (ANBSC 2024). The data include the year of confiscation, which occurs shortly after the property is seized, marking the first judicial step to remove it from its owner. Seizure and confiscation happen quickly for mafia-owned goods once sufficient evidence suggests illegal activity or money laundering. As a result, there is a short time lag between the start of investigations and the seizure and confiscation. The number of seized and redistributed goods has been increasing over time (Figure 1b).
Anti-Immigrant Attitudes
I use the panel survey conducted by Italian National Election Studies (2014) (ITANES) between 2011 and 2013, which included questions on voting behavior and political attitudes, including on immigration. The survey targeted 2,332 respondents over several waves, totaling 5,816 observations. Of these, 415 observations are in treated municipalities, belonging to 83 unique respondents interviewed over two or three waves. These individuals were in 22 of the treated municipalities (45% of the treated sample) and were asked questions before and after the union’s intervention. This allows me to assess whether the same person changed their mind on immigration by comparing their responses before and after the intervention, and to those untouched by the intervention. I use information on attitudes toward migrants, voting intentions, and trust in unions. The average anti-immigrant attitude and trust in unions are plotted in Figure 1c.
National Elections
National elections results come from the Ministero dell’Interno (2024). I group party formations into four categories consistent over time: far-left, center-left, center-right, and far-right. Ideologically, extreme parties are categorized as such, regardless of coalition with moderate parties, while parties always running in coalition are placed in the respective centrist category.Footnote 18 To account for turnout changes affecting vote share, I calculate vote share as votes divided by total eligible voters, as every person in Italy is automatically registered to vote. Figure 1d shows the evolution of vote share changes for the four party groups.
Municipalities Targeted by the Intervention
Information on municipalities reached by the intervention comes from FLAI CGIL (2016) on labor exploitation in agriculture by the union in 2016. Between 2007 and 2016, union members built and maintained relationships with immigrants in 49 municipalities across eight regions in Italy. The implementation was staggered over time (see Figure 2). Given the ongoing link between the union and migrant workers, I consider municipalities as treated in each subsequent year, though results remain robust to changes in this definition.
Number of Treated Municipalities over Time
EMPIRICAL STRATEGY
To identify the effects of the intervention, I rely on a DiD strategy comparing municipalities which were and were not targeted by the campaign, before and after it took place. For municipality i and year t, I estimate the following two-ways fixed effects model:
$$ \begin{array}{rl}{Y}_{it}={\alpha}_i+{\beta}_t+\gamma Trea{t}_{it}+{\epsilon}_{it,}& \end{array} $$
where
$ Treat $
is a municipality-specific time-varying indicator equal to 1 in municipalities and years in which the unit is targeted by the intervention. Municipality and year fixed effects guarantee that any time-invariant characteristic of the location is partialled out from the effect. For example, fixed effects account for whether the location is an agricultural area, or it has a history of xenophobia. I cluster standard errors at the municipal level.
A possible threat to identification could derive from treatment assignment being nonrandom, and in particular correlated with the level of exploitation of treated areas. If this is the case, the outcomes I consider will be already on different trends before the intervention. I account for this concern in three ways. First, I show that trends in outcomes—labor racketeering news, goods seized and destined, and voting—were largely parallel before the start of treatment in each location (Section B of the Supplementary Material). Second, using information from reports produced by the union on the level of exploitation in each location in Italy, I show that locations with high (low) levels of exploitation were not more likely to be treated earlier (or later): the intervention targeted a mix of medium and high-level areas in each roll-out phase of the campaign (Section A.1 of the Supplementary Material). Third, findings are not driven by early-treated municipalities having specific characteristics different from late-treated: results are robust to dropping the entire sample of treated observations in the first, second, third, and fourth year (Table D.1 in the Supplementary Material). Fourth, to reduce observable differences between municipalities that ever received treatment and those that did not, I compare treated units to a restricted set of most-similar control units identified with nearest neighborhood propensity score matching.Footnote 19 A balance table reveals that matching substantially minimizes differences across treated and control units, considering both the matching and the other variables (Table B.2 in the Supplementary Material).
EFFECTS OF THE INTERVENTION
Reporting on Labor Racketeering in the News
I start by asking whether the campaign was effective in increasing reporting on cases of labor exploitation in newspapers. I observe a significant 15% increase in the likelihood that any news related to labor racketeering is reported in a municipality (Figure 3a), corresponding to an increase by 0.04 news in one thousand inhabitants—a 20-fold increase with respect to the mean of the dependent variable in the control group (Figure 3b, black). This effect is also present when considering the matched control sample (Figure 3b, gray). Results are robust to running estimates separately by the newspaper covering the news (Table D.4 in the Supplementary Material) and for the total number of news, rather than news per capita (column 1 of Table D.5 in the Supplementary Material). Since the distribution of the dependent variable is skewed toward zero, I also show robustness to a Poisson and negative binomial specifications (columns 2 and 3). Results are more precise when dropping the last two posttreatment periods, when larger but higher variance effects are observed (column 5). Results are robust to adding flexible controls for important confounders such as foreign population and male unemployment (columns 6 and 7).
Effect on Labor Racketeering News and Goods Seized from Mafias
Note: Results from DiD in Equation 1. Full (propensity score matched) sample coefficients in black (gray). Panel a considers any news related to labor racketeering, b considers the number of news per capita over one thousand residents, c considers any good seized from organized crime, and d considers the number of goods. All panels include confidence intervals at 95% and 90%, municipality and year fixed effects and standard errors are clustered at the municipal level. Results in tabular form in Table 1, APSR Dataverse files (Dipoppa Reference Dipoppa2024b).
Content of the News
An important question is whether the increase in news is an automatic product of journalists reporting about the union’s campaign, or if instead reporting reveals an increase in migrants whistleblowing and mobilizing at higher rates. To answer this question, I read and classify the content of news in treated municipalities after treatment and validate my classification with that of a research assistant (details are included in Section A.3 of the Supplementary Material). The majority of news (29%) discusses cases of migrants denouncing their racketeers or police operations against labor exploitation. Second, there are demonstrations or initiatives by migrants or by the civil society to achieve change on labor exploitation (23%), third journalistic reports (22%), and finally policymaking initiatives or debates on how to fight this phenomenon (20%). A randomly selected sample of news in control locations discusses labor exploitation only in the context of more general journalistic reports about crime, without referencing protests and whistleblowing. The dynamic of migrants’ mobilization and reporting in treated areas can be understood reading examples of news items:
The strike of North African workers continues for the sixth consecutive day. These migrants, who pick tomatoes in the fields, protest against low wages and illegal hiring practices, and promise to continue crossing their arms until their requests are met. The DDA (judicial investigative unit) in Lecce has opened an investigation on the exploitation of immigrants in the fields of Nardò. There are multiple reports by workers, including people received death threats from the racketeers. (Russi Reference Russi2011)
The arrests were in the air for months. Last summer, the workers’ fight in Masseria Boncuri, a shanty town home to migrants enduring the inferno of the farmlands, garnered significant attention. They bear 10-12 working hours under a scorching sun, paid 20-25 euros at best, below the poverty line. (Colluto Reference Colluto2012)
The unions’ campaign did not receive direct media coverage, but unionists are often interviewed in the context of journalistic reports, and in this case they sometimes mention the campaign. Mentions of the union in the news (searching for “union,” “Flai,” or “Cgil”) are balanced across the treated and control sample, indicating that the union did not play a role in differential media reporting on episodes happening in treated locations.
Was It the Migrants or the Union?
Another important question is whether the intervention was driven by migrants, who found the courage to whistleblow, or directly by the union, who talked to the police themselves. Both theoretical and empirical evidence suggests that the effect was primarily driven by migrant behavior. First, both the union reports and interviews with union leaders indicate that the campaign did not involve talking to the police or the judiciary to suggest them to investigate. This is likely because prosecuting labor exploitation is challenging without workers willing to testify, and the police and judiciary in Italy are among the most overburdened bureaucracies. Encouragement to take on additional work may result in no action, while a formal denunciation of illegality compels them to act.
Second, in Section C of the Supplementary Material, I provide qualitative evidence that treatment effectively spurred migrant activation in several ways: (i) migrants in treated locations organized public demonstrations to bring attention to the issue; (ii) they denounced their employers, contributing to judicial efforts to curb exploitative labor practices; and (iii) some migrants dedicated themselves to long-term political activism, making it their career and engaging in advocacy for migrant rights.
Third, I offer a formal test that unions alone are unlikely to have caused the higher prosecution against labor racketeering. Imagine that contact with migrants does not influence prosecution, and instead all that matters is that unions identify where exploitation takes place. Union members could request prosecution directly in all locations where they have this information, without migrants’ intervention. Using the same union reports I used for validation purposes, I observe the locations which the union identified as having medium or high levels of exploitation, but where they did not lead the same intervention involving contact with migrants. I use these locations to test if areas identified as high-exploitation but where the union did not intervene with migrants, after the period of investigation by the union, experience a similar increase in labor racketeering news. This is a placebo DiD using the same specification as the main result on labor racketeering news, but replacing the intervention with this placebo treatment. As shown in columns 2 and 3 of Table D.2 in the Supplementary Material, simply gathering information about exploitation without contact with migrants has a null effect on labor racketeering news.
Finally, I code news items about labor racketeering that also contain mentions of the word “protest.” Those instances are the most likely to capture political activation to exit exploitative regimes. The DiD analysis indicates that those news items are significantly more likely to occur in treated cities after the intervention—a 16-fold increase over the mean of the dependent variable (column 4 of Table D.2 in the Supplementary Material).
Taken together, the content of news articles, the existing qualitative evidence, the null effect of placebo interventions not involving contact with migrants, and the effect of treatment on protests, provide evidence that the increase in reporting about labor racketeering is driven primarily by migrants’ mobilization which triggered state intervention, but also by the activism of local civil society and institutions to achieve change.
Magnitude of the Intervention and Impact of the Program
Are the effects on labor racketeering news reporting realistic, given the scope of the campaign? To address this question, I utilized data collected by the union on the number of migrants reached by their intervention. Over a 3-year period from 2012 to 2014, the union engaged with 21,442 migrant workers (FLAI CGIL 2016, 242), all in conditions of moderate to severe exploitation and thus in a condition to report exploitation to the state. I compute the average number of migrants reached per location-year (N = 167.6), and impute this value to the missing location-years to estimate that a total of 47,761 migrants were likely reached by the campaign. How many of these migrants would respond to the intervention by whistleblowing or engaging in mobilization? As a benchmark, a 5-minute, in-person canvassing intervention on migrants persuaded 3.4% to turnout to vote (Pons and Liegey Reference Pons and Liegey2019). The union’s intervention was a substantially more intensive interaction, but its objective was to modify a costlier behavior than voting. If we assume the same effectiveness, the intervention might have led 14,328 migrants to whistleblow. Hypothesizing a third of the effectiveness, the intervention could have activated 4,776 migrants. Both figures are substantial and could explain the presence of 0.7 additional news pieces per year, as in the main result.Footnote 20 This is particularly true when considering that political activation likely extended beyond the initially treated migrants to involve their peers, potentially leading to a larger number of activated individuals than those directly reached by the union.
Finally, I test if displacement of labor racketeering took place in neighboring municipalities. Evidence is weakly consistent with this possibility: municipalities bordering treated locations experience a (nonsignificant) drop in news related to labor racketeering. As migrants exit racketeering in treated areas, neighboring areas display either no change or a slightly stronger grip of racketeers and a consequent insignificant drop in reporting and prosecution (Table D.3 in the Supplementary Material). Treatment effects are instead similar excluding these cities from the analyses.
Organized Crime Prosecution
The second dimension of interest is state capacity to prosecute racketeers: I ask if the intervention effectively increased state prosecution against organized crime, which is often the intermediary in the most severe cases of exploitation. The results suggest this is the case: treated municipalities experience a 12% increase in the likelihood of having any goods seized to organized crime in the years following the intervention (Figure 3c). This corresponds to 1.16 to 1.3 more goods seized from organized crime—thirteen times more than the average in the control group considering the full sample (Figure 3d, black), and four times more considering the matched sample (gray coefficient). Seizing firms to mafias is a rarer event—for example, there are 44,462 seized properties in the database but only 5,365 seized firms. While less precise, there is also evidence that the number of firms seized from organized crime increased in treated municipalities (Table D.6 in the Supplementary Material).
Properties Destined to Public Use
After properties are seized from organized crime, the investigations are concluded, and the goods are conclusively confiscated, they can be reassigned for public use. These goods can either be used as offices or be assigned to local administrations, who can give them for free to cooperatives and associations. While these properties represent freebies for local governments, distributing them is expensive, bureaucratically complex, and complicated by the fear that criminal groups might retaliate against those endowed with their former properties. Often, this means that properties seized from mafias are left unused (Cisterna Reference Cisterna2012).
At the same time, financing cooperatives of agricultural workers is considered a best practice to curb exploitative practices in agriculture, since it creates legal employment options for agricultural workers (Guidelines on preventing labor exploitation in agriculture, Ministry of Labor 2021). I hypothesize that, if administrations face political pressure for adopting measures against exploitation, they might resort to goods seized from mafias by assigning them to cooperatives working in agriculture.
In line with this hypothesis, treated municipalities experience a significantly higher number of destined goods: 1.08 goods are destined to social use in treated municipalities, against an average in the control group of 0.037 (a 28-fold increase from the control group mean). This increase is largely driven by goods assigned to agricultural cooperatives: 0.26 more goods are destined to agricultural cooperatives in treated municipalities with respect to 0.008 in the control group (a 47-fold increase from the mean, Figure 4).Footnote 21 The effect is unlikely to be the automatic product of the increase in seized properties: while goods suspected to be mafia owned can be rapidly seized, for those goods to be reassigned there is a long multiple steps process, including concluding investigations, assigning the good to the ANBSC, then to local governments, making it compliant with regulations, and deciding its destination. The average time between the seizure and the final destination of a property is 2,023 days—5–6 years (Cisterna Reference Cisterna2012). These findings are in line with Boeri, Di Cataldo, and Pietrostefani’s (Reference Boeri, Di Cataldo and Pietrostefani2024) conclusion that the destination of mafia goods can, in small part, compensate the territory for the damages it suffered from organized crime.
Effect on Properties Destined for Social Use
Note: Results from DiD in Equation 1. Properties destined for any social use in solid lines, for agricultural use in dashed lines. Full (propensity score matched) sample coefficients in black (gray). Confidence intervals at 90% and 95%, municipality and year fixed effects are included. Standard errors are clustered at the municipal level. Results in tabular form in Table 2, APSR Dataverse files (Dipoppa Reference Dipoppa2024b).
Attitudes and Voting, Individual-Level Data
A campaign that generated higher media coverage of labor racketeering, prosecution of mafia members, and local governments mobilization, might have also impacted citizens’ view on immigration and partisan preferences. In the last decade in Italy, both the center-right and center-left governments promoted restrictive policies on immigration. In particular, the center-left government proposed and sealed a pact to intercept migrants trying to cross the Mediterranean and send them to Libya, an unprecedented initiative which violated international conventions requiring states to assess migrants’ right to asylum. This decision was criticized by several institutions, including nonpartisan ones like the United Nations, which called the pact “inhumane” (The Guardian 2017). Given the stance of center-left governments, voters with more pro-immigration attitudes should be more likely to vote for parties that maintained a liberal approach toward immigration—namely, the far-left.
Empirical Strategy
I consider the effects of the intervention on the public opinion of natives by capitalizing on the ITANES survey’s inclusion of individuals in some treated locations (22 municipalities, 45% of the treated sample) before and after treatment. This analysis relies on a smaller sample (depending on the outcome, approximately six thousand observations), and it only includes the years 2011–13, a fraction of the treated period. It should thus be taken with caution. However, due to the fortuitous overlap of the panel survey rollout and of the union intervention, this sample allows to observe changes in the same respondents’ immigration and voting preferences before and after treatment. This analysis is thus representative of the impact of treatment on individuals, rather than at the municipal level as in the rest of the analyses. For respondent r in wave w, I estimate
$$ \begin{array}{rl}{Y}_{wt}={\rho}_r+{\omega}_w+\lambda Trea{t}_{rw}+{\nu}_{wt,}& \end{array} $$
where
$ \rho $
and
$ \omega $
are respondents and wave fixed effects and
$ \lambda $
is the coefficient of interest, capturing the change in voting and anti-immigrants attitudes after the intervention.
Figure 5a shows that the same individual is 2–3% more likely to report voting for far-left pro-integration parties, and is 9% less likely to report anti-immigration sentiments (Figure 5b, solid line). A possible alternative explanation could be that the increase in far-left vote is not related to immigration but to the union’s activity, which supports left-wing parties. However, if this was the case, we would also see a positive effect on center-left parties’ vote share, which is not observed. Second, if the union was so influential as to sway voters, we should expect trust in this institution to increase. Instead, the effect on trust in unions is negative and null (Figure 5d, dashed line).
Effect on Anti-Immigrant Attitudes and Far-Left Voting, Individual-Level Data
Note: Results from DiD in Equation 2. Voting intentions for Rifondazione Comunista (Comunisti Italiani) in black (gray). Anti-immigrants attitudes (trust in unions) in black (gray). Confidence intervals at 90% and 95%, respondent and wave fixed effects are included and standard errors are clustered at the municipal level. Data from ITANES survey. Results in tabular form in Table 3, APSR Dataverse files (Dipoppa Reference Dipoppa2024b).
Voting, Municipal-Level Data
I confirm the results on voting intentions using municipal-level data on voting in national elections. I present results for parties’ vote share and change in vote share from the previous election year. Conceptually, the change in vote share better captures how large a variation in electoral performance occurred between elections, giving a more precise indication of the extent by which the intervention reshaped a municipality’s political landscape. Results are presented in Figure 6. Starting from the coefficient of interest (far-left, in blue), there is a significant increase in far-left votes, the only political group which consistently supported pro-immigration policies. These parties experienced a positive change with respect to the previous elections that is much larger than the average change in the control group and corresponds to a 65%–79% increase from the previous election year. In levels, this effect corresponds to a 1% increase in far-left vote share. The positive change is visible in each of the two periods after treatment (Figure B.2 in the Supplementary Material). Considering the other parties, the center-left experienced a significant drop in vote share but an insignificant effect on the change in votes. This coalition lost votes posttreatment, consistently with the possibility that voters partly shifted from the center to the far-left due to immigration. The change from each previous election year is however not statistically significant. The center-right does not display consistent signs of change across specifications. I do not interpret results for the far-right parties—those are largely unstable across specifications, ranging from positive to negative, and there is no support for the parallel trends assumption for this outcome (Figure B.2 in the Supplementary Material). Instead, across robustness tests, the coefficients for the far-left party are consistently positive and significant. If the intervention is responsible for the change in vote share for the far-left, we should observe larger effects in municipalities that were treated closer to elections. In Table D.8 in the Supplementary Material, I test the same specification on the subsample of municipalities which were treated less than 1 year before elections: as expected, the change in votes is larger considering this subsample.
Effect on Voting, Municipal-Level Data
Note: Results from DiD in Equation 1. Full (propensity score matched) sample coefficients are displayed in black (gray). For the coefficient of interest, vote for the far-left, coefficients for the full (propensity score matched) sample are in blue (green). Confidence intervals at 90% and 95%, municipality and year FE are included and standard errors are clustered at the municipal level. Data on national elections 1994–2018 from the Ministry of Interior. Results in tabular form in Tables 5 and 6, APSR Dataverse files (Dipoppa Reference Dipoppa2024b).
Robustness
Relaxing the Linearity Assumption
For the first three outcomes, the distribution of the dependent variable is skewed toward zero, as those tend to be rare events. I account for this skew by presenting results using a Poisson and negative binomial regression model for news (column 4 of Table D.5 in the Supplementary Material), goods seized to mafias (columns 1 and 2 of Table D.7 in the Supplementary Material), firms seized from mafias (Table D.6 in the Supplementary Material), and goods destined to social use (columns 3 and 4 of Table D.7 in the Supplementary Material). Results are robust to this alternative specification.
Relaxing the Treatment Homogeneity Assumption
Recent literature on dynamic DiD designs shows that these models exploit comparisons between early-treated and late-treated units which can bias the estimation of treatment effects if those are not constant across groups or times (Goodman-Bacon Reference Goodman-Bacon2021). To account for treatment effects heterogeneity bias, I rerun event studies for all outcomes using four new estimators proposed in the recent DiD literature. Results in Figures D.1–D.4 in the Supplementary Material are consistent across specifications. Treatment effects display a gradual increase over time, with news and seized goods peaking a few years into the intervention. The pattern is consistent with the possibility that initially only a few migrants denounce exploitation, but as they succeed, others follow suit. In contrast, policymaking responses (goods destined) and voting show a more immediate reaction, aligning with a different process in which the intervention impacts public opinion.
Relaxing the Parallel Trends Assumption for Voting Outcomes
The parallel trend assumption implicitly implies that it is possible to control for selection effects in which units enter treatment, as treatment post-periods are assumed to have evolved along the same lines as control pre-periods absent treatment. Using fixed effects and matching on observables alleviate this concern. Still, synthetic DiD approaches (Arkhangelsky et al. Reference Arkhangelsky, Athey, Hirshberg, Imbens and Wager2021) have proven particularly effective at compensating for the potential lack of parallel trends by reweighting units to match their pre-exposure pre-trends. For each outcome, I present results in a synthetic DiD approach: results are consistent with the main analysis (Table D.11 in the Supplementary Material). Unlike other outcomes, voting is only observed in election years, reducing the coefficients available for assessing the common trends assumption. In an alternative specification, I compare within treated units: this strategy reduces the sample size and drops the last year of treatment, but it allows relaxing the assumption that trends between treated and control group are parallel, since it uniquely relies on within-treated units. The identifying assumption is no strategic selection into treatment timing. As discussed, the treatment decision was decentralized to the local labor union, and areas with the highest levels of exploitation were not systematically targeted first. Results in Table D.12 in the Supplementary Material show again a strongly positive and significant coefficient for far-left parties.
Duration of Treatment
Interviews with activists in the campaign indicated that, once a location was treated, unionists maintained a presence in the area.Footnote 22 In Figure D.5 in the Supplementary Material, I test the robustness of the findings to modifying the duration of treatment to less years, until only considering units as treated in the same year in which treatment started. Results are robust for all outcomes across all treatment duration.
Size of Treated Sample
Results are robust to using p-values from Fisher randomization inference (Table D.9 in the Supplementary Material). This test accounts for the fact that the number of treated observations is relatively small, but it also relies on the assumption that treatment is assigned randomly, which in this case might be excessively strong. I also test the robustness of results to using Conley–Taber confidence intervals. This method relaxes the asymptotic assumption that both the number of treated and untreated units is large and only relies on a large control group, as in this article. To construct the Conley–Taber confidence intervals, information from the untreated group is used to consistently estimate the distribution of the DiD point estimator. As Table D.10 in the Supplementary Material reports, CT confidence intervals are similar to those in the main specification, and for most outcomes this specification reduces their size.
Simultaneous Interventions
A classic concern in DiD designs is that simultaneous interventions might happen at the same time as treatment and drive results. Since this is a setting where different locations are treated in eight different points in time, it seems unlikely that other groups independent from the union decided to treat the same set of locations in the same exact order chosen by the union. This might have happened by coincidence in some locations, but I show that removing any of the treated locations, results are identical: none of them in particular drives the findings (Figure D.6 in the Supplementary Material).
Multiple Hypothesis Testing
Since I estimate the effect of treatment on multiple outcomes, I account for multiple hypothesis testing. I compute sharpened p-values that factor in the potential rate of false rejections due to testing a large number of coefficients. Results reported in Table D.13 in the Supplementary Material confirm the significance of the estimates. Finally, I show that all results are robust to the inclusion of region times years flexible controls, accounting for region-specific time trends which might impact the outcomes of interest (Table D.14 in the Supplementary Material).
CONCLUSIONS
This article studies the consequences of undocumented migrants empowerment to denounce and exit from labor exploitation. An intervention providing migrants with information and incentives to denounce their employers was effective in increasing news reporting about cases of exploitation, particularly news reporting about migrants denouncing their exploitative employers or migrants’ mobilization initiatives. In line with racketeers often belonging to criminal organizations, higher whistleblowing of labor exploitation spilled over into higher rates of mafia prosecution, with a significant increase in seizure of mafia-owned properties. In this respect, this article points to an important policy implication: cracking down on migrants’ exploitation directly undermines organized crime, by unveiling their activities and reducing their profits. One limitation of this study is that it cannot observe whether retaliation against undocumented migrants took place as, unfortunately, these data are unobserved.
I further document that intervention raised awareness about the vulnerable condition migrants were subject to among local residents and institutions, as revealed by news talking about initiatives against racketeering, and by local governments assign properties to agricultural cooperatives—a system to offer migrants legal employment.
Finally, this intervention did not produce a backlash against pro-immigration parties, as in other examples of policies favoring migrant integration. Instead, treated municipalities experience reduced anti-immigration attitudes and an increase in vote share for pro-immigration parties. This finding is in line with recent literature showing that highlighting migrants’ fragility can lead natives to become more accepting of migrants (Bansak, Hainmueller, and Hangartner Reference Bansak, Hainmueller and Hangartner2016). This article contributes to this literature by showing that not only attitudes toward migrants improve when citizens are exposed to their condition of vulnerability, but that also voting behavior responds accordingly.
As labor exploitation becomes increasingly prevalent in developed countries (ILO 2022, 22), understanding under what conditions migrants can emancipate from exploitative regimes is of primary importance. This article sheds light on the determinants of migrants empowerment by showing that information and incentives to whistleblow can lead to the empowerment of marginalized undocumented communities.
This study presents promising results to address situations of severe exploitation, as it improved migrants’ welfare at no political cost for the political side promoting it. In fact, the most recent government guidelines for fighting labor racketeering were developed in conjunction with the unions and incorporated several elements of this intervention.Footnote 23 Nonetheless, a few caveats are worth noticing. First, while news items do not mention any such instances, it is possible that the intervention might have produced retaliation by criminals against migrants. More carefully collected information than those received and reported by newspapers would be needed to assess whether retaliation took place. Second, while the intervention was effective at the scale at which it was led, scaling it up might produce different effects: systematic reporting by migrants could only be achieved if migrants who whistleblow are effectively rewarded with a humanitarian residence permit, a strategy which would require governments to grant temporary legal status to a larger number of migrants and which might produce different effects on public opinion. On the other hand, if reporting labor racketeering to the police became the equilibrium, exploiting migrants might become too costly and stop being a profitable business, with positive spillovers on legal competition, fiscal entries, and state capacity more in general. The question of whether scaling up this intervention would be overall beneficial for the state, for natives, and for migrants remains open. More broadly, this study underscores the importance to consider state actors at large when devising policy interventions. While governments might have (or believe to have) misaligned electoral incentives to lead pro-migrant interventions, other actors, such as unions, might have incentives to act and empower marginalized groups like undocumented migrants.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424001138.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/G126SC.
ACKNOWLEDGEMENTS
For suggestions that have improved this article, I am grateful to Robert Blair, Donghyun Danny Choi, Gianmarco Daniele, Guy Grossman, Saad Gulzar, Jan Pierskalla, Vincent Pons, Stephanie Zonszein, and to the participants of APSA, MPSA, BSE Summer Forum, PDRI Conference, Public Choice Meetings, Economics of Organized Crime Workshop, and University of Bari and Padova. The directional hypotheses predicted in this article were not preregistered.
CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The author claims exceptions to the APSA’s Principles and Guidance for Human Subject Research (2020) and provides reasoned justification in the appendix. Although drawing on research with human participants, the author claims exemption from organizational ethical review and provides reasoned justification in the appendix.
Open access
Comments
No Comments have been published for this article.