What makes free-range broiler chickens range? In situ measurement of habitat preference

ANIMAL BEHAVIOUR, 2003, 66, 151–160doi:10.1006/anbe.2003.2172

What makes free-range broiler chickens range? In situmeasurement of habitat preference

MARIAN STAMP DAWKINS*, PAUL A. COOK†, MARK J. WHITTINGHAM*, KATHERINE A. MANSELL* & AMY E. HARPER*

*Department of Zoology, University of Oxford†Premier Fresh Foods, Lincs, U.K.

(Received 13 June 2002; initial acceptance 21 August 2002;final acceptance 19 November 2002; MS. number: 7368)

In commercial free-range broiler (meat chicken) systems, many birds never leave the houses, makingthem ‘free-range’ in name only and suggesting that the environment provided is not preferred habitat.We investigated the factors that affect the numbers of birds ranging in 40 flocks, each of 20 000 growingbroiler chickens, Gallus gallus domesticus, in 14 houses on seven commercial farms and at three seasons ofthe year. The number of birds found to be ranging outside was positively correlated with the amount oftree cover the range area contained, the time of day and the season of the year. Few birds were seenranging in the winter but even in summer, the maximum number observed outside during daylight hoursat any one time was less than 15% of the total flock. We measured habitat preference of the chickens,using a statistical method previously used for wild birds, and argue that this ‘in situ’ measurement ofhabitat preference could be widely used in a variety of situations and that it thus constitutes an importantnoninvasive and nonintrusive method for assessing animal welfare on commercial farms.

2003 Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour.

Correspondence: M. S. Dawkins, Department of Zoology, University ofOxford, South Parks Road, Oxford OX1 3PS, U.K. (email:[email protected]). P. A. Cook is now at the Food AnimalInitiative, Field Station, Northfield Farm, Wytham, Oxford OX2 8QJ,U.K.

A problem with both free-range broiler (meat) chickensand egg layers is that, although birds kept on a commer-cial scale are given access to a large outside area, many ofthem do not come outside the houses at all or at best staywithin the immediate environment of the houses (Weekset al. 1994; European Commission 2000). They are thusfree-range in name only. In the European Union eggs andmeat may be labelled ‘free-range’ if birds have access to arange but consumers may in practice be eating ‘free-range’ products where the animals have never left thehouse. This suggests that many existing free-range sys-tems are not providing the outdoor environment that thebirds themselves want. We examined the factors that areassociated with increased ranging in growing broilers,Gallus gallus domesticus, with a view to recommendingwhat might be done to encourage birds to leave theirhouses and so to be properly ‘free-range’. To do this, wetook methods previously used to study habitat prefer-ences in wild birds and applied them to chickens oncommercial farms, thus showing that techniques devel-oped by ecologists can be a valuable tool in animalwelfare studies. Our second aim was to relate rangingbehaviour to production parameters such as mortality

0003–3472/03/$30.00/0 2003 Published by Elsevie151

and carcasses judged to be unfit for human consumption,and so to test the hypothesis that ranging is associatedwith decreased mortality and better health. If increasedranging is associated with better health of the birds, thiswould give welfare grounds for encouraging them torange.

Although there are some studies on free-range layingbirds (e.g. Grigor & Hughes 1993), little has been docu-mented about the behaviour of free-range meat chickensin commercial conditions, and there are important differ-ences between the two types. Laying hens will live up to70 weeks and have access to the outdoors starting at pointof lay (ca. 16 weeks). Broiler chickens, in contrast, havebeen bred to grow much faster, and intensively housedbirds are usually killed at 6 weeks of age or less. Even theslower growing broiler breeds used for outdoor systemslive for only 2–3 months and have access to the outdoorsfor half this time. This means that work on laying hensmay not be directly applicable to the much younger birdsof broiler breeds and, for this reason, we felt it wasimportant to look at free-range systems for meat chickenswith their specific requirements in mind. We also felt itwas important to look at birds kept in fully commercialfree-range farms, since the scale of the houses and thelarge group sizes (ca. 20 000) typical of commercialsystems provide an environment that is so different fromsmaller groups that extrapolation is difficult. Quantitativeobservations of the way birds use outdoor areas at

r Ltd on behalf of The Association for the Study of Animal Behaviour.

152 ANIMAL BEHAVIOUR, 66, 1

different times of the day and at different times of theyear are also needed to develop systems that will optimizebird welfare, since their habitat preferences on a sunnysummer day may be different from those in winter.

Weeks et al. (1994) observed groups of 60 chicks of afast-growing meat breed reared in deep-litter pens or withaccess to pasture, but for only 1 h, between 1000 and1100 hours. They found little difference between thegroups and on the basis of this concluded that broilerchicks ‘lacked motivation to use space’. The recentEuropean Union report on broiler welfare (EuropeanCommission 2000) used this to conclude that modernbroiler chickens have been bred so they do not want tomove. However, if provided with environmental enrich-ments such as straw bales, broiler chicks move aroundmuch more, not just by perching on the bales but also bybeing more active away from the bales (Kells & Dawkins2001). Novel objects also stimulate movement in broilerchicks (Newberry 1999). Furthermore, the behaviour ofjunglefowl, G. g. spadiceus, and chickens is known to varythroughout the day (Collias & Collias 1967; Dawkins1989), so range use might also be affected by time of day,with mid-morning being one of the least likely times forbirds to venture out. Our study is the first to look at thefactors affecting ranging behaviour in broilers in commer-cial conditions that took into account the behaviour ofbroiler chickens at different times of day and differenttimes of the year.

We studied 40 flocks of free-range broilers that werebeing supplied to a supermarket chain. Since the birdswere being reared for normal production our results aredirectly applicable to market conditions. It is also import-ant for completeness and commercial applicability of thestudy that we observed the birds throughout the day,from the time the pop-holes were opened in the morningto when they were closed at night and at three seasons ofthe year: spring (March–May), summer (June–August) andwinter (October–January). We made use of the fact that,in the interests of product uniformity, the birds werebeing kept on 14 similarly sized paddocks, each withidentical houses and management regimes but that therewere fortuitous differences between the houses in theamounts and type of vegetation (mature trees, bushes,hedges) that were growing in the paddocks. Thesenaturally occurring differences allowed a correlation to bedrawn between the vegetation present and the extent towhich the birds ranged. There were inevitably additionaldifferences between the 14 houses, such as the nature ofthe site and the precise way in which the farm managersinterpreted the company’s guidelines. These wereanalysed as ‘farm’ effects (see Results). We identified thefactors affecting ranging (1) by comparing the numbers ofbirds ranging in the 14 paddocks in relation to thevegetation, especially areas occupied by trees and bushes,(2) by correlating the numbers of birds ranging withfactors such as temperature and sun and (3) by measuringthe spatial distribution of birds in relation to the distri-bution of habitat types within each paddock to indicatehabitat preference. For the latter approach, we used stat-istical methods previously applied to measure habitatpreference in wild birds (Morris et al. 2001; Whittingham

et al. 2001); we identified the choices the birds had madeabout which parts of the range they ‘liked’ in a com-pletely nonintrusive way by looking at their spatial dis-tribution in relation to features of the habitat that wereavailable to them. We hope that this method will betaken up as a tool for the noninvasive, nonintrusivemeasurement of welfare suitable for use on commercialfarms for both extensive and intensive systems.

METHODS

Farms and Birds

We made our observations on seven farms, each ofwhich had two broiler houses and their associated pad-docks, belonging to Premier Poultry Ltd in Lincolnshire,U.K., from March 2000 to January 2001. All 14 houseswere of identical design and construction, all were builtin 1999 and all had the same area of outside range(ca. 20 250 m2). All were operated to the same manage-ment system approved by the RSPCA Freedom Foods(RSPCA 1997). Each house contained approximately20 000 chickens and was equipped with identical cuplessnipple drinkers, Hi-lo pan feeders, balanced ventilationsystems (to prevents draughts at pop-holes), activehumidity controls, rain simulation (‘misters’) and a‘veranda’ area (see below). The environment of thehouses was controlled with a Flockman automatic moni-toring system. All houses were stocked with the samebreed of female chicks from the same hatchery, suppliedwith food from the same mill, collected by the samecatching teams and killed in the same processing plant.This made them ideal for the study, because we couldrelate differences between houses in ranging behaviour tothe main way in which the houses differed, primarily infortuitous differences in the vegetation characteristics ofthe ranging area, and factor out between-farm effectsstatistically. All ranging areas had areas of grass that wereregularly mown but differed in the numbers of maturetrees, bushes and hedges. Some had extensive areas ofmature woodland, some had no mature trees at all, somehad trees close to the house, others only well away fromthe house.

Each house measured 76.2�23.2 m with 12 externalpop-holes along each side. The birds were always fedinside the houses. A single dark period was provided inevery 24-h period from when the chicks were 1 day old.This started at 1 h and increased to 12 h by day 4, and wassupplemented with a 10-min period of dawn (automaticincreasing light intensity) and a 10-min period of dusk.The houses contained two sets of pop-holes, the internalones allowing access to a veranda area that was naturallylit (through plastic mesh), not heated and contained nofeeders or drinkers. Each pop-hole was 50 cm high and240 cm wide. The veranda area was 2.5 m wide andextended along the full length of both sides of all thehouses. The external set of pop-holes allowed access on tothe outside ranging areas. Chicks were placed in thehouses at 1 day old and initially brooded with both setsof pop-holes closed. The internal pop-holes were openedat about day 14 (depending on prevailing weather

153DAWKINS ET AL.: FREE-RANGING CHICKENS

conditions), allowing the birds to acclimatize prior to thesecond external set being opened no later than day 28.Around the outside of each house was a 2-m-wide path ofshingle to prevent the birds becoming too muddy. Rowsof small saplings had been planted in all range areas butthese had not grown well. Moveable sunshades (table-shaped wooden structures ca. 5�2 m and 1 m high) werepositioned throughout the range.

The 14 houses were part of a cycle designed so that asupermarket chain could have a continuous supply eachweek of birds of the right age and weight. A fortunateeffect of this need for a uniform product was that growingconditions were kept as uniform as possible. All birdswere pullets and were of a Ross-derived strain (SherwoodWhite) developed by Premier to be slower-growing thannormal commercial breeds. At 50 days the average weightof the birds was 2.04 kg. The birds were killed at 56 days.All observations were carried out when the birds werebetween 45 and 53 days old. Each flock was observed ononly one occasion, so no direct comparison between thebehaviour of that flock at different ages was possibleusing the data we collected. The farm managers wereasked to keep diaries of the history of the flocks (e.g. ageat which the birds first moved to different parts of therange, attacks by predators). Although this provided use-ful background information, we could not analyse itstatistically. Each farm was visited on three occasions:‘spring’ (March–May), ‘summer’ (June–August) and‘winter’ (November–January), when three different flockswere in place. On any one day, the two houses on thesame farm were both observed. The two houses on eachfarm contained birds that differed by 4–6 days of age fromeach other, and we took this into account in the analysis.A total of 40 different ‘flocks’ was observed during thisstudy (14 in both spring and summer, but only 12 in thewinter because of logistic difficulties).

Observations

On any one observation day, an observer recorded theranging behaviour of birds from the two houses onone farm, between the time the pop-holes were opened(usually between 0800 and 0900 hours, at the discretionof the farm manager) until the birds went into the housesat night (between 1800 and 2000 hours). A requirementfrom the European Union for birds to be described as‘free-range’ is that they have access to the range for atleast 8 h per day. The easiest way for the farm managers toget the birds back into the houses at night was to shut thepop-holes while the birds were being fed in the lateafternoon. However, these particular farm managers tooka great interest in the behaviour and welfare of their birdsand many of them did not like to keep the birds inside for16 h and so left the pop-holes open until much later. Wetook these differences in management practice intoaccount by viewing them as ‘farm’ effects for statisticalanalysis. Observations of ranging behaviour wererecorded photographically, using a Canon 1�USM-1camera with a wide (23-mm) angle lens and an APS filmthat recorded date on each photograph. Once an hour,and for at least 10 h per day, the observer would move

round the ranges of both houses and take photographs atset places, attempting to photograph all birds that wereoutside the houses or (where this was not possiblebecause the birds were hidden by vegetation or unneces-sary because there were so few birds out) to count thebirds by eye. A total of 1720 photographs were used forthe analysis.

Description of Habitats

Scale maps of the farms were supplied by PremierPoultry and areas of short grass, bushes, hedges and treeswere measured and drawn on to these maps, which werethen scanned into a computer and the area occupied bydifferent habitat types measured using NIH software. Weinitially characterized the different paddocks by the areasof trees and bushes they contained. Some paddocks con-tained no trees or bushes, others considerable numbers oftrees and bushes (broad-leaved but a variety of species)and we used these data to compare the numbers of birdsseen outside in the different paddocks. For a moredetailed analysis, we used the maps of each paddock tomeasure the areas occupied by six habitat types definedpartly by what they were (grass, concrete, etc.) and partlyby their nearness to the house. The advantage ofthe method described is that it allows flexibility in thedefinition of ‘habitat’. These were (1) shingle, (2) shortgrass near house (defined as within 6 m of a house), (3)short grass within 1 m of the perimeter fence, (4) othershort grass, (5) bushes, trees, long grass or hedge and (6)concrete. Since relatively few birds used the woodensunshades that were present in all paddocks, these werenot counted as a habitat type. The orientation ofthe houses was recorded to the nearest of 16 bearings(Table 1).

Weather Conditions

Every hour at the time the photographs were taken, theweather conditions were recorded as sun in or out. Forthe analysis, ‘sun’ was taken as the percentage of totalobservation hours that the sun had been recorded as‘out’ when the photographs were taken. In addition, arecord was taken of whether it was raining. For theanalysis, however, we obtained more accurate weatherinformation from the Leeds Weather Centre and theMeteorological Office in London. This provided data ondaily rainfall (mm), mean daily temperature and windspeed.

Events

Events that caused movements of the birds were alsorecorded. These included the arrival of the feed lorry, ahelicopter flying overhead and grass mowing. Inaddition, diaries kept by the farm managers of theprogress of each flock as the birds grew (e.g. attacks bycrows, Corvus corone) and verbal reports were used torecord incidents taking place when the observers werenot present.


Production Data

Data on percentage mortality during the growingperiod while the birds were on the farm were obtainedfrom the routine crop reports kept by the company. Inaddition, these reports provided information about thepercentage of birds found to be dead on arrival at theprocessing plant and the percentage of carcasses rejectedby the plant operators as unfit for human consumptionbecause of bruising or other defects. If increased rangingbehaviour results in an increase in the welfare of thebirds, lower mortality and downgrading can be predicted.In addition, information was obtained on the amountof food consumed in each house, with the company’sexpectation that increased ranging might decrease theamount of food consumed, since the birds would beobtaining some of their food outdoors. Unfortunately,not all houses had individual records of feed consump-tion so that only a limited comparison of feed conversionratios (FCRs) was possible (10 houses, spring and summerobservations only).

Statistical Analysis

Comparisons between and within flocksThe principal response variables used for comparison

between houses were the mean number of birds recordedas being outside each house per day (based on the meanobserved at each hour) and the maximum numberrecorded outside during any of the hourly observationperiods. We square-root transformed both of thesemeasures since these were count data. The 14 houses wereeach observed on two separate occasions: March–May

2000 (‘spring’) and June–August 2000 (‘summer’) and 12also on a third occasion, October 2000–January 2001(‘winter’). Because the birds were killed at 56 days, thismeant that for each season, the observations were on adifferent set of 20 000 birds and the 40 flocks were takenas independent. A GLM analysis was performed withseason and house as factors.

Time of dayThe above analysis used one data point for each flock

but was based on the means of 1 day’s observation. Forthe time of day analysis, we ignored these means andlooked at the numbers of birds seen outside at differenttimes of day, based on observations carried out for at least10 consecutive hours. Because these were observations onthe same flocks, they were treated as repeated measures. AGLM analysis was performed with season and time of dayas factors.

Environmental variablesA regression analysis with tree area, percentage sun,

mean daily temperature, mean daily wind speed andmean daily rainfall as predictors was performed with bothmean number of birds outside and maximum number ofbirds outside as the response variables. One data point perflock was used to ensure independence of data.

Production dataA GLM analysis was performed with the production

variable (percentage mortality during the growing period,percentage dead on arrival at the processing plant, per-centage rejected at the processing plant as unfit forhuman consumption or feed conversion ratio) as theresponse variable and the mean number of birds rangingas a covarying variable.

Table 1. Orientation of the broiler houses on the different farms

House Orientation of house

Farm 1House 1 N–SHouse 2 E–W

Farm 2House 1 E–WHouse 2 N–S

Farm 3House 1 NE–SWHouse 2 NE–SW

Farm 4House 1 E–WHouse 2 E–W

Farm 5House 1 N–SHouse 2 N–S

Farm 6House 1 NNW–SSEHouse 2 NNW–SSE

Farm 7House 1 E–WHouse 2 E–W

Habitat preferenceWe used the observed spatial distribution of the

chickens in relation to different features of their habitatsuch as open short grass, bushes and trees to indicatehabitat preference. The method used to establish whetherbirds were selecting certain parts of their paddock overothers was that used to demonstrate habitat selectionin wild birds (Green et al. 2000; Morris et al. 2001;Whittingham et al. 2001). We undertook nine separateanalyses to investigate habitat selection in three seasons(winter, spring and summer) and at three times of day(early, mid-day and afternoon) during each season. Birdscounted in different seasons were different individualsbecause one ‘batch’ of chickens remained in a paddockfor only 10 weeks. However, chickens counted at differenttimes of day but in the same season might have been thesame individuals. To account for this problem we used aBonferroni correction to adjust our significance value forcomparing chickens counted at different times of day (seeAppendix).

First, we tested whether chickens selected habitats atrandom. If they did not, we compared relative selection


of habitats to derive an approximate ranking of habitatuse. We used GLIM macro procedures, which allowed usto test for (1) significant overall variation in habitat useand (2) significant differences in relative use betweenpairs of habitats. We used a log-linear model with aPoisson error distribution and log link function, with thenumber of chickens observed in each habitat type speci-fied as the response variable. Habitat types and paddockswere treated as factors. Ln (area) of each habitat wasdeclared as an offset variable. The offset variable controlsfor the increased number of records associated with largerareas. This model is referred to as model ‘A’:

Log density of chickens=Ni+Hi

where Ni is the ith of a set of coefficients specific to thenumber of chickens in each paddock (14 in the springand summer and 12 in the winter) and Hi is the setof coefficients specific to the habitats in each paddock(i.e. 6).

All modelling was carried out in GLIM, release 4 (NAG1993). The methods used to test for statistical significanceof overall variation in habitat use and pairwise differencesbetween habitats are summarized in the Appendix.

RESULTS

On each of the seven farms, one house was stocked withchicks 4–6 days before the other. However, an initialtwo-way ANOVA on the spring and summer data block-ing by farm showed that neither the farm (F6,20=0.33,P=0.911) nor the small age difference in the birds affectedthe numbers of birds ranging (F1,20=0.01, P=0.930). Thefact that ‘farm’ included the farm manager (includingtheir different times of opening and closing the pop-

holes) and the site (e.g. nearness to a road) suggests thatnone of these factors was an important confoundingvariable. The 14 houses were thus thereafter treated as 14independent units and no further account was taken ofbird age.

Differences Between Seasons

There was a significant effect of season on the meannumbers of birds observed outside (F1,13=42.241,P<0.001; Fig. 1). The maximum number seen outsideduring the daylight observation hours at any house was2864 during the summer (14.3% of the flock). However,after sunset, when it was too dark to take photographs orto count the birds accurately without artificial light,many more birds came out than had been seen during thedaylight hours. Thousands of birds came pouring out ofthe houses at sunset and many were still outside when itwas completely dark, even though the lights were stillon in the houses. The daytime maximum is there-fore probably an underestimate of the number of birdsthat came out at any time. As the winter data set wasincomplete (12 houses only), and few birds came outanyway, the subsequent analysis was done on the springand summer data only, with extra information beingadded about the winter data separately.

0Summer

1250

No.

of

bird

s ou

tsid

e

Spring

1000

750

500

250

Winter

Figure 1. Mean numbers of birds seen outside at different seasons ofthe year. Each point represents the mean of daily observations for 14(spring and summer) or 12 (winter) houses. The vertical lines arestandard deviations.

012

1500

Time since doors opened (h)

No.

of

bird

s ou

tsid

e

0 1 2 3 4 5 6 7 8 9 10 11

1000

500

Figure 2. Numbers of birds seen outside at different times of theday. For the first 10 h after the pop-holes were opened, each pointrepresents the combined mean for spring and summer observationsfor 14 houses. On some farms, the pop-holes were closed after 10 hand the points for hours 11 and 12 are based on a mean of only sixhouses. Vertical bars are standard deviations.

Differences with Time of Day

The numbers of birds seen outside decreased a fewhours after the doors were opened and then increasedlate in the afternoon and evening (F1,9=4.69, P<0.0001;Fig. 2).


Differences Between Houses

There were significant differences between the houses(F13,13=4.34, P=0.006) and a highly significant differencebetween the spring and summer (F1,13=48.23, P<0.001)in the mean numbers of birds coming outside. Thissuggested that it was worth looking at the characteristicsof both the seasons and the paddocks that might accountfor these differences. A regression analysis on the meannumbers of birds showed that temperature, sun and treecover all affected ranging (Table 2) but that the orien-tation of the house did not. More birds came out onwarmer days, on days when the sun was not out and inpaddocks with greater amounts of tree cover. Theseeffects are even clearer if the maximum rather than themean numbers are used as the response variable (Table 2).

Habitat Preferences

It was clear from this comparison between houses thatamount of tree cover in the range areas was a goodpredictor of the number of birds that would come outsidethe house. We next examined the distribution of birdswithin each free-range area to see if we could furtheridentify which parts of the range they preferred. Thisapproach made use of the fact that the observed spatialdistribution of the chickens in relation to differentfeatures of their habitat such as open short grass, bushesand trees reflects the choices they have made about wherethey like to be. We divided the data into nine subsets bytaking three seasons and three times of day (early morn-ing just after the doors were opened, mid-day and the lastobservation before the birds were shut up for the night).For all combinations of time of day and season, the birdswere nonrandomly distributed across the paddocks; inother words, the birds were actively selecting their habitatwithin what was available to them, not just by the area ofeach habitat type present. (In each case, P<0.002, butwith a Bonferroni correction for the repeated measures at

three times of day P=0.017.) The next stage was a pairwisecomparison of each of the six habitat types. Figure 3shows these comparisons for the three seasons. Not all ofthe paddocks contained all of the six habitat types (somehad no trees or bushes). The most preferred habitat in allcases was the shingle surrounding the house, followed bythe short grass near the house (Fig. 3), indicating thereluctance of the birds to leave the immediate vicinity ofthe house even when trees were present. To show moreclearly the effect of distance on habitat choice, we esti-mated the median distance of each habitat type from thehouse from the median for each paddock. Figure 4 showsthe median distance in relation to the habitat’s rank.Despite being the most distant habitat, trees and busheswere consistently chosen over short grass or short grassnear a fence, and open short grass was the least preferredhabitat despite not being the most distant (Fig. 4).

Production Data

None of the production variables (percentage mor-tality: F1,21=0.39, NS; feed conversion ratio: F1,8=0.94,NS; numbers dead on arrival at the processing plant:F1,21=0.07, NS; number of carcasses judged unfit forhuman consumption: F1,21=0.07, NS) was affected by themean or maximum number of birds ranging. However,the percentage of birds rejected as unfit for humanconsumption at the processing plant was affected byseason, with most being rejected in winter (F2,22=4.25,P=0.003).

Incidental Observations

Only one (apparent) fox attack was recorded duringthis study: the electric fences surrounding the paddocksseemed to be an effective way of keeping mammalianpredators away. However, the farm managers’ diariesrecorded several attacks by crows in which two or threebirds would cooperate to kill a chicken.

Table 2. Regression analysis on the mean and maximum numbers ofbirds coming outside (r2=35.0% and 54.2%, respectively)

Predictor r t1,21

Mean numberConstant 8.141 1.39Tree area (m2) 0.176 2.58*Sun (% hours) −0.125 −2.40*Temperature (°C) 0.921 2.50*Wind speed (km/h) 0.350 0.39House orientation 0.020 0.03Rainfall (mm) −8.660 −0.66

Maximum numberConstant 14.330 2.33*Tree area (m2) 0.210 2.91**Sun (% hours) −0.120 −2.25*Temperature (°C) 1.180 3.03**Wind speed (km/h) 0.519 1.24House orientation −0.295 −0.38Rainfall (mm) −2.390 −0.17

*P<0.05; **P<0.001.

DISCUSSION

Our main findings are that chickens prefer ranging areaswith trees, they avoid bright sun and that, within theirpaddocks, they either stay close to the house or they seektree cover. This is hardly surprising in view of the fact thatdomestic birds are descended from red junglefowl, whichinhabit the dry forests of southeast Asia and are known as‘bamboo fowl’ (Collias & Collias 1967). A wide open fieldis simply not a preferred habitat. Our results have import-ant implications for the design of free-range poultrysystems and make it clear that tree cover is somethingthat should be provided to encourage ranging. The forestswhere junglefowl occur frequently consist of thickclumps of bamboo separated by small clearings, so thatthe birds can see the approach of predators but are rarelyfar from cover into which they disappear at the slightestsign of danger (Collias & Collias 1967). The attacks bycrows reported by the farm managers in this study showthat these domestic birds are still behaving adaptively by


+

(a)

Rank scoreHabitat

Gravel (1)

Short grass nearhouse (2)

Grass near fence(3)

Other short grass(4)

Long grass,bushes, hedgesand trees (5)

Concrete (6)

+ + + + + + + + + n + +* n + +12

+4

–4

–9

–2

–1

+ + + + n n + n n

n + n n – n

n n – n* – –

n n n

n n n *

+ n +

E M A E M A E M A E M A E M A2 3 4 5 6

+

(b)

Rank scoreHabitat

Gravel (1)


Grass near fence(3)



Concrete (6)

+ + + + + + + + + n +13

+4

–6

–8

–1

–2

+ + + + n n + n n

– n

n –* – –

n n n

n n n *


+

(c)

Rank scoreHabitat

Gravel (1)


Grass near fence(3)



Concrete (6)

+ + + + + + + + +15

+10

–6

–6

–3

–6

+ + + + n+

n –

n n n

n n n

*


+ + +

+ + + *

n

–*n n n

–*–*

+ + + + + +

n n n

+ + + + +

n n n

n n nn

+

Figure 3. Results of chicken habitat selection analyses. Pairwise habitat combinations for each paddock at three times of day in (a) spring, (b)summer and (c) winter. E: early; M: mid-day; A: afternoon; + selected or − avoided at P<0.017 (Bonferroni corrected for three repeat tests,i.e. tests from the same chickens in the same paddocks at three times of day); *P<0.05; n indicates no preference. Both focal habitats occurredon at least 10 paddocks. Consistent trends in habitat use between times of day are indicated by larger symbols. For example, gravel wasselected over short grass near the house at all three times of day. Habitats were assigned an approximate ranking score by adding togetherthe number of times a habitat was significantly selected (P<0.017) over another habitat (+’s) and subtracting the number of times that habitatwas avoided (P<0.017) relative to another habitat (−’s). N=14 paddocks in spring and summer, and 12 in winter.


–1015

20

Median distance from house (m)

Ran

k of

hab

itat

typ

e

0

15

10

5

0

–5

105

GR

SGN

C

SGD

F

TR

Figure 4. Rank of each habitat type in relation to its median distancefrom the house. Rank of habitat type is the median rank for all threeseasons. A habitat type’s distance from the house is the median forall 14 houses measured from the maps by taking the midpoint of themaximum and minimum distances for that habitat type. GR=gravel/shingle; SGN=short grass near house; F=grass within 1 m of fence;SGD=short grass distant from house; TR=trees, bushes, hedge orlong grass; C=concrete.

seeking cover even in an agricultural setting. Attack fromabove was a danger for their wild ancestors and it is still adanger for them now (Stahl et al. 2002).

It is still not clear exactly what features of trees are mostimportant to chickens since trees are multipurpose: theyprovide shade from the sun, dry areas for dust bathing,and shelter from aerial predators overhead and fromground predators approaching from the side. They alsoprovide food and act as wind breaks. Newberry &Shackleton (1997) showed that Brown Leghorn chickenspreferred partial, visually discontinuous cover, whichthey referred to as the ‘Venetian blind’ effect. Theyargued that the birds were seeking security cover ratherthan thermoregulation. Nevertheless, in the presentstudy, both temperature and sun affected ranging behav-iour. Warm overcast summer days were associated withthe greatest numbers of birds coming out of the housesduring daylight hours, although even greater numberscame out in summer as the sun was setting and it wasbecoming too dark to photograph the birds. This suggeststhat one way to encourage ranging is not to close thedoors later at night and to allow the birds to be out forlonger. Good fox protection is obviously essential butachievable. At least at night in the U.K., the birds are safefrom aerial predators and do not need to be chased inside.Chickens were least likely to be out at mid-day both inwinter and in bright summer sun. Again, the role of treesas wind breaks and sun shades as well as cover is stronglysuggested.

In evaluating the welfare benefits of commercial free-range systems a crucial first step is to establish whether

increased ranging behaviour is associated with lowermortality and fewer downgraded carcasses. No obviousdifferences were recorded in this study but that may bebecause the numbers of birds ranging in any of thehouses was small anyway and also because there weredifferences between the houses that could not obviouslybe related to differences in ranging behaviour. In anycase, the differences between the houses in the amount oftree or bush cover they possessed were relatively smalland fortuitous. A more experimental approach with big-ger differences between the paddocks is needed to estab-lish whether tree cover does significantly affect mortalityand carcass quality.

We were unable to identify individual birds in thisstudy because any marking or tagging could have com-promised their market value but clearly another crucialcomparison for the future is between birds that regularlyrange outside and those that remain inside for much orall of their lives. The greatest number of rejected carcassesoccurred during the winter, again emphasizing thepossible role of climatic factors in bird welfare.

Paradoxically, the numbers of birds reported outside inthis study may be lower than for some other commercialsystems precisely because of a specific welfare featureprovided by Premier Poultry: the ‘verandas’. A shaded,covered area, from which the birds could see the outsiderange without actually leaving the house, may haveprovided them with such an attractive ‘tree-like’ area thatit was strongly preferred to either short grass or distantreal trees that could be reached only by a considerablejourney without cover. It is striking, although not surpris-ing, that few birds ranged outside in the winter months,at any time of day. If commercial free-range is to providesuitable outdoor habitat for chickens, thought needs tobe given not just to the habitat available in summer butalso to what is available in winter. Trees may assume evengreater importance in winter than in summer as windbreaks and protection against wind and rain.

If studies of animal behaviour are to make a genuinecontribution to the welfare of animals kept on commer-cial farms, it is essential to develop methods that can beapplied on a commercial scale. Laboratory studies ofsmall numbers of animals, although a useful startingpoint, cannot replicate the conditions and problemsexperienced by farm animals kept in much larger groupsand in conditions constrained by market forces. We haveshown that borrowing methodology from a different field(in this case farmland bird ecology) is a fruitful way of‘scaling up’ and studying the habitat preferences of largenumbers of commercial broilers with minimal distur-bance and in the conditions they actually experience onfarm. The same method could be used for indoor or moreintensive systems such as commercial broiler or layinghouses. For example, the interior of a laying house couldbe thought of as consisting of different ‘habitats’ such asthe nestbox areas, the areas within a certain distance ofthe pop-holes and the preferences of the birds measured.This would give an objective measure of what parts of thehouse the birds like or dislike and the extent to whichenvironmental ‘enrichments’ constitute an improvementfrom the animal’s point of view.


Acknowledgments

We are grateful to the RSPCA for providing a granttowards the cost of this study and to Premier Poultry fortheir wonderful cooperation at all stages. Particularthanks go to Andrew Gibson for help with the observa-tions and for answering endless questions and to TomHolland, Anna Miller, Simon Ashworth, Rob Wardilland Mally Homden. Many thanks go to Rhys Greenfor providing the macro to use in the habitat pref-erence analysis. K.M. was supported by a BBSRC M.Sc.studentship.

References

Collias, N. E. & Collias, E. C. 1967. A field study of the red junglefowl in North Central India. Condor, 69, 360–386.

Dawkins, M. S. 1989. Time budgets in red junglefowl as a baselinefor the assessment of welfare in domestic fowl. Applied AnimalBehaviour Science, 24, 77–80.

European Commission 2000. The Welfare of Chickens kept for MeatProduction (Broilers). Report of the Scientific Committee on AnimalHealth and Animal Welfare. http://europa.eu.int/comm/dg24/health/sc/scah/index–en.html.

Green, R. E., Tyler, G. A. & Bowden, C. G. R. 2000. Habitatselection, ranging behaviour and diet of stone curlew (Burhinusoedicnemus) in southern England. Journal of Zoology, 250, 161–184.

Grigor, P. N. & Hughes, B. O. 1993. Does cover affect dispersaland vigilance in free range domestic fowls? In: 4th EuropeanSymposium on Poultry Welfare (Ed. by C. J. Savory & B. O. Hughes),pp. 246–247. Potters Bar: UFAW.

Kells, A. & Dawkins, M. S. 2001. The effect of a ‘freedom food’enrichment on the behaviour of broilers on commercial farms.Animal Welfare, 10, 347–356.

Morris, A. J., Whittingham, M. J., Bradbury, R. B., Wilson, J. D.,Krykos, A., Buckingham, D. L. & Evans, A. D. 2001. Foraginghabitat selection by yellowhammers (Emberiza citrinella) nesting inagriculturally contrasting regions in lowland England. BiologicalConservation, 101, 197–210.

NAG 1993. General Linear Interactive Modelling. Oxford: TheNumerical Algorithms Group Ltd.

Newberry, R. C. 1999. Exploratory behaviour of young domesticfowl. Applied Animal Behaviour Science, 63, 311–321.

Newberry, R. C. & Shackleton, D. M. 1997. Use of visual cover bydomestic fowl: a Venetian blind effect? Animal Behaviour, 54,387–395.

RSPCA 1997. Welfare Standards for Chickens. Horsham: RSPCA.Sokal, R. R. & Rohlf, F. J. 1995. Biometry. New York: W. H. Freeman.Stahl, P., Ruette, S. & Gros, L. 2002. Predation on free-ranging

poultry by mammalian and avian predators: field loss estimates ina French rural area. Mammal Review, 32, 227–234.

Weeks, C. A., Nicol, C. J., Sherwin, C. M. & Kestin, S. C. 1994.Comparison of the behaviour of broiler chickens in indoor andfree-ranging environments. Animal Welfare, 3, 179–192.

Whittingham, M. J., Percival, S. M. & Brown, A. F. 2001. Habitatselection by golden plover Pluvialis apricaria chicks. Basic andApplied Ecology, 2, 177–191.

Appendix

Testing for overall variation in habitat useModel A was constructed to derive maximum-

likelihood estimates of chicken densities (number of birds

per unit area of that habitat) in each habitat relative toone reference habitat, b1 (where the chicken densitycoefficient for reference habitat b1=1). These densitycoefficients were then used to derive a measure of overallvariation (V) in density between all habitat types, usingthe formula:

where bi is the ln density coefficient from model A (seeMethods) for habitat i and wi is either the total number ofobserved chickens, or the expected number of chickensbased on available area (whichever is the greater), tohabitat i from all paddocks. Summation �i is across allsix habitat types. The derived (expected) values werecalculated using the formula:

where ai is the area of habitat i within a paddock, A is thetotal area of that paddock and F is the total number ofchickens counted in all habitats from that paddock.Summation �n is across all paddocks.

Performing a randomization procedure tested the stat-istical significance of V. The labels of the habitats withineach territory were shuffled randomly and model A wasfitted to derive V from the randomized data, which werethen compared with the score from the actual data. Thisprocedure was performed 1000 times and the number ofoccasions on which the randomly derived V scoreexceeded or was equal to the actual V score was obtained.This value, when divided by 1000, gives the probabilitythat nonrandom habitat selection had taken place.

Testing for significant differences in density of usebetween pairs of habitats

In pairwise habitat tests, model A was used to derivemaximum-likelihood estimates of density (observed) inone habitat relative to the second habitat, but only forthose paddocks with both habitats present. A random-ization procedure was then performed to test thestatistical significance of the observed relative density byderiving a randomized maximum-likelihood estimate ofrelative density. This was achieved by randomly shufflingthe labels of the pair of habitats being tested for eachpaddock where both habitats were present (the otherhabitat labels were left unaltered) and refitting model A.This procedure was performed 1000 times and thenumber of occasions on which the randomized relativeforaging density was equal to or exceeded the value forthe observed relative foraging density was obtained. Thisvalue was divided by 1000 to obtain the probability thatforaging density was significantly different between thetwo habitats. This procedure was repeated for all pairwisecombinations of habitats. Habitats were assigned anapproximate ranking score by adding together thenumber of times a habitat was significantly selected over


another habitat (+’s) and subtracting the number oftimes that habitat was avoided relative to another habitat(�’s). Figure 3 gives the results of pairwise habitatcombinations, along with the rank scores.

Pairwise comparisons between habitats were carried outat three times of day (early, mid-day and afternoon), and

so it was likely that the same chickens may have beencounted up to three times. Therefore we adjusted oursignificance level for accepting a difference betweenhabitat types using the Bonferroni correction for threerepeat tests at the 5% level, i.e. P<0.017 (Sokal & Rohlf1995).

What makes free-range broiler chickens range? In situ measurement of habitat preference

Documents

Transcript of What makes free-range broiler chickens range? In situ measurement of habitat preference