Preprocessing alternatives for compositional data related to water, sanitation and hygiene

Area of Science:

• Environmental Science
• Public Health
• Statistics

Background:

• Sustainable Development Goals (SDGs) 6.1 and 6.2 track global access to water, sanitation, and hygiene (WASH) services.
• WASH service level data are compositional (sum to 100%) and often contain zero values, missing data, and outliers.
• Current statistical methods may not adequately address these data irregularities, potentially biasing WASH progress estimates.

Purpose of the Study:

• To evaluate imputation methods for addressing zero, missing, or both types of values in WASH compositional data.
• To propose robust statistical alternatives for handling outliers in WASH data, complementing existing monitoring programs.
• To enhance the accuracy and reliability of global WASH service level assessments.

Main Methods:

• Utilized compositional data analysis (CoDa) techniques.
• Applied isometric log-ratio (ilr) transformation for statistical analysis.
• Evaluated various methodological imputation alternatives for zero and missing values.
• Assessed robust statistical approaches for outlier detection and treatment.

Main Results:

• Identified and evaluated multiple imputation strategies for handling zero and missing values in WASH CoDa.
• Demonstrated the impact of different statistical treatments on WASH data accuracy.
• Presented comparative analyses of proposed methods using illustrative case studies.
• Highlighted the importance of robust statistical methods for reliable WASH data.

Conclusions:

• Standard statistical methods may yield biased WASH service level estimates due to data irregularities.
• Imputation and robust outlier detection methods are crucial for accurate WASH data analysis.
• Adopting advanced CoDa statistical techniques can improve global WASH monitoring and decision-making.