赤池のベイズ型情報量規準を用いた乳用種雄牛評価におけるモデル選択

概要

論文の詳細を見る
Model selection in sire evaluation by Akaike's Bayesian information criterion (ABIC) was examined with Japanese Holstein first lactation data. The first half data, collected from 1975 to 1984, were used to evaluate sire merits by four sire models and four maternal grandsire models and to compute ABIC. Fixed effects were herd-year effect, calving month effect and age at calving effect. Predicts of the pair of sire and maternal grandsire were computed from the sum of sire merits and half of grandsire merits. The last half data, collected from 1985 to 1987, were used to compute actual average values of the pair of sire and maternal grandsire. Rank correlations between actual average values of the pair and predicts of the pair were obtained. For milk yields, the rank correlations increased as the values of ABIC were smaller. For fat percentage and SNF percentage, the rank correlations of maternal grandsire models were larger than those of sire models, and the values of ABIC of maternal grandsire models were smaller than those of sire models. The rank correlations within maternal grandsire models or within sire models were almost the same values for fat percentage and SNF percentage.