Suitability of random regression models for growth of Madras Red sheep under a field performance recording system

Abstract

Growth, being a repeatable trait, is found to be more suitable for random regression models (RRM). Especially in a field performance recording system the nature of data is heterogenous and the structure of data causes problems in genetic evaluation through normal univariate models. In this study, RRM was used to estimate variance components for growth in farmers’ flocks of Madras Red sheep, reared under the ICAR-Network Project on Sheep Improvement-Madras Red Field Unit (NWPSI), Government of India. Data on body weight collected over a period of six years in this project was used for the study. General linear model (GLM) with ANOVA for repeated measures was used to understand the effect of non-genetic factors including, center, season, sex and period. For the RRM, along with the significant non-genetic effects, orders of polynomial (k) fit up to 4, including the constant term, were considered for the random effects of sire and individual permanent environment. Error variances were modeled as homogenous and heterogenous classes. The heterogenous classes were modeled as a step function with four and ten different classes of age. The 10 class heterogenous error class model with order of fit 4 for both the random effects was found to have the best fit and the sire variance ratio estimated through this model for weights at 3, 6, 9 and 12 months of age were 0.195 ± 0.022, 0.27 ± 0.027, 0.04 ± 0.007 and 0.39 ± 0.047. The study indicated that RRM was found to be suitable for growth data measured over wide range of ages in farmers’ flocks. Precise estimates of variance components could be estimated for most part of the growth curve and this model avoids the error due to adjustment of data usually done for body weights at specific ages. In order to obtain precise estimates of genetic parameters through RRM, data recorded should be evenly distributed over all the age classes.