Establishment of a multilevel linear model to analyse the factors affecting piglet litter performance at birth

This study aimed to establish a feasible model for analysing factors affecting piglet litter performance at birth. Data of 61,984 litters were collected from 16 herds, and general linear model (GLM), multilevel Poisson regression model (MPM) and multilevel linear model (MLM) were established to compare their goodness of fit for these data. Influencing factors of piglet litter performance at birth were analysed using the established optimal model. Results showed the intraclass correlation coefficients of total born piglets (TBP), piglets born alive (PBA), low-birth-weight piglets (LBW), and average birth weight of piglets (ABW) reached 27.89%, 23.88%, 24.66% and 22.27%, respectively (p < .05). Akaike's information criterion and Bayesian information criterion in MLM of TBP, PBA, LBW and ABW were lower than those in GLM. Pearson residuals in MPM increased to nearly 1 after introduction of a discrete scale factor, and the p values in MPM were similar to those in MLM. Analyses of MLM indicated crossbred sows with good management supplemented with oregano essential oil and farrowing at warm season had higher TBA, PBA and ABW, but lower LBW than other sows (p < .05). In conclusion, MLM is superior to GLM and can replace MPM in analysing discrete data with hierarchical structure in pig production. More importantly, other potential influencing factors of litter performance at birth can be analysed using the established MLM in the future.