Evaluation of a Method to Analyze Pig Live Weight Data from Animal Sorting Technologies

Abstract

A method to analyze pig BW data collected by an animal sorting technology scale without individual pig identification was evaluated. Data for ten 1000-pig grow-finish barns were simulated by a stochastic model. The BW at each age was modeled as a predicted BW plus a daily residual error (mean = 0; SD = 1.4kg) and a within-day residual error (mean = 0; SD = 0.98kg). The number of times each pig was weighed was simulated to vary among pigs and to vary daily for each pig. The number of daily BW measurements taken per pig had an overall mean of 5.40 and SD of 1.23. Two types of data sets were simulated. A complete data set had biweekly data from 70 to 196 d of age. A truncated data set was simulated to reflect typical serial marketing of pigs and included biweekly data from 70 to 154 d of age and weekly truncated data from 161 to 175 d of age. The percentile means of the BW data were calculated and assigned a percentile identification of 1 to 100. The percentile means were fit to a mixed model nonlinear function (Bridges) with two random effects predicted for each percentile. The fitting of either the complete or truncated data sets reproduced the mean and variance of BW at each age and predicted age to achieve 120kg of BW. The serial marketing of pigs may produce small biases in the prediction of the actual percentile means. The fitting of the mean percentile data can be used to model the underlying mean and variation of BW growth.