Application of Multiple Imputation for Missing Values in Three-Way Three-Mode Multi-Environment Trial Data.

Area of Science:

• Agricultural Science
• Biostatistics
• Data Science

Background:

• Missing values frequently occur in three-way, three-mode multi-environment trial (MET) data within plant breeding.
• Accurate estimation of these missing values is crucial for robust data analysis and reliable breeding decisions.

Purpose of the Study:

• To propose modified models for estimating missing observations in MET data.
• To develop and evaluate a novel hierarchical clustering approach for missing data imputation.
• To compare the performance of different multiple imputation (MI) methods.

Main Methods:

• Four MI techniques were employed: multiple agglomerative hierarchical clustering, normal distribution model, normal regression model, and predictive mean matching.
• The normal distribution, normal regression, and predictive mean matching models were implemented using both Bayesian and non-Bayesian analyses.
• The hierarchical clustering approach utilized a clustering procedure with randomly selected attributes and nearest neighbor imputation.

Main Results:

• Bayesian analysis models showed slightly higher accuracy than non-Bayesian models but were more computationally intensive.
• The novel multiple agglomerative hierarchical clustering method demonstrated superior overall performance in accuracy and efficiency.
• Performance was assessed by imputing varying proportions of missing data in complete datasets.

Conclusions:

• Multiple agglomerative hierarchical clustering is an effective and accurate method for handling missing data in MET datasets.
• The study provides valuable insights into imputation strategies for complex agricultural datasets.
• This method can improve the reliability of plant breeding programs by addressing data gaps effectively.