Improving Municipal Solid Waste Services: Insights into Efficiency, Productivity, and Recycling in Brazil

Abstract

This study analyzed the interactions between efficiency, productivity, and the policy of recyclable materials recovery in municipal solid waste (MSW) services in Brazil, using data from the National Sanitation Information System (SNIS) for 199 MSWs, encompassing a balanced longitudinal panel from 2013 to 2022. In this context, the study employed the dynamic slacks-based measure (DSBM) model to assess the intertemporal efficiency and productivity of MSWs. This model is an additive extension of traditional data envelopment analysis, as it does not focus solely on the overall efficiency of operators. Additionally, multiple correspondence analysis (MCA), multiple independent sample tests, and generalized estimating equations (GEE) were utilized to identify potential interactions among the variables of interest. The main contribution of this study lies in its analysis of dynamic intertemporal efficiency, considering the indirect influence of operating expenses as a carry-over, which impacts the continuity of efficiency across periods. Insights from the results indicated an average efficiency of 61.9% and an accumulated productivity loss over the analyzed period. The recovery of recyclable materials exhibited a negative interaction with both the efficiency of MSW operators performing selective collection and the overall technological advancement of MSW systems. This underscores the importance of integrating this field of study within the circular economy framework and fostering recycling initiatives. Factors such as population density, gross domestic product (GDP), and the COVID-19 pandemic influenced efficiency and, in certain aspects, productivity. The findings provide insights suggesting that policymakers and regulatory agencies should consider these results to enhance eco-efficiency and eco-productivity policies, particularly in alignment with the Sustainable Development Goals (SDGs) and current Brazilian legislation.

1. Introduction

Improper disposal methods, including the use of covered dumps without bottom protection, open dumps, and discarding waste in rivers, lakes, and along roadsides, are common in Brazil and many developing countries. In this sense, municipal solid waste (MSW) remains a chronic issue in these regions, often due to poverty and a lack of funds for adequate treatment methods [1], which can be considered a critical situation in relation to compliance with United Nations Agenda 21.

Factors such as the growth of urbanization, economic development, and improvement of living standards in developing countries are associated with a significant increase in the amount of solid waste generated [2,3]. It is relevant to emphasize that the generation of urban solid waste is considered a natural product of the aforementioned factors. However, the inadequate management of these wastes, including their disposal and final treatment, is responsible for various social and environmental consequences [4,5].

In the economies of emerging countries such as Brazil, Russia, India, China, and South Africa (BRICS), landfills represent the primary method of municipal solid waste disposal. However, many operate irregularly or even clandestinely. Practices such as source separation, recycling, composting, and reducing waste generation are essential to reduce the volume of solid waste sent to these landfills. This helps to extend the useful life of landfills, reduce the need for new disposal areas, and reduce investments [6].

In this context, studies focusing on the evaluation of service efficiency have become increasingly prevalent. Insights derived from studies on the efficiency of solid waste management (SWM) services highlight that this scenario remains relatively less explored in the literature compared to other services, such as the electricity sector [7,8]. In this regard, developing and underdeveloped countries face greater challenges in the inadequate management of solid waste, attributed to nearly absent infrastructure and inadequate processing channels, the scarcity of financial and human resources, as well as political obstacles [9].

In developing regions of Asia and Africa, continents with the highest concentration of emerging countries [10], Zhang et al. [10] examined, through a literature review, the main challenges of SWM in Asian and African countries from 2013 to 2023. Their findings indicate that a lack of public awareness and action, funding sources, infrastructure, and other cultural factors are responsible for delaying proper waste disposal. According to the authors, three main aspects hinder the optimization of SWM: socioeconomic challenges, infrastructure deficits, and cultural factors. The study emphasizes the importance of integrating this field of study into the circular economy framework and promoting recycling initiatives.

Serrona et al. [11] add that poor planning, financial insufficiency, lack of technical knowledge, and ineffective laws are barriers that hinder the implementation of effective and economically viable environmental measures for waste management programs. In the case of Brazil, located in Latin America, the third largest emerging continent [10], it is understood that the formulation and adoption of recent legislation contribute to this scenario.

Previous studies aimed to evaluate the efficiency of operators in the solid waste sector from various perspectives, such as the eco-efficiency and eco-productivity of municipalities in solid waste management in Chile [3,12] and Italy [13]; the assessment of efficiency in the prevention and treatment of industrial solid waste [14,15]; integrated structures and solid waste management systems [16,17]; and other factors influencing waste management performance [18,19,20,21]. Insights from these findings suggest that increased food recycling, waste diversion to composting, and bypassing landfill disposal reduces undesirable indicators, offsetting economic costs and enhancing municipal solid waste efficiency.

For the reasons stated in this context and considering the recovery of recyclable materials as the starting point of the full life-cycle process within industry from a circular economy and eco-efficiency perspective, and examining how recycling practices impact the efficiency and productivity of solid waste management services, this study aims to assess the interactions between intertemporal dynamic efficiency, productivity, and recyclable material recovery policies in Brazilian MSW, offering insights into how these collection policies interact with service performance (efficiency and productivity).

In this regard, the present study distinguishes itself from previous research by employing a data envelopment analysis (DEA) approach using the slacks-based measure [22] within a dynamic intertemporal framework, as originally introduced by Tone and Tsutsui [23] and referred to as the dynamic slacks-based measure (DSBM). Furthermore, it examines the effects of recyclable material recovery using a second-stage explanatory approach.

According to lo Storto [24], DEA provides a single efficiency score that considers multiple inputs and outputs, creating an empirical production function based on observed data, without making any restrictive assumptions about the function that transforms inputs into outputs. However, traditional DEA models have a disadvantage in omitting the intertemporal (dynamic) structure of decision-making units (DMUs) [23]. Thus, the structure proposed for assessing the efficiency of solid waste service providers in Brazil allows the evaluation of the overall efficiency of DMUs and provides additional assessment in an intertemporal (dynamic) manner, wherein ‘carry-overs’ connect the analysis periods (terms).

Considering what was observed in a previous literature review, the current recycling scenario in Brazil requires actions to encourage the implementation of recycling programs in an integrative manner with recycling stakeholders in the planning and management of these activities. The foundations of the circular economy can act as a catalyst for actions to achieve the Sustainable Development Goals (SDGs), assisting in the recovery of economic growth while supporting sustainability policies that recover energy and materials from waste, thereby producing balanced economic, social, and environmental performance. Therefore, the recycling process is essential for the proper management of MSW [1].

Recycling not only offers potential landfill cost savings, but also reduces the volume of waste sent to landfills or incinerators, which helps decrease soil, water, and air contamination through lower emissions. By adopting a full life-cycle approach, recycling supports regulatory authorities in more effectively managing natural resources, including raw materials, land use, and the quality of water and soil; on the other hand, selective collection is more costly in Portugal, and it is unclear whether part of these expenses stems from local administrations’ attempts to pass costs to the receiving industry (paper recycling) or if they result from inefficiencies of waste management [25].

This study presents several contributions to the existing literature, as well as political and practical implications. Firstly, by employing DSBM, this model not only broadens the scope of the existing literature, but also offers an analysis of an emerging country like Brazil. Secondly, the article highlights its contribution to the Sustainable Development Goals, particularly SDG 12 (Responsible Consumption and Production) and SDG 6 (Clean Water and Sanitation).

By providing an assessment of the efficiency of solid waste service providers, this study can be used to inform policy discussions with public managers aimed at sustainable development in the country and other developing nations, where policies for promoting recycling are incipient [26]. Furthermore, it facilitates discussions on reducing improper solid waste disposal, promoting public–private partnerships, and expanding international debates through global cooperation.

Furthermore, the study sheds light on critical issues related to public policy formulation, particularly in the context of the goals established by the new sanitation framework, including Brazilian Federal Laws 14,026/2020 [27], 12,305/2010 [28], and 11,445/2007 [29], as well as regulations aimed at promoting recycling processes, such as Law 14,260/2021 [30] and Decree 12,106/2024 [31]. However, these regulations have not proven effective, as our findings reveal a scenario of inadequate waste disposal and greater inefficiency among operators with selective collection systems, highlighting gaps in implementation and incentives for the recycling industry.

2. Solid Waste Services in Brazil

Brazil ranks 16th in the Latin American and Caribbean region, which includes 42 countries, in terms of solid waste generation (kg/per capita per day) [32]. In line, the public sector in Brazil is decentralized among three levels, the Union (central government), 26 Brazilian states plus the Federal District, and 5570 municipalities; among their competencies is the management of the MSW sector.

Municipalities under direct public administration are responsible for more than 94.6% (4637 of 4900 participants) of the management, which includes 73.7% of the population. Public companies are responsible for 3% of management bodies, autarchies represent 2.7%, and mixed economy companies with public administration represent 0.7%. The household solid waste collection coverage rate for the urban population, in general, reaches 98.7%, and the country’s average rate is 90.5%.

Regarding the coverage rate of household solid waste collection for the urban population, it reaches 98.7% overall, with an average rate of 90.5% for the country. Analyzing by regions, the North (96.2%), Northeast (97.7%), Midwest (98.2%), South (99.3%), and Southeast (99.4%) regions are observed. It is noteworthy that the North, Northeast, and Midwest regions have averages lower than the national average.

According to information from Panorama dos Resíduos Sólidos no Brasil of Associação Brasileira de Empresas de Limpeza Pública e Resíduos Especiais (ABRELPE) [33], the majority of MSW collected is properly disposed of in sanitary landfills. However, the improper dumping of MSW, when compared from 2010 to 2019, has not shown significant variations, decreasing from 43.2% of improper disposal in 2010 to 40.5% in 2019. This means that tons of MSW ended up in open dumps or controlled landfills that lack the necessary systems and measures to protect human health and the environment from harm and degradation.

In addition to data from ABRELPE, information from the Diagnóstico do Manejo de Resíduos Sólidos Urbanos [34], provided by the Sistema Nacional de Informações sobre Saneamento (SNIS) affiliated with the national government, corroborates the temporal stagnation of inadequate disposal practices concerning the collected municipal solid waste mass. As depicted in Figure 1, between 2012 and 2019, the percentage of SWS subjected to inadequate disposal exhibited minimal variation, reaching its lowest point in 2015 at 21.7%.

It is important to note that discrepancies between data from ABRELPE and the Diagnóstico do Manejo de Resíduos Sólidos Urbanos are attributed to methodological differences, including data collection, projections, and sample size. However, it is noteworthy that both sources align in showing the stagnation of inadequate waste disposal practices over time.

In Brazil, Law 12,305/2010 [28], which establishes the National Solid Waste Policy (PNRS), recognizes the economic and social potential of recyclable solid waste. Accordingly, this law prioritizes various aspects of recycling within Brazilian solid waste management. Law 11,445/2007 [29] sets standards for universalization, efficiency, and financial sustainability in solid waste services. Finally, Law 14,026/2020 [27] designates the end of 2033 as the target for universalizing these services, in alignment with Agenda 21, and, more recently, Decree 12,106/2024 [31], which regulated Law 14,260/2021 [30], which establishes incentives for the recycling industry.

However, recycling in Brazil remains relatively modest (collected = 93.0%; recycled = 1.47%; composted = 0.168%; landfilled = 61%; dumping = 39%) compared to that observed in developed countries, highlighting significant opportunities for expanding the actors involved in recycling processes and establishing structured municipal programs aimed at advancing expansion efforts [1], which represents a difficult challenge to be achieved, considering that the evident and widespread omission of Brazilian municipalities in fulfilling duties relating to the recovery of urban solid waste [35].

3. Efficiency, Productivity, and Recycling of Solid Waste Service

For at least three decades, numerous studies have focused on evaluating the efficient management of resources within the public sector from diverse perspectives and contexts [8,12,36,37,38,39,40,41,42,43,44,45]. In the MSW sector, despite being late when compared to the other sectors (e.g., energy, natural gas, water, wastewater, health, education, etc.,) recent studies have been found that present important insights.

Pérez-López et al. [39] contributed to the sector by analyzing cost efficiency by comparing the different ways (direct municipal, contracted municipal, inter-municipal cooperation, and contracted private production) of managing the service in Spanish locations. Analyzing 771 municipalities with populations ranging from 1000 to 50,000, during the period from 2007 to 2010, the authors show that the efficiency of service delivery calculated for the order-m method varies according to the size of the population, where municipalities with larger populations (above 20,000 inhabitants), the use of outsourced forms of management would be the most appropriate, while in municipalities with smaller populations, inter-municipal cooperation presents better results.

Also in the Spanish environment, Benito et al. [40] verified the impact of governance aspects (transparency and citizen participation) and other variables on the level of efficiency of MSW services. Analyzing 43 Spanish operators during the period from 2014 to 2018, the authors found low levels of efficiency in garbage collection and street cleaning services and, incredibly, transparency and citizen participation have little impact on efficiency. Other results indicate that population density, tourist activity in the municipality, and the strength of the local government are also significant in explaining the efficiency of the sector.

In China, Fan et al. [41] evaluated the efficiency of solid waste collection services in 30 provinces, covering the period from 2008 to 2017. By a stochastic frontier (SFA) approach, the average efficiency value was 0.372. Regarding influencing factors, the proportion of the population aged 15 to 64 indicates having a greater positive impact on efficiency, as well as GDP per capita, added value of the tertiary industry, and the level of education.

Studies for the Brazilian situation, such as those by Cetrulo et al. [43], Oliveira et al. [42], and Franco et al. [44] addressed the efficiency of the sector. In general, the cited studies conclude that the collection of urban solid waste needs to be more efficient, and new configurations would bring environmental, economic, and health gains. In addition, national policies (PNRS) have not achieved the desired changes, indicating that the country faces a huge problem in the proper management of municipal solid waste.

In this context, the problem of inadequate MSW disposal (Figure 1), the absence of selective collection and financial incentives for recycling practices, as well as the lack of political measures to increase the sector’s efficiency and productivity over the years, does not appear to be changing. Gonçalves et al. [45], using a dynamic network analysis, indicate an average inefficiency of around 42% in the sector from 2012 to 2022, with productivity remaining stagnant during this period. The study also shows that electoral cycles, GDP, population density, and COVID-19 influenced the combination of operational and economic efficiencies.

Incorporating selective collection (selective collection) and recycling into eco-efficiency measurement, Araya-Córdova [46] indicate improved efficiency among Chilean municipalities that adopt these practices. However, fewer than 5% of municipalities are efficient, which is consistent with the less than 50% probability of adopting recycling programs. Additionally, performance is influenced by exogenous factors, such as population, density, per capita waste generation, and tourism.

lo Sorto [47] presented convergent findings in Italy’s Apulia region, where low cost-efficiency and gains in productivity were observed from 2010 to 2017 during a transitional phase. These results later improved by including eco-efficiency and eco-productivity indicators such as totals for sorted waste (selective collection), unsorted waste, organics, plastic, metal, paper and cardboard, glass, and other materials in DEA estimates, underscoring the importance of environmental indicators in efficiency and productivity analysis.

Connecting cost efficiency with recycling, and waste-picker policies in the Indian cities of Delhi, Pune, and Mumbai results in government cost reductions of approximately USD 13,700 per day, USD 200,000 per year, and USD 9 per ton, respectively. This is due to waste-pickers handling collection, sorting, and transport at no cost to government agencies, and as noted in the literature, this also mitigates environmental costs and reduces landfill capacity issues [48]. Conversely, Cruz et al. [25] suggest that in Portugal, the cost per ton is EUR 86.1 higher for selective collection compared to undifferentiated collection.

Continuing on studies on environmental concerns, in China, the food recycling rate positively impacted MSW efficiency and productivity gains following the implementation of the 2017 waste separation policy. Public satisfaction, driven by economic and environmental factors, was also critical to efficiency, with its influence increasing as the recycling rate rose [49].

Hu et al. [37] evaluated the efficiency of recycling activities in Japan using the Network DEA model. The study analyzed the efficiency of 47 Japanese administrative regions between 2013 and 2017, breaking the process into two stages: waste collection (stage 1) and treatment and processing of recyclables (stage 2). The results indicated that waste collection efficiency is lower than processing efficiency, with metropolitan regions performing better than rural ones, especially in stage 1.

Recent research highlights efficiency in the recycling sector within the context of the circular economy, as in the study by Parte et al. [36]. Using methods such as data envelopment analysis and Tobit regression, the study revealed that larger companies are more efficient, supporting the hypothesis of economies of scale. Variables such as financial risk (rating score, Z-score) and capital structure influence efficiency. Recycling is identified as a strategic sector for the circular economy and the 2030 Agenda, with practical implications for public policies and sustainable business strategies.

Additionally, other studies discuss the role of various factors relevant to evaluating recycling efficiency and performance, such as Rijal et al. [38], who developed a recycling performance assessment procedure by sub-grouping MSWAs based on prominent local demographic characteristics, such as population density, proportion of elderly residents, tourism index, and so forth. The authors emphasized that the demographic characteristics of an area can have varying impacts on recycling performance due to differences in culture, politics, economics, and other factors.

Targeting the discourse into the scope of recovering recyclable materials in the Brazilian scenario, Rutkowski et al. [50] suggest that the use of waste pickers provides cost reductions and the efficiency of the system increases with the optimization of the use of trucks associated with the transfer system and efficiency gains in the operation of efficient collection means that take into account population density.

In this context, there is evidence of a relationship between efficiency and external factors such as population density, income, and COVID-19 in the efficiency of solid waste management. Regarding the effect of recovery of recyclable materials and recycling on the efficiency and productivity of solid waste services, the literature suggests improvements in efficiency and productivity, although the consensus remains undefined. Thus, this study contributes to the existing literature by discussing the scenarios and factors that affect the efficiency/inefficiency and productivity of selective collection in the services provided in Brazil and other emerging countries. In these nations, challenges such as inadequate infrastructure, low public awareness, and weak recycling policies hinder the proper provision of services.

4. Materials and Methods

4.1. Data Collection

The necessary information to calculate efficiency scores was collected from a database linked to the Brazilian federal government. The Sistema Nacional de Informações sobre Saneamento (SNIS) [51] compiles institutional, administrative, operational, managerial, economic–financial, accounting, and quality information related to the provision of basic sanitation services in urban areas across four components of the basic sanitation concept, according to Law 11,445 [29]: drinking water supply; wastewater; urban cleaning and solid waste management; and urban stormwater drainage and management. Municipalities are responsible for providing all input data related to solid waste [52].

The systematic procedure for data collection was as follows: (a) accessing the solid waste module (http://app4.mdr.gov.br/serieHistorica/, accessed on 6 November 2024), (b) searching for municipal management entities, (c) setting filters for reference year (2013 to 2022), (d) selecting operators with legal statuses such as direct public administration, autonomous agencies, public consortia, private companies, and public companies, (e) selecting operators geographically located in the Midwest, Northeast, North, Southeast, and South regions, and (f) selecting municipalities located in all states.

Given this, our study utilizes a balanced longitudinal panel, covering the period from 2013 to 2022. The sample includes service providers with more established characteristics of information completion, as well as providers from the main Brazilian cities, which involve a larger population. Our sample covers 67% of the population from all municipalities participating in the 2022 SNIS database, operating in 24 Brazilian states. The final sample analyzed 199 MSWs (DMUs).

4.2. Intertemporal Dynamic Efficiency and Productivity Model

According to the Relatório Nacional de Gestão de Resíduos Sólidos 2019 [53], only 45% of Brazilian municipalities charge fees for solid waste and urban cleaning services, with revenues covering just 44.1% of urban cleaning expenses. This revenue gap contributes to government deficits, and with fee increases unlikely in the short term, expense reduction in this public utility service emerges as the most immediate option. These expenses carry-over (intertemporal link) across periods, reflecting the minimum resources (workers, vehicles, etc.) required to maintain service levels. Given this premise, an ‘input-oriented’ focus is suitable, as total MSW expenditures are crucial in the intertemporal transfer of efficiency across periods (terms), making it essential for this phenomenon to be captured.

Traditional DEA models do not capture the effect of carry-over activities between consecutive periods, which is why the DSBM model was designed to address this intertemporal phenomenon [23]. In summary, the model proposed by Tone and Tsutsui [23] enables the assessment of overall efficiency across the entire analysis period and allows for the evaluation of dynamic and intertemporal changes over time. In this context, the DSBM model has been applied to assess efficiency in public service sectors, as seen in [54] and [55], where it was used to evaluate economic and financial efficiency in processes related to universal access and revenue collection for Brazilian water and sewage public services, respectively.

In addition, Addae et al. [56] utilized it to assess water-use efficiency across 11 West African countries. Lu et al. [57] focused on evaluating environmental energy efficiency among 48 high-income economies (including China), concentrating on environmental impact and emission reduction strategies. Guo et al. [58] analyzed efficiency based on CO₂ emissions from fossil fuels within Organization for Economic Co-operation and Development (OECD) countries and China. Hu et al. [59] investigated renewable energy generation efficiency in Japan, while Hu et al. [60] focused on renewable energy generation efficiency within Asian economies, and Gao et al. [61] explored efficiency through the water–energy–food nexus across Chinese provinces.

For the DSBM to estimate DMU efficiencies, at least one variable must be selected as a carry-over (link) function, with the remaining variables serving as typical DEA inputs and outputs. As per Tone and Tsutsui [23], the links can be classified as follows: (1) desirable (good) link, where the carry-over is treated as output, and any deficit indicates inefficiency; (2) undesirable (bad) link, where the carry-over is treated as input, and any surplus is considered inefficient; (3) discretionary (free) link, a carry-over under the DMU’s control that does not directly reflect efficiency but has an indirect impact across terms; and (4) non-discretionary (fixed) link, an uncontrolled carry-over that indirectly affects efficiency due to its continuity across terms. Given that operational expenses act as a linking activity, with increases leading directly to inefficiency, the DSBM classifies them as an undesirable (bad) link for this work.

Table 1. Definition of DSBM efficiency variables.

Acronyms	Description	DSBM Specification	References
WORKjt	Total number of (paid) workers in SWS, at the end of the reference year + temporary workers. Unit: number of workers. Data source: SNIS code = TB015 + TB017 + TB018 + TB019.	Input ${=x}_{1jt}$	[41,62,63]
VEHjt	Sum of vehicles (trucks, tractors, water vehicles, and animal traction) used in the SWS. Unit: number of vehicles. Data source: SNIS code = CO054 + CO081.	Input ${=x}_{2jt}$	[41,63]
OPEXjt	Annual value of the result of the sum of all operational expenditures with SWS. Data source: SNIS code = FN214 + FN217 + FN218. Unit: BRL (Reais).	Carry-over Undesirable (bad) $=z_{1jt}^{bad}$	[39,40,64]
SWCjt	Annual value of the sum of the total quantities of household waste and urban cleaning public waste collected by all agents. Data source: SNIS code = CO119. Unit: tons.	Output ${=y}_{1jt}$	[40,41,62]
PROjt	Number of active water consumption units, which were in full operation on the last day of the reference year (as a proxy for the total of the properties representing the set of: houses, apartments, businesses companies, public offices, and others within the area of SWS). Unit: consumption units. Data source: SINIS code = AG003.	Output ${=y}_{2jt}$	[65,66,67]

Notes: j = MWS (DMU) and t = year (term); OPEX were deflated by the Índice Nacional de Preços ao Consumidor Amplo (IPCA); OPEX as an undesirable (bad) transfer because it indicates undesirable transfers, where undesirable links are treated as inputs and its value is restricted to be no greater than observed. The study by Gonçalves et al. [45] applies a similar approach to variables related to expenses with MSW.

provides a detailed description of efficiency and productivity variables, highlighting their definitions, support based on previous literature, and notations aligned with the symbolic configurations outlined in Tone and Tsutsui [23]. Figure 2 depicts the dynamic DEA model DSBM employed to analyze Brazilian solid waste services.

The mathematical formulations and explanations within the DSBM model are transcribed from Tone and Tsutsui [23] (pp. 146–150), and Table 1 presents the variables’ correspondence with terms from the authors’ equations. The Malmquist score (productivity) is calculated using the same composition of efficiency variables, and productivity indices for 2013 were calculated using data from 2012. Productivity was also decomposed into its respective components: frontier shift (technological progress) and catch-up (efficiency changes). The DEA-Solver-PRO: Professional Version 11.0 software was used to calculate the efficiency studied and the productivity score. At the end of this process, four variables were obtained: the efficiency score of MSW j in year t (Eff_jt); the Malmquist productivity score of MSW j in year t (Prod_jt); the technical progress measured by the frontier shift score of MSW j in year t (Front_jt); and the efficiency gains measured by the catch-up score of MSW j in year t (Catch_jt).

The choice between constant returns to scale (CRS) and variable returns to scale (VRS) models was based on [68], following a paired t-test approach for two related groups: one with efficiency estimated generated using CRS, and another with VRS. When the mean efficiency estimated significantly differs, the VRS model is preferred. If the efficiency estimated does not show a normal distribution, the Wilcoxon signed-rank test is used as a non-parametric alternative for related samples, equally to that conducted by [54,55]. Regarding input–output protocols, the model met adequacy requirements with 199 DMUs, 2 inputs, 1 carry-over (considered an input), and 2 outputs. Kendall’s tau-b checks correlation (coefficient and significance) to confirm isotonicity, according to concerns raised by [69].

4.3. Relationships Between the Recovery of Recyclable Materials, Efficiency, and Productivity

4.3.1. Multiple Correspondence Analysis (MCA) and Multiple Independent Sample Tests

The MCA technique is applied for an initial analysis of perceptual interactions between the recovery of recyclable materials, efficiency, and productivity. In this context, Eff_jt, Prod_jt, Front_jt, and Catch_jt were categorized and ranked by interval according to Tukey’s hinges percentiles, as follows:

Eff_jt classification: values equal to 1 were classified in the ‘Efficient’ group, after which Tukey’s hinges percentiles were calculated using results with any level of inefficiency, producing classifications from the most to the least inefficient, following quartile order in the categories: ‘Largest inefficiencies’, ‘Moderate inefficiencies’, and ‘Smallest inefficiencies’.

Prod_jt classification: values equal to 1 were grouped as ‘Stability’. Tukey’s hinges percentiles were calculated for two groups (gains for values greater than 1 and losses for values below 1). Following the same quartile-based separation for each group, classifications were generated as: ‘Largest losses’, ‘Moderate losses’, ‘Small losses’, ‘Small gains’, ‘Moderate gains’, and ‘Largest gains’.

Front_jt classification, and Catch_jt classification: Values equal to 1 were classified as ‘No change’ and ‘No movement’, respectively, with Tukey’s hinges percentiles calculated similarly to Prod_jt. For frontier shift, this resulted in classifications of ‘Greatest decline’, ‘Moderate decline’, ‘Minor decline’, ‘Minor improvement’, ‘Moderate improvement’, and ‘Greatest improvement’, while for catch-up, the classifications were ‘Greatest setback’, ‘Moderate setback’, ‘Minor setback’, ‘Minor advance’, ‘Moderate advance’, and ‘Greatest advance’.

The recyclable material recovery rate was also classified (Rrec_jt classification) for this analysis, with zero values placed in the ‘Zero recovery’ group and non-zero values split into two groups based on the median: ‘Lower’ (50% with recovery rate < median) and ‘Higher’ (50% with recovery rate ≥ median). In this way, MCA was used to assess how these categories of recyclable material recovery align with categories related to efficiency and productivity.

Interactions between efficiency values, productivity, and recyclable recovery rates were further examined in a second round of perceptual analysis using multiple independent sample tests. Considering the classification of recyclable recovery rates, as previously defined into three groups, and the variables Eff_jt, Prod_jt, Front_jt, and Catch_jt, Kruskal–Wallis, median, and Jonckheere-Terpstra tests were independently conducted for each group. This second stage highlights associative interactions between recovery classifications and all efficiency and productivity outcomes derived through DSBM, completing this initial analysis of interactions based on pooled data.

4.3.2. Explanatory Econometric Model

Section 3 highlighted that selective collection tends to positively influence MSW efficiency, as supported by most of the literature referenced in this study, and that exogenous factors such as population density, income, and COVID-19 effects were relevant in evaluating MSW efficiency. The analyses outlined in Section 4.3.1 adopted a pooled data structure to present initial insights into the interactions between efficiency, productivity, and recyclable recovery rates.

In this third stage, an explanatory econometric analysis with a longitudinal panel data structure was developed (one for all jt cases and another for cases where Rrec_jt > 0). Four econometric models were proposed for each dependent variable: Eff_jt, Prod_jt, Front_jt, and Catch_jt. The independent variables include Rrec_jt (variable of interest) and the exogenous variables population density (Den_jt), GDP per capita (GDP_jt), and the period affected by the COVID-19 pandemic (COVID19_jt), as per the empirical foundation and descriptions provided in

Table 2. Definition of explanatory variables.

Acronyms	Description	Expectation	References
Rrecjt	Recovery rate of recyclable materials (excluding organic matter and waste) relative to the total amount collected. Data source: SNIS code = IN031. Unit: %.	(+)	Variable of interest
Denjt	Relation between the total number of inhabitants and the area of the municipality in km2. Unit: inhabitants/km2. Data source: Instituto Brasileiro de Geografia e Estatística (IBGE).	(+)	[40,45,70,71,72,73,74]
GDPjt	Ratio of the municipality’s gross domestic product (GDP) divided by the municipal population. Unit: BRL/Inhabitants. Data source: IBGE.	(+)	[45,71,72,73,75]
COVID19jt	Proposal for controlling the impact of COVID− 19 during the years 2020 to 2022. Unit:Binary.	(−)	[45]

Notes: j = MWS (DMU) and t = year (term). GDP were deflated by the Índice Nacional de Preços ao Consumidor Amplo (IPCA).

.

Formulations (1)–(4) represent the explanatory models for each case.

$$\mathrm{Model} 1\hspace{1em}{Eff}_{jt}={\beta }_1{Rrec}_{jt}+{\beta }_2{Den}_{jt}+{\beta }_3{GDP}_{jt}+{\beta }_4{COVID19}_{jt}+{\epsilon }_{jt}$$

(1)

$$\mathrm{Model} 2\hspace{1em}{Prod}_{jt}={\beta }_1{Rrec}_{jt}+{\beta }_2{Den}_{jt}+{\beta }_3{GDP}_{jt}+{\beta }_4{COVID19}_{jt}+{\epsilon }_{jt}$$

(2)

$$\mathrm{Model} 3\hspace{1em}{Front}_{jt}={\beta }_1{Rrec}_{jt}+{\beta }_2{Den}_{jt}+{\beta }_3{GDP}_{jt}+{\beta }_4{COVID19}_{jt}+{\epsilon }_{jt}$$

(3)

$$\mathrm{Model} 4\hspace{1em}{Catch}_{jt}={\beta }_1{Rrec}_{jt}+{\beta }_2{Den}_{jt}+{\beta }_3{GDP}_{jt}+{\beta }_4{COVID19}_{jt}+{\epsilon }_{jt}$$

(4)

To estimate Models 1 through to 4, we utilized generalized estimating equations (GEE), which are the longitudinal panel data-based version of generalized linear models (GLM) [76]. GEEs was selected in this study due to their suitability for the nature of the dependent variables Eff_jt, Prod_jt, Front_jt, and Catch_jt. Several applications of GEE as an explanatory econometric model exist in contexts where the dependent variable is derived from data envelopment analysis, such as in Jalali Naini et al. [77] for efficiency in the Iranian insurance sector, Dohmen et al. [78] for hospital efficiency in the Netherlands, Barbosa et al. [79] and Dantas et al. [80] for Brazilian football club efficiency, Barbosa et al. [54] and Barbosa et al. [55] for the efficiency of Brazilian water utilities, and most recently, within the focus sector of this study, Gonçalves et al. [45] for the efficiency of Brazilian municipal solid waste services.

For GEE model estimation, certain configurations are necessary and applicable in this study: (1) distribution family (e.g., Gaussian, binomial, gamma), (2) link function (e.g., identity, log, logit, probit), and (3) correlation structure within the group or working correlation matrix (e.g., exchangeable, independent, unstructured, autoregressive, stationary). The configuration that yields the lowest quasi-likelihood under the independence model criterion (QIC) is the most robust and selected for explanatory purposes [81].

In Model 1, Eff_jt is classified as efficient or inefficient, fitting a binomial distribution family, as seen in [45,54,55,79]. We tested identity and logit link functions. For Models 2 through to 4, we assumed that Prod_jt, Front_jt, and Catch_jt follow a Gaussian distribution family with an identity link function, as proposed in [80]. We evaluated all possible combinations of correlation matrices, distribution families, and link functions, selecting the parameterization that produced the lowest QIC to ensure robustness in the analysis. To reduce the impact of outliers, a 5% winsorization process was applied to all variables except for Eff_jt, which was excluded due to the binomial distribution assumed in the GEE model.

5. Results and Discussion

Table 3. Descriptive statistics of DSBM efficiency variables.

Year	WORKj				VEHj				OPEXj
	Mean	Std. Dev.	Min.	Max.	Mean	Std. Dev.	Min.	Max.	Mean	Std. Dev.	Min.	Max.
2013	446	1784	4	23,438	23	57	1	515	34,068,432	138,099,528	38,658	1,762,331,354
2014	449	1894	4	25,191	26	73	1	719	36,681,773	156,144,497	89,298	2,035,287,398
2015	423	1735	3	22,957	25	66	1	736	35,997,180	153,567,001	101,384	2,027,128,388
2016	431	1845	3	24,521	26	70	1	652	36,045,939	155,759,851	115,982	2,057,013,915
2017	408	1701	3	22,586	23	53	0	531	34,582,776	151,484,925	89,268	2,011,724,120
2018	414	1676	4	22,314	24	53	1	512	36,594,046	156,586,163	68,407	2,076,430,101
2019	411	1672	4	22,171	26	64	1	596	38,216,872	164,477,902	83,500	2,173,975,757
2020	426	1692	3	22,154	26	64	1	596	38,377,389	159,070,172	80,080	2,077,658,177
2021	430	1650	2	21,571	26	56	1	554	36,369,605	142,727,421	79,872	1,838,305,487
2022	446	1691	2	22,039	26	55	1	533	42,543,158	166,743,559	92,636	2,160,320,745
Year	SWCj								PROj
	Mean	Std. Dev.	Min.	Max.					Mean	Std. Dev.	Min.	Max.
2013	87,496	310,329	168	3,465,187					72,824	201,475	244	2,219,225
2014	88,972	305,125	126	3,324,526					74,779	204,559	256	2,234,347
2015	89,304	294,508	148	3,273,256					77,064	214,635	261	2,396,224
2016	83,370	276,194	158	3,146,851					78,493	217,203	247	2,423,676
2017	80,573	261,847	109	3,016,556					79,722	218,700	245	2,433,242
2018	83,665	269,533	131	3,079,233					80,849	220,467	278	2,450,920
2019	88,541	284,245	155	3,223,437					82,813	230,242	278	2,613,356
2020	92,362	292,596	174	3,229,449					85,948	237,654	0	2,715,431
2021	90,685	282,012	84	2,985,664					88,082	239,570	267	2,717,529
2022	84,130	265,931	85	3,133,457					85,957	214,399	276	2,189,367

presents the descriptive statistics for variables used to assess DSBM efficiency in waste management across the years 2013 through to 2022. However, the mean of OPEX in 2022 stands out, reflecting a year still impacted by COVID-19 and signaling both stability in productivity and a negative effect during the pandemic period.

The standard deviations shown in Table 3 tend to reject the assumption of a normal distribution for WORK_jt, VEH_jt, OPEX_jt, SWC_jt, and PRO_jt, which was confirmed by the Kolmogorov–Smirnov and Shapiro–Wilk tests, both yielding significance values of p < 0.01. In this context, Kendall’s tau-b nonparametric correlation should be applied to verify isotonicity protocol adherence.

Table 4. Kendall’s tau-b correlations coefficients between DSBM efficiency variables.

		WORKjt	VEHjt	OPEXjt	SWCjt
WORKjt	Coefficient		0.710 **	0.808 **	0.793 **
	sig. (two-tailed)		0.000	0.000	0.000
VEHjt	Coefficient	0.710 **		0.702 **	0.707 **
	sig. (two-tailed)	0.000		0.000	0.000
OPEXjt	Coefficient	0.808 **	0.702 **		0.823 **
	sig. (two-tailed)	0.000	0.000		0.000
SWCjt	Coefficient	0.793 **	0.707 **	0.823 **
	sig. (two-tailed)	0.000	0.000	0.000
PROjt	Coefficient	0.786 **	0.708 **	0.827 **	0.864 **
	sig. (two-tailed)	0.000	0.000	0.000	0.000

Note: **. Correlation is significant at the 0.01 level (two-tailed).

presents the results of these correlations, all significant at the 0.01 level, with the lowest observed correlation between VEH_jt and SWC_jt (input versus output) at 0.707, confirming that a proportional increase in input results in a proportional rise in output, and thus meets isotonicity protocol requirements.

For choosing between CRS or VRS assumptions, efficiency was estimated under both returns.

Table 5. Nonparametric tests of differences between DSBM scores under CRS and VRS assumptions.

Wilcoxon Signed Ranks Test
	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
Z (Based on negative ranks.)	−10.961	−11.534	−11.634	−11.517	−11.583	−11.570	−11.482	−11.548	−11.557	−11.297
Asymp. sig. (two-tailed)	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000

displays the nonparametric Wilcoxon signed ranks tests, comparing DSBM efficiency scores estimated under each returns assumption year by year in the panel. The results support selecting VRS as all significance values were p < 0.01 across all years, leading to the rejection of the null hypothesis H₀, indicating that efficiency estimates differ when calculated under different returns to scale assumptions and establishing VRS as the more robust choice, following the rule proposed by [68].

Regarding the specifications of the DSBM for estimating efficiency and productivity, the ‘input-oriented’ option is explained in Section 4.2, and Table 4 provides the results justifying the use of VRS. Consequently, the intertemporal dynamic efficiency and productivity model follow the parameters of being ‘input-oriented’ with ‘Variable Returns to Scale (VRS)’.

Table 6. Descriptive statistics of Eff_j, Prod_j, Front_j, and Catch_j by year and overall.

	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	Overall
Effj											Average
Mean	0.628	0.659	0.642	0.637	0.642	0.628	0.585	0.582	0.586	0.602	0.619
Efficient (% MSW)	18.1%	20.6%	19.6%	16.6%	15.6%	19.1%	14.1%	19.1%	17.6%	19.1%	4.0%
Prodj											Geometric Mean
Mean	1.069	1.025	1.020	1.027	1.050	0.985	1.036	1.132	1.013	0.919	0.997
Gain (% MWS)	58.8%	54.8%	53.3%	52.3%	55.8%	45.7%	54.8%	57.3%	47.2%	23.6%	43.7%
Loss (% MWS)	40.7%	44.7%	45.7%	47.7%	44.2%	54.3%	44.2%	41.2%	52.3%	75.4%	55.8%
Frontj
Mean	0.955	0.976	1.044	1.009	1.022	0.998	1.102	1.104	0.997	0.870	0.995
Improv. (% MWS)	30.7%	35.2%	73.9%	66.3%	55.8%	53.3%	63.3%	68.8%	49.2%	7.5%	33.2%
Decline (% MWS)	68.8%	64.3%	25.6%	33.2%	44.2%	46.7%	35.7%	29.6%	50.3%	91.5%	66.8%
Catchj
Mean	1.122	1.054	0.985	1.016	1.033	0.991	0.956	1.016	1.031	1.066	1.002
Advance (% MWS)	60.3%	49.7%	34.2%	44.2%	46.2%	42.2%	35.7%	40.2%	46.2%	55.8%	47.7%
Setback (% MSW)	26.1%	33.7%	50.3%	43.2%	43.7%	44.7%	52.8%	47.7%	40.2%	29.6%	44.2%

presents the descriptive statistics of efficiency and productivity results (by year and overall).

Analyzing the results in Table 6, the average intertemporal dynamic efficiency over the ten-year panel is 61.9%, with only 4% of MSWs achieving the efficiency frontier each year. In the year-by-year evaluation, 2020 registered the lowest average efficiency, while 2019 had the fewest MSWs reaching the efficiency zone. In contrast, 2014 showed the highest average efficiency and the most DMUs on the efficiency frontier.

From the productivity perspective, measured by the Malmquist score (Prod_jt), the results indicate an accumulated average productivity loss near the stability zone over the ten years, with the largest loss occurring from 2021 to 2022 at −8.1% (0.919–1). This decline is attributed to a greater technological regression (Front_jt = frontier shift score) recorded in the same period [= −13% → (0.87–1)], where only 7.5% of MSWs experienced technological improvement. The Catch_jt (catch-up score) showed minimal fluctuation throughout the years, with an accumulated average efficiency gain close to a no-movement condition.

Notably, 17.9% of MSWs reached the efficiency frontier at least once during the panel period (

Table 7. Descriptive statistics of Eff_jt, Prod_jt, Front_jt, and Catch_jt classifications.

Effjt Classification			Prodjt Classification			Frontjt Classification			Catchjt Classification
	Freq.	%		Freq.	%		Freq.	%		Freq.	%
Largest inefficiencies	409	20.6	Largest losses	244	12.3	Greatest decline	244	12.3	Greatest setback	205	10.3
Moderate inefficiencies	815	41.0	Moderate losses	488	24.5	Moderate decline	487	24.5	Moderate setback	410	20.6
Smallest inefficiencies	409	20.6	Small losses	244	12.3	Minor decline	244	12.3	Minor setback	205	10.3
Efficient	357	17.9	Stability	12	0.6	No change	12	0.6	No movement	265	13.3
			Small gains	251	12.6	Minor improvement	252	12.7	Minor advance	227	11.4
			Moderate gains	500	25.1	Moderate improvement	500	25.1	Moderate advance	451	22.7
			Largest gains	251	12.6	Greatest improvement	251	12.6	Greatest advance	227	11.4
Total	1990	100.0		1990	100.0		1990	100.0		1990	100.0

). The data also reveal a balance among DMUs in terms of productivity gains and losses, technological decline and improvement, as well as setbacks and advances toward the efficiency frontier.

Regarding the recovery rate of recyclable materials (Rrec_jt), the categorization from Section 4.3.1 was distributed among MSWs as follows: zero recycling = 417 (21.0%); ‘Lower’ = 747 (37.5%); and ‘Higher’ = 748 (37.6%), indicating that over 20% of MSWs in the sample conduct indiscriminate collection. Data for the Rrec_jt variable were unavailable for 78 observations (3.9%), which were excluded from the analyses. Figure 3 and Figure 4 illustrate the perception charts generated by the multiple correspondence analysis technique, showing associations for (Eff_jt classification × Prod_jt classification × Rrec_jt classification) and (Front_jt classification × Catch_jt classification × Rrec_jt classification), respectively.

Figure 3 shows various interaction perceptions from MCA with stacked panel data. Notable patterns include higher levels of the recovery rate of recyclable materials (Rrec) located in the fourth quadrant, intersecting with the moderate inefficiency category, and ‘Lower’ Rrec levels positioned in the third quadrant, aligning closely with the lowest inefficiency levels. Efficient MSWs are situated in the second quadrant, at a somewhat greater distance from MSWs engaging in indiscriminate collection. These interactions yielded a chi-square of 75.358, which was significant at 1% for the association between Eff_jt classification and Rrec_jt classification, indicating a certain alignment between selective collection and inefficiency.

Productivity, measured by the Malmquist score, follows a more distant trajectory from Rrec than efficiency, with moderate and high productivity gains in the third quadrant, aligning with low selective collection levels, though with some separation. MSWs practicing indiscriminate collection (zero) appear in the second quadrant, close to the lowest productivity gains. Productivity losses are found in quadrants separate from the three Rrec_jt categories. These results suggest a lack of association between productivity and the recovery rate of recyclable materials, confirmed by a chi-square of 20.455 with a significance level > 5% for the association between Prod_jt classification and Rrec_jt classification.

Regarding the association between productivity and efficiency, productivity losses correlate in the first quadrant with the highest inefficiencies, while efficient MSWs are closer to modest productivity gains. This understanding is supported by a chi-square of 153.808, significant at 1%, for the association between Eff_jt classification and Prod_jt classification.

In Figure 4, perceptions of interactions between selective collection and productivity components decomposed into frontier shift and catch-up are highlighted by two key findings: (1) the proximity between higher Rrec levels and smaller technological declines (in the fourth quadrant), with significant associations between category groups, as evidenced by chi-square values of 54.833 and 45.793, both with sig. < 0.01 for Front_jt classification × Rrec_jt classification and Catch_jt classification × Rrec_jt classification, respectively; and (2) an inverse association regarding productivity contributions in the interactions between Frontier-shift and Catch-up, supported by a Chi-square of 429.396 with sig. < 0.01 for the association between Front_jt classification and Catch_jt classification.

This suggests an inverse association related to benefits to productivity, where efficiency gains may reduce the technology frontier and vice versa, a phenomenon common due to the natural compensation between these productivity components, emphasizing the complexity of interactions between efficiency changes and technological advances.

Diving deeper into the analysis of interactions between efficiency, productivity, and recyclable material recovery, a non-parametric test approach for three independent recovery samples (zero or indiscriminate collection, lower, and higher) is presented in

Table 8. Nonparametric tests of Eff_jt, Prod_jt, Front_jt, and Catch_jt by Rrec_jt classification.

	Recovery of Recyclable Waste Classification				Test—Asymp. Sig.
	Classif.	N	Mean	Mean Rank	Kruskal–Wallis	Median	Jonckheere–Terpstra
Efficiency score	Zero	417	0.633456	996.27
	Lower	747	0.663958	1067.83	0.000	0.000	0.000
	Higher	748	0.559180	823.15	H0 rejected
Malmquist productivity score	Zero	417	1.035847	978.26
	Lower	747	1.027039	954.11	0.639	0.932	0.393
	Higher	748	1.019867	946.76	H0 retain
Frontier shift score	Zero	417	1.002799	941.27
	Lower	747	1.015845	977.81	0.400	0.717	0.765
	Higher	748	0.999977	943.71	H0 retain
Catch-up score	Zero	417	1.038830	977.49
	Lower	747	1.020588	952.33	0.675	0.529	0.467
	Higher	748	1.026247	948.96	H0 retain

.

From this perspective, higher inefficiencies are associated with elevated levels of recyclable material recovery, as H₀ was rejected in all tests conducted, aligning with the MCA visualizations shown in Figure 3. However, productivity measures—whether by Malmquist or decomposed—do not appear to have significant associations with recyclable material recovery. This is also consistent with the MCA results shown in Figure 3 yet diverges from the significant associations depicted in Figure 4.

More robust interactions are analyzed from an econometric perspective, using a longitudinal panel structure and incorporating exogenous control variables established in the previous literature.

Table 9. Descriptive statistics of econometric variables.

Variable	Obs.	Mean	Std. Dev.	Min	Max
All Observations
Rrecjt	1990	4.6642	6.1135	0.0000	20.3500
Denjt	1990	578.1476	1118.9330	8.6579	4186.1910
GDPjt	1990	35.1491	17.7141	11.2934	81.0735
COVID19jt	1990	0.3000	0.4584	0.0000	1.0000
Cluster with recovery rate greater than 0
Effjt	1495	0.6115	0.2393	0.1251	1.0000
Prodjt	1495	1.0118	0.1964	0.6630	1.4664
Frontjt	1495	1.0032	0.1258	0.7659	1.2708
Catchjt	1495	1.0195	0.1997	0.6652	1.4669
Rrecjt	1495	6.2085	6.3375	0.0100	20.3500
Denjt	1495	653.8314	1170.9020	8.6579	4186.1910
GDPjt	1495	37.1208	17.8387	11.2934	81.0735
COVID19jt	1495	0.3251	0.4686	0.0000	1.0000

presents the descriptive statistics of the explanatory econometric model variables, including dependent variable descriptions for the clusters that do not perform indiscriminate collection, which are not shown in Table 6.

Table 10. Results of the GEE regression models (all observations selective and indiscriminate).

	Model 1	Model 2	Model 3	Model 4
	Eff	Prod	Front	Catch
Rrec	−0.001430	−0.000502	−0.000630	0.000118
	−1.07 (0.285)	−0.98 (0.325)	−2.57 (0.010) ***	0.24 (0.813)
Den	0.000055	0.000005	0.000007	−0.000003
	3.31 (0.001) ***	1.8 (0.072) *	4.96 (0.000) ***	−1.14 (0.253)
GDP	−0.001372	−0.000152	0.000254	−0.000457
	−2.31 (0.021) **	−0.99 (0.321)	3.28 (0.001) ***	−2.97 (0.003) ***
COVID-19	−0.031244	−0.026441	−0.018829	0.005265
	−2.41 (0.016) **	−3.53 (0.000) ***	−4.93 (0.000) ***	0.63 (0.532)
Cons	Marginal effects	0.029450	0.005711	1.033675
		4.55 (0.000) ***	1.86 (0.063) *	156.43 (0.000) ***
Robustness analysis
Wald chi2(9)	27.99	25.72	74.09	10.71
Prob > chi2	0.000 ***	0.000 ***	0.000 ***	0.030 **
QIC (smallest)	2611.230	82.587	33.244	84.109
QICu	2610.450	88.037	40.295	89.678
Correlation	Exchangeable	ar9	ar3	ar6
Family	Binomial	gaussian	gaussian	gaussian
Link	Logit	Log	Log	Identity
Number of observations: 1990
Number of groups: 199
Variance inflation factor (mean): 1.05

Note: *** significance at 1%; ** significance at 5%; * significance at 10%.

and

Table 11. Results of the GEE regression models (selective collections cluster).

	Model 1	Model 2	Model 3	Model 4
	Eff	Prod	Front	Catch
Rrec	−0.002895	−0.000502	−0.001034	0.000807
	−2.12 (0.034) **	−0.98 (0.325)	−3.36 (0.001) ***	1.46 (0.145)
Den	0.000053	0.000005	0.000006	−0.0000002
	3.08 (0.002) ***	1.8 (0.072) *	3.74 (0.000) ***	−0.05 (0.959)
GDP	−0.001501	−0.000152	−0.000012	−0.000423
	−2.35 (0.019) **	−0.99 (0.321)	−0.13 (0.894)	−2.23 (0.026) **
COVID-19	−0.035468	−0.026441	−0.030293	0.020089
	−2.28 (0.022) **	−3.53 (0.000) ***	−5.94 (0.000) ***	2.26 (0.024)**
Cons	Marginal effects	0.029450	1.016779	0.023074
		4.55 (0.000) ***	217.38 (0.000) ***	2.61 (0.009) ***
Robustness analysis
Wald chi2(9)	30.32	18.75	86.34	12.17
Prob > chi2	0.000 ***	0.001 ***	0.000 ***	0.016 **
QIC (smallest)	1971.545	62.009	26.737	64.124
QICu	1971.887	67.330	33.209	69.417
Correlation	Exchangeable	Exchangeable	Exchangeable	Exchangeable
Family	Binomial	Gaussian	Gaussian	Gaussian
Link	Logit	Log	Identity	Log
Number of observations: 1495
Number of groups: 180
Variance inflation factor (mean): 1.09

Note: *** significance at 1%; ** significance at 5%; * significance at 10%.

display the results of the explanatory econometric models, covering all observations and specifically the cluster of MSWs engaged in recyclable material recovery, respectively.

Considering the broader econometric perspective presented in Table 10, which includes MSWs that perform only indiscriminate collections, recyclable material recovery does not significantly influence efficiency; however, it negatively interacts with technological progress (frontier shift). Population density has a positive impact on both efficiency and technological advancement. Regarding income, GDP is negatively related to efficiency and changes over time (catch-up), yet it supports improvements in technological progress.

The COVID-19 pandemic period (2020–2022) had a negative impact on the efficiency and productivity of MSW management services, as indicated by the Malmquist score and technological advancements. These years are considered atypical due to factors such as social isolation, changes in consumption and procurement patterns, and the increased demand for MSW management services. Although the impact was accounted for through the inclusion of a specific variable, it is acknowledged that the analysis of these effects can be explored in greater detail in future studies, particularly regarding the nature and magnitude of the observed changes.

Summarizing the results for all MSWs, the findings suggest a lack of interaction between efficiency and selective collection, as significant explanatory power at the 5% level is concentrated in the exogenous control variables. This understanding similarly extends to productivity, as measured by Malmquist and efficiency changes across periods. Advances toward the frontier are also well explained by the exogenous variables; however, the negative interaction with material recovery allows us to suggest that recyclable material recovery does not enhance the odds of MSWs becoming more efficient and productive.

Comparing only MSWs that recover recyclable materials, Table 11 shows a negative interaction between the recovery rate and intertemporal efficiency, results consistent with findings in Cruz et al. [25] but differing from those in Sharholy et al. [48], Araya-Córdova [46], lo Sorto [47], and Xiao et al. [49]. As highlighted in the literature, environmental (exogenous) variables proved relevant for comparing efficiency scores, all with the same directional influences (positive or negative) as reported for results involving all observations.

In terms of productivity, findings related to the Malmquist measure continued to show only the COVID-19 period’s relevance in degrading this productivity metric. Recyclable material recovery also interacts negatively with technological progress, contrary to [43], with density exerting a mild influence. Material recovery does not interact with efficiency change, which is negatively impacted by GDP and positively affected by the COVID-19 period—the only model with this directional outcome, explained by the natural compensation between productivity components.

These findings suggest that MSWs adopting recycling policies are less efficient as they deepen their commitment to such selective collection policies and are more prone to technological decline. They also suggest the effects of density, income, and COVID-19 on efficiency, with density being significant for technological advancement and COVID-19 linked to decline.

6. Conclusions

In many countries, including those in the BRICS, the primary method for managing solid waste involves sending it to landfills. Thus, it seems logical that the fewer solid waste materials disposed of in landfills, the lower the transportation, maintenance, and landfill creation costs, resulting in better efficiency and productivity for municipal solid waste services. In this regard, this study aimed to analyze the interactions between intertemporal efficiency, productivity, and the recovery of recyclable materials in the Brazilian MSWs.

What distinguished this work from the previous literature was the use of a dynamic intertemporal approach, with the premise that operational expenditures (OPEX) represent unwanted connections from one year to the next, as well as the analysis of the recovery of recyclable materials as an explanatory factor for the resulting efficiency and productivity.

The sector’s average inefficiency (39%) and accumulated productivity loss over ten years raise concerns about compliance with SDGs 12 and 6. Contrary to expectations, the study’s results suggest that the more solid waste recovery policies implemented by MSWs, the less efficient the management of solid waste becomes. Regarding productivity, the recovery policy appears to be causing technological regressions, despite advancements in Brazilian legal frameworks. Insights from these findings raise concerns about generating incentives to advance recycling policies, as they cast doubt on the economic efficiency and technological advancements concerning the dilemma between indiscriminate and selective collection, and whether the recovery of recyclable materials will prove economically viable.

Contextual factors played a significant role, with population density and income levels influencing efficiency, while the COVID-19 pandemic negatively impacted both efficiency and productivity, reinforcing their importance in shaping benchmarking policies for municipal solid waste services.

Policy and practical implications can be drawn from the results of this work, such as:

(a)

Coincidentally or not, the worst year for productivity was 2022, coinciding with the first full year under Law 14,260/2021 [30], enacted on 8 December 2021, aimed at generating incentives for the Brazilian recycling industry. Insights suggest that, as of 2022, we have not yet seen positive impacts from the law from a scientific perspective. Whether the newly issued regulatory decree for this law (Decree 12,106/2024 [31]) will provide sufficient incentives to change the undesirable trend remains uncertain.

(b)

Law 11,445/2007 [29], which established minimum regulatory requirements, including cost efficiency, is another framework that our results call into question. Insights reveal that our findings do not reflect practical improvements in the continued pursuit of increased efficiency, particularly given that 2033 is the deadline set by Law 14,026/2020 [27] for the universalization of basic sanitation services in Brazil, of which solid waste management is a part.

(c)

Regulatory agencies, municipal managers, and social control bodies can use these findings to delve deeper into the efficiency and productivity landscape of Brazilian solid waste services to organize effective public and regulatory policies.

This study focused on the consumption of resources aimed at managing specific volumes of solid waste and the households served, without exploring the dimension of municipal revenues associated with selective waste collection, such as industrial fees and carbon credit transactions, among others. This limitation highlights a significant gap for future research.

Therefore, it is recommended to conduct studies analyzing the relationship between revenue collection from urban solid waste services and the efficiency of service delivery, also considering their impacts and contributions to the recycling industries. Such research is particularly relevant for urban solid waste operators in emerging countries that are still in the early stages of implementing revenue collection mechanisms for these services. Additionally, the application of stochastic frontier analysis (SFA) to compare and validate the results obtained in this study emerges as another promising direction for future research.

In the political–institutional sphere, normative studies are suggested to evaluate strategies for strengthening incentives for recycling. In contexts similar to Brazil, it is crucial to deepen the debate on defining benchmarks for the sector. This includes, among other aspects, analyzing the allocation of public funds to recycling initiatives and the role of regulatory agencies in promoting service efficiency, topics that demand further attention and contributions.