Environmental Determinants of Poverty: Evidence from Sub-Districts in Bandung Regency, Indonesia

1. Introduction

Poverty remains one of the main issues in global development and public health. Although progress in poverty reduction has been ongoing for several decades, the rate of decline has significantly slowed down. Poverty, Prosperity, and Planet Report 2024 estimates that nearly 700 million people, or about 8.5% of the world's population, still live in extreme poverty, while around 3.5 billion people, or 44% of the world's population, still live below the standard of US$6.85 per person per day relevant for upper-middle-income countries [1]. These conditions indicate that poverty is not only an issue of income but also relates to the community's ability to access basic services, a healthy environment, and decent living conditions. In the context of sustainable development, poverty reduction cannot be separated from efforts to improve environmental quality and public health.

The relationship between poverty, environment, and health is multidimensional. Communities with lower socio-economic conditions tend to face limited access to infrastructure and basic services, while exposure to unhealthy environments can increase the risk of disease, healthcare costs, loss of productivity, and economic vulnerability. These conditions can create a mutually reinforcing relationship between poverty and poor environmental conditions. One of the most obvious dimensions is access to water, sanitation, and hygiene (Water, Sanitation and Hygiene or WASH). The World Health Organization (WHO) places safe and adequately available water as a fundamental component of health, human development, and poverty reduction [2]. Thus, the quality of the residential environment needs to be considered as part of a broader approach to poverty and public health.

Access to safe drinking water holds a strategic position in that relationship. In 2022, around 2.2 billion people worldwide did not have access to safely managed drinking water services, while at least 1.7 billion people used drinking water sources contaminated with feces [2]. Unsafe water can become a medium for the transmission of various diseases, including diarrhea, cholera, dysentery, hepatitis A, typhoid, and polio. Conversely, access to safe and easily reachable water sources can reduce disease exposure, save time spent obtaining water, lower healthcare expenses, and support community productivity [2,3]. Therefore, access to safe drinking water is not only an indicator of basic services but also has health and socio-economic dimensions relevant to poverty.

Sanitation is another component of WASH that is closely related to health and socio-economic conditions. Inadequate sanitation can cause environmental pollution and increase exposure to pathogens, while limited sanitation infrastructure can exacerbate health disparities and quality of life. Research on the sustainability of sanitation infrastructure shows that sanitation and toilets are essential infrastructure for public health, social stability, poverty reduction, inequality reduction, and sustainable community development [4]. However, the issue of sanitation is not only related to the existence of facilities but also to their quality, management, maintenance, and utilization. Therefore, differences in the coverage of sanitation access between regions can be one of the relevant environmental characteristics in explaining variations in the socio-economic conditions of the community.

Environmental problems also arise through waste management. Urban and residential waste that is not well-managed can cause pollution, increase the risk of exposure to biological and chemical agents, disrupt environmental quality, and create conditions that support the development of disease vectors. A systematic review of urban solid waste management found evidence of a relationship between exposure to certain waste processing or disposal facilities and various health impacts, including respiratory disorders, mortality, neonatal health impacts, and mental health [5]. Nevertheless, epidemiological evidence still has limitations, and not all waste management indicators can directly represent the level of community exposure. In the context of regional research, the coverage of waste collection remains relevant as an indicator of basic environmental service capacity, but it must be distinguished from the volume of waste collected, the frequency of collection, or the effectiveness of final processing.

Slum settlements exhibit a concentration of various forms of social and environmental vulnerabilities in the same space. Slum settlements generally face a combination of population density, low housing quality, limited infrastructure, limited WASH access, and socio-economic barriers. A review of WASH and housing inequalities in slum areas of low- and middle-income countries shows that access barriers not only stem from infrastructure limitations but also from social, cultural, economic, governance, policy, and environmental factors [6]. These barriers can lead to health consequences as well as socio-economic burdens for the residents of slum settlements. Thus, the existence of slum settlements can be viewed as one of the indicators of environmental vulnerability that is potentially related to poverty.

The relationship between WASH, slum settlements, and poverty is also a concern in the context of developing countries, including Indonesia. Research in Indonesia shows that WASH studies have developed, but there are still imbalances in the focus and distribution of research. A systematic review of 272 WASH publications in Indonesia found that the topic of water is more extensively researched compared to sanitation and hygiene, while most studies focus on social themes and behavioral determinants. Research on program implementation, behavior change interventions, and financial aspects of WASH is still relatively limited [7]. These findings indicate a need to expand WASH research from merely identifying access or behavior to analyzing how environmental service conditions relate to socio-economic issues at the regional level.

In the context of Indonesia, poverty remains an important issue despite national indicators showing a downward trend. The Central Statistics Agency (BPS) reported that the percentage of poor people in Indonesia in September 2024 was 8.57%, or around 24.06 million people [8]. In March 2025, this percentage decreased to 8.47%, with the number of poor people around 23.85 million [9]. However, the national aggregate decline does not mean that the problem of poverty has been resolved, as there are differences in conditions between regions and between urban and rural areas. Therefore, a more detailed geographical analysis is needed to understand the environmental factors that may be related to variations in poverty.

West Java is one of the regions relevant in that context. By September 2024, the percentage of poor residents in West Java reached 7.08%, with approximately 3.67 million poor people [10]. The urban poverty rate of 6.65% is still lower than the rural poverty rate of 8.85%, indicating significant spatial variation in the characteristics of poverty in the province [10]. The difference indicates that a development approach using only provincial aggregate indicators has the potential to obscure the heterogeneity of socio-economic and environmental conditions at the district or sub-district level.

Bandung Regency is one of the important areas in West Java to study because it has heterogeneous regional characteristics, including urban, peri-urban, and rural areas. In March 2024, the percentage of poor residents in Bandung Regency was 6.19%, with approximately 239.87 thousand poor residents [11]. Although the figure is lower than the poverty rate in West Java in September 2024, the aggregate figure for the district does not show how poverty is distributed among sub-districts. Differences in settlement characteristics, basic infrastructure, population density, and environmental conditions between sub-districts can result in varying levels of socio-economic vulnerability. Therefore, using sub-districts as the unit of analysis can provide more detailed information for region-based intervention planning.

The availability of sub-district level data provides an opportunity to conduct such analysis. The 2024 Village Potential Statistics of Bandung Regency published by BPS presents information on regional potential, infrastructure, and various challenges faced by villages. BPS explains that the district/city level publication is an aggregation of data at the sub-district level [12]. The source allows for the formation of environmental indicators at the sub-district level, including waste collection coverage, access to drinking water, access to sanitation, and the presence of slum settlements. The use of these area-based indicators enables research to link residential environmental conditions with poverty levels within a single empirical framework.

Although research on poverty, WASH, environmental health, and slum settlements has developed, there are still several research gaps. First, WASH research in Indonesia is still relatively dominant in focusing on behavioral determinants and access aspects, while studies linking several dimensions of WASH simultaneously with poverty at the level of relatively small administrative regions are still limited [7]. Second, research on waste management more often examines health impacts or environmental exposure, while its relationship with variations in poverty across regions has not been extensively analyzed in a single model together with water, sanitation, and slum settlements [5]. Third, studies on slum areas often emphasize aspects of health, housing, access inequality, or quality of life, but the integration of slum settlement indicators with other basic environmental services in explaining variations in poverty at the sub-district level is still relatively limited [6]. Fourth, some studies use household data or case studies of specific areas, whereas a district-based approach can provide a spatial perspective on environmental inequality and poverty within an administrative region.

Based on this gap, the novelty of this research lies in the integration of four residential environmental indicators—Waste Collection Coverage, Safe Drinking Water Access, Access to Improved Sanitation, and Slum Settlement Coverage—into a single empirical model to analyze variations in poverty levels across sub-districts in Bandung Regency. This approach expands the study of WASH and environmental health from an individual or household focus to a spatial environmental–poverty nexus perspective at the sub-district level. Furthermore, the use of secondary government data available at the sub-district level provides an opportunity to generate empirical evidence that can be replicated and more easily linked to regional development policies.

Theoretically, this research uses the perspective of social and environmental determinants of health, which views health and well-being as the result of interactions between socio-economic conditions, the physical environment, residential characteristics, and access to basic services. In this framework, poverty and the environment are not viewed as two completely separate phenomena. Poor communities may experience limited access to environmental services, while an unhealthy environment can increase health risks and the economic burden on society [2,6]. This framework provides a conceptual basis for examining whether differences in the coverage of basic environmental services and settlement conditions are related to differences in poverty levels between regions.

Based on this framework, this study aims to analyze the relationship between waste management, access to safe drinking water, sanitation, and slum settlements with the poverty level in 31 sub-districts in Bandung Regency, West Java, in 2024. The research questions posed are: (1) does the coverage of waste collection relate to the poverty level between sub-districts; (2) does access to safe drinking water relate to the poverty level; (3) does access to sanitation relate to the poverty level; and (4) does the coverage of slum settlements relate to the poverty level? The research also examines whether the four environmental indicators are simultaneously related to the poverty level.

Based on previous theories and empirical findings, the research hypothesis is formulated as follows. H1: the coverage of waste collection is negatively related to the poverty level. H2: access to safe drinking water is negatively related to the poverty level. H3: access to sanitation is negatively related to the poverty level. H4: the coverage of slum settlements is positively related to the poverty level. Simultaneously, the research proposes the hypothesis that the coverage of waste collection, access to safe drinking water, access to sanitation, and the coverage of slum settlements are significantly related to the poverty level.

2. Materials and Methods

2.1. Research Design

This research uses a quantitative approach with a cross-sectional observational design to analyze the relationship between residential environmental conditions and poverty levels at the sub-district level. The cross-sectional design is used because all research variables are observed during the same period, namely the year 2024, without providing any intervention or treatment to the observation units. This approach is suitable for research aimed at identifying patterns of statistical relationships between variables based on observational data at a single point in time.

This research does not use an experimental design because the researchers do not apply treatments to the observed regions or communities. The research is also not a purely descriptive study because, in addition to describing data characteristics, it tests the simultaneous relationships between several environmental indicators and poverty levels using a regression model. This type of quantitative approach allows for systematic analysis of inter-regional variations and can be replicated using the same dataset.

The selection of a regional observational design is also relevant to the characteristics of the Village Potential Data (Podes). Podes is a regional data source that provides information on village potential, infrastructure, and various conditions faced by the region. For Bandung Regency, the publication of the 2024 Bandung Regency Village Potential Statistics is an aggregation of data at the sub-district level [12]. Thus, the research design is directed toward the analysis of the environmental–poverty nexus at the sub-district level, rather than at the individual or household level.

2.2. Research Location

The research was conducted in Bandung Regency, West Java Province, Indonesia. Bandung Regency was chosen as the research location because it has heterogeneous regional characteristics and consists of several sub-districts with different socio-economic conditions, basic infrastructure, and residential environmental characteristics.

The geographical unit of the research is all the sub-districts located within the administrative area of Bandung Regency in 2024. Based on the administrative structure used in the research dataset, there are 31 sub-districts that serve as the units of observation. By using all sub-districts as the unit of analysis, the research allows for the identification of variations in environmental conditions and poverty between regions without selecting a sample of sub-districts.

The use of sub-districts as the unit of analysis also allows for the alignment of environmental data from Podes and socio-economic data from local governments at a single geographical level. BPS Kabupaten Bandung stated that the publication of the 2024 Kabupaten Bandung Village Potential Statistics at the district/city level is an aggregation of data at the sub-district level [12].

2.3. Population and Unit of Analysis

The research population consists of all sub-districts in Bandung Regency in 2024. Since all sub-districts available in the research area are used as observations, the research applies total enumeration, which means including all population units that meet the data criteria in the analysis.

Thus, the research did not use random sampling, stratified sampling, or purposive sampling. The use of all sub-districts is intended to reduce the risk of sampling error at the regional level and provide a more comprehensive picture of the variation in environmental conditions and poverty in Bandung Regency.

The unit of analysis for the research is the sub-district, while several environmental indicators used were initially collected at the village/sub-district level and then aggregated to the sub-district level. This aggregation approach is consistent with the characteristics of the 2024 Podes Kabupaten Bandung publication, which presents district-level information as a result of aggregating village/sub-district data [12].

Therefore, the interpretation of the research results must be conducted at the regional level. The regression coefficients cannot be interpreted as direct relationships at the individual or household level. In other words, this research is an ecological/cross-sectional analysis at the sub-district level.

2.4. Data Sources and Data Collection

The research uses secondary data obtained from official government sources. Primary data collection through questionnaires, interviews, household observations, or experiments was not conducted. The use of secondary data was chosen because the research indicators are already available through the statistical and administrative systems of the government at the regional level.

The main source of environmental data is the BPS Kabupaten Bandung through the Statistics of Village Potential in Kabupaten Bandung 2024. The publication presents data on regional potential, infrastructure, and various conditions faced by villages, as well as providing aggregation at the sub-district level [12]. Podes data also includes variables that are directly related to the presence of slum settlements; the metadata of Podes microdata 2024, for example, lists variables such as the presence of slum settlements, the number of slum settlement locations, the number of residential buildings, and the number of families in slum settlements [13].

Poverty data and other socio-economic indicators are obtained from official sources of the local government and/or the Bandung Regency BPS according to the availability of indicators at the sub-district level. The use of official sources is intended to ensure data traceability and consistency in statistical definitions.

Data collection is carried out through the following stages: (1) identification of official data sources; (2) extraction of relevant variables; (3) alignment of sub-district names; (4) examination of reference years; (5) examination of measurement units; (6) examination of missing values; and (7) merging all variables based on sub-district identity.

The data cleaning stage is carried out before estimation to ensure that each sub-district has one observation for each variable. Differences in the spelling of region names, if found between sources, are standardized based on the official administrative names of Bandung Regency.

2.5. Variable Operationalization

The research uses one dependent variable and four independent variables. The dependent variable is the poverty rate, while the independent variables include waste collection coverage, safe drinking water access, access to improved sanitation, and slum settlement coverage.

Table 1. Operational Definition of Variables.

Code	Variable	Operational Definition	Expected Sign
Y	Poverty Rate (%)	Persentage of poor population in each sub-district	—
X1	Waste Collection Coverage (%)	Percentage of villages/sub-districts with a waste management system where waste is subsequently collected, according to the data source definition	−
X2	Safe Drinking Water Access (%)	Percentage of villages/urban areas with access to safe drinking water based on data source indicators	−
X3	Access to Improved Sanitation (%)	Percentage of villages/sub-districts with access to improved sanitation/healthy latrines based on data source indicators	−
X4	Slum Settlement Coverage (%)	Percentage of villages/urban neighborhoods with the presence of slum settlements based on data source indicators	+

Note: Operational definitions were compiled by the authors based on the original definitions and data documentation provided by Statistics Indonesia (BPS) and the Bandung Regency Government [11,12,13,14].

Table 1. Operational Definition of Variables.

Code	Variable	Operational Definition	Expected Sign
Y	Poverty Rate (%)	Persentage of poor population in each sub-district	—
X1	Waste Collection Coverage (%)	Percentage of villages/sub-districts with a waste management system where waste is subsequently collected, according to the data source definition	−
X2	Safe Drinking Water Access (%)	Percentage of villages/urban areas with access to safe drinking water based on data source indicators	−
X3	Access to Improved Sanitation (%)	Percentage of villages/sub-districts with access to improved sanitation/healthy latrines based on data source indicators	−
X4	Slum Settlement Coverage (%)	Percentage of villages/urban neighborhoods with the presence of slum settlements based on data source indicators	+

Note: Operational definitions were compiled by the authors based on the original definitions and data documentation provided by Statistics Indonesia (BPS) and the Bandung Regency Government [11,12,13,14].

The operational definitions X1–X4 follow the indicator structure in government data sources and are not intended to replace the technical definitions of each statistical agency. This is important because the terms safe drinking water, improved sanitation, and slum settlement can have different operational definitions depending on the statistical system used.

Specifically for the slum settlement indicator, BPS statistical metadata shows that slum areas are related to the poor quality of housing and infrastructure, including density, access to basic infrastructure, and environmental conditions [14]. In Podes 2024, the presence of slum settlements is one of the variables available at the village/sub-district level [13]. Therefore, X4 in this study is not interpreted as the percentage of slum area in hectares, but as an indicator of coverage based on the presence of slum settlements.

Similarly, X1 should be interpreted as an indicator of waste management service coverage at the regional level, not as a percentage of the volume of waste successfully transported to the landfill. This definition limitation is necessary to ensure that statistical interpretation does not exceed the actual information contained in the data.

2.6. Model Specification

The relationship between environmental conditions and poverty levels is estimated using the following multiple linear regression model:

$$Povert{y}_i={\beta }_0+{\beta }_{1}Wast{e}_i+{\beta }_{2}Wate{r}_i+{\beta }_{3}Sanitatio{n}_i+{\beta }_{4}Slu{m}_i+{\epsilon }_i$$

with: Poverty_i= poverty level of sub-district i; Waste_i= coverage of waste collection in sub-district i; Water_i= access to safe drinking water in sub-district i; Sanitation_i= access to sanitation in sub-district i; Slum_i= coverage of slum settlements in sub-district i; β₀= constant; β₁-β₄= regression parameters; ε_i= error term.

Based on the research conceptual framework, the coefficients X1, X2, and X3 are expected to be negative, while X4 is expected to be positive. That interpretation is based on the assumption that the improvement in the quality of basic environmental services can be related to better socio-economic conditions, while the presence of slum settlements reflects the accumulation of environmental and social vulnerabilities.

However, because the research design is cross-sectional and observational, the regression coefficients are interpreted as conditional associations, not as evidence of causality. This is especially important because the relationship between poverty and environmental conditions can be bidirectional.

2.7. Econometric Estimation

Parameter estimation is performed using the Ordinary Least Squares (OLS) method. OLS is chosen because the dependent variable is the percentage of poverty level, and the research aims to estimate the linear relationship between poverty level and several environmental indicators.

Estimation is based on 31 observations of sub-districts. With four independent variables and one constant, the model has 26 degrees of freedom for the residual. The regression coefficients are estimated simultaneously so that the influence of each environmental indicator on the variation in poverty levels is calculated while controlling for the presence of the other three indicators in the model.

The coefficient of determination R² is used to measure the proportion of variation in poverty levels that can be explained by the independent variables in the model. Because the number of observations is relatively small compared to the number of parameters, adjusted R² is also reported as a measure that takes into account the number of predictors and the sample size.

A multivariate approach is necessary because WASH and environmental characteristics are interrelated systems. WASH literature also emphasizes the importance of examining the interactions of various factors within the water, sanitation, and hygiene service system, rather than just one indicator in isolation [7]. By incorporating four indicators simultaneously, the research can identify the partial relationships of each variable after considering other environmental variables.

2.8. Descriptive and Inferential Statistical Analysis

The analysis begins with descriptive statistics to illustrate the distribution of each variable. The statistics used include minimum, maximum, mean, median, standard deviation, and coefficient of variation if necessary. Descriptive statistics are used to determine the level and variation of poverty and environmental conditions between sub-districts.

Next, a correlation analysis is conducted to evaluate the direction and strength of the bivariate relationship between the poverty level and each environmental indicator. Correlation analysis is also used as an initial check for the possibility of very strong relationships between independent variables.

Inferential analysis was conducted using multiple linear regression. The overall model significance is tested using the F-test, while the significance of each coefficient is tested using the t-test. The significance level is set at α = 0.05. Thus, the coefficients are considered statistically significant if the p-value < 0.05.

In addition to the p-value, the study reports the 95% confidence interval for each coefficient. Reporting confidence intervals provides information about estimation uncertainty and does not solely rely on significance decisions based on a 5% threshold.

2.9. Robustness and Diagnostic Tests

Because the research uses cross-sectional data from 31 regional units, assumption testing and model diagnostics are an important part of the estimation procedure. Testing is conducted to ensure that the regression results are not primarily caused by multicollinearity, heteroscedasticity, problematic residual distribution, or highly influential observations.

2.9.1. Multicollinearity

Multicollinearity is examined using the Variance Inflation Factor (VIF) and tolerance. VIF is used to identify whether the independent variables have a linear relationship that is too strong with each other. This examination is important because the indicators of water, sanitation, waste, and slum settlements can conceptually be interrelated.

If a high VIF value is found, the relationships between indicators will be further examined, and the interpretation of individual coefficients will be done carefully. Multicollinearity testing is not used as an automatic reason to remove variables, as the decision must consider the theoretical foundation and research objectives.

2.9.2. Heteroskedasticity

Heteroskedasticity is tested using the Breusch–Pagan test or the White test. This test is necessary because the variance of the error term can differ between districts due to differences in area size, socio-economic characteristics, density, or level of development. If heteroscedasticity is detected, statistical inference will use heteroscedasticity-robust standard errors. Thus, the conclusion regarding the significance of the coefficients does not heavily rely on the assumption of homoscedasticity.

2.9.3. Residual Diagnostics

The residual model is examined to detect systematic patterns that may indicate specification errors. The examination includes the distribution of residuals, residuals versus fitted values, and observations with extreme residuals. The normality test of residuals can be used as an additional check, but it should not be the sole basis for assessing the model's adequacy. With a relatively small number of observations, the interpretation of normality is conducted alongside graphical examination and other diagnostic statistics.

2.9.4. Influential Observations

Because the study only uses 31 sub-districts, the presence of one or several extreme observations can have a relatively large impact on the OLS coefficients. Therefore, leverage, studentized residuals, and Cook's distance are used to identify influential observations. If there are observations with high influence, the main estimation is still conducted using all districts because all units constitute the population of the research area. However, a sensitivity analysis was conducted by comparing the results of the main model and the model after the influential observations were examined. Observations are not removed just because they produce extreme values, but will only be reviewed if there are indications of data errors or disproportionate statistical influence.

2.9.5. Robustness Analysis

As a robustness check, the OLS coefficients will be compared with the estimation results using robust standard errors if heteroskedasticity is found. The comparison focuses on changes in the sign of the coefficient, the magnitude of the coefficient, the confidence interval, and statistical significance.If the main conclusion remains consistent after using robust standard errors, the results are considered more stable against violations of the homoscedasticity assumption. On the other hand, if the significance changes, the results will be reported transparently and the interpretation will be more cautious.

2.10. Data Quality, Validity, and Reliability

Because the research uses secondary data, the validity of the survey instrument and Cronbach's alpha are not applied. There are no questionnaires or psychometric scales developed by the researchers. Data validity is addressed through source validation and data consistency checks.

First, each variable is traced back to its official source and indicator definition. Second, the values of each district are checked to ensure there are no values outside the theoretical percentage range, which is 0–100. Third, the consistency of district names and codes is checked before merging the datasets. Fourth, the reference year of each variable is verified to ensure that all observations represent the 2024 period. Fifth, missing data, duplication, or unit discrepancies are checked before analysis.

This approach is important because the research uses aggregated data from several administrative sources. Podes 2024 itself provides a data structure that allows for the systematic identification of environmental variables at the village/sub-district level, including the presence of slum settlements [13]. In addition, the BPS statistical system uses processing and validation stages as part of the statistical data production process, so official data sources provide a stronger foundation for secondary research compared to unverified data sources.

2.11. Software and Reproducibility

The initial dataset processing was carried out using Microsoft Excel, primarily for data cleaning, regional coding, completeness checks of observations, and the preparation of the final dataset. Regression estimation and econometric testing were conducted using EViews. The statistical output was then used to compile descriptive tables, correlation tables, regression results, diagnostic tests, and research interpretation.

The final dataset will be organized in a long/wide structure format, allowing each row to represent one sub-district and each column to represent one research variable. Because the research uses all sub-districts and secondary data sourced from government publications, there is no direct interaction with human respondents. Therefore, the research does not require informed consent or primary data collection.

3. Results

3.1. Descriptive Statistics

This study uses 31 sub-districts in Bandung Regency as the unit of analysis. Descriptive statistics show a significant variation in poverty levels and environmental condition indicators between sub-districts. The poverty rate has an average value of 10.31%, with a standard deviation of 4.59, and minimum and maximum values of 2.19% and 18.09%, respectively. This condition indicates that the poverty rate varies significantly between regions in Bandung Regency.

The average percentage of waste collected (X1) is 52.42%, with a standard deviation of 36.92%. The minimum value reaches 0%, while the maximum is 100%, indicating high heterogeneity in the coverage of waste collection services. Access to safe drinking water (X2) has an average of 80.43%, while access to healthy latrines (X3) has an average of 81.29%. Although both indicators have relatively high averages, each still shows variation between sub-districts. Meanwhile, the average of the slum settlement indicator (X4) is 52.26%, with a standard deviation of 34.70%. The raw data shows that all indicators have significant geographical variation between sub-districts.

Table 2. Descriptive Statistics of Study Variables.

Variable	N	Mean	Std. Deviation	Minimum	Maximum
Poverty Rate (Y)	31	10.31	4.59	2.19	18.09
Waste Collection (X1)	31	52.42	36.92	0	100
Safe Drinking Water (X2)	31	80.43	26.28	15.38	100
Improved Sanitation (X3)	31	81.29	27.22	0	100
Slum Settlement (X4)	31	52.26	34.7	0	100

Note: Authors' calculation based on the 2024 sub-district dataset.

Table 2. Descriptive Statistics of Study Variables.

Variable	N	Mean	Std. Deviation	Minimum	Maximum
Poverty Rate (Y)	31	10.31	4.59	2.19	18.09
Waste Collection (X1)	31	52.42	36.92	0	100
Safe Drinking Water (X2)	31	80.43	26.28	15.38	100
Improved Sanitation (X3)	31	81.29	27.22	0	100
Slum Settlement (X4)	31	52.26	34.7	0	100

Note: Authors' calculation based on the 2024 sub-district dataset.

3.2. Correlation Analysis

Pearson correlation analysis is used as an initial analysis to see the direction and strength of the relationship between variables. The results of the calculations based on the research dataset show that the poverty level has a relatively strong negative correlation with X1 (waste collection coverage), with a coefficient of −0.746. Conversely, the relationship between the poverty level and X2 is relatively weak and negative (−0.131), while the relationship between the poverty level and X3 is almost nonexistent (−0.012). The relationship between the poverty level and X4 is positive and relatively weak (0.267). Among the independent variables, the highest correlation is found between X2 and X3 at 0.785. This correlation indicates that districts with higher access to drinking water also tend to have higher access to sanitation. Nevertheless, the correlation has not yet indicated any serious multicollinearity issues, as confirmed by the VIF results in the diagnostic stage.

Table 3. Pearson Correlation Matrix.

Variable	Poverty (Y)	Waste Collection (X1)	Safe Drinking Water (X2)	Improved Sanitation (X3)	Slum Settlements (X4)
Poverty (Y)	1	−0.746	−0.131	−0.012	0.267
Waste Collection (X1)	−0.746	1	0.061	−0.034	−0.201
Safe Drinking Water (X2)	−0.131	0.061	1	0.785	−0.161
Improved Sanitation (X3)	−0.012	−0.034	0.785	1	0.143
Slum Settlements (X4)	0.267	−0.201	−0.161	0.143	1

Note: Pearson correlation coefficients calculated from 31 sub-district observations in Bandung Regency, West Java.

Table 3. Pearson Correlation Matrix.

Variable	Poverty (Y)	Waste Collection (X1)	Safe Drinking Water (X2)	Improved Sanitation (X3)	Slum Settlements (X4)
Poverty (Y)	1	−0.746	−0.131	−0.012	0.267
Waste Collection (X1)	−0.746	1	0.061	−0.034	−0.201
Safe Drinking Water (X2)	−0.131	0.061	1	0.785	−0.161
Improved Sanitation (X3)	−0.012	−0.034	0.785	1	0.143
Slum Settlements (X4)	0.267	−0.201	−0.161	0.143	1

Note: Pearson correlation coefficients calculated from 31 sub-district observations in Bandung Regency, West Java.

The correlation findings provide initial indications that X1 has the strongest relationship with poverty compared to the other three environmental indicators. However, since the correlation is bivariate, conclusions regarding the partial relationships of each variable are subsequently based on the results of multivariate regression.

3.3. Multiple Linear Regression

The multiple linear regression model is estimated using the Ordinary Least Squares (OLS) method with 31 observations. The estimation results show that the model is statistically significant overall, with an F-statistic = 8.815244 and Prob(F-statistic) = 0.000123. Thus, the null hypothesis that all independent variable coefficients are simultaneously equal to zero is rejected at the 5% significance level. The R-squared value of 0.575586 indicates that approximately 57.56% of the variation in poverty levels between districts can be jointly explained by X1, X2, X3, and X4. The Adjusted R-squared value of 0.510292 indicates that after accounting for the number of predictors and sample size, the model is still able to explain approximately 51.03% of the variation in poverty levels.

The estimation results show that X1 (the coverage of waste collection) is the only independent variable that has a statistically significant effect on the poverty level. The coefficient of X1 is −0.089379 with a t-statistic of −5.506229 and a p-value <0.001. Thus, an increase of one percentage point in the coverage of waste collection is associated with a decrease in the poverty rate of about 0.089 percentage points, with other variables controlled. X2 has a coefficient of −0.013172 but is not significant (p = 0.7460). X3 has a coefficient of 0.001360 and is not significant (p = 0.9723), while X4 has a coefficient of 0.014305 and is also not significant (p = 0.4647).

Table 4. Results of Multiple Linear Regression.

Variable	Coefficient	Std. Error	t-Statistic	p-value
Constant	15.19885	2.430622	6.25307	<0.001
Waste Collection (X1)	−0.089379	0.016232	−5.506229	<0.001
Safe Drinking Water (X2)	−0.013172	0.040236	−0.327365	0.746
Improved Sanitation (X3)	0.00136	0.038727	0.035118	0.9723
Slum Settlement (X4)	0.014305	0.019276	0.74214	0.4647

Model statistics: R² = 0.575586; Adjusted R² = 0.510292; F-statistic = 8.815244; Prob(F-statistic) = 0.000123; N = 31.

Table 4. Results of Multiple Linear Regression.

Variable	Coefficient	Std. Error	t-Statistic	p-value
Constant	15.19885	2.430622	6.25307	<0.001
Waste Collection (X1)	−0.089379	0.016232	−5.506229	<0.001
Safe Drinking Water (X2)	−0.013172	0.040236	−0.327365	0.746
Improved Sanitation (X3)	0.00136	0.038727	0.035118	0.9723
Slum Settlement (X4)	0.014305	0.019276	0.74214	0.4647

Model statistics: R² = 0.575586; Adjusted R² = 0.510292; F-statistic = 8.815244; Prob(F-statistic) = 0.000123; N = 31.

The obtained regression equation is:

$$Y=15.19885-0.089379X_1-0.013172X_2+0.001360X_3+0.014305X_4+\epsilon$$

The results show that the direction of the relationship between X1 and X2 is consistent with the hypothesis, while X4 also has a positive direction consistent with the hypothesis. X3 has a positive direction that contradicts the hypothesis. However, among the four variables, only X1 has a statistically significant relationship.

3.4. Diagnostic Tests

Diagnostic testing is conducted to evaluate the feasibility of the OLS model before further interpretation. The results of the White heteroskedasticity test yielded a Chi-square p-value of 0.6566 and an F p-value of 0.7781. Both values are greater than 0.05, so there is not enough statistical evidence to reject the homoscedasticity hypothesis.

The Breusch–Godfrey Serial Correlation LM test produced a Chi-square p-value of 0.3402 and an F probability of 0.4210. The results do not show evidence of residual correlation up to the second lag. However, because the research data is cross-sectional, the serial correlation test is treated as an additional check.

The residual normality test yielded a Jarque–Bera statistic of 5.814993 with a p-value of 0.054612. The value is slightly greater than 0.05, so the normality of the residuals is not rejected at the 5% significance level. However, because the value is relatively close to the 0.05 threshold, the result is interpreted with caution.

Table 5. Diagnostic Tests.

Test	Statistic	p-value	Conclusion
Breusch–Godfrey LM	2.1562	0.3402	No evidence of residual correlation
White Test	11.3715	0.6566	No evidence of heteroskedasticity
Jarque–Bera	5.815	0.0546	Normality not rejected at 5%
Durbin–Watson	2.2814	—	Additional diagnostic

Note: Diagnostic statistics obtained from EViews 13.

Table 5. Diagnostic Tests.

Test	Statistic	p-value	Conclusion
Breusch–Godfrey LM	2.1562	0.3402	No evidence of residual correlation
White Test	11.3715	0.6566	No evidence of heteroskedasticity
Jarque–Bera	5.815	0.0546	Normality not rejected at 5%
Durbin–Watson	2.2814	—	Additional diagnostic

Note: Diagnostic statistics obtained from EViews 13.

3.5. Multicollinearity

The results of the Variance Inflation Factor show that the Centered VIF values for X1, X2, X3, and X4 are 1.046, 3.257, 3.236, and 1.303, respectively. All values are below the practical threshold of 5, so there are no indications of serious multicollinearity in the model. The relatively higher VIF values for X2 and X3 are consistent with the correlation between access to drinking water and sanitation. However, the values are still within the acceptable range for OLS estimation.

Table 6. Variance Inflation Factor.

Variable	Centered VIF
Waste Collection (X1)	1.046
Safe Drinking Water (X2)	3.257
Improved Sanitation (X3)	3.236
Slum Settlement (X4)	1.303

Note: Centered VIF values obtained from EViews 13.

Table 6. Variance Inflation Factor.

Variable	Centered VIF
Waste Collection (X1)	1.046
Safe Drinking Water (X2)	3.257
Improved Sanitation (X3)	3.236
Slum Settlement (X4)	1.303

Note: Centered VIF values obtained from EViews 13.

3.6. Influential Observations

The analysis of influential observations was conducted using RStudent and the Hat Matrix through the Influence Statistics facility in EViews. This analysis is important because the study only uses 31 sub-districts, so a single observation can have a relatively large impact on the OLS estimation. The results of the Influence Statistics show that most RStudent values are within a relatively moderate range. However, there are several observations that need to be checked. Observation 30 shows an RStudent value of around −3.92, which is an extreme value compared to other observations. Several other observations also show absolute RStudent values close to or exceeding 2. From the leverage side, the Hat Matrix plot shows that Observation 20 has a leverage of about 0.642, while Observation 23 has a leverage of about 0.387. Both observations are above the screening threshold leverage of:

$${\displaystyle \frac{2p}{n}}={\displaystyle \frac{2\left(5\right)}{31}}=0.3226.$$

Thus, Observation 20 and Observation 23 are categorized as observations with relatively high leverage. However, high leverage does not by itself indicate that an observation has a significant impact on the regression results. This is evident in Observation 20, which has very high leverage but a relatively small residual. Conversely, Observation 30 has a very large residual, making its contribution to changes in the model estimation more significant.

Table 7. Influential Observation Diagnostics.

Observation	Leverage (Hat Matrix)	RStudent	Cook's Distance*	Interpretation
17	0.178	1.94	0.147	Potentially influential
20	0.642	−0.396	0.058	High leverage
23	0.387	0.899	0.103	High leverage
30	0.13	−3.916	0.296	Most influential

*Cook's Distance calculated from the EViews residual and Hat Matrix statistics.

Table 7. Influential Observation Diagnostics.

Observation	Leverage (Hat Matrix)	RStudent	Cook's Distance*	Interpretation
17	0.178	1.94	0.147	Potentially influential
20	0.642	−0.396	0.058	High leverage
23	0.387	0.899	0.103	High leverage
30	0.13	−3.916	0.296	Most influential

*Cook's Distance calculated from the EViews residual and Hat Matrix statistics.

Berdasarkan nilai $4/n$, threshold Cook's Distance penelitian adalah:

$${\displaystyle \frac{4}{31}}=0.129.$$

With those criteria, Observation 17 and Observation 30 passed the threshold, with Observation 30 showing the highest value. Observation 20 and 23 are more prominent based on leverage, but do not exceed the Cook's Distance threshold. These findings indicate that influence diagnosis needs to consider several measures simultaneously. Observation 30 is the observation that needs the most attention, as it has an extreme studentized residual and the largest Cook's Distance. Conversely, Observation 20 is primarily an observation with high leverage.

3.7. Sensitivity Analysis

To evaluate whether the regression results are sensitive to influential observations, a re-estimation was conducted by temporarily excluding certain observations. The main analysis in the available EViews output is the estimation without Observation 23, which is the observation with high leverage.

After Observation 23 was removed, the number of observations became 30. The R² value decreased from 0.575586 to 0.543665, while the Adjusted R² changed from 0.510292 to 0.470652. Although there was a change in goodness-of-fit, the model remained simultaneously significant, with an F-statistic of 7.446093 and a p-value of 0.000429.

Most importantly, the coefficient X1 remains negative and significant. In the full model, the coefficient of X1 is −0.089379 (p < 0.001), whereas after Observation 23 was removed, it became −0.086660 (p < 0.001). The change is relatively small and does not alter the substantive conclusions of the study.

Table 8. Sensitivity Analysis.

Variable	Full Sample (N=31)	Excluding Obs. 23 (N=30)
Constant	15.19885***	14.24246***
X1 – Waste Collection	−0.089379*	−0.086660*
X2 – Safe Drinking Water	−0.013172	0.009144
X3 – Improved Sanitation	0.00136	−0.012126
X4 – Slum Settlement	0.014305	0.014229
R²	0.575586	0.543665
Adjusted R²	0.510292	0.470652
F-statistic	8.815244***	7.446093***
Observations	31	30

Note: *** $p<0.001$. The sensitivity model excludes Observation 23.

Table 8. Sensitivity Analysis.

Variable	Full Sample (N=31)	Excluding Obs. 23 (N=30)
Constant	15.19885***	14.24246***
X1 – Waste Collection	−0.089379*	−0.086660*
X2 – Safe Drinking Water	−0.013172	0.009144
X3 – Improved Sanitation	0.00136	−0.012126
X4 – Slum Settlement	0.014305	0.014229
R²	0.575586	0.543665
Adjusted R²	0.510292	0.470652
F-statistic	8.815244***	7.446093***
Observations	31	30

Note: *** $p<0.001$. The sensitivity model excludes Observation 23.

The changes in X2 and X3 after Observation 23 was removed did not alter the conclusion because both variables remained insignificant. X4 also shows high coefficient stability, from 0.014305 to 0.014229. Thus, the results of the sensitivity analysis provide evidence that the main finding regarding the negative relationship between the coverage of waste collection and the poverty rate does not strongly depend on Observation 23.

3.8. Additional Sensitivity Check for the Most Influential Observation

Because Observation 30 has the highest RStudent and Cook's Distance values, an additional sensitivity check was conducted on that observation. Re-estimating by excluding Observation 30 resulted in an R² of 0.7253 and an Adjusted R² of 0.6813. The model remains simultaneously significant with an F-statistic of 16.5017 and a p-value <0.001.

The coefficient of X1 became −0.10452 with a p-value <0.001. Thus, although the magnitude of the coefficient changed from −0.08938 to −0.10452, the direction of the relationship remained negative and its significance remained very strong. X2, X3, and X4 remain insignificant.

Table 9. Sensitivity Check for Observation 30.

Variable	Full Sample	Excluding Obs. 30
X1 – Waste Collection	−0.089379***	−0.104520*
X2 – Safe Drinking Water	−0.013172	−0.006336
X3 – Improved Sanitation	0.00136	0.005639
X4 – Slum Settlement	0.014305	0.008597
R²	0.575586	0.725295
Adjusted R²	0.510292	0.681342
F-statistic	8.815244***	16.501682***
Observations	31	30

Note: *** $p<0.001$. The additional sensitivity model excludes Observation 30, which was identified as the most influential observation based on the combined residual and influence diagnostics.

Table 9. Sensitivity Check for Observation 30.

Variable	Full Sample	Excluding Obs. 30
X1 – Waste Collection	−0.089379***	−0.104520*
X2 – Safe Drinking Water	−0.013172	−0.006336
X3 – Improved Sanitation	0.00136	0.005639
X4 – Slum Settlement	0.014305	0.008597
R²	0.575586	0.725295
Adjusted R²	0.510292	0.681342
F-statistic	8.815244***	16.501682***
Observations	31	30

Note: *** $p<0.001$. The additional sensitivity model excludes Observation 30, which was identified as the most influential observation based on the combined residual and influence diagnostics.

The results reinforce the main finding that X1 continues to have a negative and significant relationship with the poverty rate even after the most influential observation was removed. Thus, the conclusion regarding X1 is not solely due to the presence of Observation 30.

3.9. Hypothesis Testing

Based on the main model estimates and sensitivity analysis, the hypothesis testing results are summarized in Table 9.

Table 10. Hypothesis Testing.

Hypothesis	Expected Relationship	Empirical Result	Decision
H1	X1 → Y (−)	β = −0.089379; p < 0.001	Supported
H2	X2 → Y (−)	β = −0.013172; p = 0.7460	Not supported
H3	X3 → Y (−)	β = 0.001360; p = 0.9723	Not supported
H4	X4 → Y (+)	β = 0.014305; p = 0.4647	Not supported
H5	X1–X4 jointly → Y	F = 8.815244; p = 0.000123	Supported

Table 10. Hypothesis Testing.

Hypothesis	Expected Relationship	Empirical Result	Decision
H1	X1 → Y (−)	β = −0.089379; p < 0.001	Supported
H2	X2 → Y (−)	β = −0.013172; p = 0.7460	Not supported
H3	X3 → Y (−)	β = 0.001360; p = 0.9723	Not supported
H4	X4 → Y (+)	β = 0.014305; p = 0.4647	Not supported
H5	X1–X4 jointly → Y	F = 8.815244; p = 0.000123	Supported

Partially, only H1 is supported. The coverage of waste collection has a negative and significant relationship with the poverty rate. H2, H3, and H4 are not supported because each variable does not show statistical significance at the 5% level. Simultaneously, H5 is supported. Thus, although only one variable is partially significant, the four environmental indicators collectively have a significant explanatory power regarding the variation in poverty levels between sub-districts.

3.10. Summary of Main Findings

The research results show that environmental conditions simultaneously have a significant relationship with the poverty levels between districts in Bandung Regency. The model explains approximately 57.56% of the variation in poverty levels, while the diagnostic tests do not show strong evidence of serious heteroscedasticity or multicollinearity.The most consistent finding is the negative relationship between the coverage of waste collection and the poverty rate. The coefficient X1 is −0.089379 and remains negative and significant when Observation 23 or Observation 30 is excluded in the sensitivity analysis. This indicates that the relationship between X1 and the poverty level is relatively robust against influential observations.

On the other hand, access to safe drinking water, access to healthy latrines, and slum settlements do not show a significant partial relationship with the poverty level in the model. These results do not mean that these three factors are not important for environmental health or community welfare, but rather indicate that, based on variations between sub-districts and the model specifications used, the statistical evidence regarding their direct relationship with poverty levels is not yet strong enough.

Overall, the research results provide empirical evidence that waste management is the environmental indicator most consistently related to variations in poverty in Bandung Regency in 2024. These findings serve as the main basis for discussions on the possible socio-economic mechanisms linking waste management quality with poverty.

4. Discussion

4.1. Main Findings in the Context of Environmental Poverty

This study aims to analyze the relationship between environmental conditions and poverty levels in 31 sub-districts in Bandung Regency, West Java. The regression results show that the four environmental indicators are simultaneously significantly related to the poverty level, with an R^2 value of 0.576 and an F-test p-value of 0.000123. However, at the partial level, only the percentage of waste collected (X1) shows a negative and significant relationship with the poverty level, while access to safe drinking water, access to healthy latrines, and slum settlements do not show a significant relationship.

The findings support the view that poverty is not only related to a lack of income but also to limited access to environments and basic services that support well-being. The literature on multidimensional poverty increasingly places drinking water, sanitation, housing, and various forms of environmental deprivation as important components of well-being conditions. Studies on the measurement of multidimensional poverty show that indicators of safe drinking water and sanitation have been used as components of standard living because limited access to both reflects forms of deprivation that are not always visible through monetary poverty measures [13,14].

That perspective is also reinforced by conceptual developments regarding the relationship between poverty and the environment. Recent literature emphasizes that the environmental dimension has not yet been fully integrated into poverty measurement, even though environmental quality is intrinsically related to human well-being. The development of the environment–poverty framework shows that environmental conditions can be an important dimension in explaining the vulnerability and quality of life of poor groups [17].

Thus, the research results from Bandung Regency provide an empirical contribution to the literature by showing that environmental service indicators at the regional level do not always have the same strength of relationship with poverty. In the context of this research, waste management appears to be more consistently related to variations in poverty compared to WASH indicators and slum settlements.

4.2. Waste Collection and Poverty

The strongest finding in this study is the negative and significant relationship between the percentage of waste collected and the poverty rate. The X1 coefficient of −0.089379 indicates that an increase of one percentage point in the coverage of waste collection is associated with a decrease in the poverty rate of about 0.089 percentage points, after controlling for access to drinking water, sanitation, and slum settlements. The relationship also shows a fairly good level of robustness because the X1 coefficient remains negative and significant after high-leverage observations and the most influential observations were examined through sensitivity analysis.

These findings are in line with the research by Schmidt et al. which shows that improving waste collection services in low-income communities can increase the use of waste collection services and reduce exposure to environmental risk factors. In the study, communities with waste collection interventions experienced a reduction in the number of flies by about 60% in areas with high service coverage [18].

Theoretically, the negative relationship can be explained through several mechanisms. First, waste collection services are part of the basic public infrastructure that determines the quality of the living environment. Regions with better waste management services tend to have cleaner environments and reduce exposure to unmanaged waste. Second, waste management can be related to public health through the reduction of disease vectors, pollution, and environmental exposure. Third, the quality of public services can be a broader indicator of institutional capacity and the ability of local governments to provide basic services in socio-economically better regions.

Research on low-income communities in Pakistan shows that low-cost waste collection interventions can increase service usage and reduce exposure to environmental risks, although their effectiveness highly depends on the level of service coverage and community characteristics [18]. These findings are relevant to the results in Bandung Regency because variable X1 shows a very large variation between sub-districts, ranging from 0 to 100%. This variation provides ample opportunity to identify the relationship between waste management services and poverty levels.

However, the negative relationship cannot be directly interpreted as a causal relationship. The research data is cross-sectional in nature, and the unit of analysis is the sub-district. Therefore, there is a possibility of a reverse relationship or common determinants. Sub-districts with lower poverty levels may have higher fiscal capacity, economic activity density, infrastructure, and institutional capacity, making them more capable of providing waste management services. Thus, the relationship found may reflect a two-way relationship between socio-economic conditions and the quality of public services.

4.3. Why Safe Drinking Water Was Not Statistically Significant

Unlike X1, access to safe drinking water (X2) has a negative coefficient of −0.013172, but it is not statistically significant (p = 0.7460). Thus, the research results do not provide sufficient evidence that the variation in access to safe drinking water between districts is partially related to the poverty level after controlling for other environmental variables.

The result does not mean that safe drinking water is not important for public welfare or health. International evidence actually shows a strong relationship between the quality of water services and various health outcomes. Meta-analysis by Wolf et al., for example, shows that improving drinking water quality can reduce the risk of diarrhea, while enhancing sanitation services also provides significant health benefits [19].

Research in Indonesia also shows that access to water, sanitation, and hygiene is related to children's health. Mulyaningsih et al. used data from Indonesia and demonstrated the importance of WASH for the development of early childhood health [20]. Thus, the insignificance of X2 in this study is more accurately understood as the absence of evidence of a partial relationship with poverty at the sub-district level and in the year of observation, rather than as evidence that access to safe drinking water has no benefits.

One possible explanation is the characteristics of the data. The average access to safe drinking water in this study has reached 80.43%, so the variation between districts may not sufficiently represent the differences in water quality or safety substantively. In addition, the access percentage indicator may not necessarily capture the dimensions of actual water quality, service continuity, affordability, distance to water sources, and service reliability.

Literature on water poverty also shows that water poverty is multidimensional and cannot be represented by just one access indicator. Studies on multidimensional water poverty indicate that aspects of resources, technology, management, welfare, and environment can provide different perspectives on the condition of water poverty [21].Thus, the insignificance of X2 may indicate that access indicators alone are not sufficient to capture the complexity of the relationship between water and poverty.

4.4. Sanitation and Poverty: Interpreting the Non-Significant Result

Access to healthy latrines (X3) has a coefficient of 0.001360 with a p-value of 0.9723, indicating that there is no statistically significant relationship with the poverty level. Besides being insignificant, the magnitude of the coefficient is very close to zero. These results differ from most literature that places sanitation as an important component of well-being and health. The systematic study by Wolf et al. shows that sanitation interventions can reduce the overall risk of diarrhea, while improvements in sanitation services at higher levels provide greater health benefits [19].

On the other hand, research on multidimensional poverty consistently uses sanitation as an indicator of living standards. Studies on multidimensional poverty in various countries show that access to sanitation is one form of deprivation that can distinguish between poor and non-poor groups [16,22]. The difference with the research results in Bandung Regency is likely due to the difference in the unit of analysis and the concept of outcome. Most previous studies analyzed households or individuals, whereas this study analyzes 31 sub-districts. The strong relationship at the household level can become weaker when the data is aggregated to the regional level because the variation between households is lost in the aggregation process.

Furthermore, variable X3 measures the percentage of access to healthy latrines, not the quality of sanitation facility usage, frequency of use, fecal sludge management, or waste disposal safety. Studies on urban sanitation show that sanitation is not just about the presence of toilets, but is related to the entire waste management system and urban governance [23].

Therefore, the results of this study should not be interpreted as sanitation being substantively unrelated to poverty. The results are more accurately interpreted as the absence of a significant partial relationship between the used indicator of access to healthy latrines and the poverty level in 31 sub-districts in 2024.

4.5. Slum Settlements and Poverty

The slum settlement variable (X4) has a positive coefficient of 0.014305, consistent with the direction of the hypothesis, but is not statistically significant (p = 0.4647). Thus, this study finds a direction of the relationship consistent with the hypothesis that areas with a higher coverage of slum settlements tend to have higher poverty levels, but the statistical evidence is not yet strong enough.

Theoretically, these results are still consistent with the perspective that poverty, housing quality, environment, and access to basic services are interrelated. The scoping review by Abdulhadi et al. shows that access to WASH and housing in slum areas is influenced by infrastructure, socio-cultural, socio-economic, governance, policy, and environmental factors. Limitations in WASH and housing can also impose health and socio-economic burdens on slum residents [24].

Recent studies on multidimensional poverty in the slums of Dhaka also show that infrastructure factors have a significant relationship with multidimensional poverty outcomes, while income-based poverty measures may fail to capture the environmental dimensions and vulnerabilities of slum areas [22].

However, there are methodological reasons why X4 is not significant in this study. First, slum settlements are not identical to monetary poverty. An area can be categorized as slum based on physical characteristics and basic services, but not all households within it have to be below the poverty line. Second, variable X4 is in the form of the percentage of slum settlement coverage at the sub-district level, so it does not capture the intensity or severity of slum conditions. Third, factors such as education, employment, local economic structure, population density, and transportation access likely also determine the level of poverty but are not included in the model. Thus, the results of X4 indicate that the relationship between slums and poverty is more complex than a simple linear relationship at the regional level.

4.6. Why Waste Collection Emerged as the Strongest Environmental Predictor

One of the important findings of this research is that out of the four environmental indicators analyzed simultaneously, waste transportation has the most consistent relationship with poverty. This is interesting because conceptually, drinking water, sanitation, and housing are also important components of a healthy environment.

One explanation is the characteristic of indicator X1, which has the largest variation in the dataset, with a range of 0–100%. This large variation allows differences in waste management services between districts to be more easily captured in the regression model. In contrast, X2 and X3 have relatively high averages, so some of the available variation may no longer represent significant differences in deprivation.

Furthermore, waste management is a public service that is very visible at the regional level. Poor waste management can result in the accumulation of waste, breeding grounds for vectors, blocked drainage, environmental pollution, and a decline in living quality. Literature on waste management in low-income communities shows that improving waste collection services can directly reduce exposure to environmental risks [18].

These findings are also in line with the literature that views WASH and waste management as interacting environmental systems. Studies on the integration of household waste management with WASH show various synergies between water management, sanitation, hygiene, and waste at the household level [23]. Thus, the research findings from Bandung Regency indicate that poverty reduction policies at the regional level should not only focus on economic transfers but also consider the quality of environmental services that are universal and region-based.

4.7. Multicollinearity and the Interpretation of Water and Sanitation

The correlation between X2 and X3 of 0.785 is the highest correlation among the independent variables. However, the Centered VIF values of 3.257 for X2 and 3.236 for X3 indicate that the relationship is not at a level suggesting serious multicollinearity.These findings are important in interpreting the insignificance of X2 and X3. The high correlation between water and sanitation indicates that these two basic services tend to be available together in some sub-districts. When both are included simultaneously in the regression, each coefficient represents the partial relationship after controlling for other indicators.

This can explain why evidence of a bivariate relationship does not always appear as a significant partial relationship. WASH literature also emphasizes that water, sanitation, and hygiene are interconnected systems, so the separation of effects of each component can become complex [19,24].

Thus, the research results should not be interpreted as evidence that water or sanitation is not important. On the contrary, the results indicate the limitations of the cross-sectional regression approach with aggregate indicators when several basic services have interrelated spatial distributions.

4.8. Robustness of the Findings

Sensitivity analysis provides support for the stability of the main findings. When Observation 23 was removed, the coefficient X1 changed relatively little from −0.089379 to −0.086660, and remained significant at p < 0.001. The overall model also remains significant, with an F-statistic of 7.446093 and p = 0.000429.

The finding is important because Observation 23 has relatively high leverage. With the direction and significance of X1 remaining after the observation was removed, the relationship between the coverage of waste collected and poverty does not appear to be fully driven by that observation.

The influence examination also shows the presence of observations with extreme characteristics. Therefore, the decision not to remove observations solely based on statistical values is the appropriate approach. In regional research, extreme observations can represent the actual conditions of a district and not measurement errors. Sensitivity analysis is used as a robustness check, not as a mechanism to obtain more significant results.

4.9. Policy Implications

The research findings have important implications for poverty reduction policies in Bandung Regency. The finding that the coverage of waste collection is the only significant environmental indicator suggests that improving waste management services can be considered as part of a strategy for social development and environmental health, rather than merely a technical waste management policy.

First, local governments can prioritize the improvement of waste collection coverage in sub-districts with relatively high poverty levels and low service coverage. This approach can combine poverty data with waste management service data, making interventions more targeted.

Second, waste management policies need to shift from merely increasing the volume of waste collected to providing regular, affordable, inclusive, and sustainable waste management services. The experience of interventions in low-income communities shows that increasing service coverage can reduce exposure to environmental risks, but the success of the program highly depends on the level of adoption and actual coverage [18].

Third, the findings regarding X2–X4 indicate that policies should not rely on a single environmental indicator. Although safe drinking water, sanitation, and slum settlements are not statistically significant in the model, all three remain important from a public health perspective. WASH literature consistently shows that water quality and sanitation are related to health outcomes, while slum areas have various environmental and social vulnerabilities [6,17].

The implication is the need for an integrated policy approach between poverty reduction, WASH, waste management, and improving settlement quality, rather than standalone sectoral interventions.

4.10. Theoretical and Empirical Contributions

This research provides two main contributions. Theoretically, the research results support the multidimensional poverty perspective, which posits that community well-being is not only determined by income but also by the ability to access basic services and an environment that supports a healthy and productive life [13,14,15].

Empirically, this research expands the literature by simultaneously examining several environmental indicators at the sub-district unit of analysis in Bandung Regency, West Java. Some previous literature on WASH and poverty uses household data, while other studies build multidimensional poverty or WASH poverty indices. This research takes a different approach by examining the relationship between environmental service indicators and regional poverty levels.

The novelty of the research primarily lies in the finding that waste collection shows a stronger and more robust relationship with poverty compared to indicators of drinking water, sanitation, and slum settlements in a cross-sectional regional model. These findings highlight the importance of explicitly incorporating waste management services into the environment-poverty analysis, as waste management is often treated as an environmental issue separate from the poverty alleviation agenda.

4.11. Limitations and Future Research

This study has several limitations that need to be considered when interpreting the results. First, the study uses cross-sectional data from 2024, so the relationships found are associations, not evidence of causality. Longitudinal research or panel data between districts is needed to test whether changes in the quality of environmental services are truly followed by changes in poverty levels.

Second, the number of observations is relatively small, namely 31 sub-districts. The sample size represents all the sub-districts in the research locus, but it still limits the statistical power and the ability to generalize the results to other areas.

Third, the environmental indicators used are aggregate measures at the sub-district level. The variable does not capture household heterogeneity, actual service quality, service continuity, affordability, or community behavior. WASH literature shows that formal access to services is not always identical to the quality or safety of the services received by the community [19,24].

Fourth, the model has not included several important factors that theoretically can affect poverty, such as education level, unemployment, economic structure, population density, transportation access, health, and the fiscal capacity of local governments. Omitted variable bias because these factors may influence the relationship between environmental quality and poverty.

Subsequent research should use panel data from sub-districts over several years to control for regional heterogeneity and examine the dynamics of the environment-poverty relationship. Future research could also use a spatial econometrics approach because poverty and the quality of environmental services likely have spatial patterns between districts. In addition, the development of an environmental multidimensional poverty index that combines water, sanitation, waste management, housing, health, and other environmental indicators can provide a more comprehensive measure of environmental deprivation.

5. Conclusions

This study examined the relationship between environmental service conditions and poverty across subdistricts in Bandung Regency, focusing on waste collection, access to safe drinking water, access to healthy sanitation, and slum settlements. The findings show that waste collection is the environmental factor most consistently associated with poverty, with better waste collection coverage associated with lower poverty levels. In contrast, access to safe drinking water, access to healthy sanitation, and slum settlements did not show statistically significant partial relationships with poverty in the estimated model. Therefore, the findings provide partial support for the initial hypotheses, particularly for the expected relationship between waste collection and poverty, while the hypotheses concerning the other environmental indicators were not empirically supported in this study.

The main contribution of this study is the empirical evidence that environmental services do not have uniform relationships with poverty at the subdistrict level. The findings extend the understanding of environmental dimensions of poverty by demonstrating the importance of considering waste collection as part of poverty-related environmental policy, while recognizing that access to drinking water, sanitation, and housing conditions may operate through more complex mechanisms that are not fully captured by aggregate subdistrict-level indicators.

From a policy perspective, the findings suggest that poverty reduction strategies in Bandung Regency should be complemented by improvements in basic environmental services, particularly the expansion and strengthening of waste collection services in areas experiencing greater socioeconomic vulnerability. However, improvements in drinking water, sanitation, and settlement conditions should continue to be integrated into broader public health and environmental policies, even though their partial relationships with poverty were not statistically significant in this study. Such an integrated approach is important because poverty and environmental deprivation are multidimensional and may reinforce one another.

This study has several limitations. The cross-sectional design limits the ability to establish causal relationships between environmental conditions and poverty. In addition, the analysis is based on subdistrict-level aggregate data, which may not fully capture differences between households within the same subdistrict. The relatively limited number of observations and the exclusion of other socioeconomic determinants of poverty may also constrain the generalizability and explanatory scope of the findings.

Future research should therefore employ longitudinal or panel data to examine changes in environmental services and poverty over time. Further studies should also incorporate socioeconomic, demographic, health, and spatial factors and consider spatial econometric or multidimensional poverty approaches. These extensions would provide a more comprehensive understanding of how environmental services interact with socioeconomic conditions and would support more targeted and evidence-based poverty reduction policies.