Hay quality and intake by dairy cows. 2. Predicting feed intake with consumer-demand models

Daily food intake is the single most important factor affecting milk production by dairy cows. However, an animal’s choice of food depends not only on the nutritional characteristics of the food in question, but also on the nutritional characteristics of other available foods. Any prediction of intake should be based on the nutritional characteristics of all foods on offer. However, when the initial food-preference experiment possesses a control-specific design (i.e. experiments that include only a limited number of control foods for comparison) it is apparent that the prediction of future food choices must include the same controls as the initial experiment underpinning the prediction model. This requirement is clearly impractical. By drawing an analogy between animal food preference and economic choice, the total and relative dry matter intake of two oaten hays was modelled on their nutritive characteristics by estimating a consumer-demand model (here a generalised additive model representation of a direct bundle good model) from experimental data offering hays to lactating cows (adj-R2 > 80%; where adj-R2 is the value adjusted for the number of predictor terms in the model). To negate the problem of control-specificity, a simplex interpolation was developed to construct and test predictions of hay intake for a second food-preference experiment (adj-R2 > 53%; correlation between predictions and actual intakes = 76%). To improve prediction accuracy and avoid control-specificity, it is recommended that future preference experiments be designed to exclude control-specificity by mimicking fractional factorial designs, supported by a two-stage approach to select a cost-effective number of comparisons. Our approach to predicting food intake may be extended to a choice between more than two foods, and to combinations of foods other than oaten hays.