Evaluation of imputation methods for microbial surface water quality studies.

Area of Science:

• Environmental microbiology
• Water quality assessment
• Statistical modeling

Background:

• Longitudinal studies assessing microbial water quality frequently encounter missing data points.
• Sample analysis deficiencies led to substantial missing data for Escherichia coli and enterococci in the Chicago Area Waterway System (2007-2009).

Purpose of the Study:

• To evaluate the performance of multiple imputation (MI) against data deletion, mean imputation, and median imputation for addressing missing microbial water quality data.
• To compare imputation methods using both simulated and original datasets with missing observations.

Main Methods:

• A simulation study was conducted using complete water quality observations with artificially introduced missing values.
• The performance of different imputation techniques was assessed by comparing results with the original dataset containing naturally occurring missing data.
• Linear regression models were employed to predict somatic coliphages density using Escherichia coli densities.

Main Results:

• Multiple imputation (MI) demonstrated the least bias compared to data deletion, mean, and median imputation methods.
• MI effectively controlled Type I error rates in the statistical analyses.
• The study identified that MI is a superior method for handling missing microbial water quality data.

Conclusions:

• Multiple imputation is recommended as a robust statistical technique for managing missing data in longitudinal microbial water quality studies.
• Further research is needed to explore the impact of varying missing data percentages and the "missing completely at random" assumption on MI performance.