The Importance of Being Thorough: How Data Analysis Choices Impact the Perceived Relationship between Pollutants and Predictors

Understanding the relationship between socioeconomic drivers and anthropogenic surface water pollution is essential for addressing global threats to aquatic ecosystems and drinking water. For this purpose, data-driven approaches, i.e., obtaining new insights directly from datasets, have become increasingly relevant. However, many environmental pollution datasets display certain characteristics, such as measurements below the limit of detection and extreme values, which influence such data analysis. Here we use a global dataset of pharmaceutical pollution of rivers to show that not only the choice of method but also data preparation related to these characteristics, particularly the treatment of values below the limit of detection, aggregation, and normalization of values, can significantly impact the results of linear regression models and group comparison tests. These results highlight the importance of considering data preparation as part of statistical modeling and carefully documenting and discussing each step. We formulate recommendations for future studies and propose a procedure to gain insights into the relationships between socioeconomic drivers and anthropogenic pollution based on the multitude of models resulting from different combinations of data preparation steps. Our recommendations support a thorough analysis of the relationship between socioeconomic drivers and environmental pollution, and thus help draw the right conclusions.