Digging into waste: an analysis of waste crime in the Italian provinces

ABSTRACT This article examines the drivers of waste crime using a panel dataset at the provincial level in Italy. The results demonstrate that while waste crime is characterized by some degree of inertia, it lacks significant spatial patterns and does not respond to general modes of deterrence. With respect to income, the human capital effect prevails over the scale effect. An inefficient public sector and inadequate waste management system result in higher waste crime rates. The hauling and storage facilities also provide several opportunities for illegal shipment and disposal, which organized crime groups can especially exploit.


INTRODUCTION
The term 'waste crime' encompasses a broad set of illegal activities conducted across the entire waste cycle (i.e., production, storage, transportation, disposal and reuse), ranging from illegal dumping to illicit trading of waste. 1 All these activities produce severe long-term concerns for human health, general quality of life and overall quality of the environment. The severity of these concerns demands a detailed investigation of the drivers of waste crime. So far there is a relative dearth of empirical studies into the supply function of waste crime. In this respect, newly available datasets and econometric techniques can aid the development of a more rigorous empirical methodology capable of testing a more extensive range of drivers.
This article explores the empirical sensitivity of waste crimes in Italy to a range of structural, economic, criminal and performance-driven determinants by extending previous empirical analyses. In particular, in contrast to prior research, by using generalized method of moments (GMM) models at the provincial level, we disclosed detailed dynamics and checked for potential spatial dependence and geo-clustering effects. In addition, the neglected problem of measurement error in such a latent phenomenon as waste crime is tackled for the first time using a tailored econometric technique, which reduces the bias in the estimates.
We consider Italy a compelling case study for this topic for three reasons: (1) severe instances of waste crime and mismanagement have occurred in some parts of the country where there is a significant mafia presence (D'Alisa et al., 2010;Massari & Monzini, 2004;Mazzanti & Montini, 2014); (2) the ready availability of data, covering more than 100 provinces over a nine-year period, which allows exploiting the substantial spatial heterogeneity over time to conduct a robust statistical analysis; and (3) the possibility to identify the impact of different waste management and environmental policies based at the regional level (Mazzanti & Zoboli, 2013).
We adopt a measure of waste crime that includes the collection of all violations that are related to waste crime. Other studies on Italy adopted different measures. To test enforcement and environmental policies mainly, D' Amato et al. (2018) used a subset of the violations included in our measure, that is, only those collected from the Forestry Corp. To test the environmental Kuznets curve, Germani et al. (2020) considered a broad measure of environmental crime that also included illegal construction and crimes against archaeological, wildlife and forest heritage. In contrast to these works, we use a measure that is not as broad as to include all environmental crimes and not too narrow to exclude other potential sources of waste crime. In particular, this analysis is more functional in constructing general policies against waste crime than addressing specific instances of such a complex phenomenon.
The paper is organized as follows. The following section presents the review of extant literature. Section 3 discusses the data on which our analysis is based, while Section 4 delineates the methodology. Section 5 presents the results of our regression analyses and is followed by a set of tests on spatial dependence and an evaluation of the robustness of the results in Section 6. Section 7 concludes.

LITERATURE REVIEW
The factors that affect waste crime are manifold and can involve all stages of waste management. The extant literature on this topic is both limited and dispersed to the best of our knowledge, while a much-needed rigorous review is still lacking. 2 One possible reason for the relative dearth of empirical studies in this area is the victimless and invisible nature of this type of crime. As a result, reliable proxies are not readily available. Only recently, Dell'Anno et al. (2020) attempted to provide a proxy for waste crime in Italy using a structural equation approach estimated by a partial least-square algorithm. Unfortunately, this measure is only available for large geographical units (i.e., regions instead of provinces).
According to the rational choice theory, human beings are rational actors who weight benefits and costs in deciding on a course of action. The primary theoretical explanation for waste crimes is the cost-saving opportunities provided by illegal waste disposal compared to legal alternatives. Waste crimes offer considerable benefits to offenders, which are mainly entrepreneurs, companies and organized crime groups. For instance, the provision of illegal services is particularly beneficial for the management of unsorted waste for which no specialized skills or high-tech plants and equipment are required (Di Pillo et al., 2022). We can generally distinguish between crime-stimulating factors that encourage criminal intent, namely legal disposal costs and regulation compliance, and crime-inhibiting factors that discourage criminal activities, such as generic and specific measures of deterrence. The cost-benefit mechanism is then combined with other socio-economic and demographic factors which shape the propensity to commit a criminal act. Table 1 illustrates the state-of-the-art of the main quantitative investigations that are strictly related to waste crime, providing details about data, methodology and relevant results. Despite some regularities in the findings, the empirical literature on the determinants of waste crime is also characterized by contradictory results. The empirical methodology has mainly relied on count data models such as Poisson or negative regression models to estimate the drivers of waste crime. However, these models neglect the persistence of criminal behaviour and the potential endogeneity mainly driven by the reverse causality between deterrence and illegal behaviour. GMM estimators better address these concerns, but in their standard forms they do not consider another potential endogeneity concern due to the correlation between the measurement error affecting the proxies of waste crime and certain explanatory variables, such as law enforcement variables. This issue eventually leads to biased results (Buonaccorsi, 2010;Meijer et al., 2017). Finally, the extant literature has not provided any testing for spatial patterns of waste crime, which could be another potential source of bias. Hence, we attempt to fill the existing gaps in the literature by using a more compelling analysis of the drivers of waste crime.
In what follows, we formulate a set of testable hypotheses that are partially based on the previous literature but are also drawn from the specific traits of the Italian context. These predictions can guide the empirical analysis in the subsequent sections.
The first crucial variable to evaluate is income. More affluent areas are likely to generate more waste, and, as a result, a scale effect takes place. Significant waste production levels can, on the one hand, lead to major inefficiencies in the waste management system, but on the other, they can provide more resources that are directed to manage the entire cycle more efficiently and at a lower cost (Liu et al., 2017). Therefore, the effect on the incentives for illegal waste disposal is ambiguous. Yet, income levels are generally correlated with higher education levels, which would increase both the opportunity cost of criminal behaviour and the level of environmental awareness (e.g., Almer & Goeschl, 2010;Liu et al., 2017). The resulting lowered socio-cultural acceptability of illegal conduct regarding waste can play an integral role in designing effective waste management policies and reducing the risks associated with illegal waste management. However, the human capital effect can also be misleading insofar as it may increase reporting propensity, in turn, positively impacting the proxy of reported illegal dumping (e.g., Almer & Goeschl, 2015;Kim et al., 2008). In our case, we may expect that the human capital effect exceeds the increase in reporting propensity. Indeed, the latter appears marginal in a situation where the proxy of waste crime includes a wide variety of offenses mainly uncovered by investigative activities. Finally, low-income levels can also encourage incoming illicit waste trafficking because the illegal shipment and treatment of especially hazardous waste (e.g., car tires, end-of-life vehicles, electrical and electronic equipment, as well as toxic and radioactive waste) is a vital source of income for poor regions (Bisschop, 2012;Liddick, 2010).
Hypothesis 1: Although there are contrasting effects, a higher income level should be associated with a lower level of waste crimes.
As a waste-generating factor, population density is expected to have a positive impact on waste crime rates due to the anthropic pressure it places on waste 1368 Daniela Andreatta et al. management services. However, population density also affects waste crimes via other channels. For instance, land opportunity costs in urban and densely inhabited areas can increase the costs of landfills, thereby increasing both collection and disposal taxes and, as a result, illegal dumping. Despite this, more densely inhabited areas may seek to exploit economies of scale and increase the level of competition between waste businesses to such an extent to improve waste management and, in turn, reduce both costs and illegal dumping. Finally, as the anthropic pressure increases, we may expect to see more intense informal controls, which would result in higher reporting rates and probability of detection. Therefore, the latter would be a crime-inhibiting channel while simultaneously having a positive impact on the reported crime rates. Both Kim et al. (2008) and Liu et al. (2017) found that high population density levels had a positive impact on reported crime rates, while Almer and Goeschl (2015) and D'Amato et al. (2018) determined that there was no effect. Thus, we can reasonably formulate the following hypothesis.
Hypothesis 2: A higher population density is associated with a higher level of waste crimes.
According to the rational choice perspective, law enforcement activities increase the expected costs of committing crimes. Law enforcement ranges from prevention activities, such as inspections and patrolling, to the investigative and sanctioning actions that determine the probability of being caught, tried, and convicted. In general, the literature lends evidentiary support to the deterrence effect of various law enforcement activities. Sigman (1998), Kim et al. (2008), Liu et al. (2017), andD'Amato et al.'s (2018) findings are all in line with theoretical expectations, but their findings are not always statistically significant.
Hypothesis 3: A higher level of enforcement is associated with a lower level of waste crimes.
Organized crime groups actively participate in the waste business and cover the shortage of waste treatment and disposal facilities. In Italy, mafia-type groups (Di Pillo et al., 2022;D'Amato et al., 2015) or other forms of criminal associations organized by professionals (Andreatta & Favarin, 2020) meet the demand for alternative solutions to waste management and disposal and have complete control over large parts of the waste cycle, both in its legal and illegal forms.
Hypothesis 4: The presence of organized crime groups is associated with a higher level of waste crimes.
Performance-driven factors of the waste sector can be relevant drivers of waste crimes' dynamics (e.g., Germani

Number of environmental crimes
Law enforcement (-); penalties (¼); environmentally sensitive parties (-) POLS, FE, diff-GMM, sys-GMM, bias correction model. Panel (1995-2005; units 15), Germany Almer and Goeschl (2015) Number of illegal dumping incidents of waste disposal Law enforcement (-); income (+); manufacturing density (¼); population density (¼); environmentally sensitive parties (¼) FE, sys-GMM. Panel (1995-2005units Liu et al., 2017). At a broad level, an efficient public administration can reduce overall disposal costs by finding innovative and cheaper forms of waste management. It is also an indirect measure of the efficiency of the waste sector itself, both public and private. 3 On the contrary, where the public sector is inefficient, illegal alternative methods of waste management are more likely to emerge. In the same fashion, at a sectoral level, the adoption of environmentally sound management policies through the investment in waste recycling may lower the occurrence of waste crimes, for instance, by reducing the demand for local waste landfilling. On the contrary, unrecoverable waste could exacerbate illegal disposal and trafficking. This is especially true for domestic waste, as opposed to special waste. Therefore, we can put forth the following hypotheses.
Hypothesis 5: A more efficient public sector is associated with a lower level of waste crimes.
Hypothesis 6: A higher percentage of waste recycling is associated with a lower level of waste crimes.
Finally, the specific economic fabric of Italian provinces might influence the incidence of waste crimes in these territories. The presence of more agricultural and less industrialized areas could lead provinces to experience fewer waste crimes since the agricultural sector is susceptible to environmental health. On the contrary, a relatively higher presence of businesses in the waste sector and businesses devoted to transportation and storage could increase waste crimes because these businesses are at risk for criminal infiltration (Andreatta & Favarin, 2020;Sahramäki et al., 2017).
Hypothesis 7: A higher incidence in the local economy of transportation, storage and waste sectors is associated with a higher level of waste crimes; the contrary is expected for the agricultural sector.

DATA AND VARIABLES
The proxy used for waste crimes is the number of verified violations related to the waste cycle at the provincial level and for a nine-year period (102 provinces from 2009 to 2017). Data on the violations are collected by Legambiente, one of the most recognized Italian environmental non-governmental organizations (NGOs). The types of waste-related offenses included in the number of verified violations are all quite severe and punishable with imprisonment. 4 Box 1 in the supplemental data online reports the strengths and weaknesses of this proxy. Unlike other available data sources (e.g., ISTAT), Legambiente data enable us to conduct a panel data analysis at the provincial level, taking advantage of the country's substantial crosssectional heterogeneity. 5 Legambiente collects data from all Italian law enforcement agencies, providing a comprehensive source of information on the phenomenon of waste crime. Legambiente also collects data on the number of reported and arrested individuals employed to perform robustness checks. Despite their advantages, these data also reveal some weaknesses. A potential aggregation bias may arise from the infeasibility to differentiate between various types of waste-related offenses. However, in general, the proxy encompasses a collection of crimes that are similar in terms of offense type and statutory penalty and thus captures a consistent behavioural pattern. As mentioned above, the Legambiente measure, like any other measure of crime, is susceptible to endogeneity and measurement error, which we address by employing an appropriate estimation model. Figure 1 shows both the geographical distribution of the average number of violations during the time interval under scrutiny and its rate per 100,000 inhabitants. The highest number of violations, on average, occurred in the Southern and Central Italian provinces. At the same time, the geographical distribution of the rate per 100,000 inhabitants was more homogenous, albeit with lower rates in Northern Italy. Table 2 presents both the definitions and main descriptive statistics for those variables that were tested on the proxy of waste crimes.
Following the several hypotheses listed above, we break down the variables into two sets. The economic model includes a core set of structural variables to account for omitted variable bias and to test the first three hypotheses. In this baseline model, we include income per capita (gdp), which combines both the scale and human capital effects, population density (dens), which identifies the demographic pressures on waste management systems as well as the informal controls against illegal forms of waste disposal, and finally, crime inhibiting factors that capture enforcement. While no data on specific forms of deterrence was available, in accordance with Ehrlich (1996), the percentage of total cleared crimes to total crimes (clearance), and the overall probability of convictions given arrests (conviction) were deemed to be good proxies through which to capture general deterrence.
A second set of potential explanatory variables, which are tested individually and serve to test the remaining hypotheses, includes economic, criminal, and performance-driven factors. We account for the impact of the sectoral composition of the economy on waste crimes, distinguishing between the supply and the demand of illegal waste disposal services. With respect to the supply side, the relative prevalence of businesses devoted to transportation and storage (worktransp), or the waste sector (workwaste), was chosen to capture the economic drivers of greater levels of crime or, alternatively, to provide proximity and facilities through which to complete the waste cycle appropriately. 6 Apart from some other relevant sectors (e.g., construction and manufacturing), the demand is mainly driven by the agricultural sector, which is also particularly sensitive to environmental health and is expected to reduce waste crime (expagr). We also tested two important criminal factors that, according to the hypotheses mentioned above, are expected to positively influence waste crime rates: the presence of criminal associations (crimeasso) and the presence of mafia-type criminal associations (mafia). Finally, the model included two performance-driven factors. On the one hand, the importance of waste sorting in terms of the percentage of waste recycling (recycling) would capture the specific efficiency level achieved by the waste management system. This measure is primarily targeted at municipal waste management systems. On the other hand, the percentage of patients who migrated to other regions for hospitalization compared to the total number of patients in a province (efficiency) would measure the general level of efficiency of the public sector as a whole. This measure can be used as a reliable proxy for the levels of efficiency of the private sector in panel data analyses. 7 Thus, it can also capture the levels of efficiency in the treatment and disposal of hazardous waste in a competitive market, where forms of illegal dumping can be widespread.

MODEL SPECIFICATION AND ESTIMATION METHODS
The relationship between waste crime (wcrim) and its drivers can be described in the following equation: where wcrim i,t is the natural log of the number of reported violations per 100,000 inhabitants in province i in year t; 8 X i,t . is a vector of endogenous and exogenous variables; P i and T t . are province-specific fixed effects and time effects.
Finally, b is a vector of the parameters to be estimated, a is a constant term and 1 i,t is an idiosyncratic error term.
Various types of crime have been found to be persistent over time in the literature (Fajnzylber et al., 2002). Waste crime is no exception in this respect. Indeed, all the tests conducted in the next section on the presence of persistence suggest the adoption of a dynamic specification of the model. For this reason, equation (1) shows a partial adjustment form where the lag of the dependent variable, wcrim i,t−1 , is included in the set of the explanatory variables, and its parameter, r, is equal to 1 minus the adjustment parameter. Standard panel data methods, such as pooled-ordinary least squares (OLS) and fixed effects, are inconsistent because of the correlation between the lag of the dependent variable and the error component (Nickell, 1981). The resulting bias could be particularly significant, especially in a panel in which the time dimension is short and the number of units is large, which is the case in the present study. Removing unit-specific fixed effects via first differences does not resolve the issue because the first difference transformation introduces a correlation between the differenced lagged dependent variable and the differenced errors. A commonly adopted solution to this problem is to rely on a GMM estimator.
Different types of endogeneity can bias the estimates of equation (1) unless they are appropriately controlled for.
First, the estimates may be biased if we fail to control for those omitted variables that are correlated with the explanatory variables (Cornwell & Trumbull, 1994). For instance, Mustard (2003) demonstrated that the effect of arrest rates was downwardly biased if conviction Digging into waste: an analysis of waste crime in the Italian provinces probabilities and time served in prison were not included in the empirical exercise. We addressed this issue by including clearance and conviction rates to control for deterrence.
Second, when estimating the economic models of crime, another source of endogeneity is simultaneity. For instance, investigative units and courts may tend to increase their efforts to combat increasing crime rates. Consequently, not only can enforcement variables affect waste crime, but the opposite direction of causality can also occur. By instrumenting the deterrence variables with their lagged terms, the GMM approach would enable to depurate the effect of deterrence on crime from any feedback effects and, in turn, obtain more consistent estimates. A similar issue can affect mafia-related or nonmafia-related types of crime, which can be jointly determined with the dependent variable. As a result, mafia and crimeasso are also considered endogenous variables.
Third, endogeneity can also emerge because the denominator of a deterrence variable is highly correlated or corresponds to the numerator of the dependent variable. Our estimates do not appear to be seriously affected by this problem because the numerator of the dependent variable recorded the number of waste crimes, while the denominator of the deterrence variables referred to all crimes indistinctly.
Fourth, in a similar vein to other crime measures, waste crime suffers from a measurement error problem. Measurement error is dependent on both the reporting propensity and detection probability, which, in turn, are dependent on investigative effort and the type of crime. To illustrate the effect of measurement error on the parameter estimates, consider that the observed waste crime variable is equal to wcrim i,t + v i,t , where wcrim i,t is the actual (unobservable) rate of waste crimes and v i,t is the measurement error. In cross-sectional and static panel data models, it is wellknown that measurement error in the dependent variable that is not correlated with the explanatory variables leads to a loss of efficiency in the estimates but does not affect consistency. In the case of a linear dynamic panel data specification, such as the one reported in equation (1), the problem is more complicated yet still because the lag of the dependent variable appears as a regressor. The estimation strategies commonly adopted to address Nickell's bias, such as the difference-GMM (diff-GMM) or the system-GMM (sys-GMM) methods, do not provide consistent estimations in the presence of measurement error. However, Meijer et al. (2013) illustrated that, based on the assumption that the measurement error is not serially correlated, the GMM method is capable of providing consistent estimates as long as the instrument of wcrim i,t−1 is wcrim i,t−3 and/or earlier terms. If the measurement error is serially correlated, then this result still holds until the persistence in wcrim i,t is similar to the level of persistence in the measurement error. Unfortunately, the measurement error problem in equation (1) can be even more complex than that described thus far. This is because the measurement error in the dependent variable can also be correlated with other explanatory variables. More specifically, enforcement variables, such as clearance and conviction rates, are likely to impact the reporting propensity positively. For instance, victims may be more likely to report a crime when the probability of punishment is higher or if they receive insurance compensation, thereby producing a spurious positive correlation with crime rates (Ehrlich, 1996;Levitt, 1996). However, in the present investigation, this correlation is not expected to be high because the dependent variable refers to a specific type of crime, while the deterrence variables refer indistinctly to all crimes. Given these premises, the most suitable empirical approach appears to be the GMM estimator. Regarding the choice between sys-GMM and diff-GMM, the former has superior finite sample properties, and, in addition, the diff-GMM performs poorly when time series are persistent. Arellano and Bover (1995) demonstrate how additional instruments can be introduced to increase efficiency by incorporating the system's original equation in levels. In particular, variables in levels are instrumented with appropriate lags of their own first differences. The underlying assumption is that these differences are unrelated to unobserved effects, which Blundell and Bond (1998) show that, in turn, depends on a more precise assumption about initial conditions. A sufficient condition for this to hold would be the joint mean stationarity of the dependent variable and the independent variables. The difference-in-Hansen test provides such a test required for the validity of the instruments of the level variables. In addition, even if the lagged dependent variable does not show a very high parameter, the sys-GMM is still preferable to the diff-GMM because the parameter value obtained with the former is halfway between the upwardbiased pooled-OLS estimate and the downward-biased fixed-effect estimate. In contrast, the value obtained with the latter is very close to the fixed-effect estimate.
The one-step GMM estimator is preferable to the optimal two-step because simulation studies have suggested very modest efficiency gains from the two-step approach. Especially in small samples or in the presence of sizeable heterogeneity, these efficiency gains may not materialize; on the contrary, the bias of the two-step GMM estimator can increase with the number of overidentifying restrictions (Hwang & Sun, 2018;Judson & Owen, 1999). Finally, robust standard errors permitted controlling for heteroskedastic errors. Table 3 reports the main results from the estimation of the empirical model. A wide range of specifications is considered to check the stability of the parameters and the effect of various factors on waste crime rates. All results are obtained using the one-step sys-GMM method with robust standard errors and with measurement error robust instruments, that is, the predetermined variable (i.e., wcrim(−1)) and the other endogenous variables (i.e., clearance, conviction, mafia and crimeasso) were instrumented with their second and third lags, allowing the differenced error to be uncorrelated with these lags. Notice that numerous instruments may overfit instrumented variables, failing to expunge their endogenous components. As Leonida et al. (2013) pointed out, some studies suggest that large sets of lags may lead to severe bias. For comparative and robustness purposes, the supplemental data online provides the results from the standard model, where the lagged dependent variable and the other endogenous variables are instrumented with their first lag, and the results of the baseline methodology of the pooled-OLS that makes no use of instruments and theoretically induces an upward bias in the lagged dependent variable. All estimates in Table 3 satisfy the requirement that the overall number of instruments does not exceed the number of provinces, thus preventing overfitting. Moreover, the test of the overidentifying restrictions for the model employing the whole set of instruments and the first-differenced model only, as well as the difference-in-Hansen of both models, suggested that the instruments were valid, that is, that they were uncorrelated with the error term and that the excluded instruments were correctly omitted from the estimated equation.

RESULTS
The coefficient of the lag of the dependent variable, r, is always positive and statistically significant at one percent of the confidence level, with a value in the range of 0.48-0.58. These values suggest a moderate degree of persistence in waste crime rates in the Italian provinces. The strong statistical significance of r, the Wooldridge test results conducted on the residuals of the static panel data model, and the Arellano-Bond tests support the validity of a dynamic specification that includes the once-lagged dependent variable. 9 The parameter on gdp is negative and statistically significant across all specifications, except one. This finding is in accordance with theoretical expectations (Hypothesis 1): on the one hand, richer provinces should produce more waste (scale effect); on the other hand, richer provinces have higher average educational levels, which, in turn, contribute to higher rates of environmental awareness and a higher opportunity cost associated with committing a crime (human capital effect). 10 Dens also shows a negative coefficient, although rarely statistically significant across the specifications (Hypothesis 2). The presence of economies of scale in the waste sector in more populated areas and the concomitant reduction in legal costs, and the more diffuse informal controls against illegal forms of waste disposal may (more than) offset the crime-inducing effect played by demographic pressure on waste management systems. The coefficients of the enforcement variables (i.e., clearance and conviction) are never statistically significantly associated with waste crime rates. This result does not meet theoretical expectations (Hypothesis 3), but it is not novel, as some of the previous findings reported in Table 1 suggest. We believe that the lack of significance for clearance and conviction rates is probably due to the type of deterrence they convey, that is, general rather than a specific type. General deterrence probably does not significantly impact waste crimes, which could be more responsive to specific forms of deterrence (Almer & Goeschl, 2010Sigman, 1998). A concurrent explanation may pertain to the positive effects that enforcement variables produce on reporting (D'Amato et al., 2018), which our measurement correction strategy may not have fully resolved. 11 The inclusion of additional explanatory variables allowed for the measurement of a more comprehensive set of determinants of waste crime but, ultimately, did not alter the sign, magnitude, and statistical significance of the parameters in the baseline specification.
In accordance with our expectations (Hypothesis 4), the impact of the mafia on waste crime rates proves to be positive and statistically significant. It is worth noting here that we did not find evidence for the case of crimeasso, which implies that waste crime benefits from more structured and territory-based forms of criminal associations.
The quality of the services that waste-related legal activities provide to a community may also play a key role in encouraging individuals and businesses to become engaged in some form of illegal trade. According to our estimates, the general measure of bad performance of the local public sector (inefficiency), which, as mentioned above, can also be used as a proxy for the private sector inefficiency, is positively associated with waste crimes with highly significant parameters. Similarly, the measure of performance for capturing the specific efficiency level achieved by the waste management system (recycling) is negatively related to waste crimes, although with a weakly significant parameter. Hence, both these results confirm the Hypotheses 5 and 6 that an inefficient economy as a whole, and an inefficient waste management system specifically, increase the relative costs of the legal disposal of waste, thereby acting as serious crime-stimulating factors.
A final set of drivers refers to the sectoral composition of the economy and tests the last hypothesis (Hypothesis 7). The relative prevalence of businesses devoted to transportation and storage (worktransp) has a net positive effect on waste crime rates. This sector is crucial for waste disposal, and it may attract the appetite of infiltrated businesses into the legal economy, which may eventually supply legal and illegal services. This result may also disguise a dynamic of waste as secondary raw material, mainly regarding transboundary waste movements (e.g., the impact of China's recent plastic ban on the increase in arsons). Hence, although the proximity to this type of business can reduce the costs of legal disposal, we are not surprised to find a net positive effect on waste crime rates. A similar analysis applies to the relative prevalence of businesses in the waste sector (workwaste). However, the proximity of treatment and disposal facilities may induce lower costs for legal waste services that may compensate for the provision of illegal waste services to the point that the net effect is insignificant. Agriculture is also an environmentally sensitive sector in the local economy, being particularly susceptible to the pollution and environmental degradation associated with abandoned waste and waste-polluted areas. The negative impact, although weak, of the relative relevance of agriculture compared to other sectors (expagr), is in accordance with the expectation that this sector is susceptible to environmental health. 12 Finally, we tested some potential policy-driven structural breaks: in 2013, 2014 and 2015, for the inclusion of waste-related illicit combustion in the environmental code, the introduction of TARI (a new type of municipal waste tariff), and the addition of a few waste-related criminal offenses to the penal code, respectively. The results show no significant effect.

ROBUSTNESS CHECKS
To check the robustness of the results, in the first instance, we re-estimated the specifications using an alternative measure of waste crime, that is, the log of the number of individuals who were reported to the judicial authority, which would better capture the size of crimes. Notice that we opted for the number of violations rather than the number of reported individuals as our preferred estimate of waste crime for two reasons: (1) the former proxy is widely adopted throughout extant literature; and (2) the latter proxy may be more imprecise, insofar as the number of charged individuals were reported suspects who could be quickly acquitted directly by prosecutors and judges, without ever having to go to trial. We also tested both the total number of reported violations and the number of violations per 100,000 inhabitants without the log transformation. Across all the specifications, we find similar findings regarding the direction of the impact, although some parameters show lower statistical significance than the results shown in Table 3.
As another robustness check, we examined several variations in the enforcement variables. First, the relationship between enforcement and deterrence may not be contemporaneous but, rather, subject to time lags. To test whether enforcement variables exhibit a lagged effect on criminal behaviour, we re-estimated the specifications including alternatively (1) the first lag of both clearance and conviction rates, (2) the current clearance rate and the lagged conviction rate, and (3) the lagged clearance rate and the current conviction rate. Second, we considered the contribution of another enforcement variable, namely the arrest rate. As this variable is strongly correlated to the clearance rate, we did not include it in the baseline specification. However, we checked whether the results changed once Digging into waste: an analysis of waste crime in the Italian provinces we replaced the clearance rate with the arrest rate. Third, we employed a simple principal component analysis (PCA) to identify those factors that explain the major variability in clearance, arrest, and conviction rates. The PCA suggested the presence of two orthogonal factors, which we tested as enforcement variables. In all three of these cases, similarly to the results reported in Table 3, all the different variants of enforcement variables do not show a significant deterrence effect on waste crimes. Given that all the enforcement variables solely reflect general deterrence, we can thus conclude that waste crime rates are not particularly sensitive to general deterrence measures.
Finally, as with other crime measures, waste crime rates may show some spatial correlation such that crimes committed in one province can be dependent on crimes committed in neighbouring provinces. Although some authors have stressed the importance of exploring the spatial patterning of waste crime (e.g., D'Amato et al., 2015), investigations on waste crime and, to the best of our knowledge, environmental crime more broadly, have hitherto neglected this source of potential bias in their parameter estimates. Generally, crime data analyses have a long history of modelling spatial effects, especially vis-à-vis crimes against property and individuals. Thus, we explored if the analysis should have adopted a spatially informed regression model to account for spatial correlation.
We tested whether the dependent variable showed signs of spatial autocorrelation using the global Moran's I and the Geary's C spatial autocorrelation tests and considering different spatial weight matrices. We also examined the error terms of both the static and dynamic models to test for residual spatial dependence. To this purpose, we adopted Pesaran's cross-sectional dependence test. Finally, we also applied the Moran's I test on each year's residuals to test for residual spatial dependence in the errors. The results of all these tests are presented in the supplemental data online and clearly show that spatial correlation is not a severe concern in relation to the results presented in the previous section. 13 This evidence is not completely surprising, but it was worth investigating. Indeed, Mazzanti et al. (2012) tested for the spatial autocorrelation of waste generation and landfilled waste at the provincial level in Italy. Using contiguity and distance weight matrices, they found negligible spatial effects and demonstrated that the strong decentralization of waste management and policies in Italy was the primary reason for the lack of spatial dependence. Waste crime follows similar spatial dynamics, and, to a certain extent, this rationale may also apply to the present case.
However, the absence of spatial autocorrelation does not preclude the possibility of other cross-sectional sources of dependence in the data (Ertur & Musolesi, 2017). The common correlated effects (Pesaran, 2006) is a popular approach that relies on a multifactor error structure, as are developments such as its dynamic case and nonparametric specifications and tests (Gioldasis et al., 2021). The basic idea behind the estimation of common correlated effects is to eliminate the effects of unobservable common factors by using their observable counterparts, that is, cross-sectional averages of the explanatory variables. Even though common correlated effects and their evolution have received scant attention in the criminological literature to date, waste crime research should consider such more flexible and general strategies for detecting potential cross-sectional dependence and eliminating potential sources of bias in the estimates.

CONCLUSIONS
The present empirical investigation has contributed to expanding extant knowledge of the supply of waste crime by using a dynamic panel data model at the provincial level in Italy. The empirical methodology paid particular attention to endogeneity problems, measurement error of the dependent variable, and potential subregional spatial dependence and geo-clustering effects.
The evidence indicates that waste crime rates are persistent over time butin contradistinction to other types of crimeneither show significant spatial patterns nor noticeably respond to general deterrence measures. In other words, waste crime appears to be both a structural and atypical crime whose drivers are not straightforwardly predictable. However, the empirical investigation also revealed several notable findings. The human capital effect predominated over the scale effect, which could support the delinking hypothesis in the waste crime Kuznets curve (Germani et al., 2020). This result also corroborates the proposition that an increase in the level of human capital correspondingly increases sensitivity towards environmental issues. However, the development of virtuous social norms provides only a partial explanation in this respect; conversely, when economic interests are endangered due to the presence of illegal polluting waste, as is the case in the agricultural sector, then the local community can experience a crime reduction. Crime-stimulating factors and how they intersect account for the remaining variability in waste crime rates. An inefficient waste management system increases the legal costs of the waste cycle, in turn inducing firms and individuals to either do it 'in their own (illegal) way' or approach criminal fixers to do it on their behalf. The supply side encompasses a range of services, such as illegal transportation, storage and disposal of waste, which organized crime groups tend to infiltrate.
In terms of policy implications, the Italian government should establish a set of conventional and effective policies that would trigger four important crime-inhibiting factors: (1) increasing the environmental sensitivity of the general public through educational campaigns that aim to increase the moral cost associated with committing a waste crime, along with spreading neighborhood-watch practices, especially in areas where illicit waste activities constitute an important alternative source of income; (2) increasing the efficiency of all the components involved in the waste cycle specifically, as well as the waste-related public sector branches generally; (3) enhancing recycling and the circular economy to reduce the overall cost of waste treatment; and (4) increasing the level of the specific forms of deterrence against waste criminals by strengthening the training and the technologies of environmental task forces.
The policy of designing specific forms of deterrence warrants further discussion here. As was shown in this study, mafia-type organizations have a predominant role in the occurrence of waste crime. Consequently, enforcement should identify stable criminal groups within a territory that get involved with networks that operate at the blurred boundaries between legal and illegal markets. These illegal activities enable cost savings related to waste treatment/disposal. However, they also can create considerable profits by, among other things, trafficking Western waste to third countries (e.g., clothes to India, cars to Africa, plastic to China). The complexity of the organizational structures of these groups advocates for greater levels of national and international cooperation between different authorities to dismantle these networks.
Forms of centralization/decentralization of the municipal waste management system also deserve further discussion. From a crime-prevention perspective, a decentralized waste management system that empowers municipalities with additional responsibilities for waste management and treatment could be beneficial for improving local control of the waste management chain and enforcing legal behaviour. There are, however, some caveats. Small municipalities in Italy may struggle to manage waste treatment and disposal efficiently. According to Sarra et al. (2017), municipal waste management systems' organization and economic performance account for a sizable portion of local governments' total expenditure. Numerous small municipalities in Italy operate at a suboptimal scale. Thus, a hybrid, decentralized system could help reduce costs and improve waste management effectiveness, for example, by establishing multi-municipal waste management optimal territorial areas. This would eventually result in diminished economic incentives for providing illegal services and engaging in illegal behaviour. On the other hand, Abrate et al. (2014) asserted that Italian municipalities that rely on inter-municipal joint ventures to provide waste management services to their communities do not demonstrate cost savings. In their sample of 500 small, medium and large municipalities in Italy, they found no significant relationship between cost reduction and inter-municipal services. These findings might be due to the presence in the sample of municipalities of different sizes. In sum, we can hypothesize that small municipalities may benefit from joint management operations, while large municipalities may benefit from in-house arrangements, but additional research should be conducted in the future to better assess this issue.
As a concluding remark, future studies may seek to examine the potential relation between waste crimes and money laundering. For example, Spapens et al. (2018) outlined the important role of economic proceeds in the context of environmental crime, both in relation to corporations and criminal organizations' activities, whereas Di Pillo et al. (2022) described how legal waste management costs tend to be inflated in the regions where mafia groups are present in order to facilitate money laundering activities. Focusing on the investigation of financial flows, inflated costs, and money laundering activities, and their peculiarities at regional and subregional level, may be an effective strategy from an enforcement and compliance perspective.
trafficking of waste; and Art. 260 organized activities for the illicit trafficking of waste) and in the Penal Code (Art. 434 disaster for negligence; Art. 452-bis environmental pollution; Art. 452-ter death or injury as a consequence of the crime of environmental pollution; Art. 452quarter environmental disaster; Art. 452-sexties illegal transportation and abandonment of radioactive materials; Art. 452-septies obstruction of controls; and Art. 452-terdecies omitted recovery). The Italian criminal system distinguishes between contravvenzione (Arts 256, 257, 259) and delitto (all other articles). Both contravvenzione and delitto involve forms of convictions, arresto and reclusione, respectively, with subtle differences that the layman is generally unaware of. In other words, the Italian criminal law is extremely detailed, beyond the actual behavioural patterns associated with an offense such as waste crime. 5. Istat also provides data on waste crime at the province level that are also comprehensive of several waste-related violations (i.e., number of crimes after prosecutors' decisions) in a similar fashion to those provided by Legambiente, but excluding contravvenzioni. Unfortunately, Istat data are not particularly suitable for our analysis for two reasons. (1) The time span is narrower, from 2011 to 2017. According to our empirical strategy, this interval is too short to capture the dynamics of the within-variability accord. (2) Istat data at the provincial level show several zeros, jeopardizing the between variability. Nonetheless, both a correlation analysis with Istat data and a replication of the estimations on such a shorter period of time shows an acceptable comparability with Legambiente's proxy. 6. The variable workwaste does not include the business of urban waste collection, which is mainly tailored to the size of the population within each province. 7. The proxy efficiency captures an informed and authentic behaviour about the choice of healthcare institutions, which are the most important extensions of the public sector at the local level in Italy. 8. The log-transformation is required due to the nonnormal distribution of the variable. 9. The result of the Wooldridge test for autocorrelation in the panel data in the baseline specification in Table 3, which excluded the lagged dependent variable, rejected the null of no serial correlation (F(1,101) ¼ 8.276; Prob > F ¼ 0.0049). The test results were very similar across all other specifications. Thus, we can conclude that there is first-order autocorrelation in the errors, which is suggestive of some functional misspecification in the static panel. 10. We also tested education, separately. The effect was not statistically significant. However, we preferred to drop this variable because it showed substantial correlation with per capita gross domestic product (GDP) and little within variability. 11. In the estimation results obtained using the standard set of instruments (i.e., without measurement error correction) and presented in the supplemental data online, conviction becomes positively and significantly associated with waste crime rates. This result points to the potential effect of the measurement error on conviction rates, which is mostly removed when we apply the measurement error correction method. 12. We thoroughly tested the impact of sectoral composition on waste crime. Other economic sectorsconstruction among themdid not have a significant effect on waste crime rates. We also tested a model that included both the baseline and all statistically significant explanatory variables. The results of the full model substantially confirm those obtained with the previous specifications, albeit with less statistical significance for the parameters in general. The results further corroborate the weak evidence for recycling and expagr, and also show that mafia loses statistical significance while maintaining its sign. 13. Similar results were also obtained for the number of reported individuals for waste crime.