Abstract

COTES TORRES, José Miguel  and  SANCHES, Adhemar. BAYESIAN PHENOTYPIC STABILITY ANALYSIS USING JEFFREYS'S PRIOR. Rev. Fac. Nac. Agron. Medellín [online]. 2006, vol.59, n.1, pp.3077-3088. ISSN 0304-2847.

Shukla's variance is a very useful method for the analysis of phenotypic stability, computing the genotypic variance among the genotype by environment interaction, using variance component estimation of combined analysis of variance. New methodologies like REML or ML allow work with unbalanced data but do not have a good solution for the negative variance estimate problem. In this case, the component is taken as zero and the rest of the variance is redistributed into other components with positive estimates. The use of Bayesian methodology in the variance component estimation resolves satisfactory this problem without affecting the other components. This research uses the Bayesian estimation methodology for the solution of the mixed model in order to obtain the Shukla's variance for potato production data of 10 regional trials established in the Colombian Andean Region. The noninformative Jeffreys's prior and Independence Chain algorithm were used. The burnin period was 500 iterations and 1,16x10⁵ generations of joint posterior distribution were obtained. The REML methodology found three Shukla's variances with zero estimates. The corresponding Bayesian estimates were 89,35; 377,18 and 101,12 and the 95% confidence intervals were (2,13 - 371,70); (35,26 - 1363,67) and (2,33 - 434,53), respectively. Finally, these estimates do not affect other variance estimate components.

Keywords : Shukla's variance; Jeffrey's prior; importance sampling.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )