We describe a model to predict the diet selection of a population of animal
s, based on simple assumptions about the characteristics of the individuals
in a population, including the variation between them. Individuals are cha
racterized by three parameters with biological relevance; a nutrient (prote
in) requirement, an ability to discriminate between foods of different prot
ein contents and a need to collect information about both foods. Each anima
l selects perfectly a diet that avoids both a deficiency and an excess of p
rotein, where this is possible. To construct the population two further ass
umptions are made. The first is that the values of each parameter are drawn
from uncorrelated normal distributions subject to the values being logical
ly possible. The second is that, for different mean values for the populati
on, the standard deviation is directly proportional to the mean so that the
coefficient of variation is independent of the mean. The model was used to
predict the outcomes of six hypothetical experiments, using 100 individual
s on each treatment, where the values of the three parameters were systemat
ically varied. In the experiments one food was always of low protein conten
t while the protein content of the other was the treatment variable. The qu
antitative effects of varying either the mean value of the parameters, or t
heir variation, on both the mean composition of the diet selected, and on i
ts variation, were not possible to predict without using the model. The sha
pe of the population response was different to that for any individual. Ext
ensions to the model may be able to increase its relevance to practical iss
ues of diet selection. (C) 2000 The Association for the Study of Animal Beh
aviour.