Precision of Genetic Variance and Heritability Estimates from Sorghum Populations1

Publisher Website
Google Scholar
Abstract
The emphasis of variance component and heritability estimation experiments has been on point estimation. Minimal attention has been given to confidence interval estimation. Our research was motivated by the limited use of confidence intervals and the importance of interval estimation to the statistical inference process. Our objectives were to evaluate variance component and heritability estimate precision from confidence interval estimate data and to describe the effects of experiment size and point estimate magnitude on interval estimate width. Point and interval estimates (1 ‐ α = 0.90) for the family variance component and heritability on a family mean basis were obtained from half‐sib, full‐sib, or S1 family selection experiments in four sorghum [Sorghum bicolor (L.) Moench.] populations. Agronomic traits were measured on 200 families per population grown on a Typic Argiudoll soil at Mead, NE, in 1981 and 1982 in a replications‐in‐incomplete blocks experimental design. Confidence interval estimate widths (widths) were functions of F =(family mean square)/(family ✕ year mean square) Mf/Mfy and tabled F‐values determined by degrees of freedom for Mf (dff) and Mfy (dffy) and the confidence coefficient (1 ‐ α). Ratios of widths point estimates (ratios) decreased at a decreasing rate as F increased for fixed F‐values. Widths and ratios decreased at a decreasing rate as dff and dffy increased for fixed F. Ratios for drf= dfy = 190 for variance components were 0.32 for F = 10.0 and 4.27 for F = 1.11, and for heritability were 0.05 for F = 10.0 and 4.05 for F = 1.11. Our data demonstrated the problems of wide interval estimates for large experiments and variation in precision among point estimates for experiments of fixed size.