Evaluating the Eco-Efficiency of Municipal Solid Waste Management: Determinants, Paradoxes, and Trade-Offs

Abstract

The management of municipal solid waste (MSW) plays a crucial role in advancing sustainable development and circular economy goals across the European Union. In Italy, despite improvements in separate collection, significant regional disparities in MSW performance and costs persist. This study assesses the eco-efficiency of MSW services in 5516 Italian municipalities to uncover performance gaps and their underlying drivers. Eco-efficiency is measured using a Data Envelopment Analysis (DEA) model based on the Generalized Directional Distance Function (GDDF). This model incorporates per capita cost as an input, sorted waste as a desirable output, and residual waste as an undesirable output. A second-stage quantile regression is then utilized to explore how contextual factors influence eco-efficiency across various performance levels. The results reveal significant territorial disparities, with only 0.13% of municipalities achieving full eco-efficiency. Paradoxically, higher levels of separate waste collection—typically a policy goal—are associated with increased costs, especially in more efficient municipalities, suggesting a trade-off between environmental performance and economic sustainability. Similarly, population density negatively affects eco-efficiency but may facilitate economies of scale in collection systems. These findings highlight a tension between achieving optimal sorting rates and maintaining cost-effectiveness. Policy interventions should consider these trade-offs, prioritizing basic performance in lagging areas while promoting cost-control strategies in high-performing municipalities.

1. Introduction

Municipal solid waste (MSW) management is a significant environmental and logistical challenge for Italian municipalities [1]. It involves sorting, treating, and disposing of a complex mix of waste streams. Municipal solid waste includes organic materials, recyclables, hazardous substances, and emerging waste categories like electronic components and pharmaceutical residues [2,3,4]. Managing this diversity requires sophisticated strategies that place strain on existing infrastructure and regulations. Italy’s unique geography and socio-economic landscape contribute to its complexity. Increasing volumes of waste generation, changing consumption patterns, densely populated urban areas, historic city layouts, and different regional development levels create distinct challenges for waste collection, transportation, and processing systems [5,6].

As a member of the European Union (EU), Italy must align its waste policies with the ambitious targets set by the European Commission (EC), particularly those defined in the Waste Framework Directive (Directive 2008/98/EC) and the Circular Economy Action Plan [7,8]. These directives promote a waste hierarchy that prioritizes prevention, reuse, recycling, and energy recovery over landfilling. The latter is considered the least desirable option due to its adverse environmental and health impacts [9].

Italy has made significant progress in improving its municipal solid waste (MSW) systems, particularly in increasing the rate of separately collected waste. The national average has consistently risen, with several northern regions surpassing EU recycling targets. However, significant regional disparities still exist. Southern areas of the country continue to struggle to achieve efficient waste separation and recycling, with excessive reliance on landfilling remaining a critical issue [3]. These variations not only undermine national compliance with EU directives but also compromise environmental and public health objectives.

MSW management in Italy is highly decentralized, with municipalities playing a central role in organizing and delivering waste services. The quality and effectiveness of waste collection and treatment vary widely, depending on local administrative capacity, infrastructure, public awareness, and the performance of service providers. This decentralized structure makes municipal-level analysis particularly important for understanding the strengths and weaknesses of the national system [10]. Moreover, MSW management is increasingly viewed as a strategic component of Italy’s broader environmental policy framework. It contributes to the reduction in greenhouse gas emissions, conservation of resources, and the transition towards a circular economy.

A substantial body of literature has examined MSW management from both experimental and modeling perspectives [11,12,13]. Experimental investigations have focused on optimizing collection systems, recycling technologies, and treatment processes, often highlighting the technical potential to increase recovery rates or reduce environmental impacts [14,15,16]. Modeling studies, on the other hand, have applied tools such as life-cycle assessment, cost–benefit analysis, and frontier efficiency methods to evaluate system performance [17,18,19,20]. Over the past twenty years, an increasing number of scholars have used parametric or non-parametric frontier methods to evaluate the efficiency of municipal solid waste management system [21]. While these contributions have generated valuable insights, most studies tend to emphasize either environmental outcomes (e.g., recycling or diversion rates) or cost efficiency, with limited attempts to integrate both dimensions simultaneously [22,23,24,25,26]. Furthermore, comparative studies across municipalities or regions often highlight best practices but fall short of systematically addressing the role of contextual factors—such as demographic, geographic, and institutional variables—in shaping waste management efficiency [25,27,28]. These limitations reduce the policy relevance of existing research, as it remains unclear why certain systems outperform others and under what conditions similar results can be replicated.

Although Italy has made significant progress in separate waste collection, there are still wide territorial disparities in both the results achieved and the costs incurred by municipalities. These disparities lead to varying levels of economic efficiency and environmental performance of the service. Some municipalities are able to achieve high levels of separate collection at relatively low costs, while others, despite incurring high expenses, still show insufficient collection rates [3,10]. In addition to this, geographic and demographic factors play a significant role. Population density, municipal size, altitude, and territorial extension all greatly influence the organization and costs of the service [10].

Against this background, the present study addresses two key research gaps. Firstly, it develops an integrated eco-efficiency indicator that captures both the environmental and economic dimensions of municipal solid waste (MSW) management. This overcomes the limitations of conventional performance metrics that consider these aspects in isolation. Secondly, by utilizing a large-scale municipal dataset that covers over 5500 Italian municipalities, the study surpasses case-specific or regional analyses. It provides a comprehensive overview of system performance in a decentralized context. The use of quantile regression to explore the determinants of eco-efficiency further contributes to the novelty of this work, as it allows for the identification of heterogeneous effects across the efficiency distribution. Taken together, these contributions provide a more nuanced understanding of the drivers of waste management efficiency and deliver practical insights for policymakers seeking to design interventions tailored to different local contexts.

While indicators such as recycling rates are commonly used to evaluate the performance of waste systems, they often fail to capture the efficiency with which resources are used to achieve environmental outcomes. In this context, eco-efficiency—the ability to produce desirable outputs (e.g., high recycling rates, low landfill usage) with minimal inputs (e.g., financial resources, labor, environmental burden)—emerges as a more comprehensive and policy-relevant metric [26]. Unlike traditional indicators, eco-efficiency incorporates both environmental and economic dimensions, offering a nuanced perspective on service performance.

This paper utilizes Data Envelopment Analysis (DEA), a commonly used non-parametric method in operational research, to assess the eco-efficiency of MSW management services in 5516 Italian municipalities. DEA is ideal for this study as it can evaluate various inputs and outputs, determining the relative efficiency of comparable decision-making units. This study aims to construct a robust composite eco-efficiency indicator that reflects how efficiently municipalities convert resources into positive environmental outcomes by applying DEA to municipal-level data. The study then identifies a set of exogenous factors that influence the level of eco-efficiency in municipal solid waste management using a quantile regression approach.

2. Materials and Methods

2.1. Performance Measurement in MSW Management

Performance management in the MSW sector has become increasingly important in recent years, driven by environmental regulations, the pursuit of cost-efficiency in public services, and the societal shift toward circular economy models [29]. The concept of eco-efficiency has become increasingly relevant in the context of MSW management due to the dual imperative of maintaining cost-effectiveness while addressing environmental sustainability [30]. Originally introduced by the World Business Council for Sustainable Development (WBCSD), eco-efficiency refers to creating more value with less environmental impact, essentially combining economic efficiency with ecological responsibility [31]. In the field of MSW, eco-efficiency reflects the ability of municipal systems or service providers to optimize resources, reduce waste generation, and improve recycling, all while maintaining or reducing operational costs [32]. In practical terms, eco-efficiency in MSW services can be understood as the ability of a municipality to maximize desirable outcomes—such as the volume of recyclable materials collected and processed—while minimizing undesirable outcomes, such as residual or unsorted waste, and keeping service provision costs low [33]. This concept is especially significant as municipalities worldwide face increasing regulatory pressure to improve recycling rates, reduce landfill use, and transition toward circular economy models [34].

Initial investigations into MSW performance were primarily focused on economic efficiency. Researchers often used parametric techniques to model cost functions and determine the key factors influencing operational efficiency. For instance, studies such as those by Hirsch [35], Kitchen [36], and Antonioli and Filippini [37] utilized production or translog cost functions to examine the effects of service scale, waste volume, and other economic variables on the cost structures of MSW services. These models primarily considered waste as a homogenous product and did not distinguish between recyclable and non-recyclable fractions, thereby ignoring the environmental dimension.

By the early 2000s, with increasing environmental awareness and stricter regulatory frameworks in place (e.g., EU waste directives, national recycling targets), scholars started integrating environmental factors into their efficiency analyses. The work by Callan and Thomas [38], which separated cost functions for recycling and disposal, represented an early attempt to analyze the complementarities and trade-offs between economically and environmentally distinct waste streams. Similarly, studies by Dijkgraaf and Gradus [39] and Bel and Fageda [40] incorporated the percentage of recycled materials as explanatory variables in cost models, though findings regarding cost savings from recycling were inconclusive.

Although parametric methods provide robust frameworks for hypothesis testing and understanding production relationships, their reliance on functional form assumptions and single-output specifications limit their applicability in assessing eco-efficiency [41]. These models often fail to capture the multidimensionality of MSW services, as they generate multiple outputs (recyclables, residuals, service coverage) and involve diverse inputs (labor, capital, technology). Additionally, parametric approaches usually assume smooth production frontiers and homogeneous technologies, which are often unrealistic in fragmented and decentralized waste systems [42].

In response to these limitations, non-parametric techniques, especially DEA, have gained prominence. DEA allows for the construction of empirical production frontiers using observed data. It does not require the specification of a production function and can simultaneously handle multiple inputs and outputs. This makes it particularly suited for measuring the eco-efficiency of MSW systems, where both economic and environmental outputs are critical [43].

Early non-parametric studies such as Vilardell i Riera [44] and Worthington and Dollery [45] applied DEA to assess technical and scale efficiencies, focusing largely on operational metrics (e.g., amount of waste collected, labor, and equipment). However, these models often neglected environmental outputs, thus falling short of true eco-efficiency evaluations. Some scholars have emphasized the importance of evaluating both the cost and environmental dimensions of MSW services to avoid drawing misleading performance conclusions. Indeed, focusing solely on economic efficiency may penalize municipalities that invest in environmentally responsible waste management. Later contributions began to incorporate differentiated waste outputs, distinguishing between desirable (recyclable) and undesirable (residual) fractions. Sarra et al. [26] introduced a DEA model that explicitly integrated environmental outputs by considering sorted waste as a desirable output to be maximized and unsorted waste as an undesirable output to be minimized, while also aiming to minimize service costs. This represented a move towards more comprehensive performance evaluations.

Key methodological advancements included the use of Directional Distance Functions (DDFs), meta-frontier DEA Models, as well as bootstrapped and conditional DEA Models. Díaz-Villavicencio et al. [28] and Romano and Molinos-Senante [25] employed a DDF to simultaneously expand good outputs and contract bad outputs, thereby capturing the dual objective of eco-efficiency. To account for technological heterogeneity among service providers (e.g., public vs. private), Romano and Molinos-Senante [25] and lo Storto [23] adopted the meta-frontier DEA based approach. Benito-López et al. [27] and Guerrini et al. [46] employed bootstrapped and conditional DEA models to improve robustness and include exogenous factors (e.g., income levels, urban density, tourism pressure) that may affect efficiency scores, while lo Storto [10] implemented a two-stage DEA modeling approach to assess the interaction between operational efficiency and effectiveness, offering richer insights into performance drivers. These advances have enabled researchers to more accurately assess the joint economic and environmental efficiency of MSW systems and to identify the best practices for sustainable waste management.

Contextual factors are also increasingly recognized in the literature as key drivers of performance differences [28,47]. Geographic conditions, such as population density, municipal size, and topography, significantly impact service organization and cost structures. Several studies suggest that economies of scale and output density (i.e., more waste collected per capita or per area) can result in greater cost- and eco-efficiency, particularly in smaller municipalities [48,49]. However, the impact of recycling on cost remains ambiguous. Some studies report that recycling increases operational costs due to sorting, transportation, and processing requirements [50], while others find no significant cost effects or even cost savings under specific conditions [1,48,51,52,53]. In Italy, empirical research has shown that there are territorial disparities in performance and expenditure levels that result in unequal environmental outcomes and inefficiencies. Some municipalities can achieve high levels of separate collection at relatively low costs, while others fail to do so despite higher expenditures. This suggests that there are differences in institutional capacity, public participation, and service design [23,25,32,54,55,56].

Despite significant progress in the development of eco-efficiency measurement tools, there are still important research gaps that need to be addressed. Specifically, the literature has not thoroughly examined the structural trade-offs that municipalities may face when trying to balance cost-efficiency with ambitious environmental objectives. For example, efforts to increase recycling rates may result in higher operational costs, while cost-cutting measures can compromise environmental performance. These tensions are frequently overlooked in traditional efficiency assessments. Moreover, although some studies have begun to take into account contextual heterogeneity, there is still a lack of comprehensive analyses that explicitly model the complex and sometimes conflicting objectives inherent in MSW governance. This study aims to address these limitations by focusing on the trade-offs between economic and environmental dimensions of performance, providing new insights into the conditions under which eco-efficiency can be realistically achieved.

2.2. The Measurement of MSW Eco-Efficiency Using DEA

A central methodological challenge in applying DEA to eco-efficiency analysis lies in appropriately modeling undesirable outputs, such as landfilled waste or pollutant emissions. Unlike traditional (desirable) outputs that should be maximized, undesirable outputs must be minimized, yet they remain a part of the production process and cannot simply be omitted from analysis.

Several techniques have been developed to integrate undesirable outputs into DEA models [57]. The Seiford and Zhu [58] method addresses undesirable outputs by reclassifying them as inputs. This is based on the logic that these outputs consume resources or degrade performance. This approach retains the standard input-output DEA structure but may distort the production technology, as it implies a different role for these variables in the process. The Golany and Roll [59] transformation utilizes a non-linear inverse transformation, converting an undesirable output u into a desirable one by using 1/u. This method retains the variable as an output to be maximized but alters its interpretation: lower levels of the original undesirable variable result in higher transformed output values. While conceptually intuitive and mathematically convenient, this technique may introduce scaling issues, particularly when values approach zero, and it can obscure the interpretation of results. The DDF, developed by Chung et al. [60], provides a more flexible and theoretically consistent framework. Unlike radial DEA models, which assume proportional expansion or contraction of all outputs or inputs, the DDF allows for the simultaneous contraction of undesirable outputs and expansion of desirable ones in a specified direction vector. This non-radial approach is particularly well-suited for environmental applications, where the goal is often to increase recycling (desirable) while decreasing landfilling (undesirable). The DDF-DEA method has several key advantages [61]: (a) it does not require arbitrary transformations or reclassification of outputs as inputs; (b) it allows for custom specification of the direction vector, reflecting policy or managerial priorities; (c) it provides consistent handling of multiple desirable and undesirable outputs in a single model; (d) it is grounded in environmental production theory, making it especially appropriate for assessing eco-efficiency in regulated sectors such as waste management. Given these advantages, the DDF-DEA has become a preferred approach in recent eco-efficiency studies across various sectors, including energy, agriculture, and waste. In the MSW context, it provides a conceptually sound and empirically practical means of assessing how well municipalities balance economic inputs with environmental goals.

In this study, the DDF-DEA is used to assess the eco-efficiency of Italian municipalities in handling the MSW service. This method aligns with the EU waste hierarchy, which emphasizes recycling over landfilling, and considers various aspects of service performance such as economic, operational, and environmental factors. This study specifically measures the MSW eco-efficiency by applying a Generalized Directional Distance Function (GDDF), which considers both desirable and undesirable outputs in waste management [62].

Assuming there are N decision-making units (DMUs), each uses a vector of inputs $x=\left(x_1,\dots ,x_m\right)\in {\Re }_{+}^m$ to produce both desirable outputs $y=\left(y_1,\dots ,y_p\right)\in {\Re }_{+}^p$ and undesirable outputs $z=\left(z_1,\dots ,z_q\right)\in {\Re }_{+}^q$.

The GDDF enables simultaneous reduction in input, expansion of desirable output, and contraction of undesirable output. The production technology T is represented by the following directional distance function:

$${\overrightarrow{D}}_T\left(x,y,z;g_x,g_y,g_z\right)=\mathit{sup}\left\{\beta :\left(x-\beta g_x,y+\beta g_y,z+\beta g_z\right)\right.\in T\left(x,y,z\right)$$

(1)

where β and $g=\left(g_x,g_y,g_z\right)$, with g_x, g_y > 0 and g_z < 0, indicate the inefficiency and the direction vector of the inputs, desirable and undesirable outputs, respectively. In the study, it is assumed that the municipality goals are to minimize inputs, maximize desirable output, and minimize undesirable output, and consequently, the direction vectors used are g_x = 1, g_y = 1, g_z = −1. The GDDF measures eco-efficiency while ensuring that the inefficiency measure β lies within the interval (0, 1), preserving unit invariance regardless of the direction vectors measurement. The efficiency score is measured as follows

$$\mathit{min}{\displaystyle \frac{1-\frac{1}{m}{\displaystyle \sum _{i=1}^m\frac{\beta g_i}{x_{io}}}}{1+\frac{1}{p+q}\left({\displaystyle \sum _{r=1}^p\frac{\beta g_r}{y_{ro}}+{\displaystyle \sum _{s=1}^q\frac{\beta g_s}{z_{so}}}}\right)}}$$

(2)

where m, p and q, respectively, denote the number of inputs, and the number of good and bad outputs. The ratios $\beta g_i/x_{io}$, $\beta g_s/z_{so}$ and $\beta g_r/y_{ro}$ indicate the proportions of the input and bad output decrease, and the proportion of good output increase, respectively.

Equation (2) is crucial to the study as it operationally defines the eco-efficiency score. The numerator represents a decrease in inputs and undesirable outputs, while the denominator shows an increase in desirable outputs. A score closer to one indicates that a municipality can recycle more (good output) and produce less residual waste (bad output) with a lower expenditure. Lower scores suggest inefficiencies in financial resources use or waste separation. Equation (2) bridges the theoretical GDDF framework to a practical performance measure that is applicable to policy and can be compared across municipalities.

The model reflects the context of Italian municipal waste management, where an increase in sorted waste collection, and a decrease in unsorted waste, have been observed over the past decade [2]. The assumption of free disposability of undesirable outputs is in line with the idea that reducing unsorted waste can happen without affecting the production of sorted waste, particularly when households participate in separating waste at source [47].

Variable returns to scale (VRS) have been assumed in the analysis. This choice is justified for several interrelated reasons. Italian municipalities vary significantly in terms of population size, population density, geographical characteristics (e.g., urban vs. rural), governance models and local regulations. Such heterogeneity implies that not all municipalities can scale their operations proportionally, making the assumption of constant returns to scale (CRS) unrealistic. In the municipal waste sector, scale inefficiencies are likely to exist due to economies or diseconomies of scale (e.g., small towns may face high unit costs, while very large cities may encounter congestion or coordination issues) and fixed costs (equipment, facilities) not being evenly distributed across municipalities.

In the development of the eco-efficiency indicator, energy recovery from waste (such as electricity or heat generation) was not considered as an output variable. This methodological choice was justified by several considerations that are specific to the Italian context. Firstly, waste-to-energy is positioned lower in the EU waste hierarchy, which prioritizes prevention, reuse, and material recovery [63]. In Italy waste-to-energy plays a relatively minor role compared to recycling and landfilling. On average, only 1.2% of municipal solid waste per year was utilized for energy recovery during the period from 2018 to 2022 [2]. Secondly, incineration facilities with energy recovery are mainly located in Northern Italy, while many municipalities in Central and Southern regions lack access to such infrastructure [2,3]. Incorporating energy generation would skew eco-efficiency scores in favor of municipalities located in areas served by waste-to-energy plants. This would not accurately reflect variations in local waste management performance. Thirdly, energy recovery is usually managed at the supra-municipal or regional level, beyond the direct control of municipalities. According to Legislative Decree 152/2006, the approval of new waste management facility projects and the authorization of modifications to existing plants are under the jurisdiction of regional administrations rather than municipalities [64]. Therefore, its inclusion would create a mismatch between the unit of analysis (municipality-level services) and the level at which energy recovery occurs. Finally, administrative data provides standardized information on sorted and residual waste at the municipal level. However, reliably allocating energy outputs from incineration plants to specific municipalities is not feasible [2]. For these reasons, the eco-efficiency indicator in this study focuses on waste flows and costs directly attributable to municipal services. This ensures comparability across the national sample and consistency with the circular economy objectives emphasized by Italian and EU policy frameworks.

In summary, the estimation of the eco-efficiency indicator relies on the following assumptions, which are consistent with the context of municipal solid waste management in Italy: (a) input and output orientation in the linear programming model formulation as municipalities aim to minimize inputs, maximize desirable outputs, and minimize undesirable outputs (direction vectors g_x = 1, g_y = 1, g_z = −1); (b) free disposability of undesirable outputs, as unsorted waste can be reduced independently of sorted waste production, reflecting household source separation practices; (c) variable returns to scale due to the marked heterogeneity in size, density, and institutional arrangements across Italian municipalities, making constant returns to scale unrealistic; (d) exclusion of energy output quantities from the eco-efficiency indicator measurement, since the volume of waste processed for energy production is irrelevant to this assessment of Italian municipalities; (e) aggregation of all source-separated waste fractions into a single output identified as differentiated waste. In this study, the following waste fractions were considered as desirable outputs: organic (food) and green waste, paper and cardboard, lightweight multi-material, glass, wood and textiles, waste electrical and electronic equipment, bulky waste, construction/demolition waste and street sweeping waste.

2.3. Second-Stage Analysis Using Quantile Regression

To explore the factors that explain differences in efficiency across municipalities, a second-stage analysis was conducted. Eco-efficiency scores, obtained through DEA, were regressed on a set of explanatory variables. While traditional approaches such as Tobit or truncated regression models are commonly used in this context, they focus on estimating the conditional mean of eco-efficiency, potentially overlooking important variations across the distribution of scores.

Recently, quantile regression has emerged as a flexible and informative alternative for second-stage DEA. While standard regression methods estimate the average effect of explanatory variables on the dependent variable, quantile regression enables the estimation of effects at different points (quantiles) of the conditional distribution—such as the median (50th percentile), lower tail (e.g., 25th percentile), or upper tail (e.g., 75th percentile) [65]. This makes it particularly valuable for analyzing eco-efficiency scores, which are bounded and often exhibit skewed or clustered distributions due to the presence of multiple fully efficient units. Thus, quantile regression provides a more nuanced understanding of how explanatory variables influence eco-efficiency across its full distribution, particularly in the presence of bounded, skewed, or clustered data, which is typical in DEA applications, and the relationship between eco-efficiency and its determinants is suspected to be non-linear or non-uniform. One of the main advantages of quantile regression is its ability to capture heterogeneity in the effects of covariates [66]. For example, the factors influencing low-performing municipalities may differ significantly from those affecting high-performing ones. Quantile regression estimates the impact of explanatory variables at different quantiles, uncovering these distinct effects and allowing for more targeted and effective policy interventions. In the MSW sector, variables such as population density or inter-municipal cooperation might have a stronger positive influence on municipalities at the lower end of the efficiency spectrum. However, their effect may diminish or even reverse at higher efficiency levels due to diminishing returns. Several empirical studies have successfully applied quantile regression in environmental and waste management contexts. For instance, Agovino et al. [67] implemented quantile regression to evaluate the magnitude of the socioeconomic and institutional factors on the landfill disposal rate in Italy. Boubellouta and Kusch-Brandt [68] applied panel quantile regression to assess how the e-waste recycling rate in 30 European countries is influenced by economic growth, population, population density, energy intensity, energy efficiency, credit to the private sector, and the amount of e-waste collected.

Formally, the quantile regression model for the τ-th quantile (where 0 < τ < 1) of the eco-efficiency score θ_i given covariates x_i is specified as

$$Q_{\tau }\left({\theta }_i\left|x_i\right.\right)=x_i^{\prime }{\beta }_{\tau },\tau \in \left(0,1\right)$$

(3)

where $Q_{\tau }\left({\theta }_i\left|x_i\right.\right)$ is the conditional quantile τ of the eco-efficiency score ${\theta }_i$ for municipality i, $x_i$ is a vector of explanatory variables representing socioeconomic, demographic, and operational characteristics, ${\beta }_{\tau }$ is a vector of quantile-specific parameters to be estimated.

Unlike ordinary least squares (OLS), which minimizes the sum of squared residuals to estimate the mean, quantile regression minimizes a weighted sum of absolute residuals to estimate specific quantiles. Vector ${\beta }_{\tau }$ is estimated by solving an asymmetric loss function that gives different weights to over- and under-predictions based on the chosen quantile. The estimated coefficients ${\beta }_{\tau }$ show how the predictors affect the τ-th quantile of the dependent variable.

2.4. Sample

The empirical analysis in this study is based on a dataset comprising 5516 Italian municipalities, covering the entire national territory. The data pertain to the fiscal year 2022 and were collected to investigate the eco-efficiency of the MSW service across different regions of the country.

To ensure a comprehensive territorial representation, the sample includes municipalities from all three macro-areas of Italy: Northern, Central, and Southern regions (including the islands). Specifically, the sample is composed of 3271 municipalities located in Northern Italy, 706 municipalities in Central Italy, and 1539 municipalities in Southern Italy and the islands (Figure 1). This geographic distribution enables a meaningful comparison of eco-efficiency performance across diverse environmental, socio-economic, and institutional settings.

Data were obtained from publicly available administrative and statistical sources, including national databases such as the Istituto Nazionale di Statistica (ISTAT) and the Catasto Rifiuti managed by ISPRA (Istituto Superiore per la Protezione e la Ricerca Ambientale) [69,70]. Only municipalities for which complete and reliable data on MSW service inputs and outputs were available for the year 2022 were included in the final sample. Municipalities with missing, inconsistent, or outlier values in key variables were excluded to ensure the accuracy and comparability of the analysis. The largest number of municipalities is located in Northern Italy (59.30%), while the smallest number is in the Central region (12.80%).

2.5. Variables

2.5.1. Variables Used in DEA Model Specification

The variables used in the model were selected based on theoretical foundations and empirical evidence from literature. The DDF-DEA model specification includes three core variables as follows:

• per capita sorted waste (pcSW). This variable measures the amount of recyclable and recoverable waste generated per inhabitant, expressed in kilograms per year [23,46]. It serves as the desirable output in the model, reflecting the positive outcome of waste management efforts aimed at diverting waste from landfills and incineration towards recycling and material recovery. Higher values of pcSW indicate more effective sorting practices and greater alignment with environmental sustainability goals.
• per capita residual waste (pcRW) captures the amount of unsorted or residual waste produced per inhabitant, also expressed in kilograms per year [23,46]. It represents the undesirable output in the DEA framework, as it corresponds to waste that is typically sent to landfill or incineration, generating negative environmental externalities such as greenhouse gas emissions, soil and water contamination, and resource loss. Minimizing pcRW is therefore central to improving eco-efficiency in MSW services.
• per capita total waste management cost (pcTWC). This variable is used as the input in the DEA model and reflects the total cost of waste management services per inhabitant, measured in euros per year [27,32,54]. It encompasses the full range of operational expenses related to the collection, transport, and treatment of municipal waste. By evaluating how well municipalities convert this input into higher levels of desirable output (pcSW) and lower levels of undesirable output (pcRW), the model provides an integrated measure of cost-efficiency and environmental performance.

2.5.2. Exogeneous Variables

In the second stage of analysis, the following variables were used as independent (exogenous) variables in quantile regression to identify factors influencing eco-efficiency outcomes:

• sorting rate (SR), expressed as a percentage. This variable indicates the share of total municipal waste that is sorted for recycling and recovery [71]. A higher sorting rate is generally associated with more environmentally sustainable waste management practices and may reflect greater public environmental concern, stronger institutional incentives, or more efficient collection systems.
• per capita total waste generated (pcTW). This variable measures the total quantity of municipal solid waste produced per inhabitant, expressed in kilograms per year [24,72,73]. It reflects the consumption behavior of residents and the intensity of waste generation, which can influence both the scale and efficiency of local waste management services.
• population (POP). This variable measures the total number of residents in each municipality [25,52]. Larger populations may benefit from economies of scale in waste management operations, but they may also pose greater logistical and infrastructural challenges.
• municipality area (AREA) indicates the total land area of the municipality, measured in square kilometers [23,26]. Larger geographic areas may face higher service delivery costs due to the need for extended collection routes, particularly in sparsely populated or rural municipalities.
• population density (DENS). Calculated as the number of inhabitants per square kilometer, this variable captures the spatial concentration of the population [23,25,47]. Higher population density can facilitate more efficient waste collection due to shorter travel distances and greater service coverage per unit of input.
• municipality height above sea level (HEIGHT). Measured in meters, this variable represents the elevation of the municipality. It may serve as a proxy for geographic or topographic constraints that could affect transportation logistics, fuel consumption, or infrastructure deployment.
• per capita sorted waste cost (pcSWC). This variable measures the cost per inhabitant associated with the collection, transport, and processing of sorted waste fractions [25,27,47]. It reflects both operational costs and the effectiveness of systems aimed at recycling and material recovery.
• per capita residual waste cost (pcRWC) is used to measure the cost per inhabitant for the management of residual (unsorted) waste, which typically includes landfill disposal or incineration [25,27,47]. High residual waste costs may signal inefficiencies or limited access to advanced treatment facilities, while lower costs could indicate better performance or cost-effective service models.

All variables were normalized using Z-score transformation prior to analysis.

Table 1. Variables used in the analysis (I and II stages).

Stage of Analysis	Variable Name (Abbreviation)	Description	Type	Unit	Mean	Std.dev.
I	pcSW	per capita sorted waste	output (good)	kg/inhabitant/year	318.79	129.10
	pcRW	per capita residual waste	output (bad)	kg/inhabitant/year	148.05	107.63
	pcTWC	per capita total cost of MSW service	input	€/inhabitant/year	171.01	88.21
II	SR	sorting rate	exogenous	percentage/year	68.63%	15.43%
	pcTW	per capita total waste generated	exogenous	kg/inhabitant/year	469.32	174.71
	POP	population	exogenous	Inhabitants/year	8813.62	47,714.92
	AREA	municipality area	exogenous	km2	38.97	53.81
	DENS	population density	exogenous	inhabitants/km2	349.80	702.25
	HEIGHT	municipality height above sea level	exogenous	m	425.85	426.82
	pcSWC	per capita sorted waste cost	exogenous	€/inhabitant/year	68.38	37.41
	pcRWC	per capita residual waste cost	exogenous	€/inhabitant/year	41.76	35.45

provides summary statistics for the variables used in the analysis.

3. Results and Discussion

3.1. First Stage DEA

The first stage of the analysis involved measuring the eco-efficiency of MSW management services using a DDF-DEA model. This model allowed for the simultaneous consideration of one input (per capita total waste management cost), one desirable output (per capita sorted waste), and one undesirable output (per capita residual waste), under the assumption of variable returns to scale and a non-oriented specification.

The analysis revealed that only a small subset of municipalities achieved full eco-efficiency. Specifically, out of the 5516 municipalities included in the sample, only 7 municipalities (0.13%) were identified as 100% eco-efficient. These municipalities are unevenly distributed across the national territory, with 6 located in Northern Italy and 1 in Southern Italy. No fully efficient municipalities were recorded in Central Italy, highlighting potential regional disparities in waste management performance. The results are consistent with those reported in previous studies, which collectively indicate the generally poor performance of the MSW sector in terms of eco-efficiency. Molinos-Senante et al. [24], analyzing a sample of 140 Chilean municipalities, estimated an average eco-efficiency score of 0.332. Using a larger dataset of 298 municipal solid waste service providers, the same scholars found that only seven out of 298 providers achieved an eco-efficiency score of 100%, representing the industry’s best-performing frontier. Similarly, Daraio et al. [54], focusing on a sample of 89 municipalities within the optimal territorial area of the “Metropolitan City of Rome Capital,” found that only 10 municipalities demonstrated full efficiency. Romano and Molinos-Senante [25] also reported low eco-efficiency scores in a sample of 225 municipalities in Tuscany (Italy), with average values ranging from 0.523 for mixed operators to 0.697 for public operators.

Descriptive statistics of the eco-efficiency scores across Italy’s three macro-areas—North, Center, and South (including islands)—further illustrate these differences. As shown in Figure 2, Northern municipalities on average exhibit higher eco-efficiency levels, accompanied by lower variability, suggesting both better and more consistent performance. In contrast, Southern municipalities display lower average efficiency scores and greater dispersion, indicating significant room for improvement and heterogeneity in service quality and environmental outcomes.

A non-parametric comparison of the distribution of eco-efficiency scores was conducted using kernel density estimation, as shown in Figure 3. The density curves indicate a noticeable shift to the right for Northern municipalities, confirming their higher concentration of efficient or near-efficient cases. Central Italy’s density is more evenly distributed, whereas Southern Italy displays a left-skewed distribution, with a notable proportion of municipalities showing low levels of eco-efficiency.

These distributional patterns likely reflect structural and institutional differences among Italian macro-areas, and underscore the influence of contextual factors, such as economic resources, institutional capacity, and environmental awareness, on municipal performance in waste management. In Central Italy, the more even distribution of eco-efficiency scores might be associated with intermediate levels of socio-economic development and institutional capacity, resulting in municipalities performing at both high and low levels without strong polarization. On the other hand, the left-skewed distribution observed in Southern Italy suggests the presence of systemic barriers—such as limited fiscal capacity, weaker institutional arrangements, and infrastructural gaps—that constrain performance in a larger number of municipalities. At the same time, a subset of Southern municipalities achieves relatively high levels of eco-efficiency, contributing to the observed heterogeneity. This interpretation is consistent with prior studies on regional disparities in public service delivery and waste management performance in Italy.

The second-stage regression analysis will further investigate these associations.

3.2. Second-Stage DEA

3.2.1. Eco-Efficiency Grouping Across the Distribution

The sample was divided into six groups based on the 0.05, 0.25, 0.50, 0.75, and 0.95 quantiles to highlight differences in waste management eco-efficiency across the distribution. This approach helps identify patterns and disparities between low, medium, and high levels of eco-efficiency.

Table 2. Descriptive statistics of second-stage analysis variables by quantile group.

	Main Statistics					Main Statistics
Variable	Mean	Median	St.dev.	Max	Min	Mean	Median	St.dev.	Max	Min
	Group 1					Group 2
eco-efficiency	0.100	0.109	0.031	0.132	0.000	0.188	0.191	0.029	0.233	0.132
SR	32.66%	34.51%	14.12%	56.56%	0.00%	54.99%	57.87%	11.66%	72.45%	12.63%
POP	23,661	1110	174,179	2,748,109	42	12,788	2459	55,472	1,354,196	37
AREA	62.50	37.37	96.68	1287.36	1.75	47.26	24.66	63.08	547.04	0.67
DENS	335.40	29.59	953.55	6794.65	0.80	415.36	87.87	1055.21	11,927.03	1.04
HEIGHT	655.96	609.20	496.37	2386.46	2.29	509.22	370.23	466.18	2590.76	0.36
pcSWC	64.09	58.91	46.23	224.92	0.00	70.87	68.05	39.54	372.56	0.00
pcRWC	103.23	92.46	51.21	377.96	11.63	59.57	52.76	36.55	338.04	0.00
	Group 3					Group 4
eco-efficiency	0.275	0.276	0.023	0.314	0.233	0.351	0.349	0.023	0.393	0.314
SR	67.49%	70.01%	8.62%	78.33%	20.64%	75.13%	77.10%	7.02%	84.92%	19.35%
POP	7915	2884	16,099	250,369	59	6320	3104	11,103	161,748	86
AREA	40.48	23.66	54.03	653.82	1.44	33.01	19.98	41.74	449.51	1.29
DENS	299.24	121.88	542.54	6742.63	1.06	348.22	148.23	585.25	7687.15	1.56
HEIGHT	426.42	306.11	398.39	2316.19	0.46	363.03	258.37	379.35	2427.30	0.55
pcSWC	68.46	64.76	32.36	333.77	0.00	66.27	61.75	33.38	392.84	0.00
pcRWC	39.64	35.16	25.95	306.88	0.00	32.00	26.47	24.65	253.55	0.00
	Group 5					Group 6
eco-efficiency	0.448	0.443	0.037	0.531	0.393	0.629	0.592	0.107	1.000	0.531
SR	80.76%	82.72%	7.45%	92.81%	24.64%	83.85%	87.06%	9.91%	99.96%	37.98%
POP	6244	3177	12,401	196,764	81	5262	3673	8758	129,340	170
AREA	31.80	18.51	40.11	530.18	1.06	33.18	21.45	41.30	405.16	1.83
DENS	357.55	157.45	520.82	3530.07	4.80	330.79	159.01	451.06	2512.13	5.82
HEIGHT	365.85	238.56	405.90	2361.12	0.56	412.91	229.49	493.88	2494.12	1.53
pcSWC	69.41	62.37	37.25	406.33	0.00	68.68	52.92	56.68	422.12	0.00
pcRWC	27.33	20.00	27.96	539.62	0.00	25.91	18.13	31.82	371.93	1.94

displays the main statistics of the quantile groups. As expected, eco-efficiency scores increase steadily across groups, starting from low values in Group 1 (the mean is equal to 0.10) to higher levels in Group 6 (mean is 0.63). This pattern confirms the accuracy of the quantile-based classification. A clear trend emerges for the sorting rate (SR), which increases proportionally with eco-efficiency, from only 33% in Group 1 to nearly 84% in Group 6. This finding emphasizes the importance of waste sorting in driving higher eco-efficiency.

Demographic and geographic variables highlight additional differences. Less efficient municipalities (Groups 1–2) tend to have larger populations, wider areas, and greater variability in density, while more efficient groups (Groups 5–6) are generally smaller municipalities. This suggests that smaller municipalities tend to perform better in terms of eco-efficiency. Large cities face greater complexity and heterogeneity in waste management, which can lead to reduced efficiency. However, the trend of population density across groups is less clear than population or area. Group 1 shows a relatively low median density (around 30 inhabitants/km²) but with extreme variability (max > 6700). Groups 3–6 have higher medians (between 120 and 160 inhabitants/km²), but still with large ranges. Overall, medium-density municipalities appear to be more eco-efficient, while very sparse and very dense contexts face challenges.

Altitude (HEIGHT) decreases as efficiency improves, with notable variability, suggesting that topographical factors may partially impact waste management performance as waste collection and transport become more complex and costly, reducing efficiency. Group 1 (low eco-efficiency) has the highest average altitude (around 656 m), while groups 4 and 5 have averages between 363 m and 366 m. Altitude slightly increases in Group 6 (around 413 m), but it is still lower than in the least eco-efficient groups. This suggests that municipalities at higher altitudes may have developed more efficient systems to overcome logistical challenges. These municipalities also have smaller populations and areas. Despite being at higher altitudes, the smaller scale of services may reduce complexity, offsetting geographical disadvantages.

Furthermore, higher eco-efficiency despite altitude could indicate stronger local environmental policies or community participation in waste sorting, compensating for geographic challenges. The cost structure reveals a significant shift as per capita residual waste cost (pcRWC) decreases sharply with higher eco-efficiency, from €103 in Group 1 to about €26 in Groups 5–6. Conversely, per capita sorted waste cost (pcSWC) remains relatively stable across groups, indicating that efficiency gains are primarily associated with reducing the burden of residual waste.

After presenting the descriptive statistics, the pairwise correlations between eco-efficiency and the explanatory variables were also analyzed. To maintain readability and prevent overwhelming details in the main body of the text, the complete set of correlation matrices by quantile group can be found in the Appendix A (

Table A1. Correlations among variables, quantile-Group 1.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	0.726 (0.000)	1
(3)	pcTW	0.151 (0.426)	0.185 (0.074)	1
(4)	pcSWC	0.298 (0.000)	0.471 (0.000)	0.462 (0.000)	1
(5)	pcRWC	−0.392 (0.000)	−0.309 (0.000)	0.409 (0.000)	−0.173 (0.140)	1
(6)	POP	0.018 (1.000)	0.066 (1.000)	−0.007 (1.000)	0.022 (1.000)	−0.050 (1.000)	1
(7)	AREA	0.051 (1.000)	0.078 (1.000)	0.082 (1.000)	0.106 (1.000)	−0.078 (1.000)	0.798 (0.000)	1
(8)	DENS	0.111 (1.000)	0.216 (0.011)	−0.094 (1.000)	0.121 (1.000)	−0.086 (1.000)	0.237 (0.003)	0.029 (1.000)	1
(9)	HEIGHT	−0.003 (1.000)	−0.057 (1.000)	0.285 (0.000)	0.033 (1.000)	0.153 (0.401)	−0.143 (0.637)	−0.093 (1.000)	−0.358 (0.000)	1

,

Table A2. Correlations among variables, quantile-Group 2.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	0.496 (0.000)	1
(3)	pcTW	−0.138 (0.000)	−0.068 (0.846)	1
(4)	pcSWC	0.029 (1.000)	0.249 (0.000)	0.539 (0.000)	1
(5)	pcRWC	−0.245 (0.000)	−0.299 (0.000)	0.366 (0.000)	0.003 (1.000)	1
(6)	POP	−0.050 (1.000)	0.077 (0.391)	−0.005 (1.000)	0.031 (1.000)	−0.072 (0.580)	1
(7)	AREA	−0.060 (1.000)	0.038 (1.000)	0.109 (0.011)	0.156 (0.000)	−0.037 (1.000)	0.232 (0.000)	1
(8)	DENS	−0.049 (1.000)	0.073 (0.542)	−0.079 (0.325)	0.025 (1.000)	−0.065 (1.000)	0.402 (0.000)	−0.103 (0.022)	1
(9)	HEIGHT	−0.101 (0.028)	−0.148 (0.000)	0.227 (0.000)	0.081 (0.257)	0.125 (0.011)	−0.139 (0.000)	0.082 (0.228)	−0.271 (0.000)	1

,

Table A3. Correlations among variables, quantile-Group 3.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	0.279 (0.000)	1
(3)	pcTW	−0.104 (0.004)	0.040	1
(4)	pcSWC	−0.078 (0.133)	0.181 (1.000)	0.483 (0.000)	1
(5)	pcRWC	−0.098 (0.010)	−0.224 (0.000)	0.200 (0.000)	0.061 (0.833)	1
(6)	POP	−0.014 (1.000)	0.093 (0.019)	0.118 (0.000)	0.126 (0.000)	−0.022 (1.000)	1
(7)	AREA	−0.041 (1.000)	0.016 (1.000)	0.085 (0.056)	0.205 (0.000)	0.041 (1.000)	0.497 (0.000)	1
(8)	DENS	−0.007 (1.000)	0.132 (0.000)	0.046 (1.000)	0.011 (1.000)	−0.083 (0.077)	0.443 (0.000)	−0.143 (0.000)	1
(9)	HEIGHT	−0.064 (0.622)	−0.235 (0.000)	0.000 (1.000)	0.020 (1.000)	0.131 (0.000)	−0.215 (0.000)	0.074 (0.215)	−0.254 (0.000)	1

,

Table A4. Correlations among variables, quantile-Group 4.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	0.293 (0.000)	1
(3)	pcTW	−0.038 (1.000)	0.034 (1.000)	1
(4)	pcSWC	−0.017 (1.000)	0.066 (1.000)	0.538 (0.000)	1
(5)	pcRWC	−0.101 (0.006)	−0.280 (0.000)	0.283 (0.000)	0.192 (0.000)	1
(6)	POP	−0.066 (0.522)	0.093 (0.021)	0.116 (0.000)	0.119 (0.000)	−0.052 (1.000)	1
(7)	AREA	−0.044 (1.000)	−0.023 (1.000)	0.096 (0.012)	0.205 (0.000)	0.144 (0.000)	0.425 (0.000)	1
(8)	DENS	−0.016 (1.000)	0.104 (0.004)	0.029 (1.000)	−0.062 (0.775)	−0.167 (0.000)	0.371 (0.000)	−0.212 (0.000)	1
(9)	HEIGHT	−0.049 (1.000)	−0.229 (0.000)	−0.025 (1.000)	−0.032 (1.000)	0.163 (0.000)	−0.181 (0.000)	0.181 (0.000)	−0.205 (0.000)	1

,

Table A5. Correlations among variables, quantile-Group 5.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	0.240 (0.000)	1
(3)	pcTW	0.081 (0.249)	0.001 (1.000)	1
(4)	pcSWC	0.059 (1.000)	0.136 (0.000)	0.516 (0.000)	1
(5)	pcRWC	−0.093 (0.071)	−0.263 (0.000)	0.253 (0.000)	0.176 (0.000)	1
(6)	POP	0.004 (1.000)	0.091 (0.088)	0.106 (0.015)	0.074 (0.498)	−0.042 (1.000)	1
(7)	AREA	0.026 (1.000)	−0.013 (1.000)	0.063 (1.000)	0.159 (0.000)	0.114 (0.006)	0.431 (0.000)	1
(8)	DENS	−0.031 (1.000)	0.158 (0.000)	0.021 (1.000)	−0.080 (0.271)	−0.177 (0.000)	0.287 (0.000)	−0.246 (0.000)	1
(9)	HEIGHT	−0.040 (1.000)	−0.268 (0.000)	0.032 (1.000)	−0.003 (1.000)	0.224 (0.000)	−0.176 (0.000)	0.157 (0.000)	−0.236 (0.000)	1

and

Table A6. Correlations among variables, quantile-Group 6.

	Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1)	eco-efficiency	1
(2)	SR	−0.194 (0.045)	1
(3)	pcTW	0.344 (0.000)	−0.015 (1.000)	1
(4)	pcSWC	0.033 (1.000)	0.134 (0.955)	0.733 (0.000)	1
(5)	pcRWC	0.059 (1.000)	−0.184 (0.078)	0.556 (0.000)	0.629 (0.000)	1
(6)	POP	−0.030 (1.000)	0.121 (1.000)	0.082 (1.000)	0.093 (1.000)	−0.076 (1.000)	1
(7)	AREA	−0.026 (1.000)	−0.169 (0.175)	0.083 (1.000)	0.102 (1.000)	0.107 (1.000)	0.516 (0.000)	1
(8)	DENS	−0.040 (1.000)	0.208 (0.019)	−0.061 (1.000)	−0.089 (1.000)	−0.222 (0.008)	0.233 (0.004)	−0.308 (0.000)	1
(9)	HEIGHT	0.131 (1.000)	−0.472 (0.000)	−0.133 (0.983)	−0.242 (0.002)	0.066 (1.000)	−0.199 (0.033)	0.328 (0.000)	−0.242 (0.002)	1

). In this section, the most significant patterns revealed by the analysis are illustrated.

The findings of the correlation analysis indicate that eco-efficiency is only weakly related to the demographic and spatial variables POP, AREA, DENS, and HEIGHT. The probabilities suggest that almost all of these associations are statistically insignificant. Population size (POP) and municipality area (AREA) consistently show correlations close to zero across all quantile groups, with probabilities generally equal to 1.000. Population density (DENS) shows weak and inconsistent associations, consistently remaining non-significant. The only exception is municipality height above sea level (HEIGHT), which displays a modest but significant negative correlation in the second quantile group (−0.101 with a probability of 0.028). This indicates that elevation may impede eco-efficiency only for municipalities located in the lower-middle segment of the eco-efficiency distribution. In all other instances, these structural characteristics do not play a systematic role.

By contrast, several waste-related variables display stronger and statistically significant relationships with eco-efficiency. The sorting rate (SR) consistently shows positive and significant correlations across all quantile groups, with particularly strong effects in the lowest quantile group (0.726 and a probability equal to 0.000) and still moderate associations in the higher groups. This emphasizes the central role of higher sorting performance in driving municipal eco-efficiency. The per capita total waste (pcTW), although generally showing weaker coefficients, is significantly correlated with eco-efficiency in some quantiles (e.g., negative in Groups 2 and 3), suggesting that higher overall waste generation is associated with lower efficiency.

The cost-related variables reveal an interesting asymmetry. Per capita sorted waste cost (pcSWC) is positively and significantly correlated with eco-efficiency in several quantiles, especially in Group 1, with a correlation of 0.298 and a probability of 0.000. This suggests that increased spending on sorted waste management may improve performance. On the other hand, per capita residual waste cost (pcRWC) shows significant negative correlations with eco-efficiency across most quantiles. For example, in Group 1 and Group 2, correlations are −0.392 and −0.245, both with a probability of 0.000. This indicates that higher expenditures associated with residual (unsorted) waste hinder eco-efficiency.

Taken together, these results suggest that the structural and demographic characteristics of municipalities (population, area, density, height) are largely irrelevant. Instead, eco-efficiency is systematically influenced by waste-specific factors. Specifically, higher sorting rates and investments in sorted waste are positively correlated with performance, while higher levels of residual waste and the associated costs significantly reduce efficiency.

3.2.2. Quantile Regression

Due to the large size of the sample (5516 municipalities), adopting a bootstrapping technique to obtain a more robust eco-efficiency indicator in the second-stage DEA was considered unnecessary. This choice is justified by the asymptotic properties of DEA estimators, as demonstrated by Banker [74], who showed that DEA efficiency scores are statistically consistent in large samples. Specifically, the bias inherent in DEA frontier estimates diminishes as the sample size increases, and the asymptotic distribution of DEA inefficiency estimates converges to the true distribution of the inefficiency deviations. Therefore, the statistical reliability of first-stage eco-efficiency scores can be reasonably assumed without resampling techniques. To correct the non-standard distribution of quantile regression estimators, a bootstrap procedure was utilized to calculate the coefficient standard errors. This method aids in quantifying empirical variability and allows for the development of robust confidence intervals. A total of five thousand replications were performed for this analysis. The quantiles q = 0.05, 0.25, 0.5, 0.75, and 0.95 were taken into account as in previous analyses.

The quantile regression results are reported in

Table 3. Results from Quantile Regression.

Parameter	Quantiles
	q = 0.05		q = 0.25		q = 0.5		q = 0.75		q = 0.95
intercept	−0.726 (0.000) [−0.740; −0.711]	−0.736 (0.000) [−0.755; −0.717]	−0.436 (0.000) [−0.454; −0.419]	−0.451 (0.000) [−470; −0.433]	−0.142 (0.000) [−0.160; −0.125]	−0.176 (0.000) [−0.200; −0.152]	0.293 (0.000) [0.261; 0.324]	0.256 (0.000) [0.212; 0.299]	1.173 (0.000) [1.092; 1.253]	1.163 (0.000) [1.054; 1.271]
SR	0.969 (0.000) [0.921; 1.018]	0.967 (0.000) [0.917; 1.017]	0.832 (0.000) [0.791; 0.872]	0.831 (0.000) [0.794; 0.869]	0.767 (0.000) [0.733; 0.801]	0.768 (0.000) [0.733; 0.802]	0.723 (0.000) [0.683; 0.764]	0.729 (0.000) [0.688; 0.769]	0.755 (0.000) [0.702; 0.808]	0.754 (0.000) [0.700; 0.809]
pcTW	−0.182 (0.000) [−0.229; −0.135]	−0.176 (0.000) [−0.223; −0.129]	−0.026 (0.117) [−0.058; 0.006]	−0.025 (0.092) [−0.054; 0.004]	0.030 (0.025) [0.004; 0.056]	0.036 (0.006) [0.010; 0.062]	0.070 (0.007) [0.019; 0.121]	0.080 (0.002) [0.030; 0.129]	0.315 (0.000) [0.221; 0.409]	0.321 (0.000) [0.221; 0.421]
pcSWC	−0.009 (0.502) [−0.036; 0.018]	−0.010 (0.477) [−0.037; 0.017]	−0.077 (0.000) [−0.100; −0.053]	−0.078 (0.000) [−0.101; −0.056]	−0.097 (0.000) [−0.123; −0.072]	−0.098 (0.000) [−0.125; −0.072]	−0.095 (0.000) [−0.134; −0.056]	−0.100 (0.000) [−0.140; −0.060]	−0.197 (0.000) [−0.256; −0.138]	−0.194 (0.000) [−0.256; −0.132]
pcRWC	0.037 (0.001) [0.015; 0.060]	0.036 (0.003) [0.012; 0.060]	−0.001 (0.939) [−0.030; 0.028]	−0.002 (0.876) [−0.030; 0.026]	−0.023 (0.135) [−0.052; 0.007]	−0.028 (0.066) [−0.057; 0.002]	−0.045 (0.022) [−0.084; −0.006]	−0.050 (0.018) [−0.092; −0.009]	−0.125 (0.001) [−0.199; −0.052]	−0.126 (0.001) [−0.200; −0.052]
POP	−0.032 (0.193) [−0.079; 0.016]	−0.124 (0.060) [−0.254; 0.005]	−0.072 (0.146) [−0.170; 0.025]	−0.325 (0.000) [−0.440; −0.210]	−0.017 (0.659) [−0.092; 0.058]	−0.347 (0.002) [−0.565; −0.130]	0.011 (0.662) [−0.038; 0.059]	−0.227 (0.238) [−0.604; 0.150]	0.007 (0.831) [−0.056; 0.070]	−0.286 (0.246) [−0.769; 0.197]
AREA	0.003 (0.741) [−0.015; 0.021]	0.014 (0.135) [−0.004; 0.032]	−0.012 (0.166) [−0.029; 0.005]	0.008 (0.399) [−0.011; 0.028]	−0.035 (0.001) [−0.055; −0.014]	−0.013 (0.319) [−0.040; 0.013]	−0.055 (0.000) [−0.084; −0.026]	−0.039 (0.054) [−0.079; 0.001]	−0.078 (0.094) [−0.170; 0.013]	−0.043 (0.389) [−0.142; 0.055]
DENS	−0.012 (0.226) [−0.031; 0.007]	−0.020 (0.404) [−0.068; 0.027]	−0.021 (0.018) [−0.039; −0.004]	0.001 (0.947) [−0.034; 0.037]	−0.039 (0.000) [−0.056; −0.023]	−0.034 (0.203) [−0.086; 0.018]	−0.070 (0.000) [−0.095; −0.045]	−0.100 (0.039) [−0.195; −0.005]	−0.068 (0.017) [−0.125; −0.012]	−0.024 (0.827) [−0.243; 0.194]
HEIGHT	0.007 (0.354) [−0.001; 0.021]	−0.011 (0.369) [−0.037; 0.014]	0.024 (0.013) [0.005; 0.043]	0.000 (0.970) [−0.021; 0.022]	0.042 (0.001) [0.016; 0.067]	−0.003 (0.812) [−0.032; 0.025]	0.066 (0.000) [0.030; 0.103]	0.015 (0.557) [−0.034; 0.064]	0.202 (0.000) [0.111; 1.253]	0.184 (0.007) [0.051; 0.317]
POP×DENS		0.005 (0.450) [−0.008; 0.019]		0.015 (0.213) [−0.008; 0.038]		0.019 (0.274) [−0.015; 0.052]		0.017 (0.397) [−0.022; 0.056]		0.035 (0.339) [−0.036; 0.106]
POP×AREA		0.001 (0.832) [−0.011; 0.014]		0.004 (0.811) [−0.032; 0.041]		0.004 (0.903) [−0.059; 0.067]		0.002 (0.960) [−0.083; 0.088]		0.001 (0.981) [−0.103; 0.106]
HEIGHT×POP		−0.094 (0.236) [−0.251; 0.062]		−0.219 (0.001) [−0.351; −0.087]		−0.247 (0.008) [−0.428; −0.065]		−0.160 (0.252) [−0.434; 0.114]		−0.202 (0.301) [−0.585; 0.180]
HEIGHT×DENS		−0.023 (0.470) [−0.084; 0.039]		0.021 (0.433) [−0.031; 0.073]		−0.011 (0.748) [−0.079; 0.057]		−0.062 (0.303) [−0.181; 0.056]		0.058 (0.693) [−0.230; 0.346]
HEIGHT×AREA		−0.008 (0.277) [−0.023; 0.007]		0.004 (0.704) [−0.017; 0.025]		−0.002 (0.851) [−0.028; 0.023]		−0.005 (0.815) [−0.043; 0.034]		0.007 (0.893) [−0.099; 0.114]
Pseudo R2	0.540	0.541	0.477	0.481	0.409	0.412	0.335	0.336	0.257	0.259

Note: estimated p-values are reported in parentheses, while confidence intervals are provided in square brackets.

. Interaction terms were included in the regression model to capture how the effects of key territorial and demographic characteristics on waste management eco-efficiency depend on each other. Specifically, POP×DENS and POP×AREA capture demographic and territorial scale effects, showing how population size interacts with density and municipal area to affect collection eco-efficiency. Interactions involving HEIGHT (HEIGHT×POP, HEIGHT×DENS, HEIGHT×AREA) capture how altitude modifies these relationships, while also accounting for the additional logistical and organizational challenges present in high-altitude municipalities. Finally, DENS×AREA captures the combined effect of population distribution and territorial extent on service eco-efficiency. However, this last term was excluded from the regression analysis due to collinearity. Table 3 displays the estimated coefficients for each quantile, with p-values reported in parentheses and 95% confidence intervals shown in square brackets.

The pseudo-R² gradually decreases from 0.540 at q = 0.05 to 0.259 at q = 0.95, confirming that the model’s explanatory power is stronger for less eco-efficient municipalities (lower quantiles) than for highly eco-efficient ones (higher quantiles). As a result, the mean predictive accuracy is higher at the lower end of efficiency distribution. These values can be considered acceptable for this study and the methodology used.

To assess the robustness of the results obtained from the quantile regression analysis, additional analyses using boosting regression in the JASP machine learning tool and Tobit regression with bootstrapped standard errors were performed [75,76,77,78]. Appendix B contains a detailed description of the methods used and the results obtained from this supplementary analysis. However, the findings from the Tobit regression cannot be used for reference because the assumptions of the Tobit model, which include normally distributed, homoscedastic errors, and natural censoring at 0 and 1, do not accurately reflect the true distribution of the MSW service eco-efficiency.

Next, the impact of exogenous factors on the measurement of MSW eco-efficiency is explored. The separate collection rate (SR) consistently emerges as the most significant determinant of eco-efficiency. Its coefficient is always positive, highly significant, and relatively stable across all quantiles, ranging between 0.72 and 0.97. The effect is stronger at the lower quantiles, suggesting that improvements in separate collection result in the largest gains in municipalities that initially perform poorly. This pattern is supported by descriptive statistics, as the average SR increases significantly from approximately 33% in the least efficient municipalities (Group 1) to over 80% in the most efficient group (Group 6). The boosting regression provides results that are comparable to those obtained with quantile regression (see

Table A10. Feature Importance Metrics.

Variable	Relative Influence	Mean Dropout Loss
SR	81.301	0.155
pcTW	8.609	0.071
pcSWC	5.097	0.070
pcRWC	3.781	0.067
HEIGHT	0.741	0.062
POP×DENS	0.471	0.062
POP	0.000	0.061
AREA	0.000	0.061
DENS	0.000	0.061
POP×AREA	0.000	0.061
HEIGHT×POP	0.000	0.061
HEIGHT×DENS	0.000	0.061
HEIGHT×AREA	0.000	0.061

Note: mean dropout loss (defined as root mean squared error (RMSE) is based on 500 permutations.

).

Per capita total waste generation (pcTW) shows a non-linear relationship with eco-efficiency. In the lowest quantiles, the effect is negative and significant, indicating that in inefficient municipalities, higher waste generation per capita worsens performance. At q = 0.05, the coefficient is −0.182 (−0.176 in the extended model), while at q = 0.95, it is 0.315 (0.321 in the extended model). This indicates that in more eco-efficient municipalities, higher waste generation is not detrimental and may even be associated with improved outcomes, reflecting the ability of advanced systems to handle larger waste volumes. Descriptive data supports this pattern. This variable also emerged as an important factor affecting the eco-efficiency of MSW in the boosting regression analysis. Although per capita total waste does not decline significantly across quantiles, per capita residual waste (pcRW) drops sharply from about 374 kg in Group 1 to around 83 kg in Group 6. This shows that high-performing municipalities can handle higher waste volumes with much lower residual fractions and higher percentages of sorted waste (see

Table A7. Descriptive statistics of input and output variables by quantile group.

	Main Statistics					Main Statistics
Variable	Mean	Median	St.dev.	Max	Min	Mean	Median	St.dev.	Max	Min
	Group 1					Group 2
pcSW	190.52	200.90	98.65	486.86	0.00	278.77	276.90	100.82	821.30	33.26
pcRW	374.27	333.39	143.25	958.70	147.35	228.55	193.11	107.38	913.55	111.59
pcTWC	264.60	238.46	96.92	777.15	119.64	199.35	179.74	79.27	941.80	105.83
	Group 3					Group 4
pcSW	307.10	303.85	100.54	985.68	46.46	330.87	326.31	103.42	982.68	52.64
pcRW	145.15	124.68	68.25	810.70	70.30	106.64	92.52	51.11	804.58	49.55
pcTWC	167.07	150.70	77.14	1484.25	96.22	153.98	134.08	76.51	1005.12	83.02
	Group 5					Group 6
pcSW	358.27	343.46	129.44	1122.15	71.44	447.92	398.29	255.47	2088.21	67.91
pcRW	82.30	68.71	52.46	705.84	25.50	83.55	53.79	102.79	1155.70	0.15
pcTWC	151.02	125.85	81.18	838.18	68.94	148.38	106.87	142.39	1228.68	43.44

in Appendix A).

The per capita cost of sorted waste management (pcSWC) is minimal in the lowest quantiles, but it becomes negative and highly significant from q = 0.25 onwards, reaching its peak impact at the upper quantiles (between −0.194 and −0.197, depending on the regression model, at q = 0.95).

This indicates that increased spending on separate collection does not enhance performance in areas with low eco-efficiency. Additionally, among the more eco-efficient municipalities, higher costs for separate collection actually impede eco-efficiency improvements. This finding suggests inefficiencies in spending, but it could also imply that achieving higher levels of eco-efficiency involves significant costs related to managing the sorted fraction of waste.

A similar, though weaker, pattern is observed for residual waste costs (pcRWC). The coefficient is slightly positive in the lowest quantiles (between 0.036 and 0.037, at q = 0.05) but becomes negative and significant from q = 0.75 onward (between −0.045 and −0.126). This indicates that higher spending on residual waste may improve service delivery in underdeveloped MSW management systems (for instance, improving basic services), but it reduces eco-efficiency in better-performing municipalities, reflecting inefficiency.

Demographic and territorial characteristics have a weaker and less systematic impact. The population of a municipality (POP) typically has a negative or neutral effect on eco-efficiency. This effect is only significant in the extended regression model that includes interaction effects and for lower quantiles (up to q = 0.50). Population size alone does not affect the MSW eco-efficiency in the most eco-efficient municipalities. The descriptive evidence in Table 2 supports this. The mean value of population drops from over 23,000 inhabitants in Group 1 to around 5000 in Groups 5 and 6, indicating that smaller municipalities are generally more efficient. This suggests that larger municipalities may face challenges with eco-efficiency, potentially because of their organizational complexity and higher demands for service delivery.

The findings partially align with previous studies, that have highlighted contrasting evidence regarding the influence of municipal size on the efficiency of waste management services. Daraio et al. [54] found that municipal size negatively affects service efficiency when the population is below a certain threshold. In contrast, Expósito and Velasco [78] reported that the size of the served population does not significantly explain the level of efficiency in the provision of MSW services. Both the territorial extension of the municipality (AREA) and its population density (DENS) display negative coefficients in some quantiles, but their significance is not robust across the distribution. This suggests that their influence might be context-dependent rather than systematic.

Specifically, the size of the area has a weak, increasing negative effect starting from q = 0.5, suggesting that municipalities covering a larger area tend to be less eco-efficient. This is particularly evident among those with medium to high eco-efficiency, likely due to higher logistical costs. A similar result was reported by Sarra et al. [26], whose second-stage DEA identified the geographical area of the municipality as a significant factor negatively associated with service efficiency. The population density (DENS) has a negative effect on eco-efficiency starting from q = 0.25. Higher density correlates with lower eco-efficiency, possibly due to organizational challenges or more difficult waste management in dense urban areas. However, the coefficients are generally significant only in the simplified regression model that does not include interaction effects. These findings are consistent with previous studies that indicate higher population density reduces the efficiency of waste service providers [79,80]. However, they contradict more recent research which suggests that greater density results in lower service delivery costs and, consequently, higher efficiency levels [46,73,78].

Altitude (HEIGHT) displays a distinct pattern. The coefficient for the height above sea level of the municipality (HEIGHT) is positive and increases from 0.024 (q = 0.25) to 0.202 (q = 0.95) when statistically significant. This suggests that municipalities at higher elevations tend to be more eco-efficient. While this finding may seem counterintuitive, it could reflect the significance of contextual characteristics in high-altitude areas when managing urban waste. Mountainous municipalities typically have smaller populations, lower waste generation per capita, and greater community involvement in local waste practices. These factors may encourage reuse and source separation, despite logistical challenges. Additionally, lifestyles in these areas may involve reduced consumption patterns, resulting in less residual waste overall. Therefore, altitude may serve as a proxy for unobserved socio-environmental factors rather than solely a geographical influence. Further research is necessary to unravel these factors and determine if elevation has a causal or correlational relationship in waste management. Descriptive statistics provide additional insights. The average altitude of municipalities is highest in the least efficient groups (around 656 m in Group 1 and 510 m in Group 2), which also contain municipalities with the largest population sizes. However, the average altitude remains relatively high across the distribution (365–413 m in Groups 5–6). Within quantile bands, municipalities located at higher altitudes tend to perform better than those at lower altitudes.

Interaction terms add additional nuance to the analysis. Although most interactions are not significant, the HEIGHT×POP term is negative and significant in the middle quantiles. This implies that the positive association between altitude and eco-efficiency exists in small and medium-sized municipalities but diminishes, and possibly reverses, in larger mountain municipalities where demographic size amplifies the logistical challenges of altitude. These results contrast with the boosting regression findings, which indicated a slight yet discernible effect of the POP×DENS interaction on eco-efficiency and no effect of HEIGHT×POP.

3.3. Discussion

The findings of this study highlight systemic challenges, contextual paradoxes, and significant trade-offs in the eco-efficiency of municipal solid waste (MSW) management in Italy. From a policy standpoint, the fact that only 0.13% of municipalities have achieved full eco-efficiency emphasizes a considerable performance disparity within the sector. This aligns with global evidence showing that waste management systems often struggle to meet the objectives of cost-effectiveness, high recycling rates, and minimal residual disposal simultaneously. It indicates that the Italian MSW sector, despite making strides in separate collection, still grapples with structural inefficiencies and regional inequalities that hinder the attainment of EU circular economy goals.

The results reveal significant regional differences. Northern municipalities consistently perform better and achieve higher average efficiency, while Central Italy shows more moderate performance, with eco-efficiency scores clustering around the mean. This likely reflects intermediate levels of infrastructure, administrative capacity, and citizen participation. In contrast, Southern municipalities have lower average scores and greater variability, indicating weaker institutional capacity, a stronger reliance on landfilling, and uneven citizen engagement. These disparities suggest that eco-efficiency in Italy is not just a technical or financial issue, but also an institutional and governance challenge. This is highlighted by the more evenly distributed density curve for Central Italy, in contrast to the left-skewed distribution observed in Southern Italy.

Quantile regression analysis reveals that the impact of explanatory variables on urban waste management eco-efficiency varies significantly across different levels of efficiency. The first key finding is the central importance of the sorting rate (SR), emphasizing the critical role of policies that promote recycling and recovery.

The consistently positive and significant impact of separate collection across all quantiles demonstrates that separate collection remains the backbone of eco-efficient waste management. However, the effect is strongest in low-performing municipalities, where even moderate improvements in separate collecting of waste can yield substantial efficiency gains. This suggests that national and regional authorities should prioritize targeted interventions, such as financial incentives, citizen awareness campaigns, and stronger enforcement of waste separation obligations, in lagging areas, particularly in Southern Italy. Conversely, in municipalities that are already performing well, small improvements in the rate of sorted waste collection no longer result in significant efficiency gains. This suggests that policies should focus on improving the quality of sorted materials, developing markets for recyclables, and preventing waste at the source. For academics, this unequal impact emphasizes the importance of looking beyond average effects and using distributional methods like quantile regression to identify non-linearities in environmental performance.

The robustness of these results is further supported by the distribution of explanatory power across quantiles. The pseudo-R² values decrease from 0.540 at the lowest quantile to 0.259 at the highest, indicating that the model explains eco-efficiency outcomes much better in lagging municipalities than in already efficient ones. This pattern reinforces the interpretation that drivers such as sorted waste rate, per capita waste generation, and cost structures are particularly decisive in shaping the performance of poorly performing municipalities. In contrast, high-performing municipalities may be influenced by additional, unobserved factors such as technological innovations, quality of recyclables, or governance practices. From an academic standpoint, this finding further emphasizes the advantage of employing quantile regression for uncovering heterogeneous explanatory power along the efficiency distribution.

The role of waste generation (pcTW) reveals a striking paradox. In municipalities with low eco-efficiency levels, the generation of higher per capita waste volumes has a negative effect on the MSW service eco-efficiency. This generally signals poor consumption habits, lack of public awareness, or inefficient waste systems. This reinforces the urgency of demand-side policies such as waste prevention, eco-design, and consumption reduction. However, in municipalities that achieve high eco-efficiency levels, greater waste volumes are associated with better performance. This counterintuitive pattern suggests that advanced systems can exploit economies of scale in waste collection and treatment. For example, adopting more effective door-to-door collection systems can compensate for increased waste production. Additionally, these advanced systems can leverage advanced logistics and processing capacities, making sophisticated sorting technologies economically viable. Lastly, they can benefit from a mature infrastructure, transforming waste from a burden into an operational and business advantage.

This potential structural paradox suggests that once a certain threshold of separate collection is reached, further increasing the amount of separately collected waste to improve the eco-efficiency of municipal solid waste (MSW) requires an overall increase in waste production, particularly of recyclable fractions. It should also be taken into account that higher per capita waste in eco-efficient municipalities increases the volume of separated recyclables sold to recycling consortia. This generates revenues that help offset waste management costs, which further explains the positive association with eco-efficiency. However, these results should be interpreted with caution as they do not imply that producing more waste is desirable. Instead, they suggest that eco-efficient MSW systems are resilient to higher volumes of waste. For policymakers, this means that high-efficiency systems can handle larger flows without sacrificing outcomes, but this should not justify complacency toward waste prevention. For scholars, this result underscores the importance of contextual thresholds because the same driver can have opposing effects depending on baseline performance levels.

The analysis of cost structures reveals additional trade-offs. Expenditures on sorted waste (pcSWC) are nearly neutral in low-efficiency settings but become a substantial burden in high-efficiency municipalities. In these areas, increased spending decreases eco-efficiency. The coefficient of the per capita cost of managing the sorted fraction of waste (pcSWC) in the quantile regression model decreases from −0.077 to −1.97. This indicates that municipalities may reach a point where they risk entering a “green performance trap” achieving environmental goals but at unsustainable costs.

Similarly, in the lower quantiles, higher expenditures on residual waste (pcRWC) are associated with slightly positive effects on eco-efficiency, reflecting the necessity to enhance basic service delivery in less efficient municipalities. However, at the upper quantiles, the coefficient turns negative (from +0.036 to −0.126), indicating that additional spending on residual waste becomes inefficient once basic systems are already established. Therefore, residual waste costs (pcRWC) seem advantageous for underdeveloped systems as they support minimum service quality. However, they are detrimental for advanced systems, as they indicate inefficiency and an over-reliance on disposal.

Descriptive statistics in Table 2 indicate that the per capita cost of managing sorted waste remains relatively stable across efficiency distribution, ranging between €64 and €71. However, the cost of residual waste decreases significantly, from over €100 in the least efficient municipalities to about €26 in the most efficient ones. This confirms that eco-efficiency gains are mainly driven by reductions in residual waste management costs rather than changes in expenditures for sorted fractions. Additionally, the relative stability of per capita costs for sorted waste (pcSWC) across the efficiency distribution does not necessarily imply that operational costs are uniform. Rather, it reflects the balancing effect of revenues from recycling consortia. Further increasing the share of separate collection often entails rising marginal costs, while the corresponding revenues from consortia remain limited or capped. For policymakers, these findings raise an uncomfortable question: Should public policy incentivize ever-higher separation rates, even if marginal costs escalate? Or should it pursue an optimal balance between recycling and cost-effectiveness? Such a tension calls for more nuanced tools, such as eco-modulated tariffs, pay-as-you-throw schemes, and extended producer responsibility systems, that internalize both environmental and financial dimensions of MSW management. For the academic debate, the results highlight the importance of integrating cost variables into eco-efficiency analyses, which are too often limited to physical waste flows alone. From a policy perspective, this suggests that municipalities aiming to improve their eco-efficiency should prioritize strategies that minimize residual waste flows and associated costs. These strategies include prevention, providing incentives linked to the reduction in environmental impact, stronger enforcement of separation, improving the quality of collected fractions to enhance the market value of recyclables and reduce contamination, and investment in recycling markets and technological innovation such as route optimization, automated sorting, and sensor-based bins to reduce operational costs. Thus, municipalities should not rely solely on increasing the intensity of waste separate collection.

Demographic and territorial variables have weaker and context-specific effects, but it still offers valuable insights. Population size decreases eco-efficiency in municipalities that are already more efficient, indicating that organizational complexity and heterogeneity can counteract potential economies of scale. Population density also correlates negatively with eco-efficiency in certain quantiles, contradicting the assumption that density automatically lowers costs. In Italian cities, congestion, narrow streets, contamination of recycling streams, and socioeconomic diversity appear to undermine theoretical advantages. This further suggests that policy interventions in densely populated areas should focus on implementing smart waste systems (for instance, sensor-equipped bins), targeted education campaigns, and differential pricing schemes that enhance accountability and reduce contamination.

Perhaps the most paradoxical result concerns altitude (HEIGHT). Contrary to traditional expectations that higher elevation hampers efficiency due to logistical challenges, altitude is positively associated with eco-efficiency in medium-to-high quantiles. This can be explained by smaller populations, lower per capita waste generation, stronger community engagement in waste practices, and different waste composition in mountain municipalities. Institutional arrangements such as inter-municipal consortia and targeted funding may also play a role. In addition, lifestyles in these areas may involve lower consumption patterns, resulting in less residual waste overall. Importantly, the HEIGHT×POP interaction reveals that this advantage is conditional: altitude benefits small and medium-sized communities, but large mountain municipalities face compounded difficulties due to a combination of geographic constraints and organizational complexity. This helps reconcile the paradox and emphasizes that altitude is not simply a disadvantage, but rather a proxy for broader socio-environmental characteristics.

Taken together, the results contribute to both policy and academic debates. For policymakers, they emphasize that improving MSW eco-efficiency requires tailored strategies based on the municipality’s efficiency level and structural characteristics. For low-efficiency municipalities, policy interventions should focus on building foundational capacity and improving basic services. This includes investing in fundamental waste collection infrastructure, implementing education and awareness programs, and gradually transitioning from disposal-focused to prevention and recycling systems. Conversely, efficient municipalities should focus on pursuing cost rationalization without compromising environmental standards. This can be achieved by investing in technological innovations to optimize existing systems, preventing waste rather than just managing increasing volumes, monitoring cost-effectiveness metrics (e.g., cost per ton of clean recyclables), integrating waste management with circular economy markets, and linking incentives to environmental impact per euro spent. High-density municipalities also require specialized approaches, such as smart waste systems with underground or sensor-equipped bins, targeted education campaigns, and differential pricing schemes to enhance accountability and reduce contamination.

As a general policy recommendation, municipal administrators and MSW operators’ managers should acknowledge the heterogeneity in how variables impact different municipal contexts. They should avoid simplistic one-size-fits-all targets or performance indicators and consider local contextual factors in policy design. At the same time, they should aim for long-term systemic efficiency rather than focusing solely on single metrics. Particularly, adaptive waste management practices should be implemented in municipalities located in mountainous areas. For academia, this study demonstrates the usefulness of quantile regression in capturing heterogeneous effects across efficiency distribution. It also highlights the importance of cost structures in shaping eco-efficiency and the influence of environmental performance on local socio-geographic conditions.

4. Conclusions

This study evaluated the eco-efficiency of municipal solid waste (MSW) management in 5516 Italian municipalities in 2022. It applied a directional distance function DEA model combined with quantile regression. The results show that eco-efficiency remains generally low, with significant territorial disparities. Northern municipalities consistently outperform Central and Southern ones. However, even in the North, only a handful of municipalities reached the efficiency frontier.

Three main conclusions emerge. Firstly, the waste sorting rate is the most influential factor in determining eco-efficiency, particularly in municipalities that are falling behind. This confirms the importance of separate collection policies while also pointing to diminishing returns in more advanced MSW systems. Secondly, per capita waste generation has non-linear effects, as it negatively impacts performance in inefficient municipalities but positively affects it in efficient ones, showcasing economies of scale. Lastly, cost structures highlight significant trade-offs; investments in sorted and residual waste improve basic service provision in underperforming areas but can hinder eco-efficiency in advanced MSW systems, leading to questions about cost-effectiveness and policy formulation.

A central insight emerging from this analysis is the trade-off between environmental performance and economic sustainability. While increasing the rate of sorted waste is a key policy objective under EU and national frameworks, the findings reveal that municipalities with higher levels of separate collection often face significantly higher operational costs. This paradox suggests that environmental gains may come at the expense of economic efficiency unless collection systems are optimized and supported by cost-effective infrastructure and technologies.

This study also highlights a density paradox, in which urbanized and densely populated areas—usually thought to take advantage of economies of scale in service delivery—actually show lower eco-efficiency scores. This could be due to the intricate and diverse nature of waste streams in urban settings, as well as difficulties in setting up efficient door-to-door collection systems in densely populated areas. Conversely, some smaller municipalities with less complex waste compositions and more uniform populations perform better than their urban counterparts, indicating that performance outcomes are heavily influenced by specific contextual conditions. International evidence suggests that high-density cities like Berlin, Tokyo, or Singapore have successfully addressed the logistical challenges of dense urban environments through integrated waste-to-energy systems, advanced collection technologies (such as underground bins and pneumatic collection), and robust regulatory frameworks. However, while this comparison suggests that the density paradox observed in Italy could be avoided through better technological integration and systemic organization, several obstacles still hinder its implementation, particularly in linking dense urban areas with energy recovery infrastructures. These include limited urban space for collection infrastructure, high investment costs, regulatory constraints, and public opposition to local energy recovery facilities. Additionally, socio-economic and cultural heterogeneity across neighborhoods can reduce the effectiveness of standardized collection systems. Therefore, overcoming the density paradox in Italy requires not only technological upgrades and better system organization but also tailored policies that address local conditions, engage communities, and provide incentives for both efficient waste separation and the adoption of energy recovery solutions.

These results have clear implications for waste governance. For policymakers, they highlight the need for differentiated and context-sensitive strategies. This includes prioritizing improvements in sorted waste rates in lagging municipalities, managing cost escalation in advanced systems, and strengthening governance capacity in metropolitan areas. It is important to recognize the adaptive strengths of mountain communities as well.

Municipalities in the early stages of the waste transition require targeted investments and technical support to improve basic performance, especially in residual waste reduction. On the other hand, high-performing municipalities need strategies to contain costs and avoid diminishing returns. This could involve investing in smart collection systems, enhancing public engagement, or fostering local circular economy initiatives. For example, economic incentives could facilitate the step-by-step introduction of anaerobic digestion plants for organic waste in mountainous regions. This would produce biogas for local energy consumption and reduce landfill dependence. Additionally, promoting the creation of refuse-derived fuel (RDF) from non-recyclable waste could enhance the economic value of waste streams and lower disposal expenses. In crowded urban areas, utilizing advanced sorting and chemical recycling technologies, along with selectively incorporating waste-to-energy facilities, could address logistical obstacles and improve the balance between environmental performance and economic viability. Policy recommendations can be summarized as follows: (a) low-performing municipalities should focus on residual waste reduction through anaerobic digestion plants for organics, RDF production from non-recyclables, and gradual investments in energy-efficient waste-to-energy infrastructure; (b) high-performing municipalities need to prioritize cost control through digital logistics systems, smart collection technologies, and continuous improvement of recyclables’ quality and market value.

Regional disparities in performance and efficiency reflect deeper institutional and infrastructural divides, especially between Northern and Southern Italy. Bridging this gap requires coordinated multi-level governance, equitable financing mechanisms, and cross-regional knowledge transfer to ensure that all municipalities can contribute meaningfully to national and EU targets.

From an academic perspective, the study demonstrates that eco-efficiency is influenced by technical and socio-geographic factors. It also suggests that quantile regression is a valuable tool for capturing diverse and sometimes contradictory effects. The importance of this research is in highlighting that eco-efficiency is not solely determined by higher recycling rates, but rather by the complex interplay of costs, institutional capacity, and contextual variables. By integrating DEA and quantile regression, this study provides a more detailed understanding of municipal performance compared to earlier studies. This study contributes to the academic literature on efficiency measurement and provides empirical evidence to inform policy debates on waste governance.

However, it is important to note that the study does have limitations that need to be taken into account when interpreting the results. The analysis is based on municipal-level data for the year 2022. While this offers a comprehensive snapshot of the Italian MSW system, it does not account for changes over time or allow for the examination of trends in eco-efficiency, the influence of previous or future investments, or the effects of policy adjustments. A panel dataset would enable more dynamic analyses of how eco-efficiency progresses and how various factors operate in both the short and long term.

The variables used capture important structural, demographic, and cost-related characteristics but do not include institutional, organizational, and socio-economic dimensions that could impact eco-efficiency. For instance, data on waste collection methods (curbside versus drop-off), service frequency, the presence of inter-municipal consortia, availability and proximity of treatment facilities, or socio-economic and cultural characteristics of the resident population were not accessible. The absence of this information may partially explain the lower explanatory power of the model at higher quantiles. Furthermore, eco-efficiency indicators did not consider energy recovery variables (such as incineration with energy recovery, anaerobic digestion outputs, or refuse-derived fuel production) due to inconsistent municipal-level data in Italy. Energy valorization is a crucial aspect of sustainable MSW management, but these processes are often managed at a supra-municipal or regional level, making it challenging to integrate them into a municipality-based DEA model.

The DEA approach assumes similar production technology and operating conditions across municipalities, which may not always be the case in different territorial contexts. Quantile regression offers valuable insights into heterogeneity but cannot fully address unobserved heterogeneity or potential endogeneity. Although bootstrapping with 5000 replications improves robustness, the validity of results depends on the chosen model specification. The study recognizes that the effects of population density and altitude may be influenced by unobserved variables like technological endowment, cultural practices, or access to energy recovery infrastructures. While a formal moderating effect test was not possible due to data limitations (for instance, the lack of socio-economic variables), this is an important avenue for future research.

Model explanatory power varies across quantiles, with pseudo-R² values decreasing from 0.540 at the 0.05 quantile to 0.259 at the 0.95 quantile. This decline in explanatory capacity at higher quantiles does not indicate a flaw, but it rather reflects the increasing complexity of determinants of eco-efficiency. In municipalities with high performance, factors such as organizational choices, behavioral dynamics, and local policy designs play a significant role, which were not captured in this dataset.

The generalizability of the results is limited due to the context-specific nature of Italy’s municipal waste system, which has unique institutional arrangements and regulatory frameworks. While the methodological framework can be applied in other countries, the specific findings may not directly apply.

Future research should explore the longitudinal dynamics of eco-efficiency. This includes integrating broader environmental indicators, such as carbon emissions and circular economy practices, and incorporating qualitative or behavioral dimensions of waste governance. By combining efficiency frontiers with lifecycle and energy recovery indicators, research can capture the full energy-environmental value of MSW systems.

Furthermore, extending the analysis to cross-country comparisons would help clarify how institutional and policy frameworks influence eco-efficiency outcomes. This can also assess whether the paradoxes observed in Italy are present in other decentralized waste governance systems.

To enhance the research, combining DEA-quantile approaches with interaction models and utilizing richer datasets would allow for testing moderating effects more explicitly.