The magnitude of contingency
To quantify strength of contingency in NCEs, we conducted a meta-analysis of the relationship between four types of prey risk taking and a key hypothesized driver of context dependence: prey energetic state (Kotler et al . 2004; Schmitz & Trussell 2016; seeBox 3 for methods). Reduced energetic state resulted in significantly elevated prey risk taking in each behavioral dimension (Fig. 4 ). When in a compromised energetic state, prey subjected to predation risk tended to be more active (d = 0.73, ± 95% CI = 0.32-1.14; lnRR = 0.34, ± 95% CI = 0.17-0.51), spend less time in refugia (d = 0.95, ± 95% CI = 0.74-1.16; lnRR = 0.58, ± 95% CI = 0.46-0.71), invest more in foraging versus apprehension (d = 0.84, ± 95% CI = 0.50-1.19; lnRR = 0.49, ± 95% CI = 0.22-0.77), and be more isolated from other group members (d = 0.86, ± 95% CI = 0.29-1.42; lnRR = 0.42, ± 95% CI = 0.19-0.65) than conspecifics in better nutritional condition. The mean estimates of d suggest a medium/strong effect of reduced energetic state on prey activity, and strong effects on exposure, foraging investment, and isolation. We detected significant heterogeneity among studies exploring each risk-taking dimension: activity (d : Q 33 = 178.39, p < 0.001; lnRR: Q 33 = 88.04, p < 0.001); exposure (d : Q 70 = 213.60, p < 0.001; lnRR: Q 70 = 369.66, p < 0.001); foraging investment (d :Q 40 = 186.19, p < 0.001; lnRR:Q 40 = 924.42, p < 0.001); isolation ((d : Q 9 = 34.85, p < 0.001; lnRR: Q 9 = 57.06, p < 0.001).
These results underscore the marked differences that can characterize risk taking by prey individuals with divergent energetic states, and by extension the pivotal role that prey condition plays as a mediator of direct NCEs in communities. Most of the studies included in this analysis examined small-bodied taxa in laboratory or small-scale mesocosm settings (Appendix S1 in Supporting Information), and we found few investigations of the indirect community consequences of state-mediated variation in prey risk taking. Accordingly, there remains need for studies quantifying the influence of energetic state on risk taking under field conditions, especially with larger-bodied species, and the degree to which disparate responses of prey in divergent nutritional states shape indirect predator NCEs. Furthermore, prey state was the only potential driver of contingency in prey anti-predator behavior subject to enough empirical investigation to support meta-analysis. Thus, we also encourage further work addressing other potential drivers of context dependent NCEs.
Our findings also revealed considerable variability in the magnitude and direction of the effect of prey condition on anti-predator behavior. Some studies in each of the four behavioral categories showed effects opposite the majority of studies. In these studies, anti-predator behavior increased with decreasing body condition or state of the prey (7 of 34 studies of activity; 7 of 71 for exposure; 5 of 41 for foraging investment; 1 of 10 for isolation). This variability may, in part, stem from the myriad means by, and circumstances under, which the experiments of the meta-analyses were conducted. The variability in results is also consistent with the idea that, during phase two, anti-predator behavior and the way it is shaped by context (including prey state) hinge on the relative safety benefits conferred by different behavioral options available to the prey species. Either increased or decreased activity, for example, can serve as effective anti-predator tactics depending on the hunting mode of the predator against which they are deployed (Schmitz 2007). Furthermore, if individuals in a low state are more vulnerable to predation, then they may adopt a higher level of antipredator investment (Raveh et al . 2011; Makon et al .in press ). It is not surprising, then, that some studies we assessed found that sated prey exposed to predation risk became more active, and that food restriction diminished rather than exacerbated this response (e.g., Hossie & Murray 2011). Though beyond the scope of this review, meta-regressions with covariates potentially influencing the efficacy of different anti-predator behaviors (e.g., predator type) would likely reduce some of the uncertainty characterizing our findings.