Quantitative Analysis of Colombian Waste Picker’s Profile

Abstract

Even though waste pickers play a crucial role in the urban solid waste management system in developing countries, their social, familial, and labor conditions remain uncertain. In this study, we analyzed the profiles of waste pickers in Colombia using multivariate statistical methods and cluster analysis. Our findings indicate that the majority of waste pickers have only completed elementary or high school education, and most of them are not associated with any organization and belong to the subsidized health regime. We identified five profiles of waste pickers in the population. The first consists of the most experienced individuals, while the second comprises individuals with the lowest salaries. The third includes older individuals who work more hours per day. The fourth is made up of individuals who work in the informal sector, and the fifth includes individuals who work in the formal sector. Our study highlights specific actions to be taken for each profile. Improving family and household conditions, social and educational status, and job conditions can have a significant impact on their lives.

1. Introduction

Municipal solid waste (MSW) management has become a severe and pervasive problem in urban agglomerations [1]. The Organization for Economic Cooperation and Development (OECD) estimated that 0.7 billion tons of waste was generated in 2010 in its member countries alone. The MSW is expected to increase by 16% by 2030 and 36% by 2050 [2]. In addition, the World Bank (WB) has estimated that global waste generation will increase by 29% by 2030 and 69% by 2050 (a value close to 3.5 billion tons) [3]. This increase in MSW generation can be attributed to different causes, such as design errors, because the products were designed from their conception to be garbage, as is the case of wrappers and packaging [4]. Another cause is the phenomenon of consumerism. The acquisition of more goods, products, and services has contributed to an increase in the generation of waste [5]. Additionally, poor or nonexistent habits in the separation and disposal of waste contribute to the accumulation of garbage [6]. Morais et al. [7] stated that another cause of the increase in MSW generation is rural–urban migration. This phenomenon has increased in recent decades, leading to a greater concentration of people in large cities and an uncontrolled increase in MSW, causing formal waste management systems to reach saturation levels [7]. This case is also prevalent in Colombia, where almost two million refugees and migrants from Venezuela have arrived in the country [8]. Despite having formal MSW management systems in developing countries, it is estimated that 80% of MSW (municipal solid waste) goes to final disposal (landfills), 7% has poor disposal (dumpsites), 1% is used for energy generation, and 12% is recycled [9].

The main stages of an MSW management process in developing countries are depicted in Figure 1. Urban solid waste (USW) is generated from various sources such as households, industries, commercial centers, and hospitals [10,11]. The USW is accumulated in collection points under uncertain conditions, where waste pickers play a crucial role in their management [12]. These workers carry out the collection and preparation activities of the material, which will be delivered to collection centers (CCs) [13]. In these centers, the material is separated, classified, and stored for its subsequent incorporation into the production chain [14]. However, in cases where the material cannot be reused, it is transported for final disposal [15]. In this way, the work of waste pickers not only contributes to environmental protection but also to the circular economy and job creation in these regions [10].

Waste pickers are essential actors in informal waste management systems due to their contribution to improving environmental health indicators, sustainability, and the reduction in discarded material flows. This, in turn, enables an increase in the efficiency of natural resources and the closure of the loop in a circular economy through processes of reuse, recovery, and recycling [17]. According to the International Labor Organization, waste pickers represent between 15 and 20 million people worldwide [18]. However, there are no accurate and updated landscapes worldwide. Consequently, the number could be higher considering the informal recycling sector in developing countries, where much of the recycling population is concentrated. The informality rate in the solid waste sector could be as high as 90% [19]. Waste pickers may include women, children, older adults, or migrants [11]. Although informal waste pickers are prevalent in developing countries, they are also found in developed countries on a much smaller scale [20,21]. Waste pickers are exposed to challenging work conditions that can lead to health problems and car accident injuries. Despite working in major cities and industrial centers, most waste pickers are impoverished and need help to access opportunities that could improve their circumstances. These characteristics are prevalent among waste pickers in various countries, including Argentina [22], Colombia, Ecuador [23], Philippines and Indonesia [24], India [25], and Perú [26].

These main characteristics of the waste picker population have been studied by academics, public policy makers, and non-profit civil [27,28]. Thus, in Colombia, initiatives have been promoted to integrate the informal recycling sector with the formal sector, as well as public policy strategies [29]. However, these initiatives have not had a sufficient impact in terms of establishing an efficient and sustainable MSW able to improve the living conditions of the waste picker population. This has been, in part, due to the limited information about the waste picker population in terms of their main characteristics and spatial distribution [30,31]. In this regard, the “Inclusive and formal economy program alliance” (Alianza EFI, in Spanish), a consortium of twenty-one institutions among academics, business, and government actors, conducted a survey in 2019. The main purpose was to identify the main characteristics and the spatial distribution of the formal and informal waste pickers in the main cities of the Colombia territory [32].

Regarding the studies related to the waste picker population worldwide, several methods of data analysis have been employed [33,34,35]. However, most research is limited to the descriptive analysis of groups of waste pickers [36,37,38,39], and some modeling techniques for the analysis of population variables [40,41,42]. In this regard, the article presents a detailed analysis of the Colombian recycling population based on the information collected by the Alianza EFI. Initially, in this work, we applied Multiple Correspondence Analysis (MCA) to select the variables that provide the most information to the data set to avoid over-specification. Secondly, we performed dimension reduction with Principal Component Analysis (PCA) using the variables selected in the MCA to avoid redundancy and find similarities between the variables. Finally, we performed cluster analysis using the dimensions resulting from the PCA, which incorporate the relationships between the variables, to incorporate the similarities between the individuals and build the groups that would allow us to build the profiles of the waste pickers through the interpretation of these groupings. It is expected that the present study in a developing country such as Colombia can be used as a proxy study for identifying similar populations in countries with comparable characteristics.

The paper is organized as follows: Section 2 describes the methods and variables used to quantify and characterize the study population. Section 3 presents the analytical results of the proposed methodology. Finally, Section 4 offers discussions of research conducted in other regions of the world regarding the case study followed.

2. Materials and Methods

This section is divided into two subsections, the first of which describes the study area of the recycling population with a brief definition of each survey variable. The second section outlines the proposed analysis framework, as shown in Figure 2.

2.1. Participants and Procedure

The data were collected by surveying waste pickers in different country regions, such as Zipaquirá, Girardot, Bucaramanga, Ibagué, Barranquilla, Bello, Neiva, Soacha, Pereira, Medellín, Santa Marta, and Cartagena. Participation in the study was voluntary, and the waste pickers were given a contextualization of the study’s objectives.

Description of Data

The sample of participants was 1740 men, 590 women, and five other genders. We made efforts to interview as many waste pickers as possible.

Table 1. Variables for multivariate statistical methods.

Dimension	Variable	Description	Type
Education	literacy	If illiterate	Binary
	education_level	Level of education	Nominal
Job	training	Has received training to carry out the work	Binary
	residential	If you collect the material at residential location	Binary
	days_working	Days worked per week	Continuous
	transport	Type of transport	Nominal
	your own	Transport is own	Nominal
	time_recycling	Years of recycling	Continuous
	hours_working	hours worked per day	Continuous
	association	Waste picker is linked to any association	Binary
	road	If you pick up the material on a road	Binary
	residential	If you collect the material at a residential location	Binary
	route	Has an established route	Binary
	mall	If you pick up the material in a shopping mall	Binary
	office	If you pick up the material at an office	Binary
	industry	If you collect the material in an industrial area	Binary
	neighborhood	If you collect the material in a neighborhood	Binary
	other	Other collection site	Binary
	other_job	Has another job	Binary
	income	Monthly recycling income	Binary
	Reason	Reason for doing this work	Nominal
Health	condition	If you have a disability	Binary
	ss	If you have a disability	Nominal
Demographic	municipality	Municipality of birth	Nominal
	age	age of the person	Continuous
	migration	If the place of residence is different from the place of birth	Nominal
	gender	Gender	Binary
Family	head_house	If head of household	Binary
	live_family	Number of persons in the household	Discrete
	job_family	Number of people working in the household	Discrete

describes the 30 survey variables grouped into five dimensions: education—this refers to the ability to read and write, and the people’s educational level; work—this dimension contains variables associated with the MSW recovery process; health—this includes variables related to their physical condition and their link to the health system; demographics—this incorporates aspects of the population for the research. Finally, the family dimension accounts for the family structure. Likewise, Table 1 categorizes the variables according to their nature.

The qualitative data were transcribed, coded, and classified into themes for the qualitative results to be consolidated to examine how they are interrelated. This analysis of the results explains the dynamics of vulnerability in the study population.

2.2. Data Processing and Statistical Analysis

Figure 2 describes the methodology applied for the identification of waste picker profiles. First, we categorized the variables according to their nature. Second, multiple correspondence analysis was performed to extract the significant variables from the data set. Subsequently, principal component analysis was applied to obtain the relationships between the variables and cluster analysis to understand the characteristics that identify the different groups of waste pickers.

2.2.1. Multiple Correspondence Analysis (MCA)

The MCA method, to extract information about the relationship between the variables finding underlying structures in the data set, is based on the matrix of similarities (or distances) between the individuals [43]. The usual output from MCA includes the best two-dimensional representation of the data, along with the coordinates of the plotted points, and a measure (called the inertia) of the amount of information retained in each dimension [44].

The MCA formulation is explained as follows. Let X with elements x_ij, be an I J two-way table of unscaled frequencies or counts (I for the rows and J for the columns). We take I > J. The rows and columns of the contingency table X correspond to different categories of different characteristics. If n is the total of the frequencies in the data matrix X, then in Equation (1), P is called the correspondence matrix.

$$\begin{gathered} P=p_{ij}=\frac{x_{ij}}{n} \\ i=1, 2, \dots , I \\ j=1, 2, \dots , J \end{gathered}$$

(1)

Then, the vectors of rows and columns in Equations (2) and (3) are used to calculate the square root matrices in (4) and (5).

$$r_i=\sum _{i=1}^J{p}_{ij}=\sum _{j=1}^J\frac{x_{ij}}{n}$$

(2)

$$c_j=\sum _{i=1}^I{p}_{ij}=\sum _{i=1}^I\frac{x_{ij}}{n}$$

(3)

$$D_r^{\frac{1}{2}}=diag\left(\sqrt{r_1}, \sqrt{r_2}, \dots , \sqrt{r_I}\right)$$

(4)

$$D_c^{\frac{1}{2}}=diag\left(\sqrt{c_1}, \sqrt{c_2}, \dots , \sqrt{c_J}\right)$$

(5)

Correspondence analysis can be formulated as the weighted least squares problem to select $\hat{P}={\hat{p}}_{ij}$, a matrix of specified reduced rank, to minimize Equation (6).

$$\sum _{i=1}^I\sum _{j=1}^J\frac{p_{ij}-{\hat{p}}_{ij}}{r_i{c}_j}$$

(6)

2.2.2. Principal Component Analysis of Mixed Data

PCA is a mathematical procedure that transforms a set of CORRELATED variables into a set of new UNCORRELATED variables called principal components, to reduce the data size and facilitate its interpretation and analysis. The PCA of mixed data considers the mixture of qualitative and quantitative variables. For qualitative variables, it is necessary to calculate the associated quantification matrix (optimal scaling procedure). The transformed or quantified variables reflect the distance between the levels of the ordinal variables or different categories of the nominal variables, which optimize the properties of the correlation matrix of the quantified variables. Then, this matrix is concatenated with the matrix of quantitative variables to perform the Principal Component Analysis [45].

If the variables x₁, x₂, …, x_p in X are correlated and have variance–covariance matrix Σ with the pairs of eigenvalues and eigenvectors:

(λ₁, c₁); (λ₂, c₂);……; (λ_p, c_p)

where

λ₁ >= λ₂ >= …… >= λ_p >= 0

Then, the ith principal component is given by Equation (7)

Y_i = c^TX = c_i1X₁ + c_i2X₂ + … +c_ipX_p

(7)

2.2.3. Cluster Analysis

Cluster analysis groups individuals (experimental units, observation units) into clusters or groups such that individuals in the same cluster are similar to each other and different from those in other clusters. The cluster analysis aims to find a “NATURAL” structure between the observations based on a multivariate profile, maximizing the homogeneity within the cluster and the heterogeneity between the different clusters. Cluster analysis has also been called classification, pattern recognition (specifically, unsupervised learning), and numerical taxonomy [46]. The two approaches to cluster analysis are hierarchical and non-hierarchical. The two methods are detailed in [47]. Hierarchical clustering (HC) algorithms involve a sequential process. In each step of the agglomerative hierarchical approach, an observation or a cluster of observations is merged into another cluster. In this process, the number of clusters shrinks and the clusters themselves grow larger. It starts with n clusters (individual items) and ends with one single cluster containing the entire data set [46].

All the mathematical and statistical calculations were carried out using rStudio Desktop 4.2.0.

3. Results

3.1. Sample Profile

3.1.1. Nominal Variables

Table 2. Education level vs. Migration.

	Migration
Education Level	No	Yes
E	573 (29%)	520 (26%)
H	330 (17%)	238 (12%)
N	148 (7%)	144 (7%)
U	23 (1.1%)	17 (0.9%)

,

Table 3. Migration vs. Gender.

	Migration
Gender	No	Yes
Female	263 (13%)	251 (12%)
Male	810 (41%)	666 (33%)
Other	1 (0.005%)	2 (0.1%)

and

Table 4. Social security vs. Association.

Social Security
Association	C	N	S
No	68 (3%)	527 (26%)	1241 (62%)
Yes	0 (0%)	6 (0.3%)	151 (8.7%)

summarize some of the nominal variables. Table 2 shows that most people have elementary or high school education and are not migrants (46%), and only a small portion have a college education (2%). Table 3 shows that the proportion of women migrants and non-migrants is almost the same. Meanwhile, most of the men are non-migrants. Regarding the social security and the association variables in Table 4, we observed that most people are not associated and belong to the subsidized health regime (The Subsidized Regime is the mechanism through which the poorest population in the country, unable to pay, has access to health services through a subsidy offered by the State (https://www.minsalud.gov.co/, accessed on 14 March 2023)).

3.1.2. Numerical Variables

In

Table 5. Descriptive measures for numerical variables.

Measure	Time Recycling	Days Working	Live Family	Job Family	Age	Hours Working	Salary (USD)
Minimum	0.0	1.0	0.0	0.0	12.0	1	0.0
Maximum	68.0	7.0	31.0	14.0	85.0	18	1800
Mean	11.9	5.5	2.6	0.82	42.8	8.05	126

, we describe the main characteristics of the numerical variables. We can observe significant differences in the variables, for example, the minimum age is 12 and the maximum is 85. The salary is also highly variable (1 USD = 4700 COP).

3.2. Multivariate Analysis

According to what is presented in the Methodology section, multivariate statistical methods were applied with the main objective of identifying the waste picker profiles. We performed the analysis in three stages. Firstly, we applied Multiple Correspondence Analysis (MCA) to select the variables that provide the most information to the data set to avoid over-specification. Secondly, we performed dimension reduction with Principal Component Analysis (PCA) using the variables selected in the MCA to avoid redundancy and find similarities between the variables. Finally, we performed cluster analysis using the dimensions resulting from the PCA, which incorporate the relationships between the variables, to incorporate the similarities between the individuals and build the groups that would allow us to build the profiles of the waste pickers through the interpretation of these groupings.

3.2.1. Multiple Correspondence Analysis (MCA)

We applied MCA to select the variables that provide the most information (variance) to the construction of the profiles of the waste pickers. We used this method instead of the correlation matrix because our dataset includes both categorical (nominal, ordinal, dichotomous) and discrete and continuous variables. According to [48], the output from MCA includes the best two-dimensional representation of the data, along with the coordinates of the plotted points, and a measure called the inertia, which indicates the amount of information retained in each dimension. In our study, we utilized the chi-squared distance as a similarity measure.

We show the best two-dimensional representation of the complete data set in Figure 3, where the variables above the red dotted line are the ones that contribute most to the total variability. The variables that provide the most information to the construction of the profiles of the waste pickers are residence, live family, social security, and education level. Others, such as migration, age, and literacy, do not contribute significant information.

We present in Figure 3 the percentage contribution of each variable in the complete data set to the two-dimensional representation in MCA, the red dashed line indicates the mean contribution of all variables, where the variables above the red dotted line are the ones that contribute most to the total variability. The variables that provide the most information for the construction of the waste picker profiles are residence, live family, job family, transport, route, head house, social security, reason, hours working, education level, and association. Conversely, variables such as migration, age, and literacy do not provide significant information.

3.2.2. Principal Component Analysis of Mixed Data (PCA)

The PCA was performed using the variables selected in the MCA (residence, live family, job family, transport, route, head house, social security, reason, hours working, education level, and association) to reduce the size of the data for facilitating the interpretation and analysis. We used these variables to calculate the PCA for mixed data, and the biplot in Figure 4 represents the two-dimensional relationships between the eleven selected variables. We use a two-dimensional representation because the first two dimensions group more information and allow for better graphic representation.

From Figure 4, it can be observed that the variables residence, social security, education level, reason, association, live family, and job family are closely related to each other and conform to dimension 1 (x-axis). We interpret this as the dimension that describes the individual characteristics. On the other hand, dimension 2 (y-axis) groups the variables transport, route, head of the house, and hours working. This dimension describes the work profile of the individuals.

In

Table 6. Squared loadings from PCA.

	Dim 1	Dim 2	Dim 3	Dim 4	Dim 5
live family	0.59	0.01	0.07	0.01	0.00
job family	0.55	0.02	0.16	0.01	0.00
hours working	0.02	0.16	0.02	0.08	0.02
residence	0.27	0.26	0.02	0.13	0.12
association	0.22	0.00	0.01	0.07	0.04
social security	0.23	0.15	0.11	0.25	0.26
education level	0.08	0.08	0.11	0.32	0.51
transport	0.10	0.46	0.26	0.14	0.13
head house	0.02	0.42	0.06	0.00	0.01
route	0.03	0.46	0.10	0.01	0.00
reason	0.26	0.01	0.38	0.24	0.00

, we present the PCA loadings. The highest load indicates the most important variable for each dimension, the numbers highlighted in bold indicate the most important variables in each dimension. Dimension 1 takes the most information from live family and job family, while dimension 2 takes the most from transport and route, and dimension 3 takes the most from reason, and so on. We used five dimensions to represent the original thirty variables. First, we selected the eleven most important variables from MCA. Then, we calculated five new variables using PCA, which are unrelated, reducing the number of variables by over 80% and improving computational calculations and data interpretation.

3.2.3. Cluster Analysis (CA)

As previously mentioned, we performed cluster analysis using the dimensions resulting from PCA, which incorporate the relationships between the variables. This allowed us to incorporate the similarities between individuals and build groups that would enable us to create profiles of the waste pickers through the interpretation of these groupings.

First, we established the optimal number of clusters based on the Silhouette coefficient shown in Figure 5. The Silhouette coefficient measures the cohesion and separation of clusters, integrating them into an indicator where the optimal number of clusters is the one with the highest coefficient [49], the dashed line indicates that the optimal number of clusters is five. Secondly, we applied the hierarchical method using the squared Euclidean distance as the similarity measure, and the Ward distance as the agglomeration method. The dendrogram in Figure 6 shows the results of this method, each color indicates a group.

The cluster conformation was validated with the numerical variables using the Kruskal–Wallis (K-W) procedure in

Table 7. Kruskal–Wallis procedure for variables and cluster.

Variable	K-W Chi-Squared	p Value
Dim 1	1107.3	<2.2 × 10−16
Dim 2	1146.3	<2.2 × 10−16
Dim 3	752.9	<2.2 × 10−16
Dim 4	368.52	<2.2 × 10−16
Dim 5	727.49	<2.2 × 10−16

. All the variables were statistically significant for the five clusters with a confidence level of 95% (p-value less than 0.05). This means that there were 253 statistically significant differences between the clusters, which allows the differentiation of 254 groups and the creation of profiles of the waste pickers through the interpretation of these 255 groupings.

We characterized and named each cluster according to the numeric variables presented in Table 5, as follows:

• Cluster 1 (experimented): people who recycle more often, live with more people in the household, and whose household has more people who work.
• Cluster 2 (lowest income): people in this cluster have the lowest income.
• Cluster 3 (hard workers): people who are the oldest and spend more hours working per week.
• Cluster 4 (highest income): people who have the highest salary.
• Cluster 5 (comfortable): people who work fewer days and live with fewer people in the household.

Cluster 1 is named “experimented” because it has the most individuals involved in the experiment, more people live in the household, and more people are working in the household. Cluster 2 is named “lowest income” because it has individuals with the lowest salaries. Cluster 3 is named “hard workers” because it groups the hardest workers, the oldest people, and those who spend more hours working per week. Cluster 4 is named “lowest income” because it has people who have the highest salary. Cluster 5 is named “comfortable” because it has people who work fewer days and live with fewer people in the household.

4. Discussion and Conclusions

This article presents a detailed analysis of the recycling population in Colombia using multivariate statistical methods based on information collected by the Alianza EFI. The main objective of the analysis was to identify waste picker profiles, which can be used to design policies, improve working conditions, and increase productivity. The analysis was performed in three stages, starting with Multiple Correspondence Analysis (MCA) to select the most informative variables, followed by dimension reduction with Principal Component Analysis (PCA) to find similarities between the variables, and finally cluster analysis to group waste pickers with similar profiles based on similarities between the individuals. This analysis can be used as a proxy profile for similar populations in countries with comparable characteristics, and the results can be used to inform policy decisions and improve the lives of waste pickers.

The variables that provide the most information about the profiles are:

• Those related to the family and household: place of residence (residence), how many people they live with (live family), how many people in the household work (job family), and if the person is the head of the household (head house).
• Social and educational status: social security status (ss) and education level.
• Job conditions: the reason they work in recycling (reason), hours working, the type of transport they use to carry out their activity (transport), and if the person has a defined route to carry out their work (route).

Other investigations found the importance of variables related to the familial and household composition [36], job [40,41,50], and social and educational conditions [37,42].

The profile of waste pickers in Colombia was established based on statistical multivariate analysis. Initially, we found that the most important variables to define the profiles are residence, family status, job status, transportation, route, head of household, social security, reason for waste picking, hours worked, education level, and association. As these variables are related, we estimated new variables that were unrelated and reduced the dimensionality. We managed to go from having 30 variables to having 5 dimensions to profile individuals. Finally, we grouped the individuals based on their similarity and built five groups with the profiles that we describe below.

Cluster 1 comprises the most experienced individuals who live with more people in their households and have more working members. Cluster 2 groups individuals with the lowest salaries. Cluster 3 includes the hardest workers, as it consists of the oldest people who spend more hours working per week. Cluster 4 is composed of people with the highest salaries. Finally, Cluster 5 is named “Comfortable” because it includes people who work fewer days and live with fewer people in their households.

Other investigations have carried out profiles of waste pickers [36,41,42,51].

Parizeau [41] analyzed the profiles based on the responses of a survey including variables related to working conditions and practices, living conditions, health, social capital, access to social services, home and community life, and demographic information. The author used the statistical techniques Analysis of Variance ANOVA, chi-squared analysis, correlation, and t-test analysis and found that these workers may be engaged in exploitative vertical social capital relationships, their labor relies on low-paid waste work that exposes them to hazardous materials and conditions, they have insecure access to social entitlements, their human capital development often requires trade-offs with other assets (and notably their labor), child labor is a common household asset, and they often rely on their homes as a productive resource. The author’s analysis was limited to univariate and bivariate methods but could benefit from including multivariate and grouping analyses that could establish relationships between variables simultaneously and possible groupings between individuals.

Uddin et al. [50] applied a structured questionnaire survey, key informant interviews, and focus group discussions to assess the social, economic, and environmental situation of local informal recyclers. The authors in the study used univariate methods such as frequency tables and pie charts to analyze the variables included in the study. They did not apply inferential or multivariate methods. According to the findings, the majority of this population in Mongolia faces a number of difficulties, including homelessness, extremely cold weather, and a lack of a formal identity document (ID card). This demographic also frequently struggles with issues including alcoholism, social isolation, unemployment, and a lack of external support for recycling efforts. Among the occupational health risks faced by two-thirds of informal recyclers are stomach disorders, skin conditions, kidney and liver issues, back pain, wounds, burns, and bone fractures.

Borges et al. [36] showed the socioeconomic, demographic, and social security conditions of waste pickers in Brasilia, Bangalore, and Kolkata and compared the profiles of these workers across the three cities. The study involved calculating frequency tables and comparing percentages and numbers between cities. However, no inferential or multivariate analyses of either variables or individuals were conducted based on the available evidence.

The study conducted by Sarkar et al. [51] presents a vulnerability assessment of rag pickers in Delhi, with a focus on socioeconomic and occupational health issues. The study analyzes the socioeconomic profile of the pickers, taking into account their difficulties, expectations, and working conditions, using a database. In relation to working conditions, they identified four different profiles of waste pickers in the city of Delhi:

• Who carry a sack on their back and collect whatever has any resale value.
• Who carry a large sack slung in two partitions across a bicycle and keep the items separate.
• Who use a tricycle and collect over 50 kg of waste per day.
• Who work for waste dealers.

The same authors point out that the health risks faced by waste pickers are twofold: poverty and the nature of their work. Waste pickers are among the most disadvantaged and underprivileged members of the urban population, and they commonly experience undernutrition, growth impairment, anemia, tuberculosis, and other bacterial and parasitic illnesses.

In order to create profiles of various populations, multivariate analyses have been used in fields such as medicine [52], microbiology [53], ecology [54], and genetics [55]. These analyses have allowed for the establishment of differences and similarities between the individuals or experimental units studied and their characteristic variables. However, there are not many applications of this type of analysis in vulnerable populations such as waste pickers. The establishment of the most important variables and the different profiles of waste pickers is crucial because it allows for specific actions to be taken for each group of individuals. Improving the family and household conditions, social and educational status, and job conditions can have a significant impact on their lives. It is important to design policies that address the specific needs of waste pickers and improve their working and living conditions.

According to the improvement actions to upgrade the living conditions of this population, various investigations have recommended the creation of recycling co-operatives or other forms of collective organizations [42,50,56,57]. Institutionalizing their activities would enhance the scope of their work and provide better working conditions. They could be organized with the help of civil society groups around micro-enterprises related to recycling. This would also help restore their self-esteem, apart from assuring their livelihood [50].

For future work, it is possible to apply classification methods to determine the variables that generate the separation between clusters. This will help to focus the actions to improve the living conditions of each group of waste pickers. Another research opportunity is to study the effect of belonging to an association on the improvement of their working conditions.