Premium
Selection of Stable Cultivars Using Phenotypic Variances
Author(s) -
Xie C.,
Mosjidis J. A.
Publication year - 1996
Publication title -
crop science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.76
H-Index - 147
eISSN - 1435-0653
pISSN - 0011-183X
DOI - 10.2135/cropsci1996.0011183x003600030007x
Subject(s) - statistics , coefficient of variation , biology , linear regression , selection (genetic algorithm) , stability (learning theory) , mathematics , gene–environment interaction , variance (accounting) , correlation , correlation coefficient , residual , genotype , genetics , geometry , accounting , algorithm , machine learning , artificial intelligence , computer science , business , gene
Genotype‐environment interactions reduce the correlation between phenotype and genotype and decrease selection progress. Type 1 stability measures (simple variance across environments S 2 i and related coefficient of variability CV i ) are biased because they do not distinguish the effects of locations and years. We propose the unbiased estimate of phenotypic variance ( V p ) and phenotypic coefficient of variability (PCV i ) as an alternative to S 2 i . The V p , unlike S 2 i and Type 4 stability measure MS Y/L (years‐within‐location mean square), includes location, year within location, and error variance components from analysis of variance based on a linear model. The F ‐test can be used to classify cultivars into location unstable, year unstable, or both location and year unstable, or vice versa. The actual model used in data analyses was based on the mean across replications in each environment (P ij .) because the plot (replication) data were not available. Hence, the corresponding V p and PCV i were denoted as V ′ P and PCV ′ i . The V ′ P underestimates V p . Monte Carlo simulation was used to evaluate the performance of V ′ P and S 2 i . A wheat ( Triticum aestivum L.) data set was used to demonstrate the calculation of the above parameters. The results from rank correlation showed that V ′ P was correlated to MS Y/L, the regression coefficient ( b i ), PCV ′ i , and S 2 i ( P < 0.05). The V ′ P was not correlated to the residual mean square of deviation from the regression (δ 2 i ). The V ′ P and S 2 i had similar sampling errors. When less than 18 locations and 6 years are used, V e is particularly useful since S 2 i is biased.