These empirical and theoretical studies suggest how economic variables might influence the way in which animals combine recent and long-term experience, yet behavioral ecologists could do much more. Specifically, no single study has manipulated environmental change and sampling error in a factorial way. In addition, we need more basic theoretical work. We need models of short-term maximization to account for effects like those observed by Cuthill and colleagues, and we need to link these studies with the mechanistic basis of animal memory (see chap. 3).
On a field edge, a starling hunts for insects in clumps of short grass. As it forages, its success or failure provides it with information about whether a particular clump is rich or poor. But starlings seldom forage alone, and the successes and failures of flockmates also provide clues about resource quality. A growing body of evidence suggests that "neighbors" can provide information about food resources. In most ofthis chapter, we have assumed that an animal's information comes from its direct experience in the environment—its successes and failures, the food it sees, the cues associated with food, and so on. For many animals, however, "the group" represents a central aspect of the environment, and so it comes as no surprise that an animal may use the actions of its groupmates as a source of information. In addition, behavioral ecologists have long thought that groups can improve feeding rates, and information transfer among group members can account for at least part ofthis facilitation effect (see, for example, Krebs et al. 1972; Lack 1968; Ward and Zahavi 1973). Recent work (Clark and Mangel 1984, 1986; Templeton and Giraldeau 1995; Valone 1989; Valone and Giraldeau 1993) has sharpened our questions about the distinction between public and private information.
Consider again the orthogonal tracking problem outlined in the previous section, but imagine this time that two individuals exploit our simplified environment ofvarying and stable resources. Ifindividual A samples according to the model, then individual B can avoid the costs of sampling by watching individual A's behavior. This problem is a game theoretical one (see boxes 1.3 and 1.4). Inman (Inman 1990; Krebs and Inman 1992) has studied this problem theoretically and experimentally. He argues that the only stable equilibria ofthis game occur when one individual samples at the individual optimum and the other parasitizes the sampler's actions. Intermediate "shared sampling" equilibria are unstable because if one individual increases its sampling rate, the other should decrease its sampling rate, leading inevitably to the stable "sampler-parasite" situation. Inman tested these predictions empirically by testing four pairs of starlings in both "alone" and "paired" conditions. In the paired condition, one individual lowered its sampling rate while the other sampled at the same rate as in the "alone" condition.
The simplest question that one can ask about public information is whether foragers use information from neighbors. This question suggests simple studies in which one manipulates the presence (or absence) of conspecifics. Investigators have conducted several studies of this type with intriguing results. Templeton and Giraldeau (1996) studied foraging starlings in a patch exploitation situation. They paired subjects with a trained stooge, who either gave up quickly or exploited the patch fully. Surprisingly, Templeton and Giraldeau found that the geometry of the experimental patch determined whether the stooge influenced patch-leaving behavior. When the experimental patches were linear arrays (egg-carton-like arrays of food wells) and the subject could exploit them systematically without information from the stooge, the stooge's behavior had no effect on patch leaving. However, when the patches were square arrays of food wells, the stooge's behavior did affect patch leaving, presumably because the subject had more difficultly implementing a simple exploitation rule. In another example, Smith et al. (1999) found that foraging crossbills were (in effect) better "empty patch" detectors when paired with two conspecifics, while a single conspecific did not improve their performance.
Results like these suggest that studies of public information must address subtle issues. These issues parallel the methodological problems in the closely allied field of social learning (see, for example, Galef 1988; Shettleworth 1998). For social animals, the presence of conspecifics influences behavior in many ways. A key challenge for students of public information is to disentangle the informational and noninformational effects ofsociality. One approach to this problem would combine social and nonsocial treatments with direct manipulations of the value of information. For example, one might create fixed and varying environments and test a focal animal in alone and paired conditions in a factorial way within these environments. If public information influences behavior, we would expect an interaction between the information treatments and the group size treatments. Recent work by Dornhaus and Chittka (2004) on the honeybee dance language provides a masterful example of how we might study the information value of social interactions.
2.7 The Behavioral Ecology of Information and Cognition
Information problems connect behavioral ecology with basic behavioral mechanisms such as learning, memory, and decision making. The mechanisms in question cover a broad swath of animal biology that includes sensory biology, neurobiology, psychology, and cognitive science, which taken together represent an enormous and important research enterprise. Students of foraging information are building connections with these mechanistic research programs in two ways. First, behavioral ecologists can use knowledge of behavioral mechanisms to constrain their models. For example, Kacelnik and his colleagues (Gibbon et al. 1988; Kacelnik et al. 1990) have incorporated the scalar property of animal time estimation (animals remember long intervals less accurately) into foraging models to provide a mechanistic account ofrisk sensitivity, patch exploitation, and animal preferences for immediacy. According to this view, the scalar property reflects a basic property ofthe neural timing system (Gibbon et al. 1997) that constrains foraging behavior. This approach assumes that some mechanism constrains animals to have less accurate representations of long time intervals and works out the consequences for foraging behavior. The second, and more challenging, type ofconnection occurs when behavioral ecologists use economic principles to provide novel insights into questions about behavioral mechanisms. An obvious example is signal detection theory, in which an economic model led to the rejection of the mechanistic idea of absolute sensory thresholds. In addition, the growing number of mechanistically based models in cognitive science and neuroscience provide new opportunities for behavioral ecologists. For example, neural network models of learning (Montague et al. 1995, 1996; Sutton and Barto 1981) may provide tools to generalize the simple models of tracking discussed in this chapter.
The models presented here rely on the mathematical machinery of statistics and stochastic processes, but foraging animals do not face information problems that precisely parallel the estimation and testing problems ofclassic statistics. In introductory statistics courses we learn to estimate quantities and make hypothesis tests. Foraging animals do not need estimates or hypothesis tests; they need to make decisions about how to feed. The development of signal detection theory presented earlier makes this point clearly. A statistician faced with a mixture of tasty and noxious beetles would take a sample and estimate the probability that the sample came from the tasty distribution. While this calculation is relevant to a beetle-eating forager, it really isn't the forager's problem. The forager needs to decide what to eat, and as signal detection theory shows, the optimal position of the "eat-don't eat" threshold depends on the tasty and noxious distributions and the costs and benefits associated with eating and avoiding the two types of beetles. The relevant body of statistical theory is statistical decision theory (DeGroot 1970; Lindley 1985; Dall et al. 2005), and not the classic statistics of estimation and hypothesis testing (Getty 1995 provides an elegant example of this difference). Nevertheless, we sometimes find it useful to frame problems as "estimation problems," as we did in our discussion of parallel tracking. This can be a useful modeling strategy in situations in which we don't know, or don't want to specify, how the acquired information will be used.
To say that animals do not need to make estimates does not mean that they don't. Animals can solve the same problem in different ways. Cuthill's starlings might have a simple procedural rule, such as "I'm tired so I'll spend a long time in this patch," or they might form some neural estimate of the current travel time between patches and use this to make a more sophisticated decision about patch exploitation. These questions tread in the realm of cognitive science. If starlings maintain some sort of representation or encoding of the current travel time, then a cognitive scientist would describe this as declarative knowledge (and according to some views, it would therefore qualify as a truly cognitive process; see Shettleworth 1998). If, instead, the starling uses a simple rule, we describe this as procedural knowledge. Studies of these types of questions are difficult, but they can be quite informative, as they have been in studies of navigation (e.g., Dyer 1998). So far as I am aware, there is no general theory about when one would expect a declarative representation to be better than a procedural one, although it should be easier to trick procedural rules by testing them outside of the context where they evolved.
Animals obtain information about the state ofthe environment as they forage. Information is valuable when it can tell an animal something that changes its behavior. The theory ofsignal detection provides a framework for the analysis of problems in which a decision maker must act in the face of environmental (and neural) noise. The overlap between the signal and noise distributions and the relative costs of false alarms and misses determine the optimal discrimination strategy.
Patch sampling has been an important topic within foraging theory. The distribution of patch types determines how information will affect patch exploitation. Animals must track changing environments, and we recognize two types of tracking problems. In orthogonal tracking, an animal must change its behavior to track a resource that it is not currently exploiting, while in parallel tracking, an animal can observe changes without changing its behavior. In orthogonal tracking problems, one focuses on the sampling rate; environmental change, and the benefits associated with varying and stable resources, influence the optimal sampling rate. In parallel tracking problems, one focuses on how animals should combine past and current information. Two factors, environmental change and sampling error, influence their behavior. When the environment changes rapidly, past information should be devalued. When a sample provides a noisy estimate of the current state, then past information should be emphasized and the current sample should be devalued. Finally, foragers can obtain information from conspecifics and group members. These public information problems should be analyzed using game theory.
The approach of Dall et al.'s (2005) recent review parallels the approach taken in this chapter, but it offers a broader perspective. A recent study by McLinn and Stephens (2006) explores the framework presented here, experimentally focusing on the effects of environmental uncertainty and signal reliability. Giraldeau (1997) provides a review of information in behavioral ecology with a more empirical emphasis. Bradbury and Vehrencamp's (1998) recent book on animal communication covers many of the same issues in a different context. Gescheider (1985) provides an engaging account of psychophysics. Volumes by Egan and Swets (Egan 1975; Swets 1996) provide reviews of signal detection theory and its applications. The volume edited by Dukas (1998a) reviews the relationship between behavioral ecology and cognition. Shettleworth (1998) gives a comprehensive, biologist-friendly treatment of psychological phenomena and practice.
Was this article helpful?