Habitat associations of European hares Lepus europaeus in England and Wales: implications for farmland management


Nancy Vaughan, School of Biological Sciences, University of Bristol, Woodland Road, Bristol BS8 1UG, UK (fax +44 117 9257374; e-mail nancy.vaughan@bristol.ac.uk).


  • 1Numbers of European hares Lepus europaeus have declined throughout Europe as a result of agricultural intensification. Ecological research to inform agricultural management policy is needed. We aimed to identify agricultural land management practices that may benefit this species, which is of conservation concern and of value as a game animal.
  • 2A postal survey of farmers was used to investigate relationships between the abundance of hares on farmland and current land management, the abundance of a possible competitor (rabbit Oryctolagus cuniculus) and the abundance of two predators (buzzard Buteo buteo and fox Vulpes vulpes).
  • 3Questionnaires were sent to 3000 farms in England and Wales selected from the annual agricultural census database held by the UK government's Department for Environment, Food and Rural Affairs and the National Assembly for Wales. One-thousand and fifty farmers (35%) responded. Non-respondents were statistically similar to respondents, so the questionnaire data were considered representative of the opinions of farmers in England and Wales.
  • 4Hares were relatively common on arable farms, especially on those with wheat Triticum aestivum, beet Beta vulgaris or fallow land. They were less common on pastural farms, where the likelihood of seeing hares was increased if improved grass, woodland or, in some cases, arable land was present. The association of relatively frequent sightings of hares with arable land was consistent at four spatial scales (farm, parish group, county and region).
  • 5Hares were seen rarely where foxes were seen frequently. Hares were generally only hunted on farms where they were common. Hence, records of numbers of hares shot may be used as indices of hare abundance but only in areas where hares are common.
  • 6Forty-two per cent of farmers believed that hares were declining. Hare numbers were most likely to be increasing on arable farms.
  • 7Synthesis and applications. Changes in land management that provide year-round cover and forage may make farms more attractive to hares. To benefit hares, pastural farms should have some woodland, improved grass and arable crops; arable farms should have wheat, beet and fallow land.


The European hare Lepus europaeus Pallas, 1778 is a game animal that has been introduced to many parts of the world (Mitchell-Jones et al. 1999). Throughout its geographical range in Europe, the hare is most common in intensively farmed arable areas and less common in non-arable areas (e.g. pasture, uplands and woodland; Tapper & Parsons 1984; McLaren, Hutchings & Harris 1997; Klansek et al. 1998). However, hare populations may decline if agriculture becomes too intensive (Schröpfer & Nyenhuis 1982; Tapper & Barnes 1986; Slamečka 1991; Panek & Kamieniarz 1999). Records of numbers of hares shot (hunting records) suggest a decline throughout Europe (Pielowski & Pucek 1976; Tapper 1992; Mitchell-Jones et al. 1999), which is largely blamed on agricultural intensification (Tapper & Barnes 1986; Slamečka 1991).

As a result of the decline, the hare is protected under Appendix III of the Convention of the Conservation of European Wildlife and Natural Habitats (Bern Convention; Mitchell-Jones et al. 1999). It is classed as a ‘priority species of conservation concern’ by the UK government, and therefore has a Biodiversity Action Plan (BAP; Anonymous 1995). In the BAP for the hare, factors causing the species’ decline are given as: conversion of grass to arable, loss of habitat diversity in the agricultural landscape, and changes in planting and cropping regimes, such as a move from hay to silage and autumn planting of cereals. The population estimate for the hare is given as 818 500–1 250 000, and the BAP target is to double spring numbers of hares in Britain by 2010. The aim of this study, which was funded as a result of the BAP, was to establish how agricultural land management might benefit hares.

Although the ultimate cause of declines in numbers of hares is believed to be habitat change through agricultural intensification, it is unclear what the proximate cause is. Populations of hares may be limited by low quality or scarcity of resources (Frylestam 1980; Hackländer, Tataruch & Ruf 2002) or by predation by the red fox Vulpes vulpes (Linnaeus, 1758; Lindström et al. 1994). We quantified associations between the abundance of hares and current farm management practices at four spatial scales (farm, parish group, county, region). In addition, we related the abundance of hares to the abundance of a potential competitor [the rabbit Oryctolagus cuniculus (Linnaeus, 1758); Chapuis 1990] and two predators, both of which are increasing throughout Europe [the buzzard Buteo buteo (Linnaeus, 1758); Snow & Perrins 1998; and the fox; Lindström et al. 1994; Reynolds & Tapper 1995; Mitchell-Jones et al. 1999]. We also investigated hunting of hares and perceived changes in abundance of hares since 1980. We conducted a large-scale postal survey of farmers, and we analysed data from this source alongside data from two large independent land-use databases.


data from the june census

We derived data on agricultural land use from the annual agricultural census database for 1999 (‘June census’) held by the UK government's Department for Environment, Food and Rural Affairs (DEFRA) and the Geographical Information (GI) Services Branch of the National Assembly for Wales.

We pooled the June census data for parish groups, counties and regions. In England there were 1263 parish groups (mean 10 526 ha of agricultural land, range 0·9–77 804 ha). On average each parish group contained 116 farms. In Wales, the ‘small area’ was comparable: there were 235, each containing 100–200 farms. ‘Parish group’ is used here to refer to both parish groups and small areas. There were 87 English counties or unitary authorities (mean 2128 farms per county, range 42–11 765). We used the eight English Government Office regions (excluding London, which only had 400 farms; Fig. 1; mean 18 352 farms per region, range 5280–36 813). Wales contained 28 016 farms.

Figure 1.

The regions used and locations of 1018 respondents’ farms. Responses to the question ‘how often do you see hares on your farm?’ are shown: circles = at most ‘most months’ (n = 579); squares = at least ‘most weeks’ (n = 395); triangles = question not answered (n = 44). Only farms for which postcodes were provided are shown.

For the June census, farmers provide information on areas of crops and grass on their farms and types of livestock kept. Farms are placed in categories (based on European Community farm types), which are called ‘robust farm types’ in Wales and ‘dominant farm types’ in England and in this study. Participation in the June census is obligatory for farmers, so the responses represent all land farmed in England and Wales except for common land and some minor holdings [small ‘hobby’ farms (< 6 ha) with no regular full-time worker].

We derived six general variables from the June census data (for names and explanations of all 25 general variables see Table 1). The nature of the farmland in each parish group, county and region (farm p, farm c and farm r) was derived from the total area of land in agricultural use and areas used for crops and set-aside, and grass. The average stocking density for each parish group, county and region (stock p, stock c and stock r) was calculated as the total number of livestock units (LSU) for cattle Bos taurus and sheep Ovis aries divided by the total area of grass (ha). LSU was calculated from total numbers of cattle and sheep by using the average LSU for each category (age and reproductive state; Nix 2000) weighted according to the number in each category kept in England and Wales. This was done so that stocking densities were comparable with those derived from the questionnaire, in which information about category of livestock was not requested.

Table 1.  The 25 independent and two dependent general variables derived from the questionnaire, the land class database and the June census. The 12 retained for the development of the multiple model are shown (†). O = origin of data: q = questionnaire, l = land class database, c = June census, r = RDS statistics; T = type of variable: or = ordinal, ca = categorical, co = continuous. For continuous variables, means (m) or medians (M) and range (minimum–maximum) for untransformed data and transformations used are shown. The simple likelihood-ratio test is for the difference in deviance (Δddf; d.f. = degrees of freedom) between the simple model and the null model, from the ordinal logistic regression with dependent variable see hare. **P < 0·05, ***P < 0·01, ****P < 0·001, NS = not significant (α = 0·1). For categorical and ordinal variables, the level associated with the most frequent sightings of hares is shown in bold. For continuous variables, directions of significant relationships are shown (positive = hares seen more frequently as value increases)
VariableOTVariable descriptionVariable levelsLikelihood-ratio test
see hareqorFrequency of seeing haresNever/very rarely, most months, most weeks, daily/most daysDependent variable
replyqcaReply timeNormal =≤ 2 months, late = > 2 monthsDependent variable
farm fqcaHabitat in farmA = > 50% arable, P = > 50% pasturalΔd1 =  150****
farm pccaFarmland in parish groupA = > 50% arable, P = > 50% pasturalΔd1 = 106****
farm cccaFarmland in countyA = > 50% arable, P = > 50% pasturalΔd1 =  65·2****
farm rccaFarmland in regionA = > 50% arable, P = > 50% pastural, O = < 50% arable and pasturalΔd2 = 75·8****
land flcaLandscape type of 100-ha square containing farmA = arable, P = pastural, U = marginal upland/uplandΔd2 = 33·7****
land clcaLandscape type in countyA= > 50% arable, P = > 50 % pastural, O = < 50% arable, pastural, and marginal upland/uplandΔd2 = 69·1****
land rlcaLandscape type in regionA = > 50% arable, P = > 50% pastural, O = < 50% arable, pastural, and marginal upland/uplandΔd2 = 30·1****
agree fqcaAgri-environment schemeNo, yesΔd1 = 23·4****
agree crco% of agricultural land in county in agri- environment schemeM = 9·5(0·69–35). Transformed to: log(x + 1). Relationship = negativeΔd1 = 23·0****
stock fqcoStocking density (LSU ha−1)M = 1·0(0·0–8·5). Transformed to: log(x + 1)Δd1 = 0·0240 ns
stock pccoStocking density in parish group (LSU ha−1)m = 1·4(0·35–2·6). UntransformedΔd1 = 2·62 ns
stock cccoStocking density in county (LSU ha−1)M = 1·3(0·89–1·8). Transformed to: log(x + 1)Δd1 = 1·43 ns
stock rccoStocking density in region (LSU ha−1)m = 1·3(0·97–1·6). Untransformed. Relationship = negativeΔd1 = 6·24**
main stockqcaMain type of stock in 2000c = > 50% of LSU cattle, s = sheep, h = horses, n = no stockΔd3 = 23·5****
fieldqcoAverage field size (ha)M = 3·5(0·33–35). Transformed to: log(x + 1). Relationship = positiveΔd1 = 288****
boundaryqcaMost common boundaryHedge, otherΔd1 = 6·31**
densityqcoNumber of habitats 100 ha−1M = 5·6(0·030–67). Transformed to: log(x + 1). Relationship = negativeΔd1 = 139****
diversityqcoShannon–Wiener habitat diversity indexM = 0·39(0·00–2·1). Transformed to: arcsine [square root(x/100)]. Relationship = positiveΔd1 = 115****
see rabbitqorFrequency of seeing rabbitsMost months or less often, most weeks, daily/most daysΔd2 = 13·8***
see foxqorFrequency of seeing foxesNever/very rarely, mostmonths, most weeks, daily/most daysΔd3 = 22·5****
see buzzardqorFrequency of seeing buzzardsMost months or less often, most weeks, daily/most daysΔd1 = 17·8****
gameqcaGamekeeperNo, yesΔd2 = 113****
hunt hareqcaHunting haresNo, yesΔd1 = 201****
hunt rabbitqcaHunting rabbitsNo, yesΔd1 = 34·1****
hunt foxqcaHunting foxesNo, yesΔd1 = 60·7****

DEFRA's Rural Development Service (RDS) and the GI Services Branch provided data on the area of agricultural land within agri-environment schemes [in which farmers are paid to manage their land for conservation: Environmentally Sensitive Areas (ESAs; England and Wales), Countryside Stewardship (CSS; England), Tir Cymen and Tir Gofal (Wales)]. We derived the variable agree c from these data (Table 1).

data from the land class database

Habitat variables at the level of 1 km2 of the National Grid for the UK (100-ha square; land f), the county (land c) and the region (land r) were derived from the land class database held by the Centre for Ecology and Hydrology (CEH; Bunce et al. 1996). Postcodes provided by questionnaire respondents were converted into grid references (Proaddress software, Global Mapping Solutions, Reading, UK). The 100-ha square represented by each grid reference was allocated to a Division Four Aggregation of CEH's land classification system (land f). These are referred to here as ‘landscape types’ and comprise arable, pastural, marginal upland and upland (Bunce et al. 1996). Counties and regions into which each farm fell were also allocated to a landscape type based on the percentage of 100-ha squares of each landscape type in each county or region (land c and land r; Table 1). County definitions differed between the June census and land class databases, and each farm was allocated to each list of counties so that it could be included at each level of analysis.

data from the questionnaire

Three-thousand farms (1·7% of all farms in England and Wales) were selected from the June census database, stratified by county and weighted in proportion to the total number of farms in each county. Minor holdings were excluded. Questionnaires were sent to farms in 48 English counties or unitary authorities as defined for the purposes of the June census (mean 2970 farms per county, range 166–11 765). The remaining 39 English counties or unitary authorities were not represented in the questionnaire and contained small numbers of farms (mean 206, range 42–568). Questionnaires with pre-addressed and prepaid return envelopes were sent to the selected farms in July 2000. Farmers were asked to complete the questionnaire even if they had no hares on their farm. They could complete it anonymously but were asked to give at least their county and postcode. Farmers were asked how often they saw hares, rabbits, buzzards and foxes on their farm; the size of their farm; the number of fields and type of boundary; how many and what type of livestock were kept; which crops were grown; whether any agri-environment schemes were in place in 2000; whether there was a gamekeeper on the farm; whether hares, rabbits and foxes were hunted; how many hares were hunted each year; whether hares had been caught or released (for restocking purposes); how numbers of hares had changed since 1980 and since 1995; whether the main business of their farm had changed since 1980 and, if so, how.

Variables derived from the answers given fell into four groups: two dependent variables, 15 general independent variables (Table 1), 15 independent variables specific to habitats (Table 2), four independent variables specific to hunting (Table 3) and four independent variables specific to change (Table 4). Hunting of hares, rabbits and foxes is defined as deliberate killing by humans (including hunting with dogs, shooting, poaching, gassing and trapping).

Table 2.  Variables specific to habitats, analysed separately for farm f=‘arable’ and ‘pastural’. Analysis not carried out if < 10% of farmers answered ‘yes’ (†) or ‘no’ (‡). Simple likelihood-ratio tests are for dependent variable see hare (Table 1); Δddf is shown with *P < 0·1, **P < 0·05, ***P < 0·01, ****P < 0·001, NS = not significant (α = 0·1). All variables are categorical; for those with significant likelihood-ratio tests, the level associated with the most frequent sightings of hares is shown in bold
VariableVariable descriptionLikelihood-ratio test
farm f = ‘arable’farm f = ‘pastural’
wheatWheat (no, yes)Δd1 = 23·7****
barleyBarley (no, yes)Δd1 = 7·94***
cerealOther cereals (no, yes)Δd1 = 0·532 NS
springIs any cereal sown in spring? (no, yes)Δd1 = 4·98**
maizeMaize (no, yes)Δd1 = 0·850 NS
rapeOilseed rape Brassica napus (no, yes)Δd1 = 7·74***
legumePeas/beans/clover Trifolium spp. (no, yes)Δd1 = 11·0***
linseedFlax Linum usitatissimum (no, yes)Δd1 = 1·67 NS
horticultureHorticultural crops (no, yes)Δd1 = 1·02 NS
beetBeet (no, yes)Δd1 = 15·6****
arableArable crops (see above; no, yes)Δd1 = 8·07***
grassGrass (including ley; no, yes)Δd1 = 0·508 NS
type grassMain type of grass (ley, improved, semi-improved, unimproved)Δd3 = 2·01 NSΔd3 = 21·6****
fallowSet-aside/fallow (no, yes)Δd1 = 16·6****
woodsWoodland/orchard (no, yes)Δd1 = 3·21*Δd1 = 4·44**
Table 3.  Variables specific to hunting, analysed only on farms where hares were hunted. Simple likelihood-ratio tests are for dependent variable see hare (Table 1). For number the median (M) and range (minimum–maximum) for untransformed data and the transformation used are shown. number has a positive relationship with see hare. Δddf is shown with ****P < 0·001, NS = not significant (α = 0·1). For shoot, the level associated with the most frequent sightings of hares is shown in bold
VariableVariable descriptionLikelihood-ratio test hunt hare=‘yes’
shootShooting (no, yes)Δd1 = 19·9****
poachPoaching (no, yes)Δd1 = 1·10 NS
beagleHunting with scent hounds (beagles/harriers; no, yes)Δd1 = 1·42 NS
numberNumber of hares killed each year. M = 5·75(0–200). Transformed to: log(x + 1)Δd1 = 42·6****
Table 4.  Simple likelihood-ratio test results of variables specific to change on dependent variable change hare. Δddf is shown with ***P < 0·01, ****P < 0·001, NS = not significant (α = 0·1). All variables are categorical; for those with significant likelihood-ratio tests, the level associated with increasing numbers of hares is shown in bold
VariableVariable descriptionLikelihood-ratio test
change hareChange in numbers of hares 1980–2000 [large decline (26%), decline (16%), stable (36%), increase (22%)]Dependent variable
categoryClassification of farm by farmer (other, sheep, cattle, crops)Δd3 = 25·9****
previousPrevious classification (other, sheep, cattle, crops)Δd3 = 12·7***
change fChange in main business of farm (no, yes)Δd1 = 0·0230 NS

robustness of data

We compared respondents and non-respondents directly by extracting pooled data from the June census for both groups. We also compared these census data to the data for all farms in England and Wales (the intended sample population) and to the data for all selected farms. The results from non-respondents were also approximated: replies from ‘late respondents’ (see variable reply in Table 1), which are statistically similar to non-respondents’ replies (Hébert et al. 1996), were compared with those from normal respondents. This was the only way to compare variables relating to wildlife and hunting that were only available from the questionnaire. Formal missing value analysis (an overall test of randomness) was conducted as an additional check for bias (Hair et al. 1998). Levels of variables with small sample sizes (< 10% of respondents) were pooled (Hosmer & Lemeshow 2000).

statistical analysis

Ordinal logistic regression (logit link function) was used for the analysis of general variables in relation to the variable see hare (Table 1). Continuous variables were transformed to normality and checked for homogeneity of variances across levels of the dependent variable. Each of the variables in turn was submitted to simple ordinal logistic regression analysis with the dependent variable see hare. The deviance (−2 log likelihood) of the null model was subtracted from the deviance of each simple model, and the difference in deviance (Δd) was tested for significance by means of a χ2 distribution table (likelihood-ratio test). Variables with significant likelihood-ratio test results were selected as candidates for inclusion in the multiple model (α = 0·1; Hosmer & Lemeshow 2000).

Multicollinearity and association among the candidate variables were assessed by means of Pearson correlations, Spearman rank correlations and χ2 analysis. Only independent variables were retained. The variable with the lowest P-value in the simple likelihood-ratio test was chosen for inclusion when several variables were associated with one another.

The multiple logistic regression model was developed manually by the backward stepwise method. Variables were omitted one at a time from the full model, and the likelihood-ratio test was used to assess improvement in the fit of the model. The Δd for the omission of each variable was tested for significance by means of a χ2 distribution table. The effect of the inclusion of each possible two-way interaction term in the final model was also evaluated by means of the likelihood-ratio test. The significance of the final model was tested against the null model by means of the likelihood-ratio test, and its overall predictive value was quantified as the percentage of pairs that were concordant (a measure of correct classification; Hosmer & Lemeshow 2000).

The same method was used for the analysis of variables specific to habitats (Table 2), hunting (Table 3) and change (Table 4), as well as for the comparison of respondents and non-respondents. All statistical analyses were carried out on SPSS software (version 10; Kinnear & Gray 2000) with α= 0·05 unless stated otherwise.


return rates and robustness of data

Overall, 1050 completed questionnaires were returned (35%); 1023 within 2 months of distribution and a further 27 3–8 months after distribution by ‘late respondents’. In total, c. 120 000 ha (1·1%) of farmland in England and Wales was farmed by the respondents. Questionnaires were returned by farmers at 0·6% of all farms in England and Wales (Table 5). Return rates were not biased geographically: in all eight English regions and in Wales similar percentages of questionnaires were returned (inline image = 7·65, NS; median 35%, range 27–38%).

Table 5.  Characteristics of farms in England and Wales, selected farms, non-respondents’ farms, and respondents’ farms. All data are from the June census and all farms are included in the calculations. Means are given ± standard errors
CharacteristicEngland and Wales 1986 n = 184 882England and Wales 1999 n = 175 238Selected farms n = 3000Non-respondents’ farms n= 2024Respondents’ farms n = 976
Mean area of farm (ha) 59 ± 0·2 61 ± 0·3 66 ± 2·1 63 ± 2·4 74 ± 3·8
% arable 27 ± 0·1% 22 ± 0·1% 23 ± 0·6% 22 ± 0·7% 24 ± 1·1%
% pasture 69 ± 0·1% 70 ± 0·1% 69 ± 0·7% 70 ± 0·8% 67 ± 1·2%
% of pasture that is rough grazing 11 ± 0·1% 10 ± 0·1% 10 ± 0·5% 11 ± 0·6%  8 ± 0·8%
% of total arable area used for wheat 23 ± 0·1% 28 ± 0·1% 28 ± 0·9% 27 ± 1·1% 30 ± 1·5%
Number of cattle per farm 48 ± 0·2 43 ± 0·2 48 ± 1·6 47 ± 1·9 51 ± 2·7
Number of sheep per farm144 ± 0·9181 ± 1·2181 ± 9·3185 ± 12·2173 ± 13·4
Stocking density (cattle and sheep on grass; LSU ha−1) 1·6 ± 0·0 1·9 ± 0·1 1·4 ± 0·1 1·5 ± 0·1 1·3 ± 0·1

From the list of 3000 selected farms and the questionnaire returns, 976 of the 1050 respondents could be identified in the June census database. The remaining 74 respondents did not provide enough information for a match to be achieved between the two databases, and so had to be pooled with the 1950 non-respondents. Respondents’ and non-respondents’ farms were similar to the selected farms and to all farms in England and Wales (Table 5 and Fig. 2).

Figure 2.

Percentages of farms in five size categories, in the whole of England and Wales (from the June census, n= 147 088 farms), the selected farms (n = 3000), the respondents’ farms (n = 976), and the non-respondents’ farms (n = 2024).

Data from the June census on respondents and non-respondents’ farms were compared in more detail using binary logistic regression with the dependent variable respond (‘respondent’ or ‘non-respondent’) and independent variables dominant farm type (type), farm size (size; Fig. 2), percentage of total arable area used for wheat (%wheat), percentage of total grassland area used for rough grazing (%rough) and stocking rate (for cattle and sheep on grass; stock). The last three variables were included as measures of farming intensity. Only type, size and %rough were significant at the simple level. In the multiple model only type remained significant, and only four of the nine types were significantly different from the reference level type=‘cereal’. These were: ‘horticulture’[coefficient ± SE =−0·72 ± 0·30, Z=−2·4, P < 0·05, odds ratio (95% confidence interval, CI) = 0·48(0·27–0·87)], ‘pigs and poultry’[coefficient ± SE = −0·81 ± 0·32, Z=−2·5, P < 0·05, odds ratio (95% CI) = 0·45(0·24–0·83)], ‘cattle and sheep in less favoured areas’[coefficient ± SE =−0·41 ± 0·17, Z=−2·4, P < 0·05, odds ratio (95% CI) = 0·66(0·48–0·92)] and ‘other’[coefficient ± SE = −0·37 ± 0·17, Z=−2·2, P < 0·05, odds ratio (95% CI) = 0·69 (0·49–0·97); coefficient ± SE for constant: −0·21 ± 0·20; the model was significantly better than the null model and the predictive value was acceptable; Δd10 = 33·8, P < 0·001; 56% of pairs concordant]. The intensity of farming was similar between respondents’ and non-respondents’ farms, although a slight bias existed in type: farmers with farms of the dominant farm types listed above were slightly less likely to respond than the overall average. ‘Horticultural’ farms, which comprised 5% of all farms in England and Wales, were represented by 3% of respondents; ‘pigs and poultry’ farms comprised 3% of farms in England and Wales and 2% of respondents’ farms; ‘cattle and sheep’ farms in less favoured areas comprised 14% of farms in England and Wales and 12% of respondents’ farms; ‘other’ farms (including specialist goat Capra hircus, horse Equus caballus and grass or turf farms) comprised 22% of farms in England and Wales and 18% of respondents’ farms.

Replies of normal respondents (Table 1) were similar to those of ‘late respondents’ (none of 26 simple binary logistic regression models produced a significant likelihood-ratio test; α= 0·1; in only two models was P < 0·2). Models had the binary dependent variable reply and independent variables, each of the general variables in Table 1 (including see hare).

Missing value analysis of all general variables showed that data were missing completely at random (Little's test:inline image = 38·0, NS). Therefore, cases or variables with missing data could be omitted from analyses without bias (Hair et al. 1998).

In summary, the questionnaire data were considered to be robust and representative of the opinions of all farmers in England and Wales.

general variables

Of the 1050 respondents, 477 (45%) saw hares ‘never or very rarely’, 120 (11%) reported seeing hares ‘most months’, 179 (17%) ‘most weeks’ and 226 (22%) ‘daily or most days’. The remainder did not reply to the question.

Simple likelihood-ratio tests showed that three variables relating to stocking density (stock f, stock p and stock c) had no effect on see hare, and could therefore be excluded from further analysis. Tests for collinearity revealed that the five continuous variables were not strongly correlated (Pearson correlation coefficients for all combinations were < 0·6). However, associations were found between the 15 categorical variables. The seven variables relating to habitat and landscape type (farm f, farm p, farm c, farm r, land f, land c and land r) were all associated and were related in the same way to see hare. Hares were seen more frequently in arable habitats or landscape types than in pastural habitats or landscape types. In mixed habitats and marginal upland or upland landscape types, hares were seen at intermediate frequency. Therefore, of these seven variables, the one with the strongest association with see hare (i.e. the lowest simple P-value; in this case farm f) was selected for inclusion in the multiple model. main stock was associated with farm f and was therefore not selected. Among associated variables relating to hunting (game, hunt hare, hunt rabbit and hunt fox), hunt hare had the lowest P-value and was therefore selected. No further associations were found. Development of the multiple model was begun with 12 variables (Table 1). Likelihood-ratio tests showed that the full model fitted the data best and that the inclusion of interaction terms did not improve model fit (Table 6).

Table 6.  The multiple ordinal logistic regression of general variables (Table 1) on dependent variable see hare. The percentage of farms within each independent variable level on which hares were seen at least ‘most weeks’ is shown; on the remainder, hares were seen at most ‘most months’[for example, 29% of farmers who had mainly pastural farms (farm f= pastural) saw hares at least ‘most weeks’, the remaining 71% saw hares at most ‘most months’]. Overall, hares were seen at least ‘most weeks’ on 39% of farms (n = 1002). Other non-significant factors in the model were: agree f, agree c, stock r, boundary, diversity, see rabbit and see buzzard. Overall, the model is significantly better than the null model (Δd16 = 296, P < 0·001; n= 709), and the predictive value is good (78% of pairs are concordant). Odds ratios are the odds of a one-step increase in score for the dependent variable (see hare, e.g. from never/very rarely to most months, or from most months to most weeks) for each level of categorical independent variables, or for an increase in continuous independent variables of one unit; CI = confidence intervals
Independent variableLevel% with see hare=‘most weeks’ or ‘daily/most days’Coefficient ± SEZ, POdds ratio (95% CI)
farm fPastural29Reference level 1·0
Arable72 0·53 ± 0·26 2·1, < 0·051·7(1·0–2·8)
field  1·8 ± 0·52 3·4, < 0·016·1(2·2–17)
density −1·1 ± 0·31−3·7, < 0·0010·32(0·18–0·59)
see foxNever/very rarely40Reference level 1·0
Most months45 0·033 ± 0·21 0·15, NS1·0(0·68–1·6)
Most weeks43−0·24 ± 0·22−1·1, NS0·79(0·51–1·2)
Daily/most days23−0·95 ± 0·31−3·1, < 0·010·39(0·21–0·71)
hunt hareNo30Reference level 1·0
Yes75  1·1 ± 0·19 5·4, < 0·0012·8(2·0–4·2)
Constant 1   −1·4 ± 0·98  
Constant 2  −0·18 ± 0·98  
Constant 3   0·56 ± 0·98  

Hares were seen most frequently on mainly or completely arable farms with large field sizes and few different habitats 100 ha−1. They were seen less frequently on farms where foxes were seen ‘daily or most days’ than where foxes were seen less often, and more frequently on farms where hares were hunted than where they were not (Table 6). In the final model, we found no relationships between the abundance of hares with that of rabbits or buzzards. Foxes were seen ‘daily or most days’ on 13% of farms; participation in fox control on these farms (56%) was similar to the average for all respondents (58%).

variables specific to habitats

Separate models were developed for farm f‘arable’ and ‘pastural’. Within each model, habitat variables were weakly or not associated with one another, so all those with significant simple likelihood-ratio test results (Table 2) were included in the full models. The model for farm f=‘pastural’ was improved by including the interaction term type grass×arable.

Arable farmers saw hares more frequently if they had wheat Triticum aestivum, beet Beta vulgaris or fallow land (including set-aside) on their farms than if they did not. Pastural farmers saw hares more frequently if they had woodland or mainly improved grass on their farms than if they did not. Farmers with mainly ley (non-permanent grass), semi-improved grass and unimproved grass saw hares less frequently, but the significant interaction term showed that if they had mainly ley or semi-improved grass their chance of seeing hares was increased if they also had arable land (Table 7).

Table 7.  The multiple ordinal logistic regression of variables specific to habitat (Table 2) on dependent variable see hare, for farm f = ‘arable’ and ‘pastural’. The percentage of farms within each dependent variable level on which hares were seen at least ‘most weeks’ is shown; on the remainder, hares were seen at most ‘most months’. Other non-significant factors in the models were for farm f=‘arable’: barley, spring, rape, legume and woods; for farm f=‘pastural’: arable (shown in table). Model performance was good for arable (Δd8 = 54·7, P < 0·001; 69% of pairs concordant; n= 227) and pastural farms (Δd8 = 43·1, P < 0·001; 58% of pairs concordant; n= 519). CI = confidence intervals; × interaction term
Independent variableLevel% with see hare=‘most weeks’ or ‘daily/most days’Coefficient ± SEZ, POdds ratio (95% CI)
farm f=‘arable’
wheatNo34Reference level 1·0
Yes77  1·4 ± 0·41 3·4, < 0·014·0(1·8–9·0)
beetNo64Reference level 1·0
Yes87  1·0 ± 0·30 3·3, < 0·012·7(1·5–4·9)
fallowNo44Reference level 1·0
Yes76 0·85 ± 0·41 2·1, < 0·052·3(1·1–5·2)
Constant 1   −3·1 ± 0·49  
Constant 2   −1·8 ± 0·46  
Constant 3  −0·93 ± 0·44  
farm f=‘pastural’
type grassImproved38Reference level 1·0
Semi-improved22 −1·5 ± 0·40−3·7, < 0·0010·23(0·11–0·51)
Ley31−0·86 ± 0·27−3·1, < 0·010·42(0·25–0·72)
Unimproved17 −1·4 ± 0·32−4·2, < 0·0010·26(0·14–0·49)
arableNo26Reference level 1·0
Yes36−0·68 ± 0·39−1·7, NS0·50(0·23–1·1)
woodsNo26Reference level 1·0
Yes33 0·38 ± 0·18 2·1, < 0·051·5(1·0–2·1)
type grass × arableImproved × yes28Reference level 1·0
Semi-improved × yes36  1·6 ± 0·77 2·0, < 0·054·9(1·1–22)
Ley × yes45  1·5 ± 0·47 3·3, < 0·014·7(1·9–12)
Unimproved × yes18 0·76 ± 0·81 0·94, NS2·1(0·4–11)
Constant 1   −1·3 ± 0·25  
Constant 2  −0·37 ± 0·24  
Constant 3   0·20 ± 0·24  

variables specific to hunting

Hares were caught for restocking on one farm (0·1%, two hares) and released on four farms (0·4%, 10–50 hares). Hares were hunted on 248 farms (24%). Simple likelihood-ratio tests were conducted on see hare with variables relating to hunting (Table 3), with the exception of coursing (hunting with sight hounds), which took place on only 11 (4%) farms on which hares were hunted. The model with independent variables number and hunt hare could not be improved by removing either variable or by including the interaction term.

High numbers of hares were hunted only on farms where hares were seen frequently. In the ordinal logistic regression model (for farms where hunt hare=‘yes’) with shoot and number on see hare, number was significant [coefficient ± SE = 2·5 ± 0·47, Z= 5·3, P < 0·001, odds ratio (95% CI) = 12(4·8–31); coefficient ± SE for constants: 1 = −1·4 ± 0·34, 2 = 0·31 ± 0·32, 3 = 1·4 ± 0·38]. The model was significantly better than the null model and the predictive value was good (Δd2 = 43·1, P < 0·001; 74% of pairs concordant). shoot was not significant in the final model.

variables specific to change

Farmers were asked whether numbers of hares had declined, remained stable or increased since 1980 and since 1995. Seventy-three per cent of respondents answered the question on changes in numbers of hares on their farm since 1980, indicating that they had been there for ≥ 20 years. The answers to the two questions were associated with one another (inline image = 632, P < 0·001; Table 8).

Table 8.  Farmers’ responses to the question: ‘how have numbers of hares changed on your farm since 1980 and since 1990?’ Numbers of farmers and percentages of total (n = 748) are shown
Since 1980Since 1995
Declined195 (26%)  5 (1%)  1 (0%)
Stable114 (15%)249 (33%) 29 (4%)
Increased 21 (3%) 23 (3%)111 (15%)

A combined variable change hare was created from answers to both questions (by adding together scores of 1 = decrease, 2 = stable, 3 = increase for each answer, to create a new score with levels 2–6, in which 2 = large decline, 3 = some decline, 4 = stable, 5 = some increase (pooled with 6 for analysis since < 10% of farmers fell into this category), 6 = large increase). change hare was examined in relation to category, previous and change f (Table 4). category and previous (farm classifications given by the farmer in answer to the question: ‘has the main business of your farm changed since 1980, and if so, how?’) were found to be associated (in both cases ‘crops’ was most likely to have experienced an increase in hares, and ‘sheep’ was most likely to have experienced a decrease) so only category was used in the final model (Table 9).

Table 9.  The simple ordinal logistic regression model with category on change hare is significantly better than the null model (see Table 3; 41% of pairs are concordant; n= 670). CI = confidence interval
Independent variableLevel% farms with change hare=‘increase’Coefficient ± SEZ, POdds ratio (95% CI)
categorySheep14Reference level 1·0
Cattle18 0·40 ± 0·231·7, NS1·5(0·95–2·4)
Crops33  1·1 ± 0·254·4, < 0·0013·0(1·9–5·0)
Other20 0·43 ± 0·271·6, NS1·5(0·90–2·6)
Constant 1   −1·9 ± 0·22  
Constant 2  −0·19 ± 0·21  
Constant 3   0·56 ± 0·21  

Farms that were mainly used to grow crops were the most likely to have experienced increasing numbers of hares. Overall, hares were described as declining on 42% of farms and increasing on 22% (Fig. 3).

Figure 3.

Distribution of values for variable change hare: circles = declining (n = 305); squares = increasing (n = 157); triangles = stable (n = 266).


validity of data and use of questionnaire surveys

This questionnaire provides data only on the opinions of respondents and on current land management practices. However, our respondents were representative of farmers in England and Wales. Our results were also consistent with those from a national hare survey, in which hares were found to be twice as numerous in arable landscape types as in pastural landscape types (Hutchings & Harris 1996).

Questionnaire studies such as this are useful and cost-effective for the collection of data on animal abundance and distribution (Reading et al. 1996) and on agricultural management (Medley et al. 1995; Macdonald & Johnson 2000). In this study, questionnaires allowed the collection of data on the abundance of a relatively rare animal over a wide geographical area, at a cost far lower than that of field assessment. In a national hare survey conducted in 1991–93, 1325 hares were seen on 6510 km of transects surveyed on foot, and sighting frequency was not high enough to allow the analysis of farm-level variation in abundance (Hutchings & Harris 1996). The main disadvantages of questionnaire studies are the lack of accurate fine detail in the data, and the need to ensure that the respondents are representative. These problems may be overcome by employing robust sampling procedures and questionnaire design, and by quantifying the extent to which the respondents are representative. The accuracy of responses may be checked in a subsample of the respondents (‘groundtruthing’; Moore et al. 1999). The comparison of respondents to non-respondents is difficult, particularly for face-to-face and telephone interviews and for postal surveys such as this, when anonymity is allowed. Authors tend to assume that respondents are (McDonald & Harris 1999) or are not (Moore et al. 1999) representative. The extent to which the respondents are representative should be quantified, for example by resurveying the non-respondents (Heydon & Reynolds 2000) or by statistical comparison of the respondents with the intended sample population (White & Whiting 2000; as undertaken here).

habitat associations of hares

Hares were seen most often on arable farms with large field sizes and low habitat density (few different habitats 100 ha−1). Small fields have been found to be beneficial for hares, but the range of field sizes differs between studies (Schröpfer & Nyenhuis 1982; field sizes 38–51 ha, Slamečka 1991; field sizes ≤ 200 ha, Lewandowski & Nowakowski 1993; field sizes 6–18 ha, Panek & Kamieniarz 1999). In our study, only 1% of farms had field sizes of > 20 ha. Increasing habitat density (up to 10 different habitats 100 ha−1) has been found to be positively associated with hares (Hutchings & Harris 1996); we found a negative relationship, but our sample consisted almost entirely of farms with ≤ 50 habitats 100 ha−1 (Table 1). It is therefore likely that optimal field sizes and habitat densities exist for hares.

On arable farms, frequent sightings of hares were associated with wheat, beet and fallow land. Wheat and beet are generally favoured by hares (Schröpfer & Nyenhuis 1982; Hutchings & Harris 1996; Marboutin & Aebischer 1996) and hares favour farms with fallow land because it provides cover and food all year round (Tapper & Barnes 1986; Lewandowski & Nowakowski 1993; Panek & Kamieniarz 1999).

On pastural farms, frequent sightings of hares were associated with the availability of woodland and improved grass, and in some cases with arable land; similar associations were found by Hutchings & Harris (1996). Hares use woodland to shelter during the day (Tapper & Barnes 1986) but may also avoid it (Marboutin & Aebischer 1996), and few hares are found in areas with large amounts of woodland (≤ 20%; Panek & Kamieniarz 1999). Thirty-five per cent of respondents’ farms had some woodland; pastural farms had a median coverage of 7% (range 0·4–48%). A fine-scale study of habitat use showed that hares avoid fields that are used by cattle (Barnes, Tapper & Williams 1983) but we found no effect of stocking density on numbers of hares.

The association of abundant hares with wheat, beet, fallow land, improved grass and some arable land in pastural farms is consistent with the hypothesis that resources limit populations of hares (Frylestam 1980; Hackländer et al. 2002). Hares feed selectively on grasses, herbs, some fruits and arable crops (particularly young cereals, but also maize Zea mays, peas Pisum sativum, beans Vicia faba, sugar beet, and ears of cereals; Homolka 1987; Chapuis 1990).

competition and predation

We found no relationship between the abundance of rabbits and hares, and so have no evidence for competition. Although they eat similar foods, the two species avoid competition by feeding in different areas where they are sympatric (Chapuis 1990).

At the simple level, high buzzard abundance was associated with low hare abundance (Table 1), although the variable see buzzard was not significant in the final model (Table 6).

The association of hare abundance and large field size, permanent cover (fallow land and woodland) and medium to low numbers of foxes is consistent with the hypothesis that predation by foxes limits numbers of hares (Lindström et al. 1994; Reynolds & Tapper 1995). Some adults, but mainly leverets, are taken by foxes. If cover is available, leverets may be able to avoid detection by predators, and predation may be reduced. Large fields may also help adult hares to avoid capture, and leverets to avoid detection, because a large uniform area is difficult to search effectively.


Although hunting effort was not quantified, the numbers of hares killed were related to their abundance, which suggests that hunting records, as used for example by Schröpfer & Nyenhuis (1982), are useful indices of hare abundance. However, interpretation of trends from hunting records is difficult (Marboutin & Péroux 1995a) and records are not available in areas where hares are uncommon and are not hunted.

changes in farmland and numbers of hares

The questionnaire covered England and Wales in 1980–2000. Major changes in agricultural practices have occurred in the UK since the Second World War, but most trends have stabilized since 1980 (Table 5). However, field size and the polarization between land use in the east and the west of the country have increased, while habitat diversity has decreased (Robinson & Sutherland 2002). The UK BAP states that the decline of the hare is due to the conversion of grass to arable land, the loss of habitat diversity, the increase in silage production and the move towards autumn planting of cereals (Anonymous 1995).

We did not find any positive effect on abundance of hares of grass in arable farms (Table 2), although arable land in pastural farms may be beneficial (Table 7; Hutchings & Harris 1996). Therefore the conversion of grass to arable is very unlikely to be detrimental to hares. We found a positive effect of habitat diversity at the simple level (Table 1), so our results are consistent with the loss of habitat diversity as a cause of decline. At the simple level, spring sowing of crops on arable farms was associated with frequent sightings of hares (Table 1). The move towards autumn planting of cereals may have led to a shortage of food during the summer when hares are breeding (Tapper & Barnes 1986; Hutchings & Harris 1996). Autumn planting of cereals in Britain has been stable since 1980 (Robinson & Sutherland 2002).

management recommendations and further work

We provide evidence of associations between the abundance of hares and current agricultural practices and predators on farms. Functional explanations of variations in abundance of hares should now be sought through measurement of demographic parameters, as recommended by Marboutin & Péroux (1995b). The direct effects of changes in habitat and numbers of predators on the abundance of hares have been tested empirically. A 3-year programme of habitat change in Slovakia, which included providing permanent cover and increasing diversity but no manipulation of fox numbers, resulted in an increase in the density of hares from 0·4 to 0·7 ha−1 (Slamečka 1991). A natural experiment caused by a sarcoptic mange epidemic in foxes showed that partial predator removal alone may result in higher numbers of hares (Lindström et al. 1994). Work is needed on possible interactions between habitat and predation. If the manipulation of habitat affects the relationship between predators and their prey, habitat management can be used as an alternative to predator control to increase numbers of the prey species (Schneider 2001). Hutchings & Harris (1996) attributed a positive association between gamekeepers and hares to enhanced habitat management, rather than to predator control.

In order to increase the value of their farms for hares, pastural farmers should be encouraged by policy-makers to have some woodland and improved grass, and to grow some arable crops in large (≤ 20 ha) fields. Arable farmers, on the other hand, should be encouraged to include in their rotations wheat, beet and fallow land, which should provide cover and food for hares all year (Panek & Kamieniarz 1999). However, removal of habitats that are of value for the conservation of biodiversity or other species (e.g. unimproved grass and hedgerows; Stoate et al. 2001; Robinson & Sutherland 2002) should be avoided. The provision of fallow land and woodland, which we recommend for hares, also benefits many species of birds, invertebrates and plant species diversity (Mineau & McLaughlin 1996; Stoate et al. 2001).

This broad-based research has shown that the questionnaire method is effective for farmland and could be used for other species of conservation concern. Easily implemented habitat management on individual farms is highly likely to benefit hares. If management is widespread, it may contribute towards the achievement of the conservation targets set for the species in the BAP (Anonymous 1995).


We thank the farmers for completing the questionnaire. We also thank Richard Brand-Hardy, Nik Cole, Phillipa Dodds, Tony Holley, Claire Horton and Philip Ray (GI Services Branch, National Assembly for Wales), Tony Hughes, Claire Johnson, Robbie McDonald, Judith Mills, Simon Poulton, Tony Robinson, Rebecca Smith, Philip Stott, Suzy Wilkinson, Alison Wray and Jane Jones (DEFRA statistics branch; census and survey), and three anonymous referees. This research was funded by DEFRA (grant number BD1436).