Factors affecting predation by buzzards Buteo buteo on released pheasants Phasianus colchicus


R.E. Kenward, Centre for Ecology and Hydrology, Winfrith Technology Centre, Dorchester, Dorset DT2 8ZD, UK (e-mailreke@ceh.ac.uk).


  • 1Information on the effects of wildlife predation on game and livestock is required to allow improved management of all organisms involved. Monitoring of prey, predators and predation mechanisms each suggests important methods, illustrated here by data from common buzzards Buteo buteo and ring-necked pheasants Phasianus colchicus.
  • 2Location data from 136 radio-tagged common buzzards, together with prey remains from 40 nest areas, records from 10 gamekeepers and vegetation surveys, were used to investigate raptor predation at 28 pens from which pheasants were released in southern England.
  • 3Among 20 725 juvenile pheasants released in 1994–95, gamekeepers attributed 4·3% of deaths to buzzards, 0·7% to owls, 0·6% to sparrowhawks, 3·2% to foxes and 0·5% to other mammals.
  • 4Fresh pheasant remains were found on 7% of 91 visits to buzzard nests, and 8% of radio-tagged buzzards had significantly more association than other buzzards with pheasant pens.
  • 5Predation by buzzards was most likely to be recorded at release pens with little shrub cover, deciduous canopies and a large number of released pheasants. The number of pheasants killed was greatest in large pens with extensive ground cover, and the highest proportion of released pheasants was killed in large pens where few were released. However, only 21% of 55 releases had > 2 pheasant kills per week.
  • 6Radio-tagged buzzards were located most often at pheasant-release pens with open, deciduous canopies. Pens were most likely to be visited by buzzards that had fledged nearby, but the proximity of buzzard nests had little influence on how much predation occurred.
  • 7Only a minority of buzzards associated frequently with pheasant pens, and predation was heavy at only a minority of sites, where pen characteristics and release factors probably made it easy for individual buzzards to kill pheasants. We suggest that the occasional heavy losses could be avoided by encouraging shrubs rather than ground cover in pens, by siting pens where there are few perches for buzzards, and perhaps also by high-density releases.


The extent and management of raptor predation on game and livestock is a controversial issue for applied ecologists. Whereas there is much recent evidence that mammal predators can reduce the abundance and distribution of birds and mammals (Marcström, Kenward & Engren 1988; Tapper, Potts & Brockless 1996), avian predators have seldom been shown to reduce wild prey populations (Lack 1954, 1966; Newton 1979, 1998; Thomson et al. 1998). Marked impacts by raptors have been recorded only on galliform birds, for example where goshawks Accipiter gentilis L. were estimated by radio-tracking to kill 56% of wild hen pheasants Phasianus colchicus L. during one winter in a Swedish study area (Kenward, Marcström & Karlbom 1981). The best evidence to date of prolonged impact by avian predators on wild prey is the decline in red grouse Lagopus lagopus scotticus Latham density in an area where raptor density was allowed to increase strongly, compared with increasing numbers of grouse in nearby areas with few raptors (Redpath & Thirgood 1997, 1999; Thirgood & Redpath 1997; Thirgood et al. 2000a,b,c).

Predatory impacts in cases where raptors kill livestock, domestic pigeons or farmed game are more easily assessed than where prey belong to wild populations. As a whole, the frequency of this predation is considered to be low (Lockie 1964; Newton 1979; Ratcliffe 1980; BTO et al. 1997). None the less, eagles Aquila sp. have caused locally severe problems for farmers (Murphy 1977; O’Gara et al. 1983; Davies 1999) and peregrine falcons Falco peregrinus Tunstall for pigeon-fanciers (Musgrove 1996). Goshawks were estimated to have killed an estimated 19% of 4300 released pheasants at an estate in Sweden (Kenward 1977). In Britain, gamekeepers reported that buzzards Buteo buteo L. caused higher losses than smaller raptors in 147 incidents at pheasant pens; the buzzards killed an average of 3·2% of the released pheasants, although on eight occasions they accounted for 10% or more (Harradine, Reynolds & Laws 1997).

Where predation is thought to conflict with human interests, it is important to understand the extent of the problem and the predatory mechanisms involved, so that remedies can meet economic and conservation objectives (Reynolds & Tapper 1996). The extent of predation from all sources can be quantified by radio-tracking the prey (Hessler et al. 1970; Dumke & Pils 1973; Keith & Windberg 1978; Angelstam 1984; Redpath & Thirgood 1997; Thirgood et al. 2000a,b,c), but with this approach it can be difficult to identify predators and predatory mechanisms that include any selection effects. An alternative approach is to monitor predators, to estimate kill-rates and observe predatory mechanisms. However, assessment of predatory impact then requires separate estimates of predator densities and prey demography, including density-dependent effects, to determine the extent to which predation is additive to other losses.

In this paper, we present comparative data from both predators and prey. Observations at 40 buzzard nesting sites provided kill-rate estimates, and radio-tracking of 136 common buzzards during 1990–95 gave data on their presence at pheasant-release pens. Levels of predation on pheasants were assessed by gamekeepers at 28 pheasant-release pens. We investigated whether the extent of predation and the independent records of the presence of buzzards could be linked to site characteristics. Our aim was to identify factors that might be managed to reduce buzzard presence and prey vulnerability.


The study was conducted in 125 km2 of Dorset, in southern Britain. The area comprised 18% woodland, 65% pasture or arable farmland and about 13% heathland. It was searched annually from 1990 to 1995 to find all buzzard nests; 40 nesting areas were identified. During 91 visits to 61 nests, we recorded 233 prey remains that were freshly dead, as well as fur and feathers of older kills. Active nests were visited one to two times during the nestling period, initially to measure the young and later to equip 157 of them with back-pack radio-tags (Biotrack Ltd, Wareham, UK). The back-pack radios that were used in 1990–91 had a battery life of 2 years; those mounted thereafter transmitted for 4 years (Walls & Kenward 1995, 1998). However, 16 radios that were mounted on tail feathers of juveniles trapped near nests after fledging in 1990–91 were moulted within a year. After leaving the nest in June and early July, radio-tagged young were tracked with programmable receivers (Advanced Telemetry Systems, Isanti, MN) twice a week until October in 1990–92, once a week over the same period in 1993–94, and once a month in 1995. Locations were collected at a resolution of 100 m, by triangulation from within 1 km of each buzzard with a three-element hand-held Yagi antenna (Biotrack Ltd, Wareham, UK), or from within 2 m in a Landrover with a mast-mounted five-element Yagi.

The locations of all 32 pens used to release pheasants during 1990–95 were noted during field work and by contacting gamekeepers (Fig. 1). Release pens were 150–12 000 m2 in area and had sides of wire mesh, 2–2·5 m high. There were often outer strands of electric fence to deter mammalian predators. The young pheasants could leave by flying through the open tops of the pens and re-enter through one-way gates at ground level (Hill & Robertson 1986). All 28 pens used for the prey study during 1994–95 were managed by 10 gamekeepers. During July, between 30 and 1000 pheasants that had been reared for 5–7 weeks in smaller enclosures were introduced into each pen. Hoodless et al. (1999) review other aspects of pheasant management.

Figure 1.

The variable densities and distances apart of pheasant pens (squares) and buzzard nests (circles, elongated for different sites in the same territory) in use during 1994 and/or 1995, within the main study area (—) in south Dorset and the boundary for location analyses (- - -); regular displacement of sites conceals true locations.

The gamekeepers visited pens at least once a day, and searched the vicinity for dead pheasants. Juveniles killed by predators were therefore found fresh, which was important for assessing the causes of death. Most mammal predation was attributable to foxes Vulpes vulpes L., which cached remains or removed kills but left feathers bitten through the shaft. Mustelids Mustela erminea L. and M. vison Schreber left carcasses with small bite marks. White faecal traces were often associated with raptor kills, which were attributed to buzzards if there were deep beak marks in the sternum and relatively little plucking. Accipiters pluck kills more extensively, but sparrowhawks Accipiter nisus L. could not remove large pieces of the sternum (Brüll 1964). Goshawks were absent. Tawny owls Strix aluco Shaw pluck few feathers, which resembles the behaviour of buzzards, but often remove the heads of pheasant poults (Lloyd 1975). Regular visits by gamekeepers helped to separate kills made by nocturnal predators (mammals, owls) from diurnal ones (buzzards, sparrowhawks); the latter were often seen at kills.

We surveyed pens during visits in May and June 1996, after checking with gamekeepers that vegetation in pens had not changed since 1994. Vegetation was measured at 50 points spaced regularly in the 28 pens, as presence of ‘ground cover’ below 0·5 m and ‘shrub cover’ at 1 m, and as canopy closure. Ground cover was typically dogs mercury Mercurialis perennis L., nettle Urtica dioecia L., bracken Pteridium aquilinum Kuhn and bramble Rubus sp., with shrub cover sometimes including nettles, bracken, bramble, rhododendron Rhododenron ponticum L., holly Ilex aquifolium L. and hazel Coryllus avellana L. Canopy species were oak Quercus robur L., hazel, sycamore Acer pseudoplatanus L., ash Fraxinus excelsior L., beech Fagus sylvatica L. and pines Pinus sylvestris L. and P. nigra J.F. Arnold. Records included whether the canopy in the pen and surrounding 200 m contained branches suitable as perches for buzzards, and the proximity of buzzard nests. Only one pen was more than 200 m from fields or parkland.

Survey data were subjected to regression-based analyses of variance and covariance in two stages, using Minitab 11 (Minitab Inc., State College, PA). Logistic regression was used to investigate the effect of pen variables on the probability of buzzards killing pheasants in each pen. Where predation occurred, general linear models were then applied to examine relationships between pen variables and the number of pheasants killed or the proportion of those released that were killed. Logarithmic transformations were used for distance, areas and numbers of pheasants, with arc-sine transformations for proportions of cover. All statistical tests were two-tailed.

Buzzard locations were analysed with the program Ranges V (Kenward & Hodder 1996). To make allowance for small discrepancies between observers in recording locations, radio-tagged buzzards were considered likely to visit pheasant-release pens if they were recorded up to 141 m away (i.e. to the furthest corner of an adjacent cell scaled to the 100-m tracking resolution). The proportion of ‘visiting pen’ locations gave one measure of association with pens for each bird. A second measure of tendency to associate with pens was estimated by comparing the geometric mean distance from pens of (a) all radio-locations with (b) random points in the convex polygon round those locations. This comparison was expressed as Jacobs’ index (Jacobs 1974). This index, D[= (b− a)/b + a − 2b × a)], tends towards +1 for strong association and −1 for avoidance, and was 0 if the mean of observed distances to pens equalled the mean distance to random points. Three categories of association with pens were defined by cluster analysis of the values from each bird on the axes given by these two measures. We used Minitab 11’s hierarchical cluster analysis with a centroid joining rule, which is appropriate for compact grouping, and the Pearson squared distances, which are appropriate for centroid-based clusters (Lance & Williams 1967).

Analyses were for the July–October period, before pheasants reached full size and were likely to disperse. However, to avoid strong associations of buzzards with pens merely because they fledged in nearby nests, we excluded locations of first-year birds before August, by which time young buzzards had moved away from their nests and some had dispersed (Walls & Kenward 1995). We did not use locations > 1 km from the study area; 1 km was the standard diameter of an autumn range (Walls & Kenward 1998). Buzzards with less than five usable locations in their first year were excluded from the analysis because they often did not provide a polygon estimate for Jacobs’ index.


The extent of predation

The 27 study pens used in 1994 contained 10 200 pheasants; in 1995 there were 10 525 in 28 pens. The gamekeepers estimated that 1971 (9·5%) of the 20 725 poults were killed by predators. They attributed 901 kills (4·3% of those released) to buzzards, 144 (0·7%) to tawny owls, 128 (0·6%) to sparrowhawks, 24 (0·1%) to corvids, 659 (3·2%) to foxes and 115 (0·6%) to other mammals. There was no significant difference between years in these categories (χ2 = 5·8, d.f. = 4, P > 0·2).

To investigate whether deaths might have been attributed in error to one particular predator instead of another at some pens or by some gamekeepers, we sought negative correlations between numbers of kills attributed to each type of predator. The largest negative correlations with numbers of buzzard kills, for either the 27 pens in 1994 or the 28 pens in 1995, were r = −0·09 (d.f. = 25, P > 0·50) for foxes, r = −0·22 (d.f. = 25, P > 0·10) for other mammals, r = −0·11 (d.f. = 26, P > 0·30) for sparrowhawks and r = −0·28 (d.f. = 26, P > 0·10) for owls.

The number of pheasants killed in each pen in 1994 correlated strongly with the number killed in 1995 (r = 0·85, d.f. = 25, P < 0·001) but the correlation between the proportion killed by buzzards in each pen each year was poor (r = 0·34, d.f. = 25, 0·05 < P < 0·1). There was considerable variation in the number of kills across the 55 releases (Fig. 2a). No juvenile pheasants were killed by buzzards on nine occasions (16%), with less than 20 kills pen−1 in another 23 of the 55 releases (42%). However, 35–40 pheasants were believed to have been killed in 12 (22%) of the releases, including two occasions when losses exceeded 20% of the released birds (Fig. 2b).

Figure 2.

The average numbers (a) and proportions (b) of pheasants estimated to have been killed by buzzards at 28 pens during 1994–95.

Fresh remains of pheasants were found in six of the 91 visits to buzzard nests (7%), in each case in a different nest, and thus involving 15% of the 40 nest areas. Assuming that kills were available in the nest as fresh remains for 1–2 days, then six kills in 91 visits represented an average kill-rate of 0·017–0·033 pheasants day−1 for each of the two buzzards that provisioned each nest during May–July. Pheasants formed a small proportion (2·6%) of the 233 prey remains (Fig. 3), which were predominantly rabbit Oryctolagus cunniculus L. Buzzards nesting close to pheasant pens did not necessarily kill pheasants: a nest found inside one of the pens contained no pheasant feathers at all.

Figure 3.

The proportion of buzzard nests at which different prey were found freshly killed during 1990–95 and the proportion of each prey category among the fresh prey remains.

Relationships between predation and release-pen variables

Before considering the factors associated with severity of predation in pens where pheasants were killed, we compared sites with and without predation (Table 1). No predation by buzzards was recorded in 1994 at pens with most shrub cover (G = 3·96, d.f. = 1, P < 0·05). The addition of canopy type to the regression (z = 1·87, P = 0·06) gave a marginal increase in significance (G = 8·26, d.f. = 2, P < 0·02). In 1995, there was no predation by buzzards at pens with relatively few pheasants (G = 5·07, d.f. = 1, P < 0·05). Addition of shrub cover to the regression (z = 1·90, P = 0·06) gave a substantial increase in significance (G = 12·79, d.f. = 2, P = 0·002; Fig. 4). Data from both years could be combined by taking the average number of pheasants released in each pen, with predation ranked as present in both years, in one year or in neither. The best model from nominal logistic regression then contained all three explanatory variables (G = 17·5, d.f. = 3, P = 0·001). Therefore, buzzards tended not to kill pheasants when there was above-average shrub cover, below-average deciduous canopy and relatively few pheasants present in the pen. These variables were not intercorrelated (the highest correlation was r = 0·18, P > 0·10).

Table 1.  Mean values, with 95% confidence intervals, of eight site variables at release pens with records of pheasants killed by common buzzards and at pens that lacked this predation. The G-values are from binary logistic regressions based on these two pen categories
 n of pensArea of pen (ha)PheasantsPercentage cover% of canopy deciduousBuzzard nests < 2 km
  • P < 0·10,

  • *

    P < 0·05.

With predation2323803021267912883912·7
by buzzards 1415–4005210–434 822–195582–9717–3777–8871–1002·2–3·3
No predation 411501851582935575501·8
by buzzards  365–3680 84–406 727–343963–10020–8237–97 0–1000·2–3·3
G for difference    1·34  1·40   0·19 0·12 3·96* 0·88 3·522·66
With predation2322663251430913184913·6
by buzzards 1350–3800234–449 988–207083–9720–4178–8971–1002·9–4·2
No predation 51540127 823904677403·2
by buzzards  506–4676 36–450 114–591066–99 9–7656–93 0–1000·1–6·3
G for difference    0·49  5·07*   1·23 0·01 1·35 1·07 2·650·22
Figure 4.

The percentage of cover at 1 m, and the number of pheasants released into pens in 1995, where buzzard predation occurred (filled circles) or was absent (open circles).

In pens where buzzards killed pheasants, the number of kills correlated strongly in each year (P < 0·001) with the area of the pen (Table 2). The second variable to enter a linear regression was the percentage of the pen with cover at ground level, in 1994 (F = 5·8, d.f. = 1, 20, P < 0·05) and 1995 (F = 11·0, d.f. = 1, 20, P < 0·01). Correlations between predation and ground cover were positive (Table 2): there were most kills in large pens with extensive ground cover (R = 0·80, d.f. = 20, P < 0·001 in 1994; R = 0·79, d.f. = 20, P < 0·001 in 1995; Fig. 5). The number of pheasants killed also increased with the number in the pen, decreased with their density, and in 1994 increased with open canopy (Table 2). However, none of these variables had a residual effect on the number of pheasants killed after pen area had been taken into account.

Table 2.  Coefficients for correlations between the extent of predation by common buzzards and eight site variables at the 23 pens where pheasants were killed in 1994 and 1995
 Area of pen (ha)PheasantsPercentage cover% of canopy deciduousBuzzard nests < 2 km
  • P < 0·10,

  • *

    P < 0·05,

  • *†

    P < 0·02,

  • **

    P < 0·01,

  • ***

    P < 0·001.

n of pheasants killed+0·74***+0·47*†−0·49*†+0·24−0·05−0·47*†−0·18+0·06
% of pheasants killed+0·15−0·55**−0·64**−0·13−0·28+0·06−0·05+0·19
n of pheasants killed+0·64**+0·41*−0·54**+0·39−0·08−0·29−0·16+0·22
% of pheasants killed+0·03−0·46*−0·37+0·07−0·16+0·35−0·03+0·34
Figure 5.

The number of pheasants killed as a function of area of pens with (filled circles) and without (open circles) predation by buzzards in 1995.

Whereas the number of pheasants killed was not related to the number released (after taking pen area into account), a reduction in the proportion killed where many pheasants were released (Table 2) was improved by adding pen area. This effect occurred in both 1994 (F = 16·6, d.f. = 1, 20, P < 0·005) and 1995 (F = 5·7, d.f. = 1, 20, P < 0·05), to give equations that explained 38–62% of variation in the proportion of birds killed (R = 0·79, d.f. = 20, P < 0·001 in 1994; R = 0·62, d.f. = 20, P < 0·01 in 1995). Thus, the proportion of pheasants killed was high where few were present in large pens, i.e. with birds at low density (Fig. 6).

Figure 6.

The proportion of pheasants killed as a function of pheasant density at pens with (filled circles) and without (open circles) predation by buzzards in 1994, plotted as a residual after taking account of ground cover.

At sites with predation, the number of kills did not correlate significantly with shrub cover. Across all sites, however, the average number of pheasant deaths attributed to buzzards was always less than 20 at the seven sites with more than 50% shrub cover, whereas 11 of the other 21 sites had at least 20 kills. None of the cover variables showed significant intercorrelations (−0·1 > r < +0·1).

After taking account of stocking density and cover variables, neither the proximity of the nearest buzzard nest nor the density of buzzard nests had a significant effect either on whether predation occurred at all or the number of kills, either in 1994 or 1995. However, an increase in the number of pheasants killed at each pen between 1994 and 1995 was significantly associated with an increase in the numbers of buzzard nests within 2 km of pens (Fig. 7; r = 0·47, d.f. = 25, P < 0·02) and was marginally associated with an increase in the number of buzzard nests within 1 km (r = 0·32, d.f. = 25, P = 0·1) but not with change in distance to the nearest nest (r = −0·07).

Figure 7.

The change between 1994 and 1995 in the number of pheasants killed at release pens in relation to whether the number of buzzard nests within 2 km of the pen decreased slightly (pale histograms) or increased (histogram shading shows extent of increase).

Association of radio-tagged buzzards with pheasant pens

During 1990–95, 136 buzzards tracked in August–October of their first year had at least five radio-locations (mean 12, spread 5–24) within 1 km of the study area. Measures of their association with pheasant pens were (i) the proportion of locations < 200 m from pens and (ii) the tendency to favour areas near pens within the convex polygon round all their locations (Jacobs’ index). Cluster analysis on these two axes divided the birds into three categories of association with pheasant pens (Fig. 8), on the basis of dendrogram discontinuity at the 90% similarity level. With the axes as two independent variables, there was then 100% classification of the three categories by ordinal logistic regression (G = 89·3, P < 0·001). One large cluster (depicted by open circles) contained 100 birds (74%) that were never recorded with 200 m of pens (thin line circles) and another 25 (18%) with < 20% of their locations at pens (thick line circles). The mean Jacobs’ index for the 100 ‘no-visit’ birds, including 76 with no pheasant pen in the polygon round their locations, was D = −0·04 (spread of −0·55 to +0·66). None of the 25 that clustered with these had D > +0·5. The remaining 11 birds (8%) formed two clusters: eight buzzards (black diamonds) had either more than 20% of their locations at pens or D > +0·66 (the maximum for no-visit birds), and three buzzards (black squares) had at least 50% of their locations at pens. The 11 birds in these two clusters associated most strongly with pheasant-release pens.

Figure 8.

The percentage of locations within 200 m of pheasant-release pens for 136 juvenile buzzards radio-tracked during August–October 1990–95, and the strength of association with pens relative to the area covered by home ranges. Symbols are open for one cluster category, of birds not found near pens (thin circles) or seldom near pens (thicker circles), and filled (diamonds, squares) for two cluster categories that associated more strongly with pens.

The 36 buzzards that were recorded at pheasant pens fledged from nests on average 603 m (95% confidence level 515–706 m) from pens, closer than the average of 1096 m (952–1205 m) of 100 no-visit buzzards (t = 6·06, d.f. = 134, P < 0·001). The difference remained similar if the analysis was restricted to 33 buzzards with a total of at least 15 radio-locations (t = 2·85, d.f. = 31, P < 0·01), which included 14 no-visit birds (1148 m, 851–1549 m) and 19 pen-visitors (708 m, 598–838), and also if confined to 31 first-year birds with at least five locations that were tracked in 1994–95 (t = 3·09, d.f. = 29, P < 0·01). Therefore, buzzards that fledged close to pheasant pens were most likely to visit them. However, among pen-visitors, there was no tendency for birds that fledged closest to pens to visit them most frequently (r = 0·05, d.f. = 34, P > 0·5). Buzzard broods averaged less than two; halving the degrees of freedom in tests, to compensate maximally for possible lack of independence between brood-mates, did not substantially affect significance levels.

After their first year, an average of seven locations (spread of 3–17) was recorded for the buzzards during July–October. Their mean distance from pens increased from an average of 1061 m (95% confidence level, 943–1194 m) in the first year to 1124 m (865–1461 m) in subsequent years for all birds. The increase was significant for the 32 individuals tracked in more than 1 year (Wilcoxon signed-rank test, z = 2·15, P < 0·05). Therefore, buzzards that survived in the study area had a slightly reduced association with pens after their first year.

There was a total of 1–24 visits to each pen during 1990–95 from all the radio-tagged buzzards. The number of records at each pen provided an index of buzzard presence there. The only single factor that affected the number of buzzard visits at each pen was whether the canopy was deciduous or conifer (F = 5·44, d.f. = 1, 26, P < 0·05). The relationship was improved by adding the extent of canopy cover (F = 8·76, d.f. = 1, 25, P < 0·01). Buzzards were recorded most often near pens with an open canopy and where the canopy was deciduous (R = 0·62, d.f. = 25, P < 0·01). The regression was improved by adding the average number of pheasants released at pens in 1994 and 1995 (F = 9·42, d.f. = 1, 24, P < 0·01). Buzzard records were mainly near pens where few pheasants were released in 1994–95. The type and extent of canopy cover, and the number of released pheasants, explained 56% of the variation in presence of radio-tagged buzzards near pens (R = 0·75, d.f. = 24, P < 0·001). Presence of tagged buzzards could not be linked directly to predation records because only two pens had more than one buzzard visit during the reduced radio-tracking period, 1994–95. There were also too few data after the first autumn to test for differences between buzzard age classes in their association with pen habitats.


In Britain, populations of most large raptors were locally restricted or extinct at least until the 1970s, as a result of habitat loss, past persecution and contamination by organochlorine pesticides. Thus, predation pressures by raptors in Britain were artificially reduced (Newton 1979). Steps taken to restore raptor populations may enhance biodiversity, but may also increase predation pressures and hence the risk of conflict between different interest groups. Such conflicts can be minimized if predation on game and livestock can be assessed objectively, and effective remedies found when necessary (Kenward 1999). This is especially important for a species like the buzzard, whose rarity in eastern England has been attributed to past persecution (Moore 1957) and continued illegal poisoning (Elliot & Avery 1991), although habitat differences might also be involved (Tubbs 1974; Taylor, Hudson & Horne 1988).

It is also important that assessments of predation are not perceived to be biased. Results in this paper were based partly on nest records, habitat survey and radio-tracking, which were collected and used in ways that should minimize bias. However, data on released pheasants might have been biased by mistaken attribution of deaths, due either to predator sign being misread or to scavenging by a different predator. Some studies of prey remains have not been able to discriminate between large raptor species that pluck and eat prey in similar ways (Jenkins, Watson & Miller 1964; Thirgood et al. 1998), but buzzards were the only large raptor in our Dorset study area. Risk of mistakes was reduced by differences in kill signs between predators, by noting whether poults were killed by day or night, and by sightings of predators at kills. Foxes, which normally consume or cache prey rather than leave it for scavengers, scavenged 16% of pheasants killed by goshawks in Sweden (Kenward 1977). Therefore, some buzzard kills may have been attributed to foxes, or missed altogether. On the other hand, some kills by other raptors, and deaths without predation, may have been attributed in error to buzzards, which also scavenge.

If misattribution errors were large and specific to particular sites or gamekeepers, then negative correlations would have been expected between kills recorded for buzzards and for other species. Such correlations were weak, which suggests that results from analysis of differences between pens were not based on misattribution-bias. Moreover, the release-pen canopy characteristics that were associated with buzzard predation also related to independent data on buzzard presence from radio-tracking. Although most of the radio-tracking records were from the years before the predation study, these pen canopy characteristics would not have altered appreciably between 1990 and 1995.

The associations between predation, pen variables and buzzard presence are most easily understood if the mechanisms that encouraged buzzards to start hunting at a site (site attraction) differed from mechanisms that made it rewarding to kill pheasants there (prey vulnerability). For example, predation was initiated most often at pens with a deciduous canopy, and radio-tagged buzzards associated most with pens that had an open, deciduous, canopy. This was probably because hunting perches were less available in the mainly young conifers than in the mainly mature deciduous trees; the exception, a site in well-spaced mature conifers that offered an abundance of perches for buzzards, had severe predation. Observations of goshawk predation on pheasants in Denmark caused Mikkelsen (1984) to recommend that perch opportunities for raptors should be minimized near pens.

With buzzards most often present in an area that had suitable hunting perches, absence of shrub cover would have made prey vulnerable. Lloyd (1975) found that 20% of shrub cover and 50% of herb cover reduced predation in pheasant pens by smaller raptors. Although shrub cover might also have reduced detection of some kills by the gamekeepers, and thus caused the extent of predation to be underestimated, it would probably not have prevented detection that some predation occurred.

The presence of radio-tagged buzzards near pens was not related to shrub cover, but was greatest near pens where few pheasants were released in 1994–95. This conflicts with findings that predation was most likely to occur where most pheasants were released, and is hard to explain unless the gamekeepers deliberately released few pheasants in 1994–95 in pens that had suffered most from buzzard presence in 1990–93. The tendency for the proportion of pheasants killed to be greatest where few were released may partly reflect the inability of a territorial predator to kill a high proportion when many prey are present.

The number of pheasants killed was best explained by large pen area with much ground cover, which is understandable if the probability of a kill was highest where young pheasants were spread out and isolated by vegetation. Raptors suffer reduced success when attacking flocks of prey, due to increased confusion when there are many targets (Page & Whitacre 1975) or enhanced detection by the prey when there are many eyes (Kenward 1978). In these circumstances, cover less than 50 cm from the ground may have hindered pheasant vigilance or pheasant escape more than it impeded an attacking buzzard.

Although further work is needed to define the mechanisms involved, it is clear that pheasants were most at risk in large pens with much ground cover, few shrubs and much deciduous canopy. Prey risk can be seen as a product of predator presence and prey vulnerability. Vulnerability may then have a feedback effect on presence through learning. Thus, an immediate benefit of steps taken to minimize predation might be complemented in the long term through encouraging a population of older predators that ignore pheasant pens.

In fact, only 4·3% of the released pheasants were reported killed by buzzards, and only 21% of 55 releases lost more than two poults a week. Moreover, despite 44% of the buzzards having a pheasant pen within their home range, an appreciable association with pens was recorded for only 8%. Data from deaths of radio-tagged buzzards supported the view that only a small proportion visited pens persistently. The maximum mortality rate for radio-tagged buzzards in their first year was 34%, with 12 of their 38 deaths caused by shooting or poisoning (Kenward et al. 2000), in all but one case near pheasant pens. About 10% of the first-year buzzards were killed (illegally) near pheasant pens.

The low proportion of buzzards visiting pens and the low presence of pheasant remains at nests suggests that they were generally unrewarding prey for buzzards to attack, which could explain the weak relationship between severity of predation and the proximity of buzzard nests. Such a relationship could also have been weakened by destruction of nests near pens. However, deliberate interference was the probable cause of only one nest failure among 208 clutches laid during 1990–97 (Kenward et al. 2000).

The death of buzzards at pens would reduce the reliability of using kill-rates at buzzard nests to check the estimates of predation at pens. The estimated kill-rate for each nesting buzzard averaged 0·025 pheasants day−1. If this average rate was applied to the estimated 240 buzzards in the study area in autumn (Kenward et al. 2000), for the period of about 3 months in which released pheasants lived near pens, 550 pheasants would have been killed. An average of 450 kills was attributed to buzzards at pens each year. Although the estimates are comparable, nests were checked mainly in the month before poults were released. Compared with the kill-rate at nests, the kill-rate in August–September might have increased for vulnerable young poults in pens, or declined if adult buzzards had less need to catch large prey outside the breeding season.

Shoots do not generally expect to harvest more than 50% of released pheasants, although some estates recover up to 70% (Hill & Robertson 1986). The estimated loss of 4·3% to buzzards was therefore a small proportion of total losses. The number of pheasants available for shooting would have been reduced appreciably at only a minority of pens. Our results suggest that the occasional heavy losses might be avoided by encouraging shrubs rather than ground cover in pens, by siting pens where there are few perches for buzzards, and perhaps also by high-density releases.

The woodland habitats used for pens were likely to have contained many owls and sparrowhawks, so the assessment of their predation at six to seven times less than for buzzards suggests that the poults were even less vulnerable to these smaller predators. In contrast, released pheasants are more vulnerable to goshawks than to buzzards. At a Swedish estate where 4300 pheasants were released annually, pheasants were killed regularly by all 11 radio-tagged goshawks whose home-ranges overlapped release sites, compared with at most 11 of 56 buzzards (20%); the goshawks killed an estimated 19% of the pheasants at rates of 0·25–0·58 pheasants hawk-day−1 (Kenward 1977) compared with 4·3% for buzzards at rates of around 0·025 pheasants day−1. Even in areas with a lower density of wild pheasants, these were killed by 10 of 22 radio-tagged goshawks (45%), with average kill-rates of 0·07–0·11 pheasants hawk-day−1 (Kenward, Marcström & Karlbom 1981). Greater predation by goshawks than buzzards was not due to differences in attraction to pens or to prey vulnerability between study areas. Within a week of two radio-tagged goshawks being released in the study area and one nearby, two found pheasant pens and one found free-range poultry before being recovered; one hawk killed 17 pheasants in 22 days (N.J. Williams & C. Duglan, personal communication). Foraging areas are some 10 times larger for goshawks than for buzzards (Kenward, Marcström & Karlbom 1981; Walls & Kenward 1998), so goshawks are much more likely than buzzards to encounter pheasant pens.


We thank many Dorset landowners, tenants and gamekeepers for help with this project, especially W. Beaumont, R. Brooks, R. Green, W. Harrison, P. Lardner, B. Mead, G. Morgan, J. Selby-Bennett, R. Vickery and E. Wadham. The work was funded by the Natural Environment Research Council, with important contributions from Biotrack Ltd. The contribution by D. G. Hall was part of a MSc thesis on wildlife management and control at the University of Reading, where he was grateful for supervision and advice from R. Sibly, C. Prescott and C. Barahona. We are grateful to A. Amar, N. Webb, S. Redpath and an anonymous referee for comments that improved an early draft of the manuscript.

Received 14 March 2000; revision received 16 March 2001