Selecting an appropriate statistical model for estimating genetic parameters: A case study of pig maternal breeds in Czechia

Abstract

A three-trait statistical model was designed for backfat thickness (BFT) and loin eye muscle depth (LMD) with accompanying variable average daily gain (ADG). Data from 82,507 Czech Large White and 37,556 Czech Landrace pigs were collected during on-farm performance testing from 2013 to 2022. Several animal models were tested with different combinations of effects and complexity. Model performance was evaluated by the linear regression (LR) method and predictivity. In Czech Large White, the preferred model comprised fixed effects of sex, birth year, herd, and ultrasound device, as well as random effects of herd-year-season (HYS), litter, and animal. In Czech Landrace, HYS was treated as fixed instead. The inclusion of maternal effects was not supported due to questionable impact on the main statistics. The selected model yielded mean absolute bias of 0.11 and 0.19, mean determination of 0.39 and 0.21, mean population accuracy of 0.36 and 0.31, and mean predictivity of 0.13 and 0.08 in Large White and Landrace, respectively. Heritability estimates were overall lower than those reported by other authors: 0.23, 0.10, and 0.10 in Large White and 0.26, 0.10, and 0.09 in Landrace for ADG, BFT, and LMD, respectively. Genetic correlations, on the other hand, reached relatively high values: 0.61, 0.61, and 0.66 in Large White and 0.71, 0.92, and 0.70 in Landrace for ADG-BFT, ADG-LMD, and BFT-LMD, respectively.