Published online by Cambridge University Press: 18 August 2016
Risk factors and variance components of pre-weaning mortality were estimated using generalized linear mixed models. Data were from 12 727 piglets born alive from 1338 litters recorded at the pig breeding farm of the University of Kiel from 1989 to 1994. Deviances due to risk factors were estimated by generalized linear model and their odds-ratios by generalized linear mixed model both with binomial errors and a logistic link. Variance components of sire, dam and litter were estimated using a logit or probit link function as well as a linear model for which estimates were transformed to the underlying continuous scale. Highest increase in deviance, indicating the risk factor, which accounts for the greatest amount of unexplained variation of pre-weaning mortality was obtained after exclusion of individual birth weight (1206) from the model, followed by year-season (217), parity-farrowing age or interval (58), genotype of piglets (56), sex (39), total number of piglets born (18) and gestation length (16). Substitution of individual birth weight successively by average piglet birth weight per litter, litter birth weight and standard deviation of birth weight within litter resulted in models with substantially lower explained variation of pre-weaning mortality. Odds of pre-weaning mortality was 1·5 times higher for males than for females and 2·0 times higher in piglets from German Landrace dams than from Large White dams. Odds increased to the fifth parity by 2·2 times the odds of the first parity or increased for the age group of dams between 850 and 949 days by 2·3 times the odds of the age group with less than 350 days. When the continuous risk factors of individual birth weight, average piglet birth weight and litter birth weight decreased or standard deviation of birth weight within litter increased by one standard deviation from the mean, the odds ratios increased by 6·0, 1·6, 0·8 and 0·4, respectively. Piglets with individual birth weights of 1·8, 1·5, 1·2 and 1·0 kg showed a rapid increase in odds ratios of pre-weaning mortality of 1·4, 2·7, 7·0 and 16·1, respectively, relative to piglets with 2·1 kg. Estimates of direct heritability for pre-weaning mortality on the linear observed, transformed underlying, logit and probit scale were 0·02, 0·06, 0·07 and 0·07, respectively. Low estimates of heritability for pre-weaning mortality, even on the underlying continuous scale, suggested low potential for improvement by selection. Therefore, selection for individual birth weight phenotypically closely associated with pre-weaning mortality was recommended to improve survival of piglets during the nursing period.