Compute realized reliabilities using Linear Regression in genetic evaluations

This document details the use of Linear Regression (LR) to estimate the accuracy of genomic predictions in livestock, specifically within the SMARTER project (EU Horizon 2020 grant 772787). It explains how realized reliability is calculated by correlating 'old' genetic evaluations at birth with 'most recent' evaluations after additional data (e.g. progeny records) are available. The method quantifies the relative increase in accuracy from birth to maturity using the formula [1/r(w,p)] - 1, where r(w,p) is the correlation between old and recent proofs. An r(w,p) of 0.8, for example, indicates a 25% increase in accuracy. High r(w,p) values suggest reliable early predictions; low values indicate unreliability. A minimum of 50 focal animals is required for valid results. The findings support improved breeding decisions in small ruminants.