A dynamic programming model was developed to evaluate conditions that promote food caching. Long-term observations (3-4 months per bird) of caching behaviour by Carolina chickadees, Parus carolinensis, are discussed in the light of the predictions from the model. Two different fitness functions were modelled: net energy-rate maximization and survival-rate maximization. Under most simulated conditions, energy-rate maximizers are predicted to cache at uniformly high rates. Two tradeoffs are important in caching decisions of survival-rate maximizers. (1) When fat levels are high, caching decisions should reflect trade-offs between time allocated to foraging and non-foraging behaviour. At the highest fat levels, the expression of non-foraging behaviour should be relatively more valuable than foraging. As a result, caching is not predicted because it increases foraging time and thus reduces time available for alternative behaviour. As fat levels drop below the maximum, foraging requirements become more important and caching rates are predicted to increase. (2) At intermediate to low fat levels, caching decisions should reflect a response to trade-offs affecting starvation risk. At very low fat levels, the forager should eat any food at the site of discovery to reduce the risk of starvation in the near term. At intermediate fat levels, the forager is not at immediate risk of starvation, so it should cache to reduce starvation risk in the long-term. These mass-dependent trade-offs should also affect diurnal patterns in caching rates: caching rates should be low at dawn, when the forager's fat reserves are at their minimum, and at dusk, when fat reserves are at their peak; caching rates should be highest at midday. The experimental results suggest that chickadees do not maximize net energy intake rates; instead, their behaviour is in broad agreement with the predictions of a survival-rate maximizer. However, one additional prediction was clearly not supported: larger birds cached less than smaller birds. Observed seasonal patterns in caching, retrieval and recaching rates are also discussed. © 1991 The Association for the Study of Animal Behaviour.
