Factors associated with grower herd performance in three New Zealand pig farms

The aim of this observational study was to identify performance
parameters, which can be used to predict market weight of a batch
of pigs on commercial farms. For that purpose, we obtained weekly
retro- and prospective production records from three New Zealand
pig farms. The observation periods on farms A, B, and C were 140,
127 and 90 weeks, respectively. As we expected the data to be
autocorrelated, we used two modelling approaches for multivariable
analysis: An autoregressive (AR) model and an ordinary least
squares (OLS) regression model (‘naive approach’). Analyses were
performed separately for each farm. Using an AR-model, we
identified four production parameters (weaning age, two sample
weights and days to market) across the three farms that were
effective in predicting market weight with accuracies greater than
70%. All AR-models yielded stationary and normally distributed
residuals. In contrast, residuals of the OLS-models showed
remaining autocorrelation on farms B and C indicating biased model
estimates. Using an AR-model also has the advantage that immediate
future observations can be forecasted. This is particularly useful
as all predictor variables (apart from ‘Days to market’) could be
obtained a month prior to marketing on all farms.