Abstract

Any animal whose form or behavior facilitates the avoidance of predators or escape when attacked by predators will have a greater probability of surviving to breed and therefore greater probability of producing offspring (i.e., fitness). Although in theory the fitness consequences of any antipredation behavior can simply be measured by the resultant probability of survival or death, determining the functional significance of antipredation behavior presents a surprising problem. In this review we draw attention to the problem that fitness consequences of antipredation behaviors cannot be determined without considering the potential for reduction of predation risk, or increased reproductive output, through other compensatory behaviors than the behaviors under study. We believe we have reached the limits of what we can ever understand about the ecological effects of antipredation behavior from empirical studies that simply correlate a single behavior with an apparent fitness consequence. Future empirical studies must involve many behaviors to consider the range of potential compensation to predation risk. This is because antipredation behaviors are a composite of many behaviors that an animal can adjust to accomplish its ends. We show that observed variation in antipredation behavior does not have to reflect fitness and we demonstrate that few studies can draw unambiguous conclusions about the fitness consequences of antipredation behavior. Lastly, we provide suggestions of how future research should best be targeted so that, even in the absence of death rates or changes in reproductive output, reasonable inferences of the fitness consequences of antipredation behaviors can be made.

Predation is clearly one of the major selection pressures that determine the form ( Endler, 1991 ) and behavior of animals ( Lima, 1998 ). For example, almost all populations of animals suffer major mortality in their first year because of predation ( Newton, 1998 ). Any animal whose form or behavior facilitates the avoidance of predators or facilitates escape when attacked by predators will clearly have a greater probability of surviving to breed and therefore greater probability of producing offspring (i.e., fitness). Although in theory the fitness consequences of any antipredation behavior can simply be measured by the resultant probability of survival or death, determining the functional significance of antipredation behavior presents a surprising problem. Put simply, death rates are difficult to measure either because observing predation events in natural systems is difficult or because ethical considerations prevent appropriate experiments. Furthermore, conclusions about the adaptive value of a morphological or behavioral trait can still be ambiguous even if death rates can be correlated with a particular behavior where all other things are equal (e.g., in an experiment). That is because in natural systems there may frequently be many other ways for animals to compensate because all other things are never equal, so that the particular behavior can actually be of little consequence for individual fitness. Such compensation may also be confounding when using death rates to determine the fitness consequences of an antipredation behavior in natural systems because although many antipredation behaviors may increase fitness they do so by reducing probability of attack, but once an animal is attacked, other behaviors may determine probability of capture. Variation in death rate then results from both probability of attack and capture, controlled by two or more behaviors that are unlikely to be independent and that probably interact. These points are best illustrated with an example. Aerodynamic theory and laboratory studies have demonstrated that fatter birds fly more slowly (reviewed in Lind et al., in press ), hence, it is frequently concluded that fatter birds of greater mass will be more at risk of capture when being chased by hawks. Yet, if fatter birds simply adopt behaviors that reduce the risk of being attacked ( Cresswell, 1999 ) or facilitate escape, then there may be no obvious relationship between the mass of a bird and mortality to predators, one major component of individual fitness ( Lind, 2004 ).

In this review, we draw attention to the problem that fitness consequences of an antipredation behavior cannot be determined without considering the potential for reduction of predation risk or increased reproductive output through other compensatory behaviors. In short we make the case that we have reached the limits of what we can ever understand about the ecological effects of antipredation behavior from empirical studies that simply correlate a single behavior with an apparent fitness consequence. Future empirical studies must involve more behaviors to consider the range of potential compensation to predation risk. This is because antipredation behaviors are a composite of many behaviors that an animal can adjust to accomplish its ends.

First, we will discuss why observed variation in antipredation behavior does not have to reflect anything about individual fitness. Second, we outline how antipredation behavior is typically studied and demonstrate that few studies can draw unambiguous conclusions about the fitness consequences of antipredation behavior. Third, we provide suggestions of how future research should best be targeted so that even in the absence of death rates or changes in reproductive output reasonable inferences of the fitness consequences of antipredation behaviors can be made. For example, we can consider multiple behaviors simultaneously, or correlate antipredation behaviors with indices of fitness, or study prey choice by predators and predator hunting behavior. We do not intend to review the field of antipredation behavior in animals, but our aim is simply to explore how we can improve empirical studies of antipredation behavior and how to avoid pitfalls in the quantification of their fitness consequences (for relevant reviews see Bednekoff and Lima, 1998 ; Brown and Kotler, 2004 ; Lima, 1998 , 2002 ; Lima and Dill, 1990 ; Ydenberg and Dill, 1986 ).

The relation between antipredation behaviors and fitness

Predation risk is usually a composite of several interacting factors, and so it is often difficult to quantify simply. Predation risk (death rate) for an animal is a function of attack frequency and its probability of being caught when attacked. Attack frequency (attack rate) incorporates the reaction of predators to the behavior of prey, for example, a functional and numerical response. All of the behaviors that a prey can adopt to modify its risk of being targeted and caught when attacked comprise prey vulnerability. The key variable in determining predation risk is probably prey vulnerability because predators that are foraging optimally will select the prey that give the maximum energy return for energy invested in capture, that is those individuals of a prey species that are the easiest to catch ( Stephens and Krebs, 1986 ). Prey vulnerability will also influence relative prey abundance and the predator's functional and numerical response because even a very common prey will not be attacked if it cannot be captured. Theoretically then, measurements of prey vulnerability should measure predation risk and there is some empirical evidence to support this ( Biro et al., 2003c ; Quinn and Cresswell, 2004 ; Sinclair and Arcese, 1995 ).

Vulnerability of prey is determined in its simplest form by the trade-off between self-maintenance and allocation of time by the prey individual to antipredation behaviors, for example, feeding versus use of refuge, mass, vigilance, flock spacing, distance from cover, or flock size. In essence, any animal that maximizes its antipredation behaviors will never be eaten, but it will of course starve and certainly never have any reproductive fitness. However, whenever an animal starts to allocate resources to some other activity than antipredation, then its vulnerability to predation increases.

This means that even if part of a population of animals shows relatively poor antipredation behavior, it does not necessarily mean that they have lower fitness. Although they may be more likely to be depredated, on average, those that survive may have more resources to allocate to produce young, on average, so that the fitness of the low vigilance strategy may be similar to a high vigilance strategy.

Vulnerability of prey in its more realistic form of course consists of many antipredation behaviors, all expressed to varying degrees dependent on the trade-off between time or energy allocated to the behavior or to an alternative fitness-enhancing behavior ( McNamara and Houston, 1986 ). Crucially, the various antipredation behaviors may not be independent, so allocation of many resources to a particular antipredation behavior may allow few resources to be allocated to another, or vice versa, resulting in an equal fitness ( Figure 1 ). For example, an animal may allocate little time to antipredation vigilance but may allocate much time to defending a territory, which instead provides safety from predators. This is what we term predation risk compensation, where an antipredation behavior that is little expressed may be compensated for by another antipredation behavior that is strongly expressed. Overall fitness may be a consequence of many behaviors, rather than just the two illustrated in Figure 1 .

Figure 1

How allocation of resources from one antipredation behavior to another may result in equal fitness despite a low allocation to one of the behaviors. Increasing body reserves do generally come with a survival cost (e.g., due to increased foraging effort), whereas increasing predator avoidance enhances survival. This graph depicts how the costs of increasing body reserves can be compensated for fully if more resources are allocated to an antipredation behavior resulting in equal fitness as when allocation to body reserves and to predator avoidance are reduced.

Figure 1

How allocation of resources from one antipredation behavior to another may result in equal fitness despite a low allocation to one of the behaviors. Increasing body reserves do generally come with a survival cost (e.g., due to increased foraging effort), whereas increasing predator avoidance enhances survival. This graph depicts how the costs of increasing body reserves can be compensated for fully if more resources are allocated to an antipredation behavior resulting in equal fitness as when allocation to body reserves and to predator avoidance are reduced.

Predation risk compensation may be further complicated because the fitness consequences of allocating the same amount of resources into different antipredation behaviors may differ. We demonstrate this problem with a simple example: if the probability that an animal escapes from a predator depends on its mass because mass compromises its running ability, then it seems reasonable to conclude that individuals with higher mass have a greater predation risk. Yet, if the probability of escaping from a predator depends more on early detection of a predator then mass will be relatively unimportant compared to vigilance ( Figure 2 ). The relative fitness consequences of allocation of resources to different antipredation behaviors may then vary depending on specific ecological circumstances ( McNamara and Houston, 1994 ) or an animal's internal state ( Houston et al., 1988 ; Mangel and Clark, 1988 ). For example, contrary to the examples above ( Lind, 2004 ), if animals forage far away from predator-concealing cover where attacking predators are always detected before they reach the prey, then endurance, escape speed, and perhaps maneuverability can be more important than early predator detection.

Figure 2

How allocation of resources to one antipredation behavior may be more important to fitness than another antipredation behavior.

Figure 2

How allocation of resources to one antipredation behavior may be more important to fitness than another antipredation behavior.

Predation risk compensation has a final level of complexity because alternative behaviors, with different costs and benefits are not just available simultaneously; there is also a range of alternative behaviors that can occur at different stages during the predation event. For example, an animal may choose to feed in a small group because small groups may be relatively inconspicuous and so infrequently attacked by predators. Yet, when a small group is attacked, the capture rate may be high relative to larger groups because the confusion and dilution effect is less in smaller groups. The overall fitness consequence of choice of group size will therefore be the consequence of both attack and capture probability that are affected in opposite ways (see for example, Cresswell, 1994b ; Lindström, 1989 ; Vine, 1971 ).

Therefore, any of the many ecological studies that measure relative vulnerability in terms of a single behavior or at a single stage during predation and conclude that, for example, animals with lower vigilance rates are more at risk than those with higher rates are potentially flawed, even if the study measures death rate as a consequence or reproductive output. This is because of the potential for compensation in other dimensions such as choice of for example feeding site or group size ( Lank and Ydenberg, 2003 ). Similarly, any laboratory study that measures variation in antipredation behavior in terms of a single behavior, constraining an animal's choice of alternative behaviors (“all other things being equal”) cannot draw any conclusions about the fitness of a particular behavior in the natural “multibehavioral” system ( Irving and Magurran, 1997 ; Lima and Bednekoff, 1999 ; Lind, 2004 ; Wolff and Davis-Born, 1997 ).

The complexity of antipredation behavior has been acknowledged before; in one of the classic theoretical analyses of antipredation vigilance, Lima (1987) concluded (with our comment in parentheses): “The model developed herein, however, suggests that this relationship (between vigilance and predation risk) may be complex even in seemingly simple situations. (…) It is more fruitful for present purposes, however, to examine potential problems with how we as investigators study the risk of predation and its effects. For instance, experimental manipulations that purport to examine scanning as it relates to one factor may actually be examining scanning as it relates to changes in two (or more) factors.” Nevertheless in many ecological studies variation in predation risk is often invoked as an explanation for observed form or behavior without any measures of death rate or inferences about reductions in reproductive output or considerations of possible other compensatory behaviors (for reviews see Elgar, 1989 ; Lima and Dill, 1990 ). In the next section we give examples of how variation in predation risk has been studied to demonstrate that despite very many antipredation behavior studies being carried out relatively few can actually draw firm conclusions with respect to fitness.

How studies have attempted to relate antipredation behavior to fitness

There are three ways that studies have been carried out to determine the fitness consequences of antipredation behavior. First, variation in antipredation behaviors can be related to death rates: this has rarely been done successfully because predation events are difficult to observe and because death rates are often a consequence of several interacting behaviors, it is difficult to make behavior-specific conclusions. Second, variation in antipredation behavior can be related to reproductive output: this has, to the best of our knowledge, rarely been done because of the difficulties of measuring antipredation behavior and reproductive output in the same individuals over a sufficiently long time and large enough scale to draw meaningful conclusions and because reproductive rates will be a consequence of several interacting behaviors. Third, variation in antipredation behavior can be related to correlates of death rate: although this approach is common, this has rarely been done successfully because few studies have considered the full potential for compensation in other behaviors that may change predation risk.

Studies that relate antipredation behavior to death rates

When discussing studies of how effectively an antipredation behavior reduces measured death rate, it is important to make the distinction between behaviors that affect the probability of attack (i.e., occur before attack) and those that affect the probability of capture (i.e., occur after attack). This is well illustrated by returning to a previous example, where small groups may be relatively inconspicuous and so infrequently attacked by predators, yet be subject to a high capture rate when attacked. A study that examined the effect of group size by measuring attack rate might conclude that group size was positively correlated with death rate, while another study that examined the effect of group size by instead measuring capture rate might conclude the opposite, that group size was negatively correlated with death rate. In reality, a more useful estimate of the fitness consequence of choice of group size would be the probability of attack multiplied by the probability of capture (see for example, table 3 in Lindström, 1989 ).

Another important distinction to be made is the one between attack and death rates. Most studies that have tried to relate antipredation behavior to death rates are flawed because they do not consider attack rate as the most important determinant of predation risk and instead they measure capture rates. We demonstrate this with another example. Juvenile redshanks Tringa totanus feeding on a risky salt marsh have a very low probability of capture for each sparrowhawk Accipiter nisus attack because of effective antipredation behaviors such as avoidance and flocking ( Cresswell, 1994a , b ). However, because juvenile redshanks are forced to feed on the salt marsh for long periods due to competition ( Cresswell, 1994a ) and energy costs ( Yasué et al., 2003 ) they are attacked frequently during the winter, and their probability of being killed during the winter is very high ( Cresswell and Whitfield, 1994 ). What probably determines fitness for redshanks in this system is the number of days that an individual is forced to feed on the risky salt marsh rather than its antipredation behavior when on the salt marsh. This shows unambiguously that one can easily ascribe erroneous significance to results obtained by measuring capture rates without taking attack rate into account.

A further reason to make the distinction between attack and capture when examining studies that relate antipredation behaviors to death rates is methodological. Studies that measure how a behavior influences the probability of capture inevitably measure death rate, and so their results in fitness terms are relatively unequivocal (at least under the conditions of the study). However, behaviors that may influence the probability of attack are less clearly directly related to fitness: for example, the fitness consequences of a low attack rate through avoiding a predator may be manifest in low body condition or higher risks elsewhere. This means that behaviors that influence the outcome of predator attacks are more likely to be observed, studied, and indeed published because their results in fitness terms are clear. This may then mean that behaviors that reduce death rate through reducing attack probability are relatively unrepresented in the literature, even though such behaviors could be theoretically far more important (for example see, Lind, 2004 ).

Table 1 gives a list of studies that have related variation in antipredation behavior during attack to fitness. Interpretation of the results of these studies seems mostly straightforward: either a behavior is more or less likely to lead to death on attack by a predator. It is important to remember, however, that if the study is at all experimental and has constrained the predator's behavior then, although a behavior may affect probability of capture, it may be of little significance because the predator in natural circumstances may never attack an individual that might respond in that way. Hence, experiments that fall under this category should require support by natural observations for results to be of relevance. Also, even in natural systems, because individuals may only spend a small proportion of their time carrying out a risky behavior, the capture rates associated with a particular behavior may not give a good indication of how important the behavior is in determining fitness for most individuals. This will be complicated by the fact that some predators preferentially target vulnerable individuals of some but not other species ( Temple, 1987 ) and because observers may direct their attention to situations where predators have a high capture rate, so making behaviors that occur at a low frequency seem more significant.

Table 1

Studies that have related variation in antipredation behavior during attack to death rates

Behaviors
 
Species
 
Reference
 
Escape performance Common lizard Clobert et al. (2000) 
 Atlantic silverside Lankford et al. (2001) 
Escape response Redshank Cresswell (1993) 
  White-tailed deer Odocoileus virginianus , Mule deer Odocoileus hemionus Lingle and Pellis (2002) 
Predator recognition  Brook trout Salvelinus fontinalis Mirza and Chivers (2000) 
Pursuit-deterrent signals Skylark Cresswell (1994c) 
 Skylark  Rhisiart, 1989, cited in Hasson (1991) 
 Thomson's gazelle Fitzgibbon and Fanshawe (1988) 
Predator-induced hatching Tree frog Warkentin (1995) 
Early detection of attack Thomson's gazelle Fitzgibbon (1989) 

 
Wood pigeon
 
Kenward (1978)
 
Behaviors
 
Species
 
Reference
 
Escape performance Common lizard Clobert et al. (2000) 
 Atlantic silverside Lankford et al. (2001) 
Escape response Redshank Cresswell (1993) 
  White-tailed deer Odocoileus virginianus , Mule deer Odocoileus hemionus Lingle and Pellis (2002) 
Predator recognition  Brook trout Salvelinus fontinalis Mirza and Chivers (2000) 
Pursuit-deterrent signals Skylark Cresswell (1994c) 
 Skylark  Rhisiart, 1989, cited in Hasson (1991) 
 Thomson's gazelle Fitzgibbon and Fanshawe (1988) 
Predator-induced hatching Tree frog Warkentin (1995) 
Early detection of attack Thomson's gazelle Fitzgibbon (1989) 

 
Wood pigeon
 
Kenward (1978)
 

Studies showing that antipredation behaviors accrue fitness benefits under attack ( Table 1 ) have shown that responding quickly on attack augments survival ( Fitzgibbon, 1989 ; Kenward, 1978 ) and that correct identification of attacking predators and appropriate responses to the predators are important for reducing the probability of death under attack ( Cresswell, 1993 ; Lingle and Pellis, 2002 ; Mirza and Chivers, 2000 ). Even embryos can benefit from responding rapidly under attack. Tree frog embryos Agalychnis callidryas increase their survival by hatching when their arboreal eggs are attacked by snakes and escape by falling into the water below ( Warkentin, 1995 ). That pursuit-deterrent signals incur fitness benefits has also been shown for Thomson's gazelle Gazella thomsoni ( Fitzgibbon and Fanshawe, 1988 ) and skylarks Alauda arvensis (Rhisiart, 1989, cited in Cresswell, 1994c ; Hasson, 1991 ). Also growth rate in Atlantic silversides Menidia menidia appears to be inversely related to survival because the fast-growing and sated fish suffer from impaired predator evasion as compared to slow-growing and hungry fish ( Billerbeck et al., 2001 ; Lankford et al., 2001 ). Similarly, locomotor endurance in juvenile common lizards Lacerta vivipara seems to be connected to growth rate and hence incur an increased predation risk, as measured by reduced survival ( Clobert et al., 2000 ; see also Le Galliard et al., 2004 ). This is somewhat counterintuitive because one could argue that larger individuals should have a relatively higher locomotor performance and would be more adept at escaping predators.

Table 2 gives a list of studies that have related antipredation behaviors that occur prior to attack to death rates. Some of the best evidence for a clear link between antipredation behavior and fitness comes from studies that relate overall activity to death rates. This approach sidesteps the main problem identified in this review, the presence of behavioral compensation confounding any single behavior relationship because activity is a composite of many antipredation behaviors. That a reduction of activity levels in response to increased predation risk incurs fitness benefits has been documented in various taxa (e.g., Schwarzkopf and Shine, 1992 ; Sih, 1986 ; Skelly, 1994 ; e.g., Anholt and Werner, 1995 , 1998 ; Biro et al., 2003a , b , 2004 ; Downes, 2002 ; Persons et al., 2001 , 2002 ; Wisenden et al., 1999 ). Although these studies show that high activity incurs an increased predation risk in general, most studies investigating the consequence of activity level on survival have not investigated antipredation behavior per se but have instead used predation risk treatments to investigate questions of general ecological significance. These studies have for example led to insights about how behavioral responses to variation in predation risk can affect population dynamics (e.g., Anholt and Werner, 1995 ; Biro et al., 2003a , b ). Measurement of activity levels is thus a reliable and important method to reach conclusions about how vulnerable individuals are relative to others. However, the combination of the diverse nature of antipredation behaviors and the impression that animals modify every aspect of their behavioral repertoires to variation in predation risk (reviewed in Lima, 1998 ) suggests that measuring only activity levels, rather than constituent behaviors, will not allow us to understand these constituent behaviors. Another problem with studying activity levels is that many animals are far too large and elusive for this to be a possible approach.

Table 2

Studies that have measured the consequences of antipredation behaviors before attack in terms of death rate

Behaviors
 
Species
 
Reference
 
Activity  Tadpole Rana catesbeiana Anholt and Werner (1995) 
  Tadpole Rana sylvatica Anholt and Werner (1998) and Skelly (1994) 
  Common garden skink Lampropholis guichenoti Downes (2002) 
  Wolf spider Pardosa milvina Persons et al. (2001) 
  Wolf spider Pardosa milvina Persons et al. (2002) 
  Mosquito larvae Culex pipiens Sih (1986) 
  Amphipod Gammarus minus Wisenden et al. (1999) 
Activity, habitat use  Rainbow trout Oncorhynchus mykiss Biro et al. (2003a , b , 2004 )  
Activity, cryptic behavior  Skink Eulamprus tympanum Schwarzkopf and Shine (1992) 
Group benefits Butterflies Burger and Gochfeld (2001) 
 Redshanks Cresswell (1994b) and Whitfield (2003) 
 Ocean skaters Foster and Treherne (1981) 
 White-nosed coatis Hass and Valenzuela (2002) 
 Wood pigeons Kenward (1978) 
 Web-building spiders Uetz et al. (2002) 
 Northern bobwhites Williams et al. (2003) 
Vigilance Thomson's gazelle Fitzgibbon (1989) 
Habitat choice Redshank Cresswell (1994a) and Whitfield (2003) 
 Blue tit, Great tit Hinsley et al. (1995) 

 
White-tailed deer, mule deer
 
Lingle (2002) and Lingle and Pellis (2002)
 
Behaviors
 
Species
 
Reference
 
Activity  Tadpole Rana catesbeiana Anholt and Werner (1995) 
  Tadpole Rana sylvatica Anholt and Werner (1998) and Skelly (1994) 
  Common garden skink Lampropholis guichenoti Downes (2002) 
  Wolf spider Pardosa milvina Persons et al. (2001) 
  Wolf spider Pardosa milvina Persons et al. (2002) 
  Mosquito larvae Culex pipiens Sih (1986) 
  Amphipod Gammarus minus Wisenden et al. (1999) 
Activity, habitat use  Rainbow trout Oncorhynchus mykiss Biro et al. (2003a , b , 2004 )  
Activity, cryptic behavior  Skink Eulamprus tympanum Schwarzkopf and Shine (1992) 
Group benefits Butterflies Burger and Gochfeld (2001) 
 Redshanks Cresswell (1994b) and Whitfield (2003) 
 Ocean skaters Foster and Treherne (1981) 
 White-nosed coatis Hass and Valenzuela (2002) 
 Wood pigeons Kenward (1978) 
 Web-building spiders Uetz et al. (2002) 
 Northern bobwhites Williams et al. (2003) 
Vigilance Thomson's gazelle Fitzgibbon (1989) 
Habitat choice Redshank Cresswell (1994a) and Whitfield (2003) 
 Blue tit, Great tit Hinsley et al. (1995) 

 
White-tailed deer, mule deer
 
Lingle (2002) and Lingle and Pellis (2002)
 

Studies measuring the effect of antipredation behaviors (other than activity) on attack rate that can draw conclusions about consequent death rates are rare, mainly because observing predation rates is difficult. Here we report studies where the prey was free to compensate behaviorally and where the predator's choice was not constrained so that attack rates were biologically meaningful (also see Table 1 ). We have also excluded studies where there was a confounding effect of the focal behavior on capture rate during an attack. That antipredation benefits from reduced attack rates that accrue from flocking have been shown for ocean skaters Halobates robustus ( Foster and Treherne, 1981 ), butterflies Lepidoptera ( Burger and Gochfeld, 2001 ), web-building spiders Meteperia incrassata ( Uetz et al., 2002 ), northern bobwhites Colinus virginianus ( Williams et al., 2003 ), wood pigeons Columba palumbus ( Kenward, 1978 ), redshanks ( Cresswell, 1994b ), and white-nosed coatis Nasua narica ( Hass and Valenzuela, 2002 ). Choice of safe habitats when foraging is also important to reduce attack rates because high attack sites incur survival costs in great tits Parus major , blue tits Parus caeruleus ( Hinsley et al., 1995 ), and redshanks ( Cresswell, 1994a ; Whitfield, 2003 ). However, this effect was not found in brambling flocks Fringilla montifringilla foraging during migratory stopover ( Lindström, 1990 ). In Thompson's gazelles individual vigilance has also been shown to be a determinant of mortality with less vigilant individuals being more vulnerable to attack by cheetahs Acinonyx jubatus ( Fitzgibbon, 1989 ).

Studies that relate antipredation behaviors to reproductive output

Variation in antipredation behavior has rarely been related to reproductive output, except in studies that directly examine behaviors that result in defense of young. Many studies have been carried out, particularly with respect to nest defense in birds ( Byrkjedal, 1987 ; see review in Montgomerie and Weatherhead, 1988 ). Other studies have examined how antipredation behavior is directly related to increasing survival of young (e.g., Laundré et al., 2001 ). We do not consider further studies into parental antipredation behaviors directed towards nest predators (e.g., nest defense) or defense of young that have measured juvenile (or clutch) survival because the results of these studies are relatively easy to interpret in terms of fitness benefits but do not necessarily tell us anything about the fitness consequences of general antipredation behavior when an animal is not directly defending its young.

This leaves few studies because of the logistical difficulties of measuring antipredation behavior and reproductive output in the same individuals over a sufficiently long time and large enough scale to draw meaningful conclusions. Consequently, most studies of antipredation behavior use nonbreeding animals. In any case, reproductive rates are a consequence of several interacting behaviors such as parental quality, territory quality, life-history trade-offs, phenotypic limitations, and so on. Allocation of time to antipredation activities is just one of several alternative strategies that will affect reproductive output, and because individuals that allocate more time to antipredation activities either before or during reproduction are likely to make life-history trade-offs, it may be necessary to monitor lifetime reproductive output. This is the holy grail of antipredation studies and has up to now only been done successfully in one species of spider and in three species of stream-dwelling insects: being able to relate time or resources allocated to antipredation behavior to lifetime reproductive output.

Results from an experimental study on the wolf spider Pardosa milvina suggest that an allocation in predator avoidance behaviors can be coupled to fitness costs in terms of lighter egg sacs that also contain fewer eggs ( Persons et al., 2002 ). This effect was probably due to the fact that individuals spending more time avoiding predators consumed less prey, resulting in less energy to invest into the egg sacs. In addition, studies examining the costs of antipredation behaviors in insects that have nonfeeding adult life stages have shown, for example, that reduced feeding due to predator avoidance translates into negative fitness consequences ( Ball and Baker, 1996 ; Peckarsky et al., 1993 ; Scrimgeour and Culp, 1994 ; see also Peckarsky and McIntosh, 1998 ) often, but not always ( Ball and Baker, 1995 ), mediated through reduced fecundity. Even though we failed in finding any studies on vertebrates relating antipredation behaviors to reproductive output, we would like to point out that perceived predation risk, a probable major determinant of antipredation behaviors, has been shown to have a profound impact on reproductive rate. Animals can probably adjust their reproductive output to perceived predation risk, and gray-sided voles Clethrionomys rufocanus might even suppress their breeding when exposed to the odor of weasels Mustela nivalis ( Fuelling and Halle, 2004 ). We believe that no field study to date has provided any link between individual benefits of antipredation behavior and reproductive success.

Studies that have considered multiple antipredation behaviors

Variation in antipredation behavior has often been related to single correlates of death rate ( Lima, 1998 ): although this approach is common, this has rarely led to robust conclusions about fitness because few studies have considered the full potential for compensation in other behaviors that may change predation risk. Predation risk varies with a plethora of single abiotic (e.g., moonlight and temperature) and biotic (e.g., information about predation risk) factors. Single factors governing antipredation behaviors may be relatively straightforward, such as the avoidance of diurnal predators by seabirds returning to their colonies at night ( Watanuki, 1986 ) or that cryptic prey animals can reduce the risk of being detected by choosing microhabitats that resembles their coloration ( Edmunds, 1974 ). Further examples include studies where information about prevailing predation risk helps animals to recognize and avoid predators effectively (for comprehensive reviews see Chivers and Smith, 1998 ; Kats and Dill, 1998 ). This has been documented in, for example, branchiopods, especially Daphnia ( Lampert, 1993 ), amphipods Gammarus minus ( Wisenden et al., 1999 ), thrips Frankliniella occidentalis , wolf spiders P. milvina ( Persons et al., 2002 ), and in fish (e.g., Mirza and Chivers, 2001 ; Smith, 1992 ). However, the measure of the behavior will only be an index of fitness unless death rates, or decreases in reproductive output are measured, and knowledge of the system is complete enough to rule out the effects of other compensating behaviors.

Because antipredation behaviors may not be independent and instead interact frequently, the fitness consequences of variation in any one behavior can be difficult to predict ( Sih et al., 2003 ). To date, there have been few studies that have specifically measured more than one or two concurrent antipredation behaviors, so the possibility exists that most studies are confounded by predation risk compensation. We list studies that have considered more than one antipredation behavior in Table 3 . The cost of reduced locomotor performance, and thus impaired predator evasion, in gravid lizards does not have to be very important for individual vulnerability because of behavioral compensation ( Cooper et al., 1990 ; Pérez-Tris et al., 2004 ; Schwarzkopf and Shine, 1992 ). Williams et al. (2003) used field and aviary experiments to study optimal group size in Northern bobwhites and found that an intermediate group size was beneficial to individuals due to its effect on for example group persistence, movement, feeding efficiency, predator detection, and individual survival. Changes in antipredation behavior can also be affected by landscape changes and how they translate into differences in group composition ( Tellería et al., 2001 ). Considering multiple antipredation behaviors has also led to insights that one cannot simply classify habitat quality based on the rate of foraging gain because antipredation behaviors also affect habitat choice. For example, Western sandpipers Calidris mauri appear to choose stopover sites based on their internal state as well as the risk and profitability connected to a specific patch ( Ydenberg et al., 2002 ). Similarly, red-necked pademelons Thylogale thetis compensate for predation risk during foraging by altering vigilance, group size choice, and distance to cover ( Wahungu et al., 2001 ). Considering multiple behaviors also makes it possible to pinpoint which behaviors are important for individual survival. For example, Main (1987) found that otherwise rare behaviors became increasingly common under predation and contributed significantly to individual survival. Sih et al. (2003) have also showed elegantly how antipredation behaviors, previously thought to be maladaptive, are better understood when other behaviors were incorporated in their study. Lastly, Kotler et al. (2004) used indirect measurements of apprehension and time allocation to illustrate how predation risk varies with energetic state and habitat characteristics in Allenby's gerbil Gerbillus a. allenbyi .

Table 3

Studies that have considered more than one dimension of antipredation behavior

Species
 
Reference
 
Caridean shrimp Tozeuma carolinense Main (1987) 
Salamander larvae Ambystoma barbouri Sih et al. (2003) 
Lizard Eumeces laticeps Cooper et al. (1990) and Schwarzkopf and Shine (1992) 
Lizard Psammodromus algirus Pérez-Tris et al. (2004) 
Northern bobwhites Williams et al. (2003) 
Western sandpiper Ydenberg et al. (2002) 
Red-necked pademelon Wahungu et al. (2001) 
Blue tits Tellería et al. (2001) 
Allenby's gerbil
 
Kotler et al. (2004)
 
Species
 
Reference
 
Caridean shrimp Tozeuma carolinense Main (1987) 
Salamander larvae Ambystoma barbouri Sih et al. (2003) 
Lizard Eumeces laticeps Cooper et al. (1990) and Schwarzkopf and Shine (1992) 
Lizard Psammodromus algirus Pérez-Tris et al. (2004) 
Northern bobwhites Williams et al. (2003) 
Western sandpiper Ydenberg et al. (2002) 
Red-necked pademelon Wahungu et al. (2001) 
Blue tits Tellería et al. (2001) 
Allenby's gerbil
 
Kotler et al. (2004)
 

Future research to link antipredation behavior to fitness

We conclude, surprisingly, that although we know much about how predation risk affects individual decision making, we know relatively little about how individual behavioral modifications translate into fitness. Studies on vertebrates that have been successful in determining fitness consequences of antipredator behavior are often painstaking field studies and when the natural systems being observed studies are not constrained by ethical issues ( Huntingford, 1984 ). In contrast, behavioral ecology experiments in the study of predation, conducted under sound ethical guidelines, are constantly forced to base conclusions and analyses on more or less well-established proxies of fitness. So what is the way forward? We propose three ways in which studies of the functional significance of antipredation behavior should be carried out: (1) studies where multiple alternative compensatory behaviors, and the constraints operating on alternative behaviors are considered; (2) studies that relate resources allocated to antipredation behavior to indices of reproductive output; and (3) studies of how predators choose prey to make inferences about prey vulnerability.

Multibehavioral studies

If we are to draw conclusions with respect to fitness from measures of antipredation behavior, we need to know the following from any empirical study: (1) the range of possible compensatory behaviors, (2) the degree to which compensatory behaviors are independent, and (3) the ecological constraints or physiological and phenotypic limitations operating on the expression of each antipredation behavior.

First, we need to know the range of possible compensation behaviors. Clearly, in some systems there may be no alternatives: for example, animals may be restricted to feed in a single location and cannot vary group size. Here individual vigilance rate may then give an index of predation risk, but only if vigilance is a reliable estimate of predator detection (cf. Kaby and Lind, 2003 ; Lima and Bednekoff, 1999 ). More typically, there will be a range of antipredation options. For example, birds can compensate to increased predation risk by changing their mass by avoiding the higher risk area and by spending more time in antipredation behaviors ( Lima, 1998 ; Lima and Dill, 1990 ; Witter and Cuthill, 1993 ). Moreover, animals may compensate for increased predation risk by decreasing their mass because reducing mass might increase their ability to escape from predators ( Kullberg et al., 1996 ; Lind et al., 1999 ; McLachlan et al., 2003 ; Roitberg et al., 2003 ; Witter et al., 1994 ) or simply avoid the predator spatially ( Biro et al., 2004 ; Durant, 2000 ; Rettie and Messier, 2001 ) or temporally ( Biro et al., 2004 ; Creel and Winnie, 2005 ; Watanuki, 1986 ) or allocate more time to antipredation behaviors ( Elgar et al., 1986 ; Whitfield, 1988 ; Wolff and Van Horne, 2003 ). The exact number or type of important related or compensatory behaviors that comprise the antipredation behaviors of an animal will probably be species specific, although some taxonomic generalizations can be made. For example, mass regulation may be important only to flying, climbing, and cursorial animals, while avoidance will not be an option for any sessile or dormant animal.

Second, we need to determine the degree to which compensatory behaviors are independent. This is because if there is a range of compensatory mechanisms that are not independent of each other, then there may be multiple values of the antipredation behaviors that compensate adequately to any level of predation risk. However, if one particular way of compensation is the best strategy, then compensation mechanisms may be independent because compensation in one behavior may occur fully before it is expressed in a second behavior. For example, under increased predation risk animals may lose mass first until all possible mass is lost and then start to compensate by another behavior such as avoidance. This means, hypothetically, that a single measure such as mass could then provide a measure of predation risk.

Third, we need to take into account the possibility that observed behavioral changes may be unrelated to predation risk due to constraints. Animals may be constrained if factors other than predation constrain fitness. For example, both theoretical and empirical studies have shown that time constraints can cause habitat shifts (e.g., a shift from an aquatic to a terrestrial habitat in metamorphosing insects) and increased risk taking by prey, even though the expected mortality and growth rate stay the same ( Johansson and Rowe, 1999 ; Johansson et al., 2001 ; Rowe and Ludwig, 1991 ; Werner and Anholt, 1993 ). Animals may also be constrained when other sources of mortality outweigh that of predation risk, such as risk of starvation, and this will vary according to ecological conditions such as predictability of foraging opportunities ( Houston and McNamara, 1993 ; Ydenberg and Dill, 1986 ). For example, the capacity for mass loss may vary between species ( Rogers, 1987 ) and between seasons within a species ( Cresswell, 1998 ). In midwinter, mass loss capacity in blackbirds Turdus merula is greater than at the end of the winter as fat reserves are highest midwinter when starvation risk is high ( Cresswell, 1998 ). Similarly, redshanks may feed in low predation risk areas until increased energy demands and declining food supply during midwinter forces them to feed in high-risk areas ( Yasué et al., 2003 ). In short, it would be wrong to conclude that no variation in an antipredation behavior indicated that the behavior did not have fitness consequences because all individuals may be forced to allocate all their resources away from the antipredation behavior because of a constraint.

Constraints do not operate equally across a population. We therefore need to measure the degree to which the compensatory mechanisms are available to all individuals in the population. There may be age-related differences in how risky behaviors are; for example, young Thomson's gazelles take a higher risk when inspecting predators than do adults ( Fitzgibbon, 1994 ). Therefore, different individuals will have different solutions to the trade-off between promoting self-maintenance or reproduction versus minimizing predation risk. This can depend on differences in their initial state (e.g., Godfrey and Bryant, 2000 ) or competitive ability (e.g., Cresswell, 2003 ), such as juveniles and females having less access to food because of dominance ( Pravosudov and Grubb, 1997 ). For example, in ground squirrels, lower vigilance of juveniles is due, at least in part, to the greater nutritional needs of young animals with consequent increases in foraging, which is largely incompatible with vigilance ( Arenz and Leger, 2000 ; Bachman, 1993 ). Similarly, only sated killifish Fundulus diaphanus can afford to join shoals for antipredation reasons because of increased foraging competition there ( Hensor et al., 2003 ).

It seems likely that animals may use any combination of antipredation compensation depending on ecological conditions, especially when conditions increase costs of allocating resources away from antipredation behavior. It might be expected that in natural systems under a reasonably constant moderate predation risk, animals that show compensation to predation risk in one dimension should show less compensation to predation risk in another, that is, a negative correlation between behaviors. However, if predation risk increases, changes in any one behavior to compensate, such as by decreasing mass, may not affect the values of another behavior, such as avoidance of predators, because an animal may aim to keep its predation risk at a fixed level. After compensation has occurred to increased predation risk, any negative correlation between different behaviors involved in predation compensation may remain (although at a different value) or may be lost if all animals in the population have fully compensated with respect to one behavior. For example, if all individuals in a population lose as much mass as possible in response to increased predation risk, there can be little or no variation in mass with respect to further compensation to predation risk. This is demonstrated by the negative correlations between antipredation behavior and antipredation related morphology in snails, intraspecifically and interspecifically. Snails with a stronger shell, which provides efficient protection from crabs, show little predator avoidance behavior ( Cotton et al., 2004 ; Rundle and Brönmark, 2001 ).

To conclude, it should be possible to demonstrate that a single antipredation behavior has important fitness consequences if it is shown that there is only one main antipredation behavior or that the antipredation behavior operates independently of other antipredation behaviors. A study cannot conclude that an antipredation behavior does not have a fitness consequence in the absence of variation of the behavior with respect to predation risk unless it can be demonstrated that there are no constraints acting on its expression. In practical terms this probably means a uniformity of lack of response across individuals varying in age, sex, or competitive ability. This has important implications for the validity of experimental studies under artificial conditions. We would like to emphasize the importance of taking multiple antipredation behaviors into account with a final example. Even though reproduction is often inferred as being costly in terms of increased risk of predation due to body mass increase ( Lee et al., 1996 ; Magnhagen, 1991 ), gravid female skinks Eulamprus tympanum do not suffer from increased mortality from predators ( Schwarzkopf and Shine, 1992 ). Despite the fact that gravid females suffered from reduced escape speed, they compensated for this by behaving cryptically, implying that vulnerability in this species is determined primarily by avoiding detection rather than escaping swiftly (see also Pérez-Tris et al., 2004 ).

An alternative approach that might prove fruitful to estimate fitness consequences of foraging under predation risk is to measure giving-up densities in controlled food patches, that is, the amount of resources remaining in a patch when the foraging individual leaves ( Brown, 1988 ). This assumes that a forager leaves a patch when the benefits of staying are less than the combined energetic, predation, and missed opportunity costs of going elsewhere ( Brown, 1988 ). This methodology is successful because it combines several important aspects of antipredation behavior ( Brown, 1999 ), for example, time allocation and vigilance ( Kotler et al., 2004 ). One reason the giving-up density works well for certain questions in predation is that it is a multivariate statistic that distils an entire vector of behaviors and decisions into a single scalar output that can be measured under diverse conditions and with animals with different states (see Olsson et al., 2002 ). However, future studies using this approach need to design studies where the individual is the statistical unit, which will require validation of the underlying assumptions as a prerequisite for the possibility of drawing sound conclusions ( Price and Correll, 2001 ).

Finally, negative results in multibehavioral studies may tell us as much as positive results. If compensation occurs in a variety of behaviors it is unlikely that any one antipredation behavior will correlate with a fitness surrogate. Any study that investigated how, for example, spacing affected the probability of capture might find no relationship because more spaced individuals compensated by being more vigilant. Such a negative result would be less likely to be published and would certainly not be published in an influential journal. If however the study also measured vigilance, the lack of a relationship between spacing and vigilance and fitness would make reasonable sense: two negative results would greatly increase our understanding of the system and would indeed be much more publishable. One consequence of this suggestion is that there are many unpublished nonsignificant relationships that apparently show no effect of variation in an antipredation behavior. If such studies could be viewed together, they could tell us much about alternative antipredation strategies and predation risk compensation.

Correlating resources allocated to antipredation behavior to measures that predict reproductive output

Animals that express an antipredation behavior at a low level may simply be allocating more resources to another fitness-enhancing activity. For example, a bird feeding to gain resources to produce eggs may reduce vigilance to a very low level, so that its food intake rate and corresponding clutch size is high. Although on average those individuals that do this may have a greater death rate, this may be more than offset by the increased fecundity. This type of trade-off and the predictions that arise from it have been explored within the theoretical discussion of state-dependent foraging, where an individual's internal state will determine the optimal allocation of time or energy to antipredation behavior ( Houston et al., 1988 ; Mangel and Clark, 1988 ). Nevertheless, as we have argued above, there are almost no empirical studies that have measured these types of trade-offs in the context of their effects on actual reproductive rates.

Correlating antipredation behavior to actual reproductive measures will be outside the bounds of most studies because measuring vigilance behavior in the nonbreeding season and the number of offspring produced by the same individual in the breeding season is very difficult. One way forward, however, is to establish which behaviors correlate with both antipredation behaviors and reproductive output. A candidate behavior linking antipredation behavior to reproductive output is likely to be competitive foraging ability ( Sutherland, 1996 ). That intake rate might be the best indicator of reproductive fitness is not certain. Nevertheless, it seems likely that the intake rate should at least correlate well with reproductive fitness. It should be made clear though that by intake rate we mean intake rate in terms of absolute foraging rate and susceptibility to interference competition (e.g., Sutherland and Parker, 1985 ). In other words, what the feeding rate of an individual is in the absence of competitors and how this declines on addition of competitors. An individual that has a high absolute foraging rate and low decline in intake rate on addition of competitors will be a good competitor and is likely to be able to allocate more resources to antipredation behavior or reproduction. In the nonbreeding season, the amount of resources that can be allocated from self-maintenance to antipredation behaviors will depend on the intake rate, and similarly in the breeding season the amount of resources that can be allocated from self-maintenance to offspring production also will depend on the intake rate.

To put this idea simply, we can discount future reproductive trade-offs when the overall effects of an antipredation behavior on intake rate are measured. For example, a vigilance reduction that does not increase energy intake is unlikely to increase future reproductive output. In this case we might be able to reasonably conclude that vigilance did indicate predation risk (in the absence of other potential compensation behaviors as outlined above). We would also learn much about the fitness consequences of antipredation behaviors by measuring variation in antipredation behavior with respect to individual variation in intake rate. If those individuals with the greatest potential intake rate (or competitive ability) allocate more resources to antipredation behavior, then we would predict strong positive correlation between the degree of expression of antipredation behaviors that affect fitness and competitive ability. Practically this may mean that dominant sex and age classes may show the highest level of an antipredation behavior. In effect this is the same as looking at variation in the expression of an antipredation behavior as the cost of expression decreases (less cost for individuals of higher competitive ability). Therefore, those antipredation behaviors that are shown only, or to the highest degree, by individuals of highest competitive ability are likely to be those that have the greatest fitness consequences.

Predator preferences and hunting behavior—the difficult shortcut

Studying the hunting behavior of predators, especially vertebrate predators hunting vertebrates, is usually considered difficult, and this is often used as a justification for drawing indirect conclusions about fitness consequences from behavioral observations of prey animals. However, although counterintuitive, studying the predators instead of the prey could actually be a more efficient way to understand the fitness consequences of prey behavior than studying the behavior of the prey themselves, especially if the goal is to understand prey vulnerability. As mentioned above, prey vulnerability may well be the key variable in determining predation risk because predators that are foraging optimally should be selective and hunt to gain the maximum energy return for energy invested during foraging (i.e., the predator should select those individuals that are the easiest to catch within a prey species). Consequently, if we understand the characteristics of the prey a predator prefers to attack, or the proximate mechanism behind the predator's choice ( Peckarsky and Penton, 1989 ), we can then draw reasonable conclusions about what makes prey vulnerable to predation and hence the benefits accrued from antipredation behaviors. For example, determining whether sparrowhawks attack the least vigilant of two mounts that vary in the state of alert (they do not—see Cresswell et al., 2003 ) or the smaller of two flocks (they do—see Cresswell and Quinn, 2004 ) suggests that choice of flock is more important than relative vigilance rate within a flock in determining attack probability. In addition, contradictory results regarding group size benefits in shoaling fish will be better understood when predator hunting behavior is accounted for ( Turesson and Brönmark, 2004 ). Pioneering work has also been done on the hunting preferences of bats to determine the vulnerability of calling frogs (e.g., Tuttle and Ryan, 1982 ), and a recent playback study provides empirical evidence of increased predation risk associated with contact calling in crested tits Parus cristatus ( Krams, 2001 ). However, one should bear in mind that conclusions drawn from indirect measures based on either the predator or prey alone should be accepted only with caution ( Blomberg and Shine, 2000 ). There is of course an additional level of complexity that needs to be mentioned, which is added by the fact that prey species are generally hunted by more than one predator species. The emergent effects of multiple predators can result either in risk reduction or risk enhancement for the prey, but not enough is yet known to draw general conclusions about when either of these two effects are expected ( Sih et al., 1998 ).

It is interesting to note that because it is necessary to show that the predator is responding to the prey in order to establish that pursuit deterrence is actually occurring, studies on pursuit-deterrent systems have resulted in collection of clear data on the fitness consequences of antipredation behaviors. This contrasts with the study of most other antipredation behaviors where the focus of study is the prey only rather than the response of the predator to changes in prey behavior. This insight leads us to an important conclusion for the future: if antipredation behaviors can be shown to affect predator behavior, then they are likely to have important fitness consequences, even if variation in death rates cannot be measured. Finally, by studying the predators, the former “fixed-risk” paradigm in the study of predator-prey interactions will be abandoned, which will tell us much about how predators themselves respond to prey. To acknowledge that predators are themselves optimal foragers that move, wait, and search for prey can lead to unexpected results concerning how effective prey decision making is in reducing predation ( Lima, 2002 ).

CONCLUSIONS

We believe that there is little point in continuing the study of antipredation behavior by measuring the variation in an antipredation behavior or by correlating the occurrence of one antipredation behavior with another in the absence of attack and capture rate data. We think that it is possible, however, to draw firm conclusions regarding how behaviors or potential strategies contribute to individual fitness. It can be done by measuring how animals may compensate for potentially risky behaviors and bodily states, then measuring to what degree the compensations are available to the individuals in the population, and finally by determining whether the compensations are independent. This could also be done by measuring variation in antipredation behavior with respect to an index of fitness such as competitive ability. Finally, another alternative is to study which prey predators choose or how predator behavior is affected by antipredation behavior. These approaches can also be more fruitful because it has become well established that individuals differ consistently in how they cope with a variety of circumstances (reviewed in Koolhaas et al., 1999 ). Hence, we think that the knowledge of how antipredation behavior arises and how risk is managed within and between individuals would benefit from fully incorporating the potential for compensation into the study of antipredation behavior. When we consider multiple antipredation behaviors, we can learn more about how and why individuals differ in their risk management, whether there are consistent individual strategies, and how personality traits may affect how individuals manage their risks.

To apply our ideas simply, future empirical studies could concentrate on measuring antipredation behaviors in to the context of individual energy budgets. This approach ensures that the full range of antipredation behaviors is considered and the relative importance of each in terms of cost can be assessed. A vital component of energy budgets, and in perhaps many cases a reasonable index for energy budget as well, are time budgets. These are straightforward to measure in natural systems. Therefore, we believe time budgets may be the easiest way to measure the full range of antipredation behaviors and the potential for compensation and so discount future reproductive trade-offs. Time and/or energy budgets are probably essential to study the fitness consequences of antipredation behavior in any case. This is because the costs of an antipredation behavior can only be assessed by determining how much energy or time is being allocated away from feeding and/or reproductive investment. And the costs of an antipredation behavior must be known if we are to understand the constraints operating on its expression, its occurrence in the population, and indeed whether an alternative antipredation behavior is more likely. So instead of measuring the costs of not showing an antipredation behavior (i.e., death rates), we can more easily measure the costs of showing the behavior (i.e., time budgets). Where costs of an antipredation behavior are high, and not compensated for by other behaviors, then it would seem reasonable to conclude that the antipredation behavior affects predation risk and fitness.

Although the best insights into the constraints and adaptations to predation risk are gained from elegant, often laboratory-based, experimental work, perhaps the only way to understand the fitness consequences of antipredation behavior is through time budgets in natural systems. The last generation of natural historians had the right approach to understand predation, and we need to get back into painstaking study of individuals in natural systems. This does not mean that we should not experiment; rather it means that we should conduct experiments within natural systems that are well understood from observation, where the animals have the opportunity to use their full repertoire of behaviors to manage predation risk. Such experiments will inevitably show a behavior accounting for a lower proportion of variance, because all other things are not held equal, but will allow that behavior's proportional contribution to overall fitness to be assessed properly. Also, such experiments will likely show no consequence of variation in a behavior much more often than in laboratory experiments, but negative results in the context of a natural system tell us much about the ecological or other constraints acting on the system, and therefore how natural selection and evolution operates. The evolution of antipredation behavior can only really be understood by considering the whole system at the same time: interactions between behaviorally complex predators hunting behaviorally complex prey in a heterogeneous environment cannot be meaningfully studied out of context of the whole system.

We thank Joel Brown for constructive criticism and Anne Magurran for giving comments on an early draft of the manuscript. We would also like to thank two anonymous reviewers who helped us improve the manuscript. W.C. is a Royal Society University Research Fellow and J.L. was supported by the Swedish Research Council.

References

Anholt BR, Werner EE,
1995
. Interaction between food availability and predation mortality mediated by adaptive behavior.
Ecology
 
76
:
2230
–2234.
Anholt BR, Werner EE,
1998
. Predictable changes in predation mortality as a consequence of changes in food availability and predation risk.
Evol Ecol
 
12
:
729
–738.
Arenz CL, Leger DW,
2000
. Antipredator vigilance of juvenile and adult thirteen-lined ground squirrels and the role of nutritional need.
Anim Behav
 
59
:
535
–541.
Bachman GC,
1993
. The effect of body condition on the trade-off between vigilance and foraging in Belding ground-squirrels.
Anim Behav
 
46
:
233
–244.
Ball SL, Baker RL,
1995
. The non-lethal effects of predators and the influence of food availability on life history of adult Chironomus tentans (Diptera: Chironomidae).
Freshw Biol
 
34
:
1
–12.
Ball SL, Baker RL,
1996
. Predator-induced life history changes: antipredator behavior costs of facultative life history shifts?
Ecology
 
77
:
1116
–1124.
Bednekoff PA, Lima SL,
1998
. Randomness, chaos and confusion in the study of antipredator vigilance.
Trends Ecol Evol
 
13
:
284
–287.
Billerbeck JM, Lankford TE Jr, Conover DO,
2001
. Evolution of intrinsic growth and energy acquisition rates. I. Trade-offs with swimming performance in Menidia menidia .
Evolution
 
55
:
1863
–1872.
Biro PA, Abrahams MV, Post JR, Parkinson EA,
2004
. Predators select against high growth rates and risk-taking behaviour in domestic trout populations.
Proc R Soc Lond B
 
271
:
2233
–2237.
Biro PA, Post JR, Parkinson EA,
2003
a. Density-dependent mortality is mediated by foraging activity for prey in whole-lake experiments.
J Anim Ecol
 
72
:
546
–555.
Biro PA, Post JR, Parkinson EA,
2003
b. From individuals to populations: prey fish risk-taking mediates mortality in whole system experiments.
Ecology
 
84
:
2419
–2431.
Biro PA, Post JR, Parkinson EA,
2003
c. Population consequences of a predator-induced habitat shift by trout in whole-lake experiments.
Ecology
 
84
:
691
–700.
Blomberg SP, Shine R,
2000
. Size-based predation by kookaburras ( Dacelo novaeguineae ) on lizards ( Eulamprus tympanum : Scincidae): what determines prey vulnerability?
Behav Ecol Sociobiol
 
48
:
484
–489.
Brown JS,
1988
. Patch use as an indicator of habitat preference, predation risk, and competition.
Behav Ecol Sociobiol
 
22
:
37
–47.
Brown JS,
1999
. Vigilance, patch use and habitat selection: foraging under predation risk.
Evol Ecol Res
 
1
:
49
–71.
Brown JS, Kotler BP,
2004
. Hazardous duty pay and the foraging cost of predation.
Ecol Lett
 
7
:
999
–1014.
Burger J, Gochfeld M,
2001
. Smooth-billed ani ( Crotophaga ani ) predation on butterflies in Mato Grosso, Brazil: risk decreases with increased group size.
Behav Ecol Sociobiol
 
49
:
482
–492.
Byrkjedal I,
1987
. Antipredator behavior and breeding success in greater golden-plover and Eurasian dotterel.
Condor
 
89
:
40
–47.
Chivers DP, Smith RJF,
1998
. Chemical alarm signalling in aquatic predator-prey systems: a review and prospectus.
Ecoscience
 
5
:
338
–352.
Clobert J, Oppliger A, Sorci G, Ernande B, Swallow JG, Garland T Jr,
2000
. Trade-offs in phenotypic traits: endurance at birth, growth, survival, predation and susceptibility to parasitism in a lizard, Lacerta vivipara .
Funct Ecol
 
14
:
675
–684.
Cooper WE Jr, Vitt LJ, Hedges R, Huey RB,
1990
. Locomotor impairment and defense in gravid lizards ( Eumeces laticeps ): behavioral shift in activity may offset costs of reproduction in an active forager.
Behav Ecol Sociobiol
 
27
:
153
–157.
Cotton PA, Rundle SH, Smith KE,
2004
. Trait compensation in marine gastropods: shell shape, avoidance behaviour, and susceptibility to predation.
Ecology
 
85
:
1581
–1584.
Creel S, Winnie JJA,
2005
. Responses of elk herd size to fine-scale spatial and temporal variation in the risk of predation by wolves.
Anim Behav
 
69
:
1181
–1189.
Cresswell W,
1993
. Escape responses by redshanks, Tringa totanus , on attack by avian predators.
Anim Behav
 
46
:
609
–611.
Cresswell W,
1994
a. Age-dependent choice of redshank ( Tringa totanus ) feeding location: profitability or risk?
J Anim Ecol
 
63
:
589
–600.
Cresswell W,
1994
b. Flocking is an effective anti-predation strategy in redshanks, Tringa totanus .
Anim Behav
 
47
:
433
–442.
Cresswell W,
1994
c. Song as a pursuit-deterrent signal, and its occurrence relative to other anti-predation behaviours of skylark ( Alauda arvensis ) on attack by merlins ( Falco columbarius ).
Behav Ecol Sociobiol
 
34
:
217
–223.
Cresswell W,
1998
. Diurnal and seasonal mass variation in blackbirds Turdus merula : consequences for mass-dependent predation risk.
J Anim Ecol
 
67
:
78
–90.
Cresswell W,
1999
. Travel distance and mass gain in wintering blackbirds.
Anim Behav
 
58
:
1109
–1116.
Cresswell W,
2003
. Testing the mass-dependent predation hypothesis: in European blackbirds poor foragers have higher overwinter body reserves.
Anim Behav
 
65
:
1035
–1044.
Cresswell W, Lind J, Kaby U, Quinn JL, Jakobsson S,
2003
. Do opportunistic predators preferentially attack non-vigilant prey?
Anim Behav
 
66
:
643
–648.
Cresswell W, Quinn JL,
2004
. Faced with a choice, sparrowhawks more often attack the more vulnerable prey group.
Oikos
 
104
:
71
–76.
Cresswell W, Whitfield DP,
1994
. The effects of raptor predation on wintering wader populations at the Tyninghame estuary, southeast Scotland.
Ibis
 
136
:
223
–232.
Downes SJ,
2002
. Does responsiveness to predator scent affect lizard survivorship?
Behav Ecol Sociobiol
 
52
:
38
–42.
Durant SM,
2000
. Living with the enemy: avoidance of hyenas and lions by cheetahs in the Serengeti.
Behav Ecol
 
11
:
624
–632.
Edmunds M,
1974
. Defence in animals: a survey of anti-predator defences. Harlow: Longman.
Elgar MA,
1989
. Predator vigilance and group size in mammals and birds: a critical review of the empirical evidence.
Biol Rev
 
64
:
13
–33.
Elgar MA, McKay H, Woon P,
1986
. Scanning, pecking and alarm flights in house sparrows.
Anim Behav
 
34
:
1892
–1894.
Endler JA,
1991
. Interactions between predators and prey. In: Behavioural ecology: an evolutionary approach (Krebs JR, Davies NB, eds). Oxford: Blackwell; 169–196.
Fitzgibbon CD,
1989
. A cost to individuals with reduced vigilance in groups of Thomson's gazelles hunted by cheetahs.
Anim Behav
 
37
:
508
–510.
Fitzgibbon CD,
1994
. The costs and benefits of predator inspection behavior in Thomson gazelles.
Behav Ecol Sociobiol
 
34
:
139
–148.
Fitzgibbon CD, Fanshawe JH,
1988
. Stotting in Thomson's gazelles: an honest signal of condition.
Behav Ecol Sociobiol
 
23
:
69
–74.
Foster WA, Treherne JE,
1981
. Evidence for the dilution effect in the selfish herd from a fish predation on a marine insect.
Nature
 
293
:
466
–467.
Fuelling O, Halle S,
2004
. Breeding suppression in free-ranging grey sided voles under the influence of predator odour.
Oecologia
 
138
:
151
–159.
Godfrey JD, Bryant DM,
2000
. State-dependent behaviour and energy expenditure: an experimental study of European robins on winter territories.
J Anim Ecol
 
69
:
301
–313.
Hass CC, Valenzuela D,
2002
. Anti-predator benefits of group living in white-nosed coatis ( Nasua narica ).
Behav Ecol Sociobiol
 
51
:
570
–578.
Hasson O,
1991
. Pursuit-deterrent signals: communication between prey and predator.
Trends Ecol Evol
 
6
:
325
–329.
Hensor EMA, Godin JJ, Hoare DJ, Krause J,
2003
. Effects of nutritional state on the shoaling tendency of banded killifish, Fundulus diaphanus , in the field.
Anim Behav
 
65
:
663
–669.
Hinsley SA, Bellamy PE, Moss D,
1995
. Sparrowhawk Accipiter nisus predation and feeding site selection by tits.
Ibis
 
137
:
418
–420.
Houston AI, Clark CW, McNamara JM, Mangel M,
1988
. Dynamic models in behavioural and evolutionary ecology.
Nature
 
332
:
29
–34.
Houston AI, McNamara JM,
1993
. A theoretical investigation of the fat reserves and mortality levels of small birds in winter.
Ornis Scand
 
24
:
205
–219.
Huntingford FA,
1984
. Some ethical issues raised by studies of predation and aggression.
Anim Behav
 
32
:
210
–215.
Irving PW, Magurran AE,
1997
. Context-dependent fright reactions in captive European minnows: the importance of naturalness in laboratory experiment.
Anim Behav
 
53
:
1193
–1201.
Johansson F, Rowe L,
1999
. Life history and behavioral responses to time constraints in a damselfly.
Ecology
 
80
:
1242
–1252.
Johansson F, Stoks R, Rowe L, De Block M,
2001
. Life history plasticity in a damselfly: effects of combined time and biotic constraints.
Ecology
 
82
:
1857
–1869.
Kaby U, Lind J,
2003
. What limits predator detection in blue tits ( Parus caeruleus ): posture, task or orientation?
Behav Ecol Sociobiol
 
54
:
534
–538.
Kats LB, Dill LM,
1998
. The scent of death: chemosensory assessment of predation risk by prey animals.
Ecoscience
 
5
:
361
–394.
Kenward RE,
1978
. Hawks and doves: factors affecting success and selection in goshawk attacks on woodpigeons.
J Anim Ecol
 
47
:
449
–460.
Koolhaas JM, Korte SM, De Boer SF, Van Der Vegt BJ, Van Reenen CG, Hopster H, De Jong IC, Ruis MAW, Blokhuis HJ,
1999
. Coping styles in animals: current status in behavior and stress-physiology.
Neurosci Biobehav Rev
 
23
:
925
–935.
Kotler BP, Brown JS, Bouskila A,
2004
. Apprehension and time allocation in gerbils: the effects of predatory risk and energetic state.
Ecology
 
85
:
917
–922.
Krams I,
2001
. Communication in crested tits and the risk of predation.
Anim Behav
 
61
:
1065
–1068.
Kullberg C, Fransson T, Jakobsson S,
1996
. Impaired predator evasion in fat blackcaps ( Sylvia atricapilla ).
Proc R Soc Lond B
 
263
:
1671
–1675.
Lampert W,
1993
. Ultimate causes of diel vertical migration of zooplankton: new evidence for the predator-avoidance hypothesis.
Arch Hydrobiol Beih Ergebn Limnol
 
39
:
79
–88.
Lank DB, Ydenberg RC,
2003
. Death and danger at migratory stopovers: problems with “predation risk”.
J Avian Biol
 
34
:
225
–228.
Lankford TE Jr, Billerbeck JM, Conover DO,
2001
. Evolution of intrinsic growth and energy acquisition rates. II Trade-offs with vulnerability to predation in Menidia menidia .
Evolution
 
55
:
1873
–1881.
Laundré JW, Hernández L, Altendorf KB,
2001
. Wolves, elk, and bison: reestablishing the “landscape of fear” in Yellowstone National Park, USA.
Can J Zool
 
79
:
1401
–1409.
Le Galliard J-F, Clobert J, Ferrière R,
2004
. Physical performance and Darwinian fitness in lizards.
Nature
 
432
:
502
–505.
Lee SJ, Witter MS, Cuthill IC, Goldsmith AR,
1996
. Reduction in escape performance as a cost of reproduction in gravid starlings, Sturnus vulgaris .
Proc R Soc Lond B
 
263
:
619
–624.
Lima SL,
1987
. Vigilance while feeding and its relation to the risk of predation.
J Theor Biol
 
124
:
303
–316.
Lima SL,
1998
. Stress and decision making under the risk of predation: recent developments from behavioral, reproductive, and ecological perspectives.
Adv Study Behav
 
27
:
215
–290.
Lima SL,
2002
. Putting predators back into behavioral predator-prey interactions.
Trends Ecol Evol
 
17
:
70
–75.
Lima SL, Bednekoff PA,
1999
. Back to the basics of antipredatory vigilance: can nonvigilant animals detect attack?
Anim Behav
 
58
:
537
–543.
Lima SL, Dill LM,
1990
. Behavioral decisions made under the risk of predation: a review and prospectus.
Can J Zool
 
68
:
619
–640.
Lind J,
2004
. What determines probability of surviving predator attacks in bird migration? The relative importance of vigilance and fuel load.
J Theor Biol
 
231
:
223
–227.
Lind J, Fransson T, Jakobsson S, Kullberg C,
1999
. Reduced take-off ability in robins ( Erithacus rubecula ) due to migratory fuel load.
Behav Ecol Sociobiol
 
46
:
65
–70.
Lind J, Jakobsson S, Kullberg C, in press. Impaired predator evasion in the life-history of birds; behavioral and physiological adaptations to reduced flight ability. Curr Ornithol.
Lindström Å,
1989
. Finch flock size and risk of hawk predation at a migratory stopover site.
Auk
 
106
:
225
–232.
Lindström Å,
1990
. The role of predation risk in stopover habitat selection in migrating bramblings, Fringilla montifringilla .
Behav Ecol
 
1
:
102
–105.
Lingle S,
2002
. Coyote predation and habitat segregation of white-tailed deer and mule deer.
Ecology
 
83
:
2037
–2048.
Lingle S, Pellis SM,
2002
. Fight or flight? Antipredator behavior and the escalation of coyote encounters with deer.
Oecologia
 
131
:
154
–164.
Magnhagen C,
1991
. Predation risk as a cost of reproduction.
Trends Ecol Evol
 
6
:
183
–186.
Main KL,
1987
. Predator avoidance in seagrass meadows: prey behavior, microhabitat selection and cryptic coloration.
Ecology
 
68
:
170
–180.
Mangel M, Clark CW,
1988
. Dynamic modelling in behavioural ecology. Princeton: Princeton University Press.
McLachlan A, Ladle R, Crompton B,
2003
. Predator-prey interactions on the wing: aerobatics and body size among dance flies and midges.
Anim Behav
 
66
:
911
–915.
McNamara JM, Houston AI,
1986
. The common currency for behavioural decisions.
Am Nat
 
127
:
358
–378.
McNamara JM, Houston AI,
1994
. The effect of a change in foraging options on intake rate and predation rate.
Am Nat
 
144
:
978
–1000.
Mirza RS, Chivers DP,
2000
. Predator-recognition training enhances survival of brook trout: evidence from laboratory and field-enclosure studies.
Can J Zool
 
78
:
2198
–2208.
Mirza RS, Chivers DP,
2001
. Chemical alarm signals enhance survival of brook charr ( Salvelinus fontinalis ) during encounters with predatory chain pickerel ( Esox niger ).
Ethology
 
107
:
989
–1005.
Montgomerie RD, Weatherhead PJ,
1988
. Risks and rewards of nest defence by parent birds.
Q Rev Biol
 
63
:
167
–187.
Newton I,
1998
. Population limitations in birds. New York: Academic Press.
Olsson O, Brown JS, Smith HG,
2002
. Long- and short-term state-dependent foraging under predation risk: an indication of habitat quality.
Anim Behav
 
63
:
981
–989.
Peckarsky BL, Cowan CA, Penton MA, Anderson C,
1993
. Sublethal consequences of stream-dwelling predatory stoneflies on mayfly growth and fecundity.
Ecology
 
74
:
1836
–1846.
Peckarsky BL, McIntosh AR,
1998
. Fitness and community consequences of avoiding multiple predators.
Oecologia
 
113
:
565
–576.
Peckarsky BL, Penton MA,
1989
. Mechanisms of prey selection by stream-dwelling stoneflies.
Ecology
 
70
:
1203
–1218.
Pérez-Tris J, Díaz JA, Tellería JL,
2004
. The loss of body mass under risk of predation: cost of anti-predatory behaviour or adaptive fit-for-escape?
Anim Behav
 
67
:
511
–521.
Persons MH, Walker SE, Rypstra AL,
2002
. Fitness costs and benefits of antipredator behavior mediated by chemotactile cues in the wolf spider Pardosa milvina (Araneae: Lycosidae).
Behav Ecol
 
13
:
386
–392.
Persons MH, Walker SE, Rypstra AL, Marshall SD,
2001
. Wolf spider predator avoidance tactics and survival in the presence of diet-associated predator cues (Araneae: Lycosidae).
Anim Behav
 
61
:
43
–51.
Pravosudov VV, Grubb TC,
1997
. Energy management in passerine birds during the non-breeding season: a review.
Curr Ornithol
 
14
:
189
–234.
Price MV, Correll RA,
2001
. Depletion of seed patches by Merriam's kangaroo rats: are GUD assumptions met?
Ecol Lett
 
4
:
334
–343.
Quinn JL, Cresswell W,
2004
. Predator hunting behaviour and prey vulnerability.
J Anim Ecol
 
73
:
143
–154.
Rettie WJ, Messier F,
2001
. Range use and movement rates of woodland caribou in Saskatchewan.
Can J Zool
 
79
:
1933
–1940.
Rogers CM,
1987
. Predation risk and fasting capacity: do wintering birds maintain optimal body mass?
Ecology
 
68
:
1051
–1061.
Roitberg BD, Mondor EB, Tyerman JGA,
2003
. Pouncing spider, flying mosquito: blood acquisition increases predation risk in mosquitoes.
Behav Ecol
 
14
:
736
–740.
Rowe L, Ludwig D,
1991
. Size and timing of metamorphosis in complex life cycles: time constraints and variation.
Ecology
 
72
:
413
–427.
Rundle HD, Brönmark C,
2001
. Inter- and intraspecific trait compensation of defence mechanisms in freshwater snails.
Proc R Soc Lond B
 
268
:
1463
–1468.
Schwarzkopf L, Shine R,
1992
. Costs of reproduction in lizards: escape tactics and susceptibility to predation.
Behav Ecol Sociobiol
 
31
:
17
–25.
Scrimgeour GJ, Culp JM,
1994
. Feeding while evading predators by a lotic mayfly: linking short-term foraging behaviours to long-term fitness consequences.
Oecologia
 
100
:
128
–134.
Sih A,
1986
. Antipredator responses and the perception of danger by mosquito larvae.
Ecology
 
67
:
434
–441.
Sih A, Englund G, Wooster DE,
1998
. Emergent impacts of multiple predators on prey.
Trends Ecol Evol
 
13
:
350
–355.
Sih A, Kats LB, Maurer EF,
2003
. Behavioural correlations across situations and the evolution of antipredator behaviour in a sunfish-salamander system.
Anim Behav
 
65
:
29
–44.
Sinclair ARE, Arcese P,
1995
. Population consequences of predation-sensitive foraging: the Serengeti wildebeest.
Ecology
 
76
:
882
–891.
Skelly DK,
1994
. Activity level and the susceptibility of anuran larvae to predation.
Anim Behav
 
47
:
465
–468.
Smith RJF,
1992
. Alarm signals in fishes.
Rev Fish Biol Fish
 
2
:
33
–63.
Stephens DW, Krebs JR,
1986
. Foraging theory. Princeton: Princeton University Press.
Sutherland WJ,
1996
. From individual behaviour to population ecology. Oxford: Oxford University Press.
Sutherland WJ, Parker GA,
1985
. Distribution of unequal competitors. In: Behavioural ecology (Sibly RM, Smith RH, eds). Oxford: Blackwell Scientific Publications; 255–273.
Tellería JL, Virgós E, Carbonell R, Pérez-Tris J, Santos T,
2001
. Behavioural responses to changing landscapes: flock structure and anti-predator strategies of tits wintering in fragmented forests.
Oikos
 
95
:
253
–264.
Temple SA,
1987
. Do predators always capture substandard individuals disproportionately from prey populations?
Ecology
 
68
:
669
–674.
Turesson H, Brönmark C,
2004
. Foraging behaviour and capture success in perch, pikeperch and pike and the effects of prey density.
J Fish Biol
 
65
:
363
–375.
Tuttle MD, Ryan MJ,
1982
. The role of synchronized calling, ambient light, and ambient noise, in anti-bat-predator behavior of a treefrog.
Behav Ecol Sociobiol
 
11
:
125
–131.
Uetz GW, Boyle J, Hieber CS, Wilcox RS,
2002
. Antipredator benefits of group living in colonial web-building spiders: the ‘early warning’ effect.
Anim Behav
 
63
:
445
–452.
Vine I,
1971
. Risk of visual detection and pursuit by a predator and the selective advantage of flocking behaviour.
J Theor Biol
 
30
:
405
–422.
Wahungu GM, Catterall CP, Olsen MF,
2001
. Predator avoidance, feeding and habitat use in the red-necked pademelon, Thylogale thetis , at rainforest edges.
Aust J Zool
 
49
:
45
–48.
Warkentin KM,
1995
. Adaptive plasticity in hatching age: a response to predation risk trade-off.
Proc Natl Acad Sci USA
 
92
:
3507
–3510.
Watanuki Y,
1986
. Moonlight avoidance behavior in Leach's storm-petrels as a defense against slaty-backed gulls.
Auk
 
103
:
14
–22.
Werner EE, Anholt BR,
1993
. Ecological consequences of the trade-off between growth and mortality rates mediated by foraging activity.
Am Nat
 
142
:
242
–272.
Whitfield DP,
1988
. Sparrowhawks Accipiter nisus affect the spacing behaviour of wintering turnstone Arenaria interpres and redshank Tringa totanus .
Ibis
 
130
:
284
–287.
Whitfield DP,
2003
. Redshank Tringa totanus flocking behaviour, distance from cover and vulnerability to sparrowhawk Accipiter nisus predation.
J Avian Biol
 
34
:
163
–169.
Williams CK, Lutz RS, Applegate RD,
2003
. Optimal group size and northern bobwhite coveys.
Anim Behav
 
66
:
377
–387.
Wisenden BD, Cline A, Sparkes TC,
1999
. Survival benefit to antipredator behavior in the amphipod Gammarus minus (Crustacea: Amphipoda) in response to injury-released chemical cues from conspecifics and heterospecifics.
Ethology
 
105
:
407
–414.
Witter MS, Cuthill IC,
1993
. The ecological costs of avian fat storage.
Phil Trans R Soc Lond B
 
340
:
73
–92.
Witter MS, Cuthill IC, Bonser RHC,
1994
. Experimental investigations of mass-dependent predation risk in the European starling, Sturnus vulgaris .
Anim Behav
 
48
:
201
–222.
Wolff JO, Davis-Born R,
1997
. Response of gray-tailed voles to odours of a mustelid predator: a field test.
Oikos
 
79
:
543
–548.
Wolff JO, Van Horne T,
2003
. Vigilance and foraging patterns of American elk during the rut in habitats with and without predators.
Can J Zool
 
81
:
266
–271.
Yasué M, Quinn JL, Cresswell W,
2003
. Multiple effects of weather on the starvation and predation risk trade-off in choice of feeding location in redshanks.
Funct Ecol
 
17
:
727
–736.
Ydenberg RC, Butler RW, Lank DB, Guglielmo CG, Lemon M, Wolf N,
2002
. Trade-offs, condition dependence and stopover site selection by migrating sandpipers.
J Avian Biol
 
33
:
47
–55.
Ydenberg RC, Dill LM,
1986
. The economics of fleeing from predators.
Adv Study Behav
 
16
:
229
–249.