Agri-environment schemes could play a key role in the reversal of farmland bird declines. The effectiveness of the Arable Stewardship Pilot Scheme for delivering farmland birds was tested in a replicated, farm-scale field trial, in two lowland farmland regions of England. Changes in numbers of birds over five years were compared between control and scheme farms. In East Anglia, productivity of Grey Partridges Perdix perdix was significantly higher on scheme than on control farms, although such an effect was not seen in the West Midlands where the species was rarer. At the field scale, wintering granivorous passerines and Skylarks Alauda arvensis responded to stewardship habitats (options) such as stubble and wild bird cover designed specifically to benefit them. However, at the farm scale, winter bird counts were higher on scheme than control farms only in the West Midlands, and not in East Anglia where the availability of set-aside and features managed for game in the wider landscape was higher. In the absence of other high-quality habitat, arable options are effective at providing good quality habitat, but the amount needed to exert an influence at the bird population level remains uncertain.
Declines in UK farmland birds are now well documented (e.g. Pain & Pienkowski 1997, Krebs et al. 1999, Gregory et al. 2003). Many farmland bird species have Biodiversity Action Plans (BAPs: Anon. 1995, 1998) and the farmland bird index, summarizing the smoothed population trends of 20 species, is used by the UK's Department for Environment, Food and Rural Affairs (Defra) as a ‘headline indicator’ of the sustainability of its policies and ‘quality of life’ in the UK (Anon. 1999). Against this background, in 2000, Defra adopted a Public Service Agreement (PSA) target to reverse the decline of farmland birds, as measured by the farmland bird index (Vickery et al. 2004).
Meeting farmland bird targets will require measures in the wider countryside beyond the designation of discrete nature conservation sites. At least in the short to medium term, management practices within agri-environment schemes (AESs) will probably be key tools (Evans et al. 2002, Vickery et al. 2004). Briefly, these schemes aim to compensate farmers for loss of income incurred by implementing measures to benefit the biodiversity or the wider environment. AESs are found throughout Europe, in both European Union (EU) and non-EU states, and their objectives are diverse: as well as biodiversity conservation, they include landscape enhancement, water quality protection, soil protection and protection of archaeological sites.
Within the EU, approximately 24.3 billion has been spent on AESs since 1994. A review by Kleijn and Sutherland (2003) found that 62 ecological evaluations of AESs had been undertaken, but that 76% of the studies were from The Netherlands and the UK, where only about 6% of the EU agri-environmental budget has been spent. Even when studies were undertaken, the research design was often inadequate to assess reliably the effectiveness of the schemes, and there are few published examples of AESs achieving biodiversity objectives (Kleijn et al. 2001). The lack of scientifically sound evaluation studies does not allow for a general judgement of the effectiveness of European AESs.
However, when targeted appropriately, ‘special’ AESs have delivered biodiversity gains on a local scale in the UK (Aebischer et al. 2000, Evans et al. 2002, Wilson et al. 2004). For example, environmentally sensitive area (ESA) agreements have helped to increase breeding wader numbers in some parts of the UK, although enhancement of habitat (higher tiers), especially on nature reserves, worked much better than habitat maintenance (lower tiers) in more general farmland (Ausden & Hirons 2002). The threatened Cirl Bunting Emberiza cirlus population has increased dramatically as a result of provision of weedy cereal stubbles and invertebrate-rich grassland via Countryside Stewardship Scheme (CSS) agreements (Peach et al. 2001).
It is likely that the biodiversity benefits of AESs will be realized only if the management prescriptions are tailored to the ecological needs of the target taxa, and studies that test for such benefits are sensitive to the likely spatial scale and temporal lag of the response of the target populations (Vickery et al. 2004). Evans et al. (2002) propose that such schemes need to be allowed to evolve through an iterative process of design, trial and revision. Flexibility, early and continued dialogue with all stakeholders, a foundation of sound science and comprehensive monitoring in the early stages of deployment of schemes are essential steps in their eventual success.
Evans et al. (2002) argue that the UK's CSS is an example of a ‘smart’ scheme that has evolved considerably since its introduction, to meet biodiversity objectives. This led, in 1998, to the deployment by the UK Ministry of Agriculture Fisheries and Food (MAFF) of the Arable Stewardship Pilot Scheme (ASPS) in two contrasting lowland English farmland areas: the West Midlands (mainly mixed arable and livestock systems) and East Anglia (predominantly arable). An array of options were made available (Table 1), based on the results of research and designed to meet the ecological requirements of farmland birds, especially seeds for wintering granivorous birds, invertebrates as chick food, and low or sparse vegetation for ground-nesting species such as Skylark Alauda arvensis and Lapwing Vanellus vanellus. The first agreements began in autumn 1998 and ran for five years.
Table 1. Options and suboptions available in the Arable Stewardship Pilot Scheme.
Limited herbicide use in cereal or linseed, followed by overwintered set-aside stubble
Overwintered cereal or linseed stubble, followed by spring/summer fallow
Suboption 1A + B
Limited herbicide use in cereal or linseed, followed by overwintered stubble and spring/summer fallow
Overwintered stubble followed by spring crop
Suboption 1A + C
Limited herbicide use in cereal or linseed, followed by overwintered set-aside stubble and a spring crop
Undersown spring cereal
Overwintered stubble, followed by undersown spring cereal
Undersown spring cereal, followed by grass ley
Suboption 2A + B
Overwintered stubble, followed by undersown spring cereal and grass ley
Crop margins with no summer insecticide
Conservation headlands with no fertilizer applications
Grass field margins by natural regeneration or sown grasses (4–12 m wide)
Uncropped wildlife strips (4–12 m wide)
Wildlife seed mixtures
Kleijn and Sutherland (2003) suggest that ecological evaluations should incorporate the collection of baseline data, random placement of scheme and ‘control’ sites in areas with similar initial conditions, and sufficient replication. Such an evaluation was instigated in the first two years of the ASPS. Although invertebrates and plants showed encouraging responses to options, vertebrates showed few responses (Anon. 2001, Bradbury & Allen 2003). Therefore, the Royal Society for the Protection of Birds (RSPB) and the Game Conservancy Trust (GCT) conducted a follow-up evaluation in 2002–3, at the end of the first 5-year agreements, to see if benefits had accrued over a longer time period. Here we present the results of work on Grey Partridge Perdix perdix (GCT) and wintering birds (RSPB).
Assessment of the effect of the scheme on birds was achieved by comparison of responses on farms in the scheme with those on control farms that had not entered the scheme. Details of sites are given elsewhere (Anon. 2001, Bradbury & Allen 2003). Stewardship agreements with farmers were drawn up in summer 1998, for implementation immediately post-harvest in 1998. The first Grey Partridge surveys were conducted post-harvest in 1998, prior to option implementation. However, the first broad bird surveys, necessarily being conducted during winter 1998/9, happened at a time after options had been implemented and when some, such as options 1 and 5, could already have influenced wintering birds. By surveying immediately post-harvest, GCT were able to estimate not just density but also productivity (see below) of Grey Partridge, the only PSA farmland bird species for which ageing of individuals by direct field observation (and therefore estimation of productivity) is reliable.
In autumn 1998 and 2002, 20 scheme sites and 19 control sites were surveyed in East Anglia and 20 scheme and 17 control sites in the West Midlands. In East Anglia, the mean (± sd) area surveyed was 169 ± 11.0 ha in 1998 and 161 ± 11.4 ha in 2002. In the West Midlands, the mean area surveyed was 126 ± 11.1 ha in 1998 and 148 ± 11.5 ha in 2002.
Each farm was surveyed during either a single early morning (2–3 h after first light) or late evening (2–3 h before dark). Surveys involved two people driving around all accessible fields. The vehicle was driven along the headland around all fields and, in large fields or in fields where the topography hindered visibility, it was driven along parallel transects up and down the field until the ground was completely covered. Fields were scanned initially with the naked eye and when a covey of partridges was seen, the vehicle was stopped and the observers used binoculars to count, age and sex the birds. For farms smaller than 200 ha, all suitable fields were surveyed, and on farms greater than 200 ha, as large an area as possible was surveyed within the time available. All fields that could be driven around were surveyed and grass fields containing livestock or fields that had been cultivated and drilled were scanned from a number of vantage points.
In the two winters 1998/99 and 2002/03, 18 scheme sites and 19 control sites were censused in East Anglia, and 19 scheme and 18 control sites in the West Midlands. Each site was censused twice each winter, once between 1 October and 31 December and once between 1 January and 31 March. In East Anglia, the mean (± sd) area surveyed in each year was 185 ± 5. 6 ha. In the West Midlands, the mean area surveyed in each year was 129 ± 6.6 ha.
Each site-census was completed in a single day. On each site visit, counts were made of all birds feeding in (or ‘hunting over’, in the case of raptors) each field. The observer walked along a predetermined number of transects across each field (each 50 m apart, aligned with the longest axis), in order to flush all birds in the field (Wilson et al. 1996, Perkins et al. 2000). This method achieves complete counts of birds except in rare cases where the vegetation is particularly tall and dense. In these rare cases, closer transects were used. Some species tend to use only the field edge, so the two outer transects were within 10 m of the field boundary (Perkins et al. 2000). Double-counting of birds was minimized by observers counting birds that were flushed to other fields or other parts of the field being censused, and ignoring them on subsequent encounter (Wilson et al. 1996, Perkins et al. 2000). Counts commenced at least 1 h after sunrise and were completed at least 1 h before sunset, to avoid missing birds because they were yet to arrive from, or had already left for, roost sites. Counts were not conducted in periods of heavy rain, strong wind or poor visibility and census routes were reversed between visits to minimize any effects of time of day on the presence and detectability of birds.
Grey Partridge densities
Two measures of density were analysed: numbers of adults gave an indication of breeding stocks, and total numbers of adults and young gave autumn densities. Variation in counts was assessed using log-linear mixed models, specifying a Poisson error structure, a logarithmic link and controlling for overdispersion. Farm identity was included as a random effect and the natural logarithm of study site area was specified as an offset, to account for the possibility of higher counts at larger sites. The analyses excluded farms with no Partridges recorded in either year and separate models were constructed for each ASPS pilot region. The model specified was;
ln(count) = constant + farm + year + farm type + (year × farm type)
where farm type was a two-level factor (control = 1, scheme = 2), as was year (1998 = 1, 2002 = 2). Testing the interaction term (year × farm type) assessed whether the change in numbers at a site between 1998 and 2002 varied between treatments. The analyses sought to test whether the management undertaken on agreement farms had increased abundance and productivity during the 5-year period between 1998 and 2002 relative to control farms. Because the change was expected to be positive, statistical tests of the corresponding null hypothesis were one-tailed (α = 0.05); for tests based on the Wald statistic with df = 1, the probability level was determined by treating its square root as a one-tailed z-test.
Grey Partridge productivity
The young-to-old ratio (Y:O) and the number of young per brood (mean brood size) provided two measures of productivity. Y:O ratios were log(x + 1)-transformed to achieve a normal distribution and then analysed using least-squares regression. The analysis was weighted by the number of adults forming the denominator in the ratio. To analyse variation in brood size, the number of young birds was specified as the response variable in a general log-linear regression, specifying ln(number of broods) as an offset. Because sample sizes were low, and because the difference between control and scheme farms was small in 1998, the analysis was based on all farms for which results could be calculated in any one year.
Winter birds farm-scale assessment
Owing to the rarity of many species, most were grouped into ecological or taxonomic guilds for analysis (Table 2). Analysis was based on mean count in each winter, and undertaken separately for each ASPS region. The models specified and assessment of significance were as for Grey Partridge. However, whereas the Grey Partridge analysis assessed only the significance of the year × farm type interaction term, if this term was not significant for winter bird guilds, it was dropped from the model and the effect of farm type assessed, to test whether a difference between treatments existed in year 1 and was maintained in year 5. This was done because of differences in the timing of counts relative to option deployment (Grey Partridge before, winter birds after option deployment) and would identify whether some options (e.g. wildlife seed mixtures and stubble) were mature enough to deliver for winter birds by the time of the first monitoring period.
Table 2. Composition of guilds used in the winter bird analyses. Numbers in parentheses are percentage of farms on which each species was recorded in year 1 and year 5.
Snipe Gallinago gallinago (26, 44)
Jack Snipe Lymnocryptes minimus (2, 0)
Woodcock Scolopax rusticola (5, 18)
Stock Dove Columba oenas (52, 71)
Woodpigeon Columba palumbus (99, 99)
Collared Dove Streptopelia decaocto (23, 32)
Wagtails and pipits
Meadow Pipit Anthus pratensis (89, 93)
Pied Wagtail Motacilla alba (69, 64)
Grey Wagtail Motacilla cinerea (7, 16)
Wren Troglodytes troglodytes (61, 99)
Dunnock Prunella modularis (87, 99)
Robin Erithacus rubecula (71, 100)
Blackbird Turdus merula (90, 99)
Fieldfare Turdus pilaris (64, 96)
Song Thrush Turdus philomelos (72, 96)
Redwing Turdus iliacus (47, 84)
Mistle Thrush Turdus viscivorus (58, 77)
Jay Garrulus glandarius (27, 37)
Magpie Pica pica (82, 86)
Jackdaw Corvus monedula (57, 67)
Rook Corvus frugilegus (75, 68)
Carrion Crow Corvus corone (97, 99)
Raven Corvus corax (5, 4)
Sparrowhawk Accipiter nisus (42, 53)
Buzzard Buteo buteo (47, 56)
Kestrel Falco tinnunculus (71, 75)
Merlin Falco columbarius (4, 1)
Peregrine Falco peregrinus (8, 4)
Skylark Alauda arvensis (99, 99)
Lapwing Vanellus vanellus (30, 30)
Golden Plover Pluvialis apricaria (17, 15)
Granivorous passerines (finches/buntings/sparrows). Note that Skylark is omitted owing to its more catholic diet (Wilson et al. 1996)
House Sparrow Passer domesticus (46, 49)
Tree Sparrow Passer montanus (32, 37)
Chaffinch Fringilla coelebs (91, 99)
Brambling Fringilla montifringilla (16, 5)
Greenfinch Carduelis chloris (68, 81)
Goldfinch Carduelis carduelis (60, 70)
Linnet Carduelis cannabina (58, 66)
Redpoll Carduelis flammea (2, 7)
Bullfinch Pyrrhula pyrrhula (18, 60)
Yellowhammer Emberiza citrinella (95, 100)
Reed Bunting Emberiza schoeniclus (59, 84)
Corn Bunting Miliaria calandra (27, 30)
Winter birds field-scale assessment
Data for winter 2002/03 were examined at the field scale, to test the effect of specific features contained within the field. Analysis was based on mean count of a species or guild in each field in winter 2002/03. Variation in mean count was assessed in a log-linear mixed model, specifying a Poisson error structure, a logarithmic link and controlling for overdispersion. Farm identity was included as a random effect, and the natural logarithm of field area was specified as an offset, to account for the possibility of higher counts in larger fields. Separate models were constructed for each ASPS pilot region. The model specified was;
ln(count) = constant + farm + farm type + field type + wildlife seed mixture + livestock + game feeder
with factors farm type (control = 1, scheme = 2), field type (stubble = 1, pastoral = 2, other arable = 3), wild bird cover (absent = 1, present = 2), livestock (absent = 1, present = 2) and game feeder (absent = 1, present = 2). All terms were first entered in a full model and a step-down model selection procedure used to obtain a minimum adequate model. The significance of exclusion of terms from the model was assessed by comparing the Wald statistic with the χ2 distribution (α = 0.05).
Autumn and adult densities
In East Anglia, autumn densities decreased significantly less between years on scheme farms than on control farms (z = 1.70, P < 0.05; Fig. 1a), although adult densities did not (z = 0.10, ns; Fig. 1b). In the West Midlands, there was no difference in between-year change between scheme and control farms of either autumn (z = −0.57, ns; Fig. 1a) or adult densities (z = −0.14, ns; Fig. 1b).
Productivity (Y:O ratio and brood size)
In East Anglia, the Y:O ratio in 1998 was slightly lower on scheme farms than on control farms (Fig. 2a). In 2002, it was four times higher on scheme farms than on control farms (t19 = 3.77, P < 0.001). The mean brood size in 1998 was slightly lower on scheme farms than on control farms (Fig. 2b), but in 2002 it was higher (t15 = 3.11, P < 0.01). In the West Midlands, the mean Y:O ratio could be calculated on 10 farms in 2002, and was similar on both types of farm (t8 = −0.40, ns; Fig. 2a). The difference in brood size was minor and not significant (t4 = 0.10, ns; Fig. 2b).
In East Anglia, the only effect was a negative effect of farm type on thrushes (z = −1.64, ns). In the West Midlands, for only Skylark was there a significant interaction between year and farm type, indicating that numbers increased more between years on scheme than on control farms (z = 1.73, P < 0.05; Fig. 3). However, across years, there were significantly greater numbers of four other guilds on scheme than on control farms. These included the key target group, granivorous passerines (z = 3.08, P < 0.01; Fig. 3), as well as wagtails/pipits (z = 1.97, P < 0.05; Fig. 3), miscellaneous insectivores (z = 1.79, P < 0.05; Fig. 3) and raptors (z = 2.42, P < 0.05; Fig. 3).
Results are presented in Table 3. Here, we focus on the two main target groups, granivorous passerines and Skylarks. For granivorous passerines, the key target group, there were strong responses at the option scale that were consistent between regions (Fig. 4). Densities were higher on stubbles than other field types and were high on fields with wildlife seed mixtures. In the West Midlands, densities were also high on fields with livestock and fields with game feeders, but this effect was not seen in East Anglia. For Skylarks, densities were again higher on stubbles than on other field types. Responses to other variables differed between regions: in East Anglia there was a negative response to feeders and a positive response to wildlife seed mixtures, whereas in the West Midlands there was a positive response to livestock presence.
Table 3. Field-scale winter bird results, giving mean percentage difference in density between each factor level and its baseline, controlling for the effects of all other factors. For stubble and pastoral fields, difference is relative to ‘other arable’ fields. For wildlife seed mixture, game feeder and stock, difference is relative to fields without those habitats. Figures in parentheses are approximate standard errors. Significant factors are highlighted in bold, with the significance value given adjacent to the mean value for stubble, in the case of field type ( * P < 0.05, ** P < 0.01, *** P < 0.001). Plovers were not recorded with sufficient frequency in East Anglia to enable analysis.
In East Anglia, where the higher densities of Grey Partridges were recorded, productivity increased much more between years on scheme than on control farms. However, there was no evidence of any difference in the rate of change in density of adult Grey Partridges between scheme and control farms, with numbers halving on both. The good productivity on scheme farms was reflected in a considerably smaller fall (around 30%) in overall autumn density from 1998 to 2002 than on control farms (fall of around 60%). According to the GCT's national Partridge Count Scheme, 2002 was a good year for Partridge reproduction in East Anglia, and followed a succession of mainly poor years since 1995 (Aebischer & Ewald 2004). If the benefits of ASPS options take effect mainly in years of good reproduction, then the lack of response for adults is perhaps not surprising. Aebischer and Ewald (2004) show how annual change in the UK Grey Partridge population is closely related to productivity, so by raising breeding success through the provision of nesting cover and insect-rich brood-rearing habitat, the ASPS has the potential to improve adult densities even though this has not yet been realized.
In the West Midlands, there was a trend to a greater decrease in densities between years on scheme than on control farms. However, the low density of Grey Partridges resulted in high variation between farms and measures of productivity and change in density were difficult if not impossible to measure precisely. It is therefore not possible to state with confidence whether the ASPS was less successful in the West Midlands.
At a farm scale, ASPS delivered enhanced numbers of a range of guilds in the mixed farming landscape of the West Midlands, but not in the arable landscape of East Anglia. This raises several questions. First, was the scheme deployed better in the West Midlands than East Anglia? The scheme's economic evaluation (http://statistics.defra.gov.uk/esg/evaluation/asps/default.asp) shows that farmers in the West Midlands were more likely to receive advice before submitting an application. Several studies have demonstrated that sound advice can improve the delivery of schemes (e.g. Kleijn et al. 2001, Ausden & Hirons 2002, Smallshire et al. 2004), so it is conceivable that the delivery potential of the scheme was enhanced in the West Midlands.
Secondly, is it possible that the West Midlands stewardship farms just happened to have higher winter bird densities than control farms anyway? It is impossible to tell for certain, because the first counts occurred after the first options had been deployed. However, we argue that this is unlikely to be the case. The greater resolution of the field-scale analysis shows that the observed effect at the farm scale was probably attributable to aggregative responses of birds to provision of specific ASPS options. These field-scale relationships were almost entirely as would be expected, for instance the response of granivorous passerines to stubbles (Wilson et al. 1996, Buckingham et al. 1999) and to wildlife seed mixtures (Brickle 1997, Boatman & Stoate 2002). Such options would be expected to have an immediate effect on winter birds, explaining the consistent effect across years one to five.
Thirdly, the field-scale analysis also revealed aggregative responses to options in East Anglia. So, why did this not translate to an observed effect at the farm scale? It is possible that, because farms are larger in East Anglia than in the West Midlands, the beneficial effect of options at the field scale was more diluted at the farm scale in East Anglia. Inspection of the proportion of the farms with each of three main winter food resources (stubble, wildlife seed mixtures and game feeders) suggests that, on ASPS farms, availability of resources was indeed greater in the West Midlands than in East Anglia (Table 4). However, in contrast, availability of feeders and wildlife seed mixtures on control farms was greater in East Anglia than in the West Midlands. It is therefore not clear whether the lack of a farm-scale difference between control and ASPS farms in East Anglia was due to greater dilution of options at the landscape scale than in the West Midlands, or to relative quality of control farms in the two regions. Indeed, for stubbles (the only habitat for which we have such data), comparison with availability in the wider countryside suggests that availability of this habitat on ASPS farms in East Anglia was actually no greater than in the wider countryside (Table 4).
Table 4. Comparison of delivery of ASPS options by scheme farms, control farms and farms in the wider countryside at the start of winter. In East Anglia, 73% of scheme farms had stubble options, 69% had wildlife seed mixture options and 50% had both options. In the West Midlands, 100% of scheme farms had stubble options, 82% had both stubble and wildlife seed mixtures. Data on stubble availability in the wider countryside, from winter 2000/01, were kindly provided by BTO (Gillings & Siriwardena 2003 in unpublished report to Defra).
Total area (ha)
Percentage area of farm = stubble
Percentage of fields with wildlife seed mixture
Percentage of fields with game feeder
c. 1.7 million ha
c. 0.9 million ha
So why were winter resources in East Anglia as common on other farms as on ASPS farms? First, the economic evaluation shows that control farms in East Anglia were as likely to be members of conservation organizations such as the Farming & Wildlife Advisory Group (FWAG) and GCT as ASPS farms, and so were likely, for example, to provide food resources for game. Secondly, as farmers were paid only for implementing new options, some who already provided good habitats may not have entered the scheme. New AESs will may resolve this problem of lack of payment for existing good practice. Furthermore, in addition to the lack of difference in option quantity between ASPS and control farms, there may be little difference in habitat quality. Only about 25% of ASPS stubbles came from low-input cereal options (see economic evaluation), the remaining stubble options probably being no different in quality to set-aside, yet low-input cereals are known to be of considerable added value (e.g. Evans 1997).
The ASPS seems to have been successful at delivering options that benefit key groups of wintering birds. However, although this could be detected as a farm-scale response in the West Midlands, this was not the case in East Anglia. A major reason for the lack of farm-scale effect in the arable landscape of East Anglia is likely to be set-aside policy, which could enable control farms to deploy stubbles and game cover over 10% of the farm area. However, the EU have already cut set-aside to 5%, and this may be further reduced in future. Under such a scenario, AES options may be more valuable as delivery mechanisms. There is evidence that Grey Partridge benefit from provision of breeding resources by ASPS, at least in East Anglia. This is yet to be assessed for other species, and the implied increase in subsequent breeding density has yet to be demonstrated on the ground.
Changes in EU policy, such as the decoupling of income support from production, will raise the economic viability of AESs for farmers. The range of options being considered for the new Entry-Level (ELS) and Higher Tier Schemes could deliver many of the resources needed by a range of species (Vickery et al. 2004). However, the success of future AESs will depend crucially on the balance between quantity and quality of resource provided (Vickery et al. 2004). For example, stubbles could well be included in the ELS, in which case they could be taken up by over 50% of farmers. However, stubble quality would be likely to be relatively low without low previous herbicide inputs. As long as sufficient coverage of high-quality options are incorporated in the ELS, we anticipate that this may play a lead role in helping deliver both BAP targets for individual species and the overall PSA target. Further research will be crucial to determine the optimum amount, and spatial configuration, of options needed by both breeding and wintering birds.
Thanks to Defra and English Nature for funding this work and to all the Defra staff, too numerous to mention here, who have supported the work throughout. Many thanks to Simon Cockayne, James Darke, Diana de Palacio, Jim Dustow, Matthew Edwards, Laura Hurt, Chris Keeling, Will Kirby, Andrew Mayo, Charles Morrison, Barry O'Dowd, Richard Penson, Stuart Priestley, David Wright and Rosa Zuniga for help with fieldwork. Will Peach gave statistical advice. Andy Evans, Phil Grice, Sue Armstrong-Brown and three anonymous referees gave helpful comments on a draft of the paper.