Vigilance and fitness in grey partridges Perdix perdix: the effects of group size and foraging-vigilance trade-offs on predation mortality

Authors


School of Biology, Bute Building, University of St Andrews, KY16 9TS, UK. E-mail: wrlc@st-and.ac.uk

Summary

  • 1Vigilance increases fitness by improving predator detection but at the expense of increasing starvation risk. We related variation in vigilance among 122 radio-tagged overwintering grey partridges Perdix perdix (L.) across 20 independent farmland sites in England to predation risk (sparrowhawk Accipiter nisus L., kill rate), use of alternative antipredation behaviours (grouping and use of cover) and survival.
  • 2Vigilance was significantly higher when individuals fed in smaller groups and in taller vegetation. In the covey period (in early winter when partridges are in flocks), vigilance and use of taller vegetation was significantly higher at sites with higher sparrowhawk predation risk, but tall vegetation was used less by larger groups. Individuals were constrained in reducing individual vigilance by group size and habitat choice because maximum group size was determined by overall density in the area during the covey period and by the formation of pairs at the end of the winter (pair period), when there was also a significant twofold increase in the use of tall cover.
  • 3Over the whole winter individual survival was higher in larger groups and was lower in the pair period. However, when controlling for group size, mean survival decreased as vigilance increased in the covey period. This result, along with vigilance being higher at sites with increasing with raptor risk, suggests individual vigilance increases arose to reduce short-term predation risk from raptors but led to long-term fitness decreases probably because high individual vigilance increased starvation risk or indicated longer exposure to predation. The effect of raptors on survival was less when there were large groups in open habitats, where individual partridges can probably both detect predators and feed efficiently.
  • 4Our study suggests that increasing partridge density and modifying habitat to remove the need for high individual vigilance may decrease partridge mortality. It demonstrates the general principle that antipredation behaviours may reduce fitness long-term via their effects on the starvation–predation risk trade-off, even though they decrease predation risk short-term, and that it may be ecological constraints, such as poor habitat (that lead to an antipredation behaviour compromising foraging), that cause mortality, rather than the proximate effect of an antipredation behaviour such as vigilance.

Introduction

Predation is one of the major selection pressures that determine the form (Endler 1991) and behaviour of animals (Lima & Dill 1990; Lima 1998) because any animal whose form or behaviour facilitates avoidance or escape from predators will have a greater probability of surviving to breed and therefore producing offspring. Although in theory the fitness consequences of any antipredation behaviour can simply be measured by the resultant probability of survival or death, even if death rates can be correlated with a particular behaviour there may frequently be many other ways for animals to compensate so that the particular behaviours can actually be of little overall consequence for individual fitness (Lind & Cresswell 2005). Empirical studies that seek to relate antipredation behaviours to fitness must therefore consider a range of behaviours to determine the potential compensation to predation risk, and quantify the trade-off between countering risks of predation and starvation (Lind & Cresswell 2005). In this study we investigate how vigilance, a widespread antipredation behaviour where an animal scans for warning of approaching predators, interacts with other antipredation behaviours, leading to consequences for fitness.

Vigilance is an almost universal antipredation behaviour (Elgar 1989; Treves 2000) and is likely to determine probability of detection of an approaching predator, with the probability of detection increasing particularly with increasing group size and distance from predator-concealing cover (Pulliam 1973; Cresswell 1994b). If animals can detect a predator's approach early they can flee in sufficient time to escape (e.g. Hilton, Cresswell & Ruxton 1999), or have sufficient time to make the most appropriate escape response (Cresswell, Hilton & Ruxton 2000; Quinn & Cresswell 2005), thereby potentially increasing their fitness. There are, however, a number of alternative antipredation behaviours that may make vigilance relatively unimportant in determining fitness. For example, vigilance may actually reduce fitness in a cryptic animal because it increases conspicuousness and exposure time. Grouping also reduces the relative importance of vigilance, through the ‘many-eyes’ effect (Pulliam 1973), reduction in the individual probability of attack (Hamilton 1971) and the ‘confusion’ effect (Neill & Cullen 1974). Avoidance of predators may also reduce the importance of vigilance (Suhonen, Norrdahl & Korpimäki 1994; Schultz & Noë 2002) and consequently prey individuals that can avoid these areas may reduce their risk of predation more effectively than maintaining high vigilance in predator areas (e.g. Cresswell 1994a). The effectiveness of vigilance as an antipredation behaviour may also depend on habitat characteristics, with vigilance increasing in more obstructed habitats (Metcalfe 1984; Whittingham et al. 2004). In this study we measured how vigilance in overwintering grey partridges Perdix perdix (L.) varied with alternative antipredation behaviours and predation risk. We related vigilance to survival to determine if increased vigilance led to greater survival because it increased predator detection, or led to lower survival because it compromised foraging, or was a relatively unimportant antipredation behaviour.

We collected data on grey partridge antipredation behaviour, survival, habitat use and predation risk during the winter from September to early April at 20 farms across lowland England. Grey partridges show a clear change in vigilance-related behaviour through the winter so that there are two distinct periods where both group size and use of both partridge- and predator-concealing cover vary significantly. At the start of the winter they form flocks (coveys) that represent one or two pairs of adults and their offspring from the preceding breeding season, or adults that have not bred successfully. Coveys often forage in open habitats far from the edges of fields (Potts 1986) where predator detection is likely to be relatively easy (e.g. Metcalfe 1984). In late winter, however, the coveys break down into pairs as males become territorial and move to edge habitats, such as hedgerows, that contain suitable nesting habitat, but where predators may be concealed.

Because of the nature of compensatory antipredation behaviour it is impossible in many cases, a priori, to make explicit predictions about the direction of relationships (Lind & Cresswell 2005): for example, there may be no relationship between vigilance and survival if individuals with lower vigilance compensate by using alternative antipredation behaviours such as use of concealing cover, predator avoidance or reducing exposure (Blumstein 1998; Blumstein et al. 2004). Therefore, in this study we investigated a number of relationships between vigilance, group size, habitat and survival, and the constraints operating on them:

  • 1How vigilance and use of taller vegetation varies with group size, vegetation height and predation risk. We predicted vigilance to be higher in smaller groups and at sites with a higher sparrowhawk kill rate (overall predation risk). We also predicted that vigilance would depend on vegetation height, and that use of taller vegetation would also depend on overall predation risk at a site and group size; the direction of the relationships depending on whether cover acts primarily as a concealment for partridges or as concealment for approaching predators.
  • 2How vigilance and use of taller vegetation is constrained by partridge density (that may limit flock size) and the onset of breeding at the end of the winter. We explored how group size was a function of density at a site, and how group size and choice of vegetation height changed between the covey and the pair period.
  • 3How survival depended on vigilance levels and how this was affected by changes in group size and use of tall vegetation. We predicted that survival would depend on vigilance and/or group size and/or vegetation height, unless other behaviours compensated for the predation risk that any value of vigilance entailed and the degree to which vigilance compromised foraging behaviour. We also predicted that survival rate would be lowest in the pair period when group size was smaller, constraining the ability to compensate for increased predation risk.

Methods

We studied partridge antipredator behaviour relative to habitat use and predation risk from raptors during the winter at 20 sites on arable farmland in lowland England over three winters (2001–03). These sites held a range of different partridge densities under different levels of raptor predation risk. At each site up to 10 grey partridges from different coveys were fitted with 10-g necklace radio transmitters (≤ 2·5% of body weight with a lifespan of 9 months and range of up to 2 km Biotrack Ltd, Wareham, Dorset), with the aim of obtaining survival data and habitat use data for a minimum of six independent individuals per site (see Appendix 1 for summary). Birds were caught with a landing net after dazzling with a lamp at night during September. Each bird was then studied for two 6-day periods, once in October–December when birds were in family coveys (‘covey period’), and once in January–March when they were in pairs (‘pair period’). In order to maintain sample sizes for habitat use and behaviour observations, birds that died part way through the study were replaced by capturing and tagging new birds.

partridge density

Partridges were counted systematically at each of the 20 sites on specific count days (see Appendix 1). Counts were done using binoculars from a vehicle driven in a criss-cross pattern across each stubble and grass field. For sown fields, observations were made from the margins (Potts 1986). For sugar beet and game covers a pointing dog was used to flush coveys. One count was performed at the start of covey monitoring and another at the end of pair monitoring in order to reflect losses over winter. Counts were divided by the site area to give a density.

partridge behaviour

Focal behavioural observations of individually radio-tagged grey partridges within coveys and pairs were made from a vehicle using a telescope (n = 732 observations from 122 radio-tagged individuals). Coveys were searched for and sampled twice on specific behavioural sampling days (3 days per site in both the covey and pair period) at times determined at random. If a covey was located with at least one radio-tagged individual present, the group size was recorded: groups were sampled only under circumstances where group size could be counted properly, or where the group was flushed after observation enabling an accurate count. Then for each radio-tagged bird present, the proportion of 1 min spent vigilant, feeding, preening and sleeping were recorded. For feeding bouts, the number of pecks was recorded simultaneously using a counter. Vegetation height was measured using a ruler after the partridges had left the location: vegetation density was ignored because vegetation was sufficiently dense in all crops and stubbles to make detectability effectively a function of height. Vigilance was defined as being in the upright posture with neck outstretched, scanning the surroundings (Beani & Dessì-Fulgheri 1998). An attempt was made to locate and sample all radio-tagged individuals present on the site during a sampling period (i.e. twice during the sampling day) but in reality the procedure resulted in about one sample per individual per sampling day.

partridge habitat use

Partridge habitat use was measured by radio-tracking 150 individuals across the 20 sites (sample sizes are larger than for the vigilance data because habitat use data did not require clear visual observation of a feeding bird). Cropped and noncropped areas were verified by ground-truthing and mapped using a GIS (MapInfo 8.1). Habitats were aggregated into five qualitative classes: cereals, edge (including hedges, scrub, trees and plantations), grass, noncereals (gamecover, oilseed rape, roots, cabbages, leeks, beans, peas) and other (buildings, water, roads, tracks, plough). Habitats were also classified into five maximum vegetation height categories (class 1: 0–5 cm, class 2: 6–20 cm; class 3: 21–100 cm; class 4: 101–300 cm; class 5: 300+ cm). Habitat classes and heights were highly correlated: height classes 1 and 2 encompassed cereal and grass, whereas height classes 3–5 encompassed edge and noncereals; the latter two habitats were considered as ‘taller’ vegetation as they were sufficiently high to completely obscure partridges. For each covey or pair that contained a radio-tagged partridge, Minimum Convex Polygon home-range areas were generated (Kenward, South & Walls 2002) using a minimum of 20 fixes per individual. These were overlaid on the GIS habitat maps and the proportions of habitat classes at radio-locations compared with availability within the home range were taken as a measure of microhabitat use. Compositional Analysis was performed on the proportional data (Aebischer, Robertson & Kenward 1993). A measure of the relative use of tall cover was obtained by calculating log((ratio of proportion edge + noncereal used to proportion cereal + grass used)/(ratio of proportion edge + noncereal available to proportion cereal + grass available)) for each individual and then taking the average across individuals within each site.

raptor predation risk index and cause of mortality

An index of predation risk was calculated by adding the number of raptor kills of all bird species including tagged and untagged grey partridges found at each site controlling for hours searched and dividing by site area to give a density of raptor kills per site. Kill searching was performed by systematically walking field margins and woodland for the first 5 hours of data-gathering at each site: all kills found were buried to prevent double-counting. Any kills found during subsequent radio-tracking for up to 60 h were added to the total, corrected for the variable tracking effort across sites. Across all 20 sites, for covey and pair periods the total number of raptor kills found was 347 (Mean per site = 17·35, SE = 1·89). Some original raptor kills may have been scavenged by mammals and thus incorrectly classified. However, scavenging rates did not vary significantly across sites: in separate tests of scavenging rates the difference between the mean percentage of carcasses with fox Vulpes vulpes (L.) sign or completely removed after 72 h at high and low fox densities was not significant (t3 = −1·86, P = 0·58). Sighting rate per hour was calculated from raptors counts during six fieldwork days during the covey period and 6 days during the pair period. Density of raptor bird kills depended only on sparrowhawk Accipiter nisus (L.) sighting rate in the covey period (F1,18 = 7·6, R2 = 0·26, P = 0·013), suggesting that bird kill rate measured mainly the predation risk that they represented. The total number of dead grey partridges found was significantly positively correlated with the total number of dead birds found at a site (F1,18 = 6·1, R2 = 0·21, P = 0·023). We used overall number of kills rather than number of killed grey partridges as our measure of raptor predation risk because we are interested in the capacity for compensation by partridges across a site. Therefore the number of grey partridge kills may not necessarily reflect the number of raptors at a site, whereas the total number of all kills does, because at sites where partridge antipredation behaviour was effective, the raptors present would be likely to shift to more vulnerable prey.

The cause of death of radio-tagged partridges and other kills was determined by interpreting the sign left on the carcass. Raptors left plucked feathers and picked the skeleton clean. However, we were unable to distinguish between raptor species, although observational data of other raptor species and raptor hunting suggested that only sparrowhawks were taking grey partridges. Foxes left broken bones, saliva and jaw marks and often teeth marks on the tag itself. Collisions with wires left the body intact, but with external lesions and impact injuries such as broken bones. Carcasses that were sufficiently intact were given a postmortem examination by a gamebird veterinarian. No disease was detected.

partridge survival

Radio-tracking allowed the status of individuals (alive or dead) to be determined in any given interval: 31 weekly survival intervals were defined over the course of each of the study winters from early September to mid April. The Kaplan–Meier method was then used to calculate a cumulative survival rate for each site from the number of individuals at risk in each interval compared with the number that die. Variance for each interval was calculated by (survival rate (1 − survival rate)/number of birds at risk) and the standard error was calculated as the square root of the variance (Kaplan & Meier 1958). Freshly tagged birds were added to the number of birds at risk in each interval and censored birds were subtracted. Tag failures, lost signals and deaths due to stress from tag fitting and shooting were censored in order to compare survival against natural mortality. Partridges were monitored for 6 days in each of the covey and pair periods to determine habitat use and a final check was made on survival status at the end of the winter, as close to 30 March as possible. Approximately 3 months separated covey and pair monitoring at each site, so when dead partridges were found during the first check after this interval of absence, the death date was assigned to the mid-point of the period since last seen alive. A generalized χ2 statistic test (Saur & Williams 1989) showed that there was a significant difference in the survival estimates between the 20 sites (inline image = 38·2, P < 0·01). Of 150 tagged grey partridges, 18% were killed by raptors and 13% were killed by foxes (unpublished data). We do not further investigate different causes of partridge mortality within this paper; instead this paper tests how vigilance variation leads to variation in survival within the context of a system where raptor predation is clearly an important component of overall mortality, and where it is reasonable to assume that vigilance is likely to influence avoidance of predation.

analysis

Aim 1: variation in vigilance and use of taller vegetation with group size, vegetation height and predation risk

We tested how vigilance varied with group size and vegetation height using a Generalized Linear Mixed Model (GLMM in SPSS 12·0 as with all subsequent analyses) with mean proportion of time spent vigilant for an individual as dependent variable and group size and vegetation height during a focal observation as covariates. Analysis was therefore at the level of the individual mean from 732 focal samples (i.e. n = 122 individual means). Only birds that were not vigilant throughout the sample were included in the analysis (mean proportion of vigilance was 0·43 ± 0·03 SE and normally distributed, P = 0·71, Kolmogorov–Smirnov one-sample test compared with a normal distribution, n = 122): birds that were vigilant throughout a sample were ignored because they were assumed to be in an alert state after a disturbance rather than showing a basal vigilance rate prior to any predator detection. We controlled for confounding variables on vigilance (e.g. see Elgar 1989) by including sex as a factor and peck rate (per second spent nonvigilant) as a covariate (to control for food availability); position in the group was not considered because most birds were edge birds. Site number and period (pair or covey) were included as random variables to control for any confounding effects of unequal sampling across sites and between the two periods. Year was not considered (and this applies to all models) because sites were sampled only within a single year and so inclusion of site as a random effect in a model also controls for variation and lack of independence due to year effects. In any case, there was no significant variation in any variable with year (e.g. mean proportion of time spent vigilant per individual, F2,114 = 2·1, P = 0·12).

We tested how vigilance varied with predation risk with a GLM of proportion of time spent vigilant (dependent variable) with raptor predation risk index (covariate) controlling for group size, vegetation height and peck rate. We then tested how use of taller vegetation types varied with predation risk with a GLM: mean log-transformed proportional use of taller cover within home ranges was the dependent variable with raptor predation risk index (covariate) controlling for group size, proportion of time spent vigilant and peck rate. Analyses were all at the level of the site (i.e. n = 20 site means). Analyses were carried out separately on the covey and the pair period because of the differences confirmed when exploring Aim 2 below.

Aim 2: change in group size and use of taller vegetation through the winter

We explored how the potential for compensation to predation risk via vigilance was constrained. Analyses were all at the level of the site (i.e. n = 20 site means). First, we tested in a GLM how group size (dependent variable) was a function of mean density of partridges at a site (covariate) controlling for peck rate per second (as a surrogate for prey availability), vigilance rate (proportion of time spent vigilant) and vegetation height. Second, we confirmed how the group size changed between the covey and the pair period by comparing mean group size in the two periods with a t-test. Third, we tested how choice of vegetation height changed between the covey and the pair period based on paired differences in site mean log ratios of radio locations vs. availability within minimum convex polygon of range following compositional analysis (Aebischer et al. 1993).

Aim 3: survival and vigilance, group size and use of taller vegetation

We tested whether mean group size, height of vegetation where individuals fed, peck rate and proportion of time spent vigilant affected survival by comparing with t-tests the site means (n = 20) for the four variables for those partridges (where data were available) that survived (n = 57) with the site means for those that died (n = 11), over each of the pair and covey periods. A backwards step-wise logistic regression was then performed to assess which of the four variables was the best predictor of survival (0/1). The importance of vigilance to survival was tested with a GLM with site means of survival rate as dependent variable with proportion of time spent vigilant, controlling for peck rate, group size and vegetation height. Analyses were carried out separately on the covey and the pair period because of the differences confirmed when exploring Aim 2 above. Finally we tested whether death rate due to raptors was higher in the pair period (when compensation through group size and choice of vegetation height was constrained) by comparing the number of deaths at a site between the covey and the pair period using a matched pairs t-test (n = 20 sites). Note that this test assumes that the underlying number of birds available to die at each site is the same in the covey and in the pair period, which due to replacement of dead birds with new tagged birds is true.

Results

variation in vigilance and use of taller vegetation with group size, vegetation height and predation risk

Vigilance decreased significantly as group size increased (Table 1): assuming average peck rates, vegetation height, and a male bird during the covey period at site 20, then proportion of time spent vigilant will decrease from 0·543 to 0·127 from the minimum group size of 1 recorded to the maximum group size of 17, a decline in vigilance of 77%. Vigilance increased significantly as vegetation height increased (Table 1): assuming average peck rates, group size, and an average male bird during the covey period, then proportion of time spent vigilant is predicted by the model to increase from 0·298 to 0·793 from the minimum vegetation height of 0 cm recorded to the maximum vegetation height (in which partridges could be viewed) of 35 cm, an increase in vigilance of 166%.

Table 1.  Generalized linear mixed model to test variation in mean proportion of time spent vigilant per individual with group size, vegetation height and peck rate per second spent nonvigilant (as a control for food availability), controlling for period in the winter when surveyed, sex and study site (both period and study site entered as random effects). Data are mean values of 122 radio-tagged individuals from 719 1-min samples
Source Type III sum of squaresd.f.FSignificanceParameter estimate
  1. 1, 2 Parameter estimates for covey period and males, respectively.

InterceptHypothesis1·5 124·7< 0·001 
Error2·337·8   
Site number (1–20)Hypothesis1·519 1·4  0·16 
Error5·392   
Period (pair or covey)Hypothesis0·1 1 1·5  0·23−0·081 ± 0·0671
Error5·392   
SexHypothesis0·1 1 1·8  0·1820·103 ± 0·077
Error5·392   
Group sizeHypothesis0·4 1 7·7  0·007−0·026 ± 0·009
Error5·392   
Vegetation heightHypothesis0·3 1 5·9  0·0170·10 ± 0·004
Error5·392   
Peck rate per second head downHypothesis0·2 1 3·4  0·07−0·14 ± 0·08
Error5·392   

In the covey period, proportion of time spent vigilant was significantly higher at sites with a greater index of raptor predation risk (Fig. 1: F1,15 = 5·6, P = 0·032; overall R2 = 0·16, model also including group size, peck rate per second nonvigilant and vegetation height). Vigilance was not affected by risk in the pair period (F1,15 = 0·3, P = 0·61). In the covey period, use of taller concealing vegetation (edge vegetation and noncereal crops) was also significantly higher at sites with a greater index of raptor predation risk (Fig. 2: F1,15 = 4·6, P = 0·047; overall R2 = 0·32, model also including group size, proportion of time spent vigilant, peck rate per second nonvigilant). Group size was marginally significant in this model and this effect became significant if the nonsignificant terms of vigilance and peck rate were removed from the model: use of taller concealing cover was significantly less when mean group size at a site was higher (Fig. 3: group size, F1,17 = 4·6, P = 0·047; index of raptor predation risk, F1,17 = 6·8, P = 0·018, overall R2 = 0·26). There were no effects of index of predation risk (F1,15 = 0·2, P = 0·69) or group size (F1,15 = 0·03, P = 0·87) on relative use of tall cover in the pair period.

Figure 1.

Residual proportion of mean time spent vigilant with predation risk across 20 sites. The residuals after a regression to control for the effects of group size are plotted (y = 0·049x − 0·114, F1,18 = 4·3, P = 0·05, R2 = 0·15).

Figure 2.

Residual preference for taller vegetation with predation risk across 20 sites. The residuals after a regression to control for the effects of group size, peck rate per second nonvigilant and proportion of time spent vigilant are plotted (y = 0·70x + 0·139, F1,18 = 4·4, P = 0·049, R2 = 0·15). Habitat preference is the log value of tall to short vegetated areas used divided by the corresponding ratio for availability.

Figure 3.

Residual preference for taller vegetation with mean group size across 20 sites. The residuals after a regression to control for the effects of index of predation risk, peck rate per second nonvigilant and proportion of time spent vigilant are plotted (y = −0·047x + 0·378, F1,18 = 4·3, P = 0·05, R2 = 0·15). Habitat preference is the log value of tall to short vegetated areas used divided by the corresponding ratio for availability.

change in group size and use of taller vegetation through the winter

Potential for compensation to predation risk via shared vigilance in larger groups was constrained because density of partridges at a site was correlated with group size (Fig. 4: F1,15 = 8·2, P = 0·012; overall R2 = 0·22, model also including proportion of time spent vigilant, peck rate per second nonvigilant and vegetation height): as density increased so mean group size at a site increased. Compensation to predation risk by vigilance was also constrained by the formation of pairs in late January or early February: mean group size during the covey period was 8·0 ± 0·43, vs. 2·6 ± 0·25 in the pair period, n = 20 sites, matched pairs t-test, t = 9·6, P < 0·001. There was also a significant twofold increase in the use of tall cover (mainly increased use of hedgerows) from the covey to the pair period: vegetation height, site means of radio locations vs. availability within minimum convex polygon of range, Wilks’λ = 0·47, F3,15 = 4·2, P = 0·02. In total, 89·9% of radio locations in the covey period were in vegetation height 0–100 cm and 10·1% in height 101–300 cm, compared with the pair period when usage was 77·5% and 22·5%, respectively; the availability of vegetation height 101–300 cm remained the same at approximately 8%.

Figure 4.

Residual mean group size with density of partridges per km2 at 20 sites . The residuals after a regression to control for the effects of proportion of time spent vigilant, peck rate per second nonvigilant and vegetation height are plotted (y = 0·096x−1·25, F1,18 = 9·2, P = 0·007, R2 = 0·30).

survival and vigilance, group size, and use of taller vegetation

Over the whole winter group size and proportion of time spent vigilant were the only significant single predictors of individual survival: individuals that survived were from larger groups and had lower vigilance levels (Fig. 5). Only group size was a significant predictor of survival probability over the winter, when both variables were considered in a logistic regression to predict survival (group size, Wald = 4·0, P = 0·047; vigilance, Wald = 0·02, P = 0·88; model also including site number; Nagelkerke R2 = 0·71, overall 89·7% correct classification of cases). There was no effect of raptor predation index on survival when included in the model (Wald = 0·6, P = 0·41) demonstrating that survival was not simply a consequence of raptor abundance. In the covey period, there was, however, a trend for a higher probability of survival at sites that had a lower mean value of vigilance controlling for group size (proportion of time spent vigilant, F1,15 = 4·4, P = 0·05, R2 = 0·25, model including peck rate per second nonvigilant, group size and vegetation height; Fig. 6). In the pair period, there were no significant relationships between mean survival rate and vigilance, peck rate per second, group size or vegetation height (maximum F1,15 = 1·6, minimum P = 0·22). However, across the 20 sites most partridge deaths to raptors occurred in the pair period (paired t-test, radio-tagged and untagged kills pooled, t1,19 = 2·2, P = 0·04, Fig. 7).

Figure 5.

The mean values of behaviours (± 1 SE) comparing grey partridges that died with those that survived over the winter: (a) group size, (b) proportion of time vigilant, (c) vegetation height, and (d) peck rate per second spent nonvigilant. n = sample sizes of birds; sample sizes are smaller for (d) because of missing values. Significant t-test values are given for (a) and (b).

Figure 6.

Residual mean survival rate and proportion of time spent vigilant at 20 sites. The residuals after a regression to control for the effects of group size, peck rate per second nonvigilant and vegetation height are plotted (y = −0·51x + 0·221, F1,18 = 4·7, P = 0·044, R2 = 0·16).

Figure 7.

Mean number of partridges (± 1 standard error) killed by raptors found in the covey and pair periods across 20 sites, radio-tagged and untagged birds pooled.

Discussion

The key results from this study are, first, that both vigilance levels and use of taller vegetation correspond to levels of risk, with vigilance being higher in small groups, in taller vegetation and at sites with higher predation risk, and use of taller vegetation being higher for small groups and at sites with higher predation risk. Secondly, that despite vigilance increasing with risk, those individuals with higher vigilance levels were less likely to survive and group size was the primary determinant of survival. Thirdly, although formation of groups allows partridges to maintain high collective vigilance and survival, formation of large groups was limited by partridge density.

vigilance, use of tall vegetation and risk

Vigilance was significantly higher when individuals fed in smaller groups as has been found in many studies (Elgar 1989) because the probability of predator detection for any given level of vigilance increases with group size (Pulliam 1973). Vigilance was significantly higher in taller vegetation, as has been found in several studies (Metcalfe 1984; Ebensperger & Hurtado 2005) Vigilance was also significantly higher for partridges at sites with a higher index of raptor predation risk (see Elgar 1989; Childress & Lung 2003; Wolff & Van Horn 2003).

In the covey period, use of taller vegetation was significantly higher at sites with a higher index of raptor predation risk with smaller groups tending to use taller vegetation more. This suggests that taller vegetation may have acted as a refuge or an area of concealment to reduce attack rate in areas where attack risk was higher and for smaller and so more vulnerable groups. Vigilance was higher in taller vegetation, however, suggesting that predation risk was not reduced relatively when partridges were in taller vegetation; why then, do partridges use tall vegetation at all when there are apparently no predation risk benefits? The vigilance cost for feeding in taller vegetation was of a similar level to the vigilance cost of being in a smaller group (see Results, ‘Variation in vigilance and use of taller vegetation with group size, vegetation height and predation risk’) suggesting that vigilance costs may not have been determining this group-size-dependent habitat choice. Instead it may be that taller vegetation is a better foraging habitat in terms of food availability: small groups (and those in more risky sites) may then have traded off an increased predation risk of feeding in raptor concealing cover with higher intake rates, possibly because intake rates in open habitats may not have been sufficient where individuals would have to maintain high vigilance levels anyway. Use of tall vegetation should also be a more favoured antipredation strategy for smaller groups if tall cover obscures partridges and reduces attack rate, because on attack there will be relatively few opportunities for early detection (tall vegetation concealing an approaching predator as well) compared with those that larger groups might gain when foraging in the open (e.g. see Pulliam 1973). That there was a significant twofold increase in the use of tall cover, in the pair period compared with the covey period is also consistent with the suggestion that use of concealing cover is a likely secondary strategy to compensate for increased predation risk when large group sizes are not available as an option.

vigilance, group size and survival

When controlling for group size, mean survival decreased as vigilance increased in the covey period. This result, along with vigilance being higher at sites with increasing with raptor risk, suggests that even though increases in vigilance decrease mortality from predation instantaneously, they may increase mortality from starvation and indeed other types of predation overall. An individual grey partridge probably responds to raptor predation risk by becoming more vigilant to avoid being the individual caught on attack. However, increasing vigilance will reduce time available for feeding and so increase the probability of starvation if foraging is unpredictable, or increase the amount of time that the individual must spend feeding and so exposed to higher levels of predation risk. It is also interesting to note that there may be an additional cost of vigilance on intake rate: peck rate per time spent nonvigilant decreased with increased vigilance (although this result was only P = 0·069, see Table 1). This may have arisen because if animals searching for cryptic prey interrupt a search task then they suffer a temporary decline in the probability of success when resuming the search (see Dukas & Clark 1995).

Why then do partridges show higher vigilance at more risky sites if selection ultimately favours individuals with low vigilance overall? The answer to this may be because selection of individuals is dependent on the behaviour of other individuals. On a daily basis in a high-risk area, those individuals that have low vigilance, all other things being equal, will be more likely to be captured on attack so all individuals are forced to adopt the highest level of vigilance that allows them to meet their daily intake needs. Individuals will vary in foraging efficiency, ability to detect predators and their body condition so there will be variation in the amount of time that individuals allocate to vigilance, with some individuals allocating more time to foraging and so reducing their starvation risk more (e.g. Whittingham et al. 2003). Therefore, for any given level of mean vigilance (e.g. high in a risky site or a small group) those individuals that can allocate more time to feeding will have a lower probability of starvation or time spent exposed to predation, and therefore have a higher probability of survival.

The argument above, that reducing vigilance time maximizes long-term survival, but compromises short-term survival, is supported by the result that over the whole winter, group size was the only significant predictor of individual survival. One way that partridges can overcome the conflict between the need to have high vigilance to avoid being captured in the short term, with the need to have low vigilance to avoid starving in the long term, is to form a group, because individuals in groups can reduce their vigilance levels greatly without compromising their ability to detect predators. The effect of raptors in our study was probably less when there were large groups of grey partridges in open habitats, where individual partridges can probably both detect predators and feed efficiently.

It is important to note that although we argue that the starvation–predation risk trade-off is important in determining mortality rate in grey partridges, we do not provide any evidence for starvation risk being high in any of the sites (e.g. starvation deaths). However, starvation deaths are unlikely even in systems where starvation risk dominates (e.g. see Cresswell & Whitfield 1994; Yasué, Quinn & Cresswell 2003; Minderman, Lind & Cresswell 2006) because animals adopt increasingly risky foraging options as they starve, leading to increased mortality from predators (Lima 1998).

constraints on vigilance and group size

The ability of the grey partridges in our study to maximize predator detection while minimizing vigilance was constrained however, because maximum group size was determined by density during the winter (covey period). Compared with many other birds (e.g. Cresswell 1994b) the grey partridges’ ability to compensate for increasing predation risk through increasing group size is limited because coveys are usually based on family units. For example, partridge group size is predominantly determined by breeding density and chick survival rate in the preceding breeding season (Potts & Aebischer 1995). Therefore, minimum vigilance levels in this study as a result of the group size effect were limited by the availability of other birds with which to form large groups. Moreover, among ground-dwelling gamebirds the antipredator benefits of flocking seemed to be constrained at approximately 11 individuals. This upper limit may be set by the negative consequences for interference competition in groups above this size (e.g. Blumstein, Daniel & Evans 2001; Williams, Lutz & Applegate 2003).

Compensation to predation risk via group size was also constrained by the formation of pairs at the end of the winter (pair period). In effect, partridge antipredation behaviour was disrupted by the demands of pair formation and nest searching prior to breeding: accepting greater predation risk is often a cost of reproduction (Newton 1998). The greater mortality to raptors in the pair period corroborates this.

applied consequences

In the intensively used landscapes of developed countries there is a conflict between conservation of predator species and conservation of prey species, against a background of environmental degradation (Thirgood et al. 2000; Ormerod 2002; Allen & Feare 2003). Given that raptor control is not acceptable because it threatens the status of species of conservation concern (Allen & Feare 2003; Greenwood, Crick & Bainbridge 2003), there is an urgent need to quantify the population effects of raptors on partridges and secondly to seek benign solutions if necessary. If, as our results imply, predation risk is mediated by habitat (e.g. see also Quinn & Cresswell 2004), and changes in the availability of habitat are practical, habitat manipulations could help reduce conflict between game management and conservation interests (Ormerod 2002).

So is there a habitat solution to mitigating predation risk? The fact that in the covey period use of concealing cover 100–300 cm tall was higher at sites with greater predation risk, and that survival was probably related to vigilance implies that manipulating cover availability to favour vigilance could increase prey survival. Providing additional cover to partridges as part of general habitat enhancements to benefit all farmland birds is an attractive idea (Stoate & Szczur 2001). However, our results do not indicate what the optimal cover for partridges is in all circumstances. For the covey period, partridges increased vigilance as cover height increased up to the limit of measurement at 35 cm tall. Thus cover of intermediate height, especially stubbles at 12–20 cm appeared suboptimal, requiring greater vigilance and therefore reducing foraging time, despite the benefits of being in a covey. Modern cereal varieties have short stems, removing the possibility of leaving a tall stubble of concealing cover above 30 cm high, so cutting shorter stubbles would decrease the level of obscuration, reduce vigilance and increase foraging time.

Our results also suggest an Allee effect (e.g. Keitt, Lewis & Holt 2001). The major determinant of survival was group size, probably mediated by the reduction in allocation of time to vigilance, so increasing time available for foraging that can occur in larger groups. At sites with smaller group sizes there may be a reduction in survival because of increased time spent foraging when individual vigilance levels remain high. Therefore maintaining a higher density of partridges to begin with at a site, as well as habitat management, may reduce the impact of raptor predation. Conversely, the importance of habitat management to reduce predation risk probably increases as the density of partridges decreases.

vigilance and fitness

We have shown that antipredation decisions made by grey partridges have consequences for individual survival. Over the whole winter, partridges had higher survival in larger groups and when they had lower levels of vigilance: few studies have demonstrated such effects because of the difficulties of controlling adequately for compensatory factors (Lima & Dill 1990; but see Williams et al. 2003). Our study supports the general theoretical principle that antipredation behaviours may reduce fitness long-term via their effects on the starvation–predation risk trade-off, even though they decrease predation risk short-term (the predation-sensitive food hypothesis McNamara & Houston 1987). Ecological constraints, such as poor habitat that increase probability of starvation, so that time allocated to antipredation behaviours severely compromise foraging, may cause mortality indirectly (see Evans 2004), rather than antipredation behaviours such as vigilance directly determining mortality.

Acknowledgements

We thank the many landowners, gamekeepers and farmers who gave access to land and supported the study. We thank Ian Newton, Johan Lind, John Quinn and Dan Blumstein for helpful comments. Mark Watson was funded by The Game Conservancy Trust. Will Cresswell is a Royal Society University Research Fellow.

Table Appendix 1..  Details of 20 independent study areas
Periods of studyStudy siteArea (km2)Partridge density spring (birds km−2)Total spring countNo. of partridges radio-tagged
Sep 00–Apr 01Dorset 122·51·5134 6
Hampshire 19·765·1250 7
Hertfordshire 12·839·8928 8
Sussex 19·832·4424 5
Sep 01–Apr 02Lincolnshire 15·336·573510
Norfolk 18·465·6748 7
Norfolk 25·598·4147 8
Norfolk 36·225·7936 8
Norfolk 46·259·2858 7
Nottinghamshire 15·6312·437010
Wiltshire 110·774·0944 6
Wiltshire 25·387·2539 7
Sep 02–Apr 03Dorset 28·062·4820 5
Hampshire 27·864·7137 6
Norfolk 54·7116·9780 8
Norfolk 63·4815·8155 8
Nottinghamshire 24·242·5911 8
Wiltshire 312·672·0526 6
Wiltshire 420·670·7716 4
Wiltshire 58·393·731 7

Ancillary