Statistical Implications of Zero Imputation for Missing Data in Cowpea (Vigna unguiculata L. Walp) Mutation Breeding: Evidence for Systematic Bias in Yield Traits

Abstract

Zero imputation for missing data in plant breeding experiments represents a widespread but statistically invalid practice that threatens the reliability of genetic evaluations, especially mutation breeding study with potential high rate of plant mortality. This study quantified the statistical consequences of zero substitution versus appropriate missing data handling (listwise deletion) in a factorial cowpea experiment involving five accessions, four ethyl methane sulfonate (EMS) concentrations, and three water withdrawal regimes. Six yield-related traits were analyzed using both approaches. Zero imputation systematically biased statistical inferences, reducing trait means by 65-85% and eliminating crucial genotype × treatment interactions (accession × EMS interactions, P = 0.0008 under proper handling, inestimable under zero imputation). Correlation magnitudes between yield components were reduced by 15-35%. Proper missing data handling identified TVU17315 as consistently superior (39.74 seeds/plant ± 7.23 SE, 33.11g hundred-seed weight ± 4.18 SE) and revealed 0.75% EMS as optimal, increasing pod production by 37.9% and seed production by 113.0% compared to controls. The 24-hour water withdrawal treatment yielded optimal results across traits. Zero imputation conflates biological failure with quantitative measurement, leading to incorrect breeding decisions and potential loss of valuable germplasm. Plant breeding programmers must adopt statistically valid missing data methodologies to ensure reliable genetic evaluations and accelerate development of improved varieties.