HABITAT PREFERENCES OF FERAL AMERICAN MINK IN THE …of rabbit warrens seemed to be strongly affected by riverside farmland management, this might eventually determine the distribution

1356

Journal of Mammalogy, 84(4):1356–1373, 2003

HABITAT PREFERENCES OF FERAL AMERICAN MINK IN THEUPPER THAMES

NOBUYUKI YAMAGUCHI, STEVE RUSHTON, AND DAVID W. MACDONALD*

Wildlife Conservation Research Unit, Department of Zoology, South Parks Road, Oxford OX1 3PS,United Kingdom (NY, DWM)

Center for Life Sciences Modelling, University of Newcastle, Porter Building, Newcastle upon TyneNE1 7R0U, United Kingdom (SR)

Habitat use by members of a wild population of American mink (Mustela vison) wasevaluated by continuous monitoring of individuals that were live trapped and radiotrackedyear round in the Upper Thames region, United Kingdom. Spatially lagged autoregressivemodels were used to investigate the relationship between population abundance and mea-sured habitat variables. Resident mink were found in places characterized by rich tree cover,plenty of scrub, rank grasses, and especially abundant rabbits, and they avoided open habitatcharacterized by farming activities. These trends were not detected, however, in eithertransient adults or juveniles. The presence of the opposite sex did not appear to influencethe presence of resident mink of the other sex. The single most important feature influencingthe presence of resident mink was the size of rabbit warrens. Warrens were, overall, themost important den sites for mink, especially for breeding females. Because the distributionof rabbit warrens seemed to be strongly affected by riverside farmland management, thismight eventually determine the distribution and local population growth of feral mink inthe Upper Thames region.

Key words: Arvicola terrestris, conservation, invasive species, land-use, mink predation, Mustelavison, Oryctolagus cuniculus, riverside habitat

The American mink (Mustela vison),which is native to North America, is theonly widely distributed nonindigenous car-nivore in the United Kingdom, and the po-tential negative impact of its predatory hab-its has led to much debate (Birks and Dun-stone 1991; Clark 1991; Linn and Chanin1978; Macdonald 1995; Macdonald et al.1999). However, in spite of its reputation asa vicious alien killer, and with exceptionsconcerning some negative relations with lo-cal populations of ground nesting sea birds(Craik 1995) and water voles (Arvicola ter-restris—Barreto et al. 1998a, 1998b; Jef-feries et al. 1989; Strachan and Jeffries1993; Woodroff et al. 1990), few studieshave been able to explore the complicated

* Correspondent: [email protected]

relationship between mink predation andother processes in natural habitats (Barretoet al. 1998a, 1998b; Macdonald et al. 1999;Sidorovich et al. 1998). Management ofwildlife populations, whether to preservethreatened species or control pests, requiresan understanding of the species’ habitat re-quirements. The best measure of habitatquality would be a test of its effects on de-mographic parameters, such as populationgrowth and carrying capacity (Garshelis2000). Earlier field studies have related thepattern of space use by mink to vegetationtypes and to the availability and distributionof food and dens (Birks 1981; Birks andLinn 1982; Clode and Macdonald 1995;Dunstone and Birks 1983; Erlinge 1972;Gerell 1970; Halliwell and Macdonald1996; Hatler 1976; Melquist et al. 1981).

November 2003 YAMAGUCHI ET AL.—HABITAT PREFERENCES OF MINK 1357

More recently, detailed quantitative studieshave reported on the habitat preferences ofmink in coastal populations in Scotland andin Argentina (Bonesi et al. 2000; Ireland1988; Previtali et al. 1998). There havebeen no such analyses, however, for river-ine habitats, which are the most frequentlyused noncoastal habitats.

An analytical complication is that thepresence of mink in one area is likely to bedependent not only on the presence of suit-able habitat features there but also on thepresence of such features (and mink them-selves) in adjacent areas. Therefore, the rec-ords of mink in nearby areas are not trulyindependent of each other, and as such, cor-relations between abundance of mink andmeasured habitat features of an area may bespurious. This autocorrelation betweenmink abundance and the lagging of habitatvariables in adjacent areas may complicateanalyses of species–habitat relationships us-ing simple linear models and correlation. Inorder to evaluate species–habitat relation-ships fully, we have specifically modeledthe spatial component introduced by habitatin adjacent sections as well as autocorrela-tion.

In this article, we investigate the habitatpreferences of free-ranging American minkin a riparian system in the Upper Thamesregion, United Kingdom, using data derivedfrom a field survey of habitat characteristicsand an intensive study of space use derivedfrom radiotracking of captured mink. Wethen use the correlation and spatial autore-gression approaches to investigate the re-lationship between mink abundance andhabitat features.

MATERIALS AND METHODS

Study area and habitat survey.—The studyarea consisted of approximately 24 km along theRiver Thames (about 518409N, 18259W) to thewest of Oxford City, Oxfordshire, United King-dom. The river was between 15 and 45 m inwidth and .1 m in depth, providing a hetero-geneous habitat fringed with trees such as wil-low (Salix fragilis). Bands of vegetation, such as

common reed (Phragmites australis) emergedfrom the water in summer. Adjacent land wasmainly pasture, but arable land and woodlandalso occurred, where abundant populations ofrabbit (Oryctolagus cuniculus) were found.

The habitat survey was carried out followingthe guidelines developed by The EnvironmentAgency (Environment Agency, in litt.). The sur-vey recorded 12 habitat variables concerningproportions of the area covered by tree, scrub-grass, grass, water grass, and open field, and ex-istence of public path, human activities, and oth-er water sources (Appendix I). Because rabbitswere the most important food source for minkin the study area (Ferreras and Macdonald1999), size of rabbit warrens was also includedin the survey to assess its effects on mink habitatpreferences. The survey was carried out on bothsides of the river #50 m from the water’s edge.This width was chosen because earlier radi-otracking of mink in the study area revealed thatnone of the animals had gone .50 m from thenearest water source. The survey for major per-manent features (e.g., tree coverage) was carriedout during winter in 1996 and for the comple-mentary seasonal features (e.g., emergent vege-tation) during summer in 1996.

Trapping and radiotracking.—Mink weretrapped in commercial, single-entry aluminummink and rat cage traps of approximately 14 by14 by 76 cm (A. Fenn and Co., Redditch,Worcestershire, United Kingdom). The studyarea was divided into 4 stretches of river, eachof which consisted of 3–6 km between 2 neigh-boring locks. A 4-week cycle of trapping wascontinued throughout the year between May1995 and August 1997. Trapping was conductedin each stretch for 1 week, and a trap was set,on average, every 200–300 m of riverbank. Amink was classified as a kit if it was observedor trapped with its mother before dispersal,which is at approximately 13 weeks of age (ear-ly August); after that, it was classed as a juvenileuntil the onset of the 1st breeding season (Jan-uary), at approximately 8 months of age, andafter that as an adult. Following Hatler (1976)and Ireland (1988), females present for $3 con-secutive months were classified as residents. Be-cause of the reported seasonal change in theirspacing patterns (Birks 1981; Dunstone 1993;Ireland 1988), males were classified as residentsonly if they were in the study area for $3 con-secutive months in the nonbreeding season

1358 Vol. 84, No. 4JOURNAL OF MAMMALOGY

(May–December). Other individuals were clas-sified as transients. Some of the captured ani-mals were fitted with waterproof radiotags at-tached to collars with integral reed switches(Biotrack Ltd., Wareham, Dorset, United King-dom), and followed with receivers (M57, Mari-ner Radar Ltd., Lowestoft, Suffolk, UnitedKingdom) connected to 3-element Yagi antennas(Biotrack Ltd.).

During radiotracking, the location of the focalanimal was recorded every 15 min, mainly be-tween dusk and dawn. When each radiofix wasrecorded, the following additional data were alsologged. Distance of the focal animal from thenearest water: within 10 m (,10 m), 10–50 m,50–100 m, and .100 m; and habitat type wherethe animal was located: rabbit burrow, willowtree, scrub, rank grass, open area, water, others,and unknown.

Den search.—In the United Kingdom, minkin freshwater habitats are predominantly noctur-nal and are inactive in the den by day (Birks andLinn 1982; Dunstone 1993). Opportunistic day-time radiotracking (den searching) revealed an-imals’ dens. When the location of a mink wasdetected by daytime radiotracking, a short blockof radiotracking was continued for the following1 h to confirm that it was stationary. Mink denswere defined as follows: places where minkwere found by den searching during daytime,places where mink were stationary for .2 con-secutive h leading up to sunrise, and placeswhere mink were located with .20 fixes (5 h)in total during the entire radiotracking period. Ineach case, the locations and features of denswere confirmed by subsequent field searches andthe vegetation type of the location was recorded.Dens were classified as breeding dens if a fe-male and dependent kits were observed togetherat the den by early July, or a female with de-veloped nipples returned to the den regularlyduring May—early July. In addition to these, forcases where breeding dens were not confirmed,the dens used regularly from the middle of Aprilonward by pregnant females were classified aspotential breeding dens.

Analysis of mink habitat relationships.—Foursets of analyses were undertaken using correla-tion and regression analysis, investigating the re-lationships between habitat characteristics andmink abundance based on the trapping success,abundance of resident mink, spatial utilization

by radiotracked mink within their ranges, andabundance of mink dens.

The entire study area was considered as a riv-er corridor and was divided into sections, eachof which consisted of a 200-m river stretch: 200m by 50 m areas on both sides of the river. Ex-cept for the analyses based entirely on radio-tracking, which treated the 2 areas along thesame river stretch separately, all analyses werecarried out on the bases of these sections. Home-range size was calculated as a length along theriver corridor on the basis of the number of thesections between the most downstream and themost upstream sections that contained eithercapture points or radiofixes of the individual. Weassumed that the mink was present in all inter-vening sections. The number of resident mink ineach section was calculated by summing thenumbers of all individuals for which the homeranges included the section. The average numberof resident mink per month per section was es-timated from the residence period of each minkin each section. For example, if there was 1mink recorded in a section for 10 months andanother for 5 months, during the 28-month studyperiod, the average number of resident mink permonth for the section is (10 1 5) 4 28 5 0.54.For breeding females, the number of femaleswith kits per breeding season per section wasused. Associations between habitat variables andlevels of preference may be prone to dependenceproblems because habitat variables are often in-tercorrelated (Aebischer et al. 1993; Garshelis,2000). Therefore, we first investigated correla-tions among habitat variables and used principalcomponent analysis (PCA) to exclude possibleredundant variables from further analyses.

We analyzed the relationship between habitatvariables and trapping success using the totalnumber of captures per section. The habitat pref-erence of radiotracked mink was analyzed on thebasis of the number of radiofixes per sectionwithin their home ranges. The relationship be-tween habitat characteristics and the abundanceof dens was analyzed based on the number ofdens per section within the animals’ home rang-es. Correlation tests, Mann–Whitney U-tests andchi-squared tests were used for these analyses.To avoid statistical problems concerning simul-taneous correlation tests, appropriate probabilityvalues were adjusted using the Bonferroni tech-nique (Rice 1989). All these tests were per-


formed using StatView 4.01 (Abacus Concepts,Inc., Berkeley).

The relationships between habitat character-istics and the abundance of resident mink werealso analyzed using correlation tests, Mann–Whitney U-tests, and chi-squared tests. Howev-er, the relationships were further investigated us-ing regression modeling.

While individual 200-m sections of river pro-vide a convenient unit of sampling for habitatfeatures such as vegetation and space use bymink, this unit does not necessarily relate to thescale of space use by the mink themselves. Onthis basis, it is unlikely that individual 200-mriver sections were used solely by 1 mink; in-deed, records in multiple sections may derivefrom the same mink. This means that recordedspace use by mink in any 1 section will reflecta range of processes. The abundance of mink islikely to be determined not only by the presenceof suitable habitat features in that section butalso by the abundance of mink in other sections(an autoregressive response) as well as by thedistribution of such habitat features in other sec-tions (a spatially lagged predictor variable). Inorder to evaluate the impact of these variableson the presence of mink abundance in adjacentsections and the presence of mink in the focalsection, we used spatial autoregression. Thismodel includes both spatially lagged predictorvariables as well as an autoregressive compo-nent.

The full model is described in its matrix formas follows: Y 5 rWY 1 Xb 2 WXg 1 u, whereY is the response variable, number of mink persection; X is a matrix of predictor variables(habitat features, rabbits); W is a weighting ma-trix representing the effects of spatial separationof the river sections; u is the vector of errors,and r, b, and g have to be estimated. Note thatthe response variable Y appears on both sides ofthe equation and the product rWY (the autore-gressive component) effectively quantifies thespatial effects of mink in adjacent sections.When explained in biological terms, Xb reflectsthe impacts of habitat features in the sectionwhere the mink abundance is estimated. Theterm WXg is the spatially lagged term for in-dependent predictor variables and reflects the ef-fects of habitat variables in all of the other sec-tions surrounding the section where the minkabundance is estimated. The term rWY is theautoregressive predictor and represents the im-

pacts of mink in other sections on records in thefocal section. W, in both the spatially laggedhabitat features and autoregressive terms, is amatrix that gives the weight of each of the sec-tions in the sample. Setting an entry in this ma-trix to 0 means that the habitat features and minkrecords it contains have no effect on the focalsection. Variation in the number of nonzero el-ements in this weighting matrix allows the in-vestigation of the importance of the autoregres-sive and spatially lagged variables on the re-sponse of interest.

Evaluation of spatial autoregressive modelsrelating the abundance of mink to habitat vari-ables, to mink in adjacent river sections, and tohabitat variables in adjacent sections was under-taken using the SPACESTATPACK package de-veloped by Pace et al. (1997) using the data seton the basis of the 200-m section. As with allspatially lagged regression models, the user hasto define the form of weighting matrix and quan-tify the relative impacts that adjacent sites con-tribute to the model. Because there were no spe-cific data available on the extent to which therecords of mink and presence of habitat featuresin adjacent river sections could influence thepresence of mink in a section, models were fittedover a range of neighborhoods with 2, 4, 6, or8 nearest river sections contributing (those hav-ing nonzero entries) to the weighting matrix W.The weight given to each section was assumedto be equal among all neighbors. Models werefitted with 3 predictor variables. These were 2summary habitat variables that were principalcomponent scores derived from a PCA of thehabitat features in each river section, and a 3rdordinal variable quantifying the relative abun-dance of rabbits in each river section. The dataon habitat features were summarized by PCA for2 reasons: first, to overcome the unit-sum con-straint imposed by the use of areas of individualhabitat features within a river section, and sec-ond, to provide a reduced set of predictor vari-ables for the regression analysis because the useof spatially lagged variables doubles the effec-tive number of predictors in each model leadingto model overparameterization. The modelswere estimated using maximum likelihood.Likelihood ratio tests were used to assess thesignificance of the parameter estimates of coef-ficients of individual lagged explanatory vari-ables and the autoregressive parameter betweenmodels and the Akaike information criterion was


used to identify the best model among the suiteof potential models. Three sets of analyses wereundertaken, 1st with the overall abundance ofresident mink of both sexes combined in eachsection and then with the records of single sexeswith the incorporation of the abundance of theopposite sex as a predictor and spatially laggedpredictor. These last 2 models were investigatedin order to assess the extent to which the ob-served abundance of one sex was influenced bythe abundance of the other. Analyses for tran-sient individuals were not undertaken becausepreliminary spatial analyses indicated that theseanimals were not distributed in relation to anyhabitat variables. Models were fitted in a step-wise fashion starting with the simplest, ordinaryleast squares, followed by models with spatiallylagged predictors, and finally with the full modelwith autoregressive components.

RESULTS

Mink.—Fifty-one mink were captured atotal of 184 times during 4,336 trap nightsbetween May 1995 and August 1997—anaverage of 23.6 trap-nights for each cap-ture—consisting of 27 males (10 residentsadults, 11 transient adults, 3 unclassifiedadults, 4 juveniles, and 2 kits) and 24 fe-males (13 residents adults, 3 transientadults, 1 unclassified adult, 11 juveniles,and 6 kits). Birks and Linn (1982) reportedthat mink radiotracked at least twice a dayrevealed more than 80% of their total homeranges within 5 days and their entire homeranges within 10 days. We included onlyindividuals that were radiotracked for either.10 days or intensively (followed, on av-erage, for .20 h a day) for .5 days duringthe tracking periods. Out of a total of 24mink to which radiotransmitters were fitted,these criteria were met by 13 animals:among these was 1 female whose range laylargely outside the study area and was ex-cluded from most analyses.

Habitat features, PCA, and mink trap-pability.—There were highly significantnegative correlations between variablespositively related to the existence of openfield and most other variables that associ-ated with natural and seminatural vegeta-

tion (Appendix II). In general, total cap-tures of juveniles and transient individualswere not correlated with habitat variables.On the other hand, total captures of residentindividuals, both males and females, weresignificantly correlated with some habitatvariables (Table 1). The number of capturesof kits was significantly and positively cor-related only with the size of rabbit warrens(Table 1). The 1st axis of the PCA was as-sociated with the existence of natural andseminatural vegetation (Fig. 1). The 2ndaxis was associated with the existence offeatures taller than rank grass.

Habitat features and the presence of res-ident mink.—In general, the number of res-ident mink was positively correlated withthe area covered by trees, scrub, and rankgrasses, and negatively with the open area(Table 2). Habitat characteristics along theedge of water alone, including the emergentvegetation, did not strongly influence thepresence of resident mink (Table 2). Themost important single habitat variable sig-nificantly and positively correlated with thenumber of resident mink was the size ofrabbit warrens. This was especially so forfemales with dependent young. The numberof females with kits was negatively asso-ciated with the presence of public footpaths(Mann–Whitney U-test; n 5 104 and 8, U5 194, P 5 0.0019) and positively relatedto the presence of other water sources near-by (,100 m; Mann–Whitney U-test: n 578 and 34, U 5 922.5, P 5 0.0016). Thenumber of resident females was negativelyrelated to the presence of human activities(Mann–Whitney U-test: n 5 91 and 21, U5 697, P 5 0.044). None of these 3 vari-ables had a significant relationship with thepresence of resident males (Mann–WhitneyU-test: P . 0.05). Also, there was no sig-nificant association between any 2 of these3 variables (chi-squared test, P . 0.5). Acorrelation between the PCA scores for riv-er sections with the number of residentmink was significant for the 1st axis of thePCA only (1st axis: r 5 20.442, P ,0.0001; 2nd axis: r 5 20.148, P 5 0.11).


TA

BL

E1.

—C

orre

lati

onbe

twee

nha

bita

tva

riab

les

and

the

num

ber

ofca

ptur

esof

min

kin

the

sect

ion,

base

don

tota

lca

ptur

e,re

side

ntm

ales

,re

side

ntfe

mal

es,

tran

sien

tm

ales

,tr

ansi

ent

fem

ales

,ju

veni

lem

ales

,ju

veni

lefe

mal

es,

and

kits

(Ken

dall

rank

corr

elat

ion’

st)

.S

tati

stic

ally

sign

ifica

ntco

rrel

atio

nsw

ere

dete

cted

byK

enda

llra

nkco

rrel

atio

nte

sts

(n5

64se

ctio

ns)

wit

hap

prop

riat

epr

obab

ilit

yva

lues

adju

sted

usin

gth

eB

onfe

rron

ite

chni

que

(Ric

e19

89).

Hab

itat

vari

able

Min

kca

tego

ry

Tot

alca

ptur

eR

esid

ent

mal

eR

esid

ent

fem

ale

Tra

nsie

ntm

ale

Tra

nsie

ntfe

mal

eJu

veni

lem

ale

Juve

nile

fem

ale

Kit

Veg

etat

ion

area

Tre

esS

crub

Ran

kgr

ass

Ope

n

0.33

7**

0.25

9*0.

179

20.

266*

*

0.24

2*0.

105

0.16

32

0.25

0*

0.25

8*0.

278*

*0.

271*

*2

0.24

6*

0.04

62

0.05

12

0.03

12

0.00

1

0.12

00.

059

20.

079

20.

048

0.04

82

0.03

70.

048

20.

032

0.24

1*2

0.00

20.

057

20.

184

20.

016

0.18

20.

103

20.

016

Veg

etat

ion

alon

gth

eed

geof

wat

er

Tre

esS

crub

0.33

0**

0.10

90.

232*

0.02

70.

175

20.

038

0.10

00.

188

0.24

0*0.

205

0.16

90.

014

0.11

20.

136

20.

148

20.

052

Oth

erva

riab

les

Em

erge

ntve

geta

tion

Rab

bit

war

rens

20.

037

0.21

3*0.

083

0.26

0*2

0.05

10.

328*

*2

0.08

92

0.05

20.

028

20.

058

20.

033

20.

062

20.

150

20.

095

20.

080

0.23

7*

*P

,0.

05,

**P

,0.

01.


FIG. 1.—Factor plot of habitat variables alongthe first 2 factors, which explain 52.7% of theoriginal variance, of a principal component anal-ysis (200-m section). The shorter the distancebetween 2 variables, the more closely correlatedthey are. Abbreviations indicate habitat variablesdescribed in Appendix I: A 5 area; B 5 lengthalong edge of water; T 5 trees; S 5 scrub-grass;G 5 grass; O 5 open; and R 5 rabbits.

TABLE 2.—Correlation between habitat variables and the number of mink present in the section.Statistically significant correlations were detected by ANOVA (n 5 112 sections) with appropriateprobability values adjusted using the Bonferroni technique (Rice 1989).

Habitat variable Resident males Resident females Females with kits

Vegetation area

TreesScrubRank grassOpen

0.441**0.1030.450**

20.519**

0.409**0.1590.377*

20.479**

20.0160.174

20.00120.053

Vegetation along the edge of water

TreesScrub

0.21420.071

0.18720.089

0.1320.207

Other variables

Emergent vegetationRabbit warrens

0.2250.469**

0.2290.462**

0.1780.273*

* P , 0.05, ** P , 0.01.

Spatially lagged and autoregressivemodels.—An ordinary least squares analy-sis relating the abundance of mink in riversections to the 1st and 2nd axes of the prin-cipal components scores, size of rabbit war-rens, and the spatially lagged habitat vari-ables showed that only 2 variables, the 1stPCA axis scores and size of rabbit warrens,were significant predictors of mink (Appen-dix III). Comparison of the log likelihoodsfor the ordinary least squares analysis mod-

els based on the PCA scores and size ofrabbit warrens with no spatial laggingshowed that inclusion of the 2nd axis of thehabitat PCA did not decrease the log like-lihood significantly (Appendix IV). This isin agreement with the preliminary linearanalyses of mink resident numbers withboth PCA habitat variables. In general, in-clusion of spatially lagged variables, bothaxes scores from the PCA, or an autore-gressive component for mink in neighbor-ing sections decreased the log likelihoodsignificantly (Appendix IV). Also, likeli-hood ratio tests comparing the autoregres-sive models with the equivalent ordinaryleast squares analysis models with spatiallylagged and simple nonspatial models weresignificant for all neighborhood ranges (Ap-pendix IV). The Akaike information crite-rion was smallest for the models fitted for4 nearest neighbors with both the PCAaxes, size of rabbit warrens, the spatial lagfor these variables, and the autoregressivecomponent for neighboring resident minkindicating that this (400 m on each side ofthe focal section) was the spatial range overwhich the spatially lagged variables had in-fluence on the mink abundance in any oneriver section (Appendix V). Analyses com-paring the full autoregressive model includ-ing spatially lagged habitat variables with


TABLE 3.—Correlation between habitat variables and the numbers of radiofixes in the section of12 radiotracked individuals (correlation coefficient or Kendall rank correlation’s t). ‘‘Length vari-ables’’ means vegetation along the edge of water. ‘‘All’’ means all individuals combined. Animals01, 02, 03, 04, and 05 are males, and 06, 07, 08, 09, 10, 11, and 12 are females. Sample size isnumbers of 200-m river section within each animal’s home range.

Habitat variable

MinkID

Samplesize

Vegetation area

Tree Scrub Rank grass Open

Length variables

Tree Scrub

Other variables

Emergentvegetation

Rabbitwarrens

010203040506070809101112

56764046524037

820213736

0.38*0.060.210.48*0.190.200.210.000.51*0.49*0.29

20.02

0.30*0.27*0.200.36*0.100.37*0.18

20.510.120.340.46*

20.00

0.25*0.210.040.31*

20.000.080.16

20.470.260.030.18

20.31

20.36*20.1820.1420.43*20.1220.1520.19

0.5320.4220.3720.34*

0.04

0.39*20.03

0.200.38*0.200.140.220.000.45*0.190.41*0.24

0.2020.00

0.040.240.120.060.27

20.6420.17

0.130.220.19

20.100.13

20.1320.16

0.0420.29

0.0320.1220.2420.3320.1920.11

0.22*0.25*0.41*0.33*0.130.47*0.250.200.190.110.39*0.05

All 173 0.05 0.24* 0.19* 20.22* 0.25* 0.03 0.01 0.46*

* P , 0.05.

an autoregressive model without spatiallylagged habitat features are also shown inAppendix IV. The models shown are forboth PCA habitat axes and the rabbit vari-able and for the 1st PCA axis and rabbitvariables for each of the 4 neighborhooddistances. For the analyses with 2 PCA hab-itat variables, there was a significant differ-ence between the simple autoregressivemodel without spatially lagged variablesand the models that included lagging, withmodels including lagged variables having asmaller maximum likelihood than thosebased on an autoregressive predictor alone.For the single PCA model, only the 2- and4-neighbor models were different. Theanalyses show that inclusion of spatiallylagged habitat variables with an autoregres-sive model explained more of the variationin the mink abundance data than did theautoregressive component alone. In otherwords, habitats in river sections were animportant predictor in their own right (i.e.,over and above an autoregressive compo-nent for adjacent mink). When the analyses

were repeated for the individual sexes, us-ing the abundance of the opposite sex as apredictor (both lagged and nonlagged),abundance of the opposite sex was not asignificant predictor in the models. Thissuggests that the presence of male and fe-male mink did not impact on the abundanceof animals of the opposite sex in river sec-tions.

Habitat use inside the home range.—There were individual differences in habitatuse; however, in general, mink used the sec-tions that had larger scrub–grass-coveredarea, had larger rank-grass area, had bankwith more tree cover, and had bigger rabbitwarrens (Table 3). They tended to avoidopen areas. Radiotracked mink, both malesand females, stayed within 10 m of thenearest water source most of the time (88%for males and 95% for females); however,compared with females, males tended to befound further from the water (more than 10m from the nearest water source) on moreoccasions, and significantly so (chi-squaredtest: d.f. 5 1, x2 5 265.6, P , 0.0001).


←

FIG. 2.—(top pie chart) Vegetative cover inhome ranges of radiotracked mink, and (middleand bottom pie charts) habitat types where ra-diotracked mink were found.

Although, on average, nearly 60% oftheir home ranges were covered by openareas, radiotracked animals were neverfound in these areas (Fig. 2). Males weremore often found in rabbit warrens and fe-males more often in willow trees and inrank grass (rabbit burrow, willow tree,scrub, grass, and the rest combined; chi-squared test: d.f. 5 4, x2 5 550.7, P ,0.0001)

Habitat features and dens.—Altogether,119 dens of 13 radiotracked individualswere found in the study area. The presenceof a den was significantly and positivelycorrelated with the size of rabbit warrens(Table 4). Dens of radiotracked individualswere exclusively found ,50 m from waterand most of them were found ,10 m fromit (84% for males and 98% for females).Females had their dens closer to water thandid males (for ,10 m category; Mann–Whitney U-test: n 5 5 and 8, U 5 7, P 50.039). In total, there were 8 dens recordedin the area .10 m from the water. Amongthem, 5 were in rabbit warrens, 2 were inscrub, and 1 was under a bridge. The radio-tracked animals had no den in open habitat(Fig. 3). Rabbit warrens were the favoriteden site for both sexes (Fig. 3). Male denswere more often found in rabbit warrensand scrub areas, female dens in rank grass(rabbit burrow, willow tree, scrub, grass,and the rest combined; chi-squared test: d.f.5 4, x2 5 9.64, P 5 0.047). Out of the totalof 77 dens detected in the home ranges of10 females (8 females successfully radio-tracked and 2 other females that were ra-diotracked only briefly), 5 dens were con-firmed to be used when they had dependentkits. Females significantly preferred rabbitwarrens for kit-rearing dens (chi-squaredtest: d.f. 5 1, x2 5 12.60, P 5 0.0004).


TABLE 4.—Correlation between habitat variables and the numbers of dens of 12 radiotrackedindividuals (correlation coefficient or Kendall rank correlation’s t). Details concerning column head-ings are presented in Table 3.

Habitat variable

MinkID

Samplesize

Vegetation area

Tree Scrub Rank grass Open

Length variables

Tree Scrub

Other variables

Emergentvegetation

Rabbitwarrens

010203040506070809101112

56764046524037

820213736

0.44*0.150.090.09

20.060.120.27

20.410.310.360.16

20.22

0.25*0.26*0.240.120.070.37*0.16

20.450.010.430.23

20.02

0.26*0.200.080.160.180.190.030.000.300.16

20.0820.40*

20.38*20.20

0.0020.1220.1620.1520.14

0.2920.2620.3820.12

0.17

0.55*0.000.21

20.000.080.200.25

20.330.270.170.29

20.05

0.0920.0920.06

0.220.03

20.010.15

20.3220.24

0.1320.07

0.31

20.140.070.12

20.0820.0920.23

0.1020.0720.2420.3020.35*20.20

0.25*0.30*0.39*0.34*0.30*0.60*0.36*0.09

20.120.300.35*

20.01

All 173 0.12 0.24* 0.13 20.23* 0.38* 20.01 20.09 0.44*

Furthermore, there were 3 other dens thatwere almost certainly used for rearing de-pendent kits. They were also all in rabbitwarrens. If these 3 dens are included asbreeding dens, females’ preference of rabbitwarrens as breeding dens becomes moresignificant (chi-squared test: d.f. 5 1, x2 521.00, P , 0.0001).

DISCUSSION

The presence of mink and habitat fea-tures.—The trapping results suggest thepresence of resident adult mink was strong-ly associated with habitat features: positive-ly with the size of rabbit warrens and theareas covered by trees, scrub, and rankgrass and negatively with the area coveredby open habitat. Resident individuals,which are considered to defend territories(Birks 1981; Gerell 1969, 1970; Ireland1988), occupy their ranges for a prolongedperiod. They may need particular habitatfeatures to survive and, in the case ofbreeding females, to breed, whereas tran-sient individuals, which stay in one area fora short period, may have less demandingrequirements. Indeed, the trapping resultssuggest that the distribution of transient

adults and juveniles in the study area is notinfluenced by the habitat features as strong-ly as is that of residents. Alternatively, tran-sients may move on because suitable habi-tat is unavailable, being occupied alreadyby residents. Most transient adult males re-corded, however, were passing through thestudy area during the breeding season, andthere is no evidence that resident males aredominant over transients at that time. Onthe contrary, reproductively successfulmales may abandon their territories andtravel in search of females (Birks 1981; Ire-land 1988). Under such circumstance, thehabitat requirement of transient adult malesmay be different from those of residentmales.

Radiotracking revealed that, in general,within their ranges, mink prefer the sectionswith more tree cover, more scrub-grass,more rank grass, and bigger rabbit warrensand they avoid the sections with more openareas (Table 3). Not a single radiofix out ofthe total of 5,152 fixes was recorded in theopen areas that, on average, comprisednearly 60% of the home ranges of the ra-diotracked individuals (see Fig. 3). Radio-tracking indicates that, on average, com-


FIG. 3.—Habitat types where dens of radio-tracked mink were found.

pared with females, males spent more timein rabbit warrens and in scrub (see Fig. 2).The size of a rabbit warren had a significantpositive relationship with the area coveredby scrub-grass, and furthermore, these twovariables were very closely associated inthe factor plot of habitat variables on thebasis of the PCA (see Fig. 1). Male minks’strong associations with rabbit warrens andscrub may be related to their prey selection,as males are reported to hunt rabbits more

than do females (Birks 1981; Birks andDunstone 1985; Ireland 1988).

Mink den and habitat features.—In ourstudy area, most mink dens were found,10 m from the water and, on average,males had a significantly greater proportionof dens further from the water than did fe-males. Ireland (1988) made similar obser-vations on coastal feral mink in Scotlandand argued that this difference between thesexes indicated the greater importance ofrabbits as prey for males than females.Birks (1981) also found that mink movedto a rabbit-rich area, away from the river,when aquatic prey became scarce. In ourThames study area, although some rabbitwarrens were ,10 m from the water, theother main potential den sites (all in hollowwillow trees) were at the water’s edge. Fiveout of 8 dens recorded in the areas .10 mfrom the water were in rabbit warrens. Weargue, therefore, that their preference forrabbit as a prey results in the males’ ten-dency to use dens further from the waterthan do females. Inside their ranges, radio-tracked individuals denned preferentially insome sections (Table 4). The strongest cor-ollary of den site was the presence of rabbitwarrens, which comprised 42% of all re-corded males’ dens and 34% of females’(see Fig. 3). Ireland (1988) reported thesame trend in a coastal habitat in Scotland(65% for males’ dens in warrens and 43%for females). Hollow willow trees on theriver bank were also important den sites formink in the Upper Thames region (Fig 3;Halliwell and Macdonald 1996), as else-where (Birks 1981; Gerell 1970).

It has been suggested that mink restricttheir foraging to the vicinity of dens andhence select dens that are close to preferredforaging areas or concentrations of preyitems (Birks 1981; Birks and Linn 1982;Dunstone 1993; Ireland 1988). In the UpperThames study area, rabbits comprise 43%of the estimated ingested energy for mink,much higher than the second-most-impor-tant prey, fish, which comprises 27% (Fer-reras and Macdonald, 1999). The use of


warrens as dens obviously fits well with thepredominance of rabbits in mink diet, es-pecially for breeding females. Breedingdens may be used for up to 40 days and thefemale mink’s prey requirement might rise5-fold while she is rearing young (Dunstone1993). Furthermore, the time taken to tendkits and the need to guard them may detractfrom foraging opportunities and furtherconcentrate the mother’s hunting activitiesclose to the den. The opportunity to huntabundant rabbits in a large warren close toher den therefore fulfills important require-ments for breeding female mink. The num-ber of kits captured was strongly influencedonly by the size of rabbit warrens, suggest-ing they are kept close to big rabbit war-rens, which are used as dens by the moth-ers. The size and distribution of rabbit war-rens may be one of the most important hab-itat factors affecting long-term minkpopulation changes in the Upper Thamesregion.

Modeling the relationship between minkabundance and habitat features.—Spatialmodels with autoregressive components arecomputationally difficult because suitablealgorithms for analyzing data sets have notbeen widely available. Furthermore, theneed for information describing the abun-dance of species and their habitats and theirrelative juxtaposition means that the datasets for such models are usually larger thanthose used in simple spatial linear modelingapproaches. In addition, there are consid-erable problems with using autoregressivemodels predictively, so their application inan applied setting is restricted. While Au-gustin et al. (1996) developed an autologis-tic approach for modeling the incidence ofspecies in whole landscapes using incom-plete census data based on the Gibbs sam-pler, autoregressive approaches have notbeen used extensively in modeling specieshabitat relationships. Our rationale for us-ing the spatial autoregressive approach herewas less to quantify the role of space andautocorrelation in determining the abun-dance of mink but rather was to assess to

what extent the relationships between spe-cies and habitat characteristics were evi-dent, after allowing for the effects of anyspatial lagging in habitat variables and au-tocorrelation in the abundance data. To thatend, inclusion of spatially lagged predictorvariables and an autoregressive componentgreatly increased the variation in minkabundance on river sections explained bythe models. The results suggest that thenumber of resident mink in a river sectionis dependent on extent of natural vegetationin the section, rabbit abundance in the sec-tion, natural vegetation, and rabbits in ad-jacent sections up to 400 m away, and alsonumber of mink present in adjacent sec-tions. Considering the autoregressive factorfirst, positive autoregressive responsescould be explained in terms of the attractionof conspecifics, but we reject this hypoth-esis because there has been no reported ev-idence suggesting that mink are attracted toeach other in this way (Dunstone 1993).Second, the spatial disposition of animalranges is often dependent on the sex andsexual status of individuals, with male an-imals often forming ranges encompassingthose of females (Sandell 1989). This ex-planation does not seem feasible given that,in the single-sex models, the number of res-ident mink in a section was not dependenton numbers of other mink of either thesame or opposite sex. The explanation forthe significant autoregressive response issimpler, in that mink are likely to rangeover more than the 200 m representing ariver section we used for analyses (Birks1981; Ireland 1988). We might expect theresident mink in any one section to com-prise part of the measured response of minkabundance in other sections, having result-ed in the inclusion of mink in adjacent sec-tions as an autoregressive predictor reflectsthis (i.e., the record in a section comprisespart of the record of an adjacent section).The fact that the best model was derivedwhen 4 neighbors were used suggests thatthe impact of space and the presence ofmink declines above 400 m from each sec-


tion. This would suggest that the abundanceof mink in any 1 section is determined tosome extent by the habitat features 400 mon either side. The most obvious biologicalexplanation for this is that mink have homeranges of the order of 1 km, but this is con-siderably less than that recorded in the riversystem, where animals had ranges in excessof 2 km (and up to about 8 km; Yamaguchiand Macdonald 2003). However, it may beworth mentioning that the smallest minkrange in the study site was about 1 km, sug-gesting that, to determine the minimumrange for a mink in our study site, this ordermay have an important meaning. It is pos-sible that other factors besides home-rangesize influence the models; what these are isdifficult to hypothesize. One possible ex-planation is that mink ranges overlap, as de-scribed above. Alternatively, it may be re-lated to the fact that different predictor var-iables had effects on mink abundance at dif-ferent spatial lags, the best fit model with400-m influence then representing an av-eraging of the effects of the individual var-iables, or that a more coarse scale, greaterthan 200 m, was appropriate for studyingthese relationships. Nonetheless, the factthat inclusion of habitat variables in neigh-boring sections increased the variation inabundance of mink explained over andabove a model with no spatial lagging butincluding an autoregressive component,suggests that the there was indeed a spatialhabitat effect. Unfortunately, the algorithmused in the present study assumed that allvariables would have influence over thesame spatial lag, and this may be biologi-cally unrealistic. Disentangling these differ-ent scaling issues with these sorts of modelsis difficult because the solution of spatialmodels with different spatial lags is prob-ably mathematically intractable.

The results suggest that the use of theautoregressive components in the modelseffectively allows us to identify at whatspatial scale mink abundance in adjacentsections have influence and may hence sug-gest the spatial scale at which habitat mink

relationships should be studied. Therefore,although the results also suggest that sim-pler univariate analyses, after all, may beuseful for investigating species’ habitatpreferences in this case, the autoregressivemodels contribute greatly to understandingthe spatial organization of the species in ri-parian systems.

ACKNOWLEDGMENTS

We gratefully acknowledge the funding ofboth the Environment Agency and the Peoples’Trust for Endangered Species. In particular, wethank A. Driver for his support. We thank G.Berry, J. Bowen, O. Burman, and A. Grogan fortheir field assistance during the habitat surveyand P. Johnson for his statistical advice. We alsothank F. Tattersall, L. Bonesi, M. Thom, H.Kruuk, and two anonymous referees for theirhelpful comments on this article.

LITERATURE CITEDAEBISCHER, N. J., P. A. ROBERTSO, AND R. E. KENWARD.

1993. Compositional analysis of habitat use from an-imal radiotracking data. Ecology 74:1313–1325.

AUGUSTIN, N. H., M. A. MUGGLESTONE, AND S. T.BUCKLAND. 1996. An autologistic model for the spa-tial distribution of wildlife. Journal of Applied Ecol-ogy 33:339–348.

BARRETO, G. R., D. W. MACDONALD, AND R. STRACHAN.1998a. The tightrope hypothesis: an explanation forplummeting water vole numbers in the Thamescatchment. Pp. 311–327 in United Kingdom flood-plains (R. G. Bailey, P. V. Jos, and B. R. Sherwood,eds.). Westbury Academic and Scientific Publishing,Otley, United Kingdom.

BARRETO, G. R., S. P. RUSHTON, R. STRACHAN, AND D.W. MACDONALD. 1998b. The role of habitat andmink predation in determining the status and distri-bution of declining populations of water voles inEngland. Animal Conservation 1:129–137.

BIRKS, J. D. S. 1981. Home range and territorial be-haviour of the feral mink (Mustela vison Schreber)in Devon. Ph.D. dissertation, Exeter University, Ex-eter, United Kingdom.

BIRKS, J. D. S., AND N. DUNSTONE. 1985. Sex-relateddifferences in the diet of the mink Mustela vison.Holarctic Ecology 8:245–252

BIRKS, J. D. S., AND N. DUNSTONE. 1991. Mink. Pp.406–415 in The handbook of British mammals. 3rded. (G. B. Corbet and S. Harris, eds.). BlackwellScientific Publications, Oxford, United Kingdom.

BIRKS, J. D. S., AND I. J. LINN. 1982. Studies on thehome range of feral mink (Mustela vison). Symposiaof the Zoological Society of London 49:231–257.

BONESI, L., N. DUNSTONE, AND M. O’CONNELL. 2000.Winter selection of habitats within intertidal forag-ing areas by mink (Mustela vison). Journal of Zo-ology (London) 250:419–424.


CLARK, H. 1991. Mink hunting on the Nene. Interna-tional Fieldsports and Conservation September–Oc-tober 1991:76–78.

CLODE, D., AND D. W. MACDONALD. 1995. Evidencefor food competition between mink (Mustela vison)and otter (Lutra lutra) on Scottish island. Journal ofZoology (London) 237:435–444.

CRAIK, J. C. A. 1995. Effects of North American minkon the breeding success of terns and smaller gulls inwest Scotland. Seabird 17:3–11.

DUNSTONE, N. 1993. The mink. T. and A. D. PoyserLtd., London, United Kingdom.

DUNSTONE, N. AND J. D. S. BIRKS. 1983. Activity bud-get and habitat usage by coast-living mink (Mustelavison). Acta Zoologica Fennica 174:189–191.

ERLINGE, S. 1972. Interspecific relations between otter(Lutra lutra) and mink (Mustela vison) in Sweden.Oikos 23:327–335.

FERRERAS, P., AND D. W. MACDONALD. 1999. The im-pact of American mink Mustela vison on water birdsin the upper Thames. Journal of Applied Ecology36:701–709.

GARSHELIS, D. L. 2000. Delusions in habitat evalua-tion: measuring use, selection, and importance. Pp.111–164 in Research techniques in animal ecology:controversies and consequences (L. Boitani and T.K. Fuller, eds.). Columbia University Press, NewYork.

GERELL, R. 1969. Activity patterns of the mink Mus-tela vison Schreber in southern Sweden. Oikos 20:451–460.

GERELL, R. 1970. Home ranges and movements of themink Mustela vison Schreber in southern Sweden.Oikos 21:160–173.

HALLIWELL, E. C., AND D. W. MACDONALD. 1996.American mink Mustela vison in the Upper Thamescatchment: relationship with selected prey speciesand den availability. Biological Conservation 76:51–56.

HATLER, D. F. 1976. The coastal mink on VancouverIsland, British Columbia. Ph.D. dissertation, Uni-versity of British Columbia, Vancouver, British Co-lumbia, Canada.

IRELAND, M. C. 1988. The behaviour and ecology ofthe American mink (Mustela vison Schreber) in acoastal habitat. Ph.D. dissertation, Durham Univer-sity, Durham, United Kingdom.

JEFFERIES, D. J., P. A. MORRIS, AND J. E. MULLENEUX.1989. An enquiry into the changing status of thewater vole Arvicola terrestris in Britain. MammalReview 19:111–131.

LINN, I., AND P. CHANIN. 1978. Are mink really pestsin Britain? New Scientist 77:560–562.

MACDONALD, D. W. 1995. European mammals: evo-lution and behaviour. Harper Collins, London, Unit-ed Kingdom.

MACDONALD, D. W., ET AL. 1999. The impact of Amer-ican mink, Mustela vison, as predators of native spe-cies in British freshwater systems. Pp. 5–23 in Ad-vances in vertebrate pest management (D. P. Cow-and and C. J. Feare, eds.). Filander Verlag, Fürth,Germany.

MELQUIST, W. E., J. S. WHITMAN, AND M. G. HOR-NOCKER. 1981. Resource partitioning and co-exis-tence of sympatric mink and river otter populations.Proceedings of the Worldwide Furbearer Conference1:187–220.

PACE, R., R. KELLEY, AND R. BARRY. 1997. Quick com-putations of regressions with a spatially autoregres-sive dependent variable. Geographical Analysis 29:232–247.

PREVITALI, A., M. H. CASSINI, AND D. W. MACDONALD.1998. Habitat use and diet of the American mink(Mustela vison) in Argentinean Patagonia. Journal ofZoology (London) 246:482–486.

RICE, W. R. 1989. Analysing tables of statistical tests.Evolution 43:223–225.

SANDELL, M. 1989. The mating tactics and spacing pat-terns of solitary carnivores. Pp. 164–182 in Carni-vore behaviour, ecology, and evolution (J. L. Gittle-man, ed.). Chapman & Hall, London, United King-dom.

SIDOROVICH, V., H. KRUUK, D. W. MACDONALD, AND T.MARAN. 1998. Diet of semi-aquatic carnivores innorthern Belarus with implications for populationchanges. Symposia of the Zoological Society ofLondon 71:177–187.

STRACHAN, R., AND D. J. JEFFERIES. 1993. The watervole Arvicola terrestris in Britain 1989–1990: itsdistribution and changing status. The Vincent Wild-life Trust, London, United Kingdom.

WOODROFF, G. L., J. H. LAWTON, AND W. L. DAVIDSON.1990. The impact of feral mink Mustela vison onwater vole Arvicola terrestris in the North YorkshireMoors National Park. Biological Conservation 51:49–62.

YAMAGUCHI, N., AND D. W. MACDONALD. 2003. Theburden of co-occupancy: intraspecific resource com-petition and spacing patterns in American mink,Mustela vison. Journal of Mammalogy 84:1341–1355.

Submitted 12 June 2002. Accepted 27 November 2002.

Associate Editor was John G. Kie.


APPENDIX I.

Habitat variables recorded.

Variable Description

Vegetation area As percentage of the area covered by the following vegetation in the sectionTrees Area covered by trees . about 10 m in height

Scrub-grass Area covered by scrub (including trees ,10 m in height) and rank grass

GrassOpen

Area covered by rank grass onlyPasture, agricultural field, or bare soil

Vegetation along the edgeof water

As percentage of the length covered by the following vegetation along the wateredge (within 5 m from the water) in the section

Trees Length covered by trees . about 10 m in height

Scrub-grass Length covered by scrub (including trees ,10 m in height) and rank grass

GrassOpenWater grass

Length covered by rank grass onlyLength covered by pasture, agricultural field, or bare soilLength covered by emergent vegetation

Other variables Variables assessed as present, absent, or scored

Public path Present or absent in the sectionHuman activities Regular human presence (e.g., houses) present or absent in the sectionOther water sources Presence of any permanent water source outside of the section ,100 m from

any edge of the sectionRabbits Rabbit abundance estimated by warren size: scored between 0 (no warren) and 3

(largest warren) in the section


AP

PE

ND

IXII

.

Mat

rix

ofco

rrel

atio

nco

effi

cien

tsfo

rha

bita

tva

riab

les.

Sta

tist

ical

lysi

gnifi

cant

corr

elat

ions

wer

ede

tect

edus

ing

AN

OV

A(n

511

5se

ctio

ns)

wit

hap

prop

riat

epr

obab

ilit

yva

lues

adju

sted

usin

gth

eB

onfe

rron

ite

chni

que

(Ric

e19

89).

Veg

etat

ion

area

Tre

eS

crub

Ran

kgr

ass

Ope

n

Veg

etat

ion

alon

gth

eed

geof

wat

er

Tre

eS

crub

Ran

kgr

ass

Ope

n

Oth

erva

riab

les

Em

erge

ntve

geta

tion

Rab

bit

war

rens

Veg

etat

ion

area

Tre

esS

crub

Ran

kgr

ass

Ope

n

0.11

90.

264*

0.26

4*2

0.68

8**

20.

526*

*2

0.81

9**

0.39

9**

0.20

30.

022

20.

271*

0.02

40.

181

20.

111

20.

000

0.26

6*0.

170

0.64

8**

20.

578*

*

20.

444*

*2

0.36

1**

20.

569*

*0.

676*

*

20.

204

20.

004

0.09

50.

047

0.32

4**

0.42

5**

0.47

1**

20.

581*

*

Veg

etat

ion

alon

gth

eed

geof

wat

er

Tre

esS

crub

Ran

kgr

ass

Ope

n

0.09

12

0.16

42

0.30

5**

20.

340*

*2

0.32

2**

20.

693*

*

0.06

32

0.03

90.

051

20.

067

0.23

10.

146

0.35

2**

20.

524*

*

Oth

erva

riab

les

Em

erge

ntve

geta

tion

0.06

7


APPENDIX III.

Log maximum likelihood estimates for full autoregressive and spatially lagged models, ordinaryleast squares analysis models with and without spatial lagging for 4 neighborhood ranges (n 5 112).Principal component analysis axis-1 (PCA 1) and PCA 1 and 2 represent the variables includingPCA 1, rabbit and other mink, and PCA 1 and 2, rabbit and other mink, respectively, and numberof neighbors; 2, 4, 6, and 8 represent the nearest 2, 4, 6, and 8 river sections to the focal section.

Numbersof

neighbors

Autoregressive models

PCA 1 and 2

Lagged No lag

PCA 1

Lagged No lag

Ordinary least-squares analysis

PCA 1 and 2

Lagged

PCA 1

Lagged

PCA 1 and 2

No lag

PCA 1

No lag

2468

2222.72212.02222.42226.2

2235.82223.42227.72230.8

2231.92216.72227.12227.9

2235.62223.82230.02229.3

2292.22282.32282.32276.2

2295.82285.82288.02282.9

2306.62306.62306.62306.6

2307.42307.42307.42307.4

APPENDIX IV.

Results of likelihood ratio tests comparing autoregressive models with ordinary least squares anal-ysis (OLS) models with and without spatial lags for predictor variables. Values approximate to chisquared.

Full autoregressive model with spatially lagged predictors compared with OLS with spatially lagged predictors(PCA 1, 2, rabbit and other mink)

Number of neighborsLikelihoodP-value

2129.0,0.001

4140.6,0.001

6119.6,0.001

8100.0,0.001

Full autoregressive model with spatially lagged predictors compared with OLS with spatially lagged predictors(PCA 1, rabbit, and other mink)


2127.8,0.001

4138.2,0.001

6121.8,0.001

8110.0,0.001

Comparison of full autoregressive models with spatially lagged predictors between PCA 1, 2, rabbit, and othermink and PCA 1, rabbit, and other mink


28.4

,0.05

49.4

,0.01

69.4

,0.01

83.4NS

OLS spatially lagged predictors compared with OLS with predictors of no spatial lags (PCA 1, 2, rabbit, andother mink)


228.8

,0.001

448.6

,0.001

648.8

,0.001

860.8

,0.001

OLS spatially lagged predictors compared with OLS with predictors of no spatial lags (PCA 1, rabbit, and othermink)

Number of neighborsLikelihood

223.2

443.2

638.8

849.0

P-value ,0.001 ,0.001 ,0.001 ,0.001

OLS (nonspatially lagged model with PCA1, 2, rabbit, and other mink) compared with OLS (nonspatially laggedmodel with (PCA 1, rabbit, and other mink)

LikelihoodP-value

1.2NS


APPENDIX IV.—Continued.

Full autoregressive model with spatially lagged variables compared with autoregressive model without laggedhabitat variables (PCA1, 2, and rabbit)


226.2

,0.001

422.8

,0.001

610.6

,0.05

89.2

,0.05

Full autoregressive model with spatially lagged variables compared with autoregressive model without laggedhabitat variables (PCA1 and rabbit)


27.4

,0.02

414.2

,0.001

64.6

.0.05

82.8

.0.05

TABLE V.

Akaike’s Information Criterion (AIC) for all models. AIC calculated as 2(log likelihood) 1 2p, where p isthe number of parameters in model. The model with the smallest AIC is the best.

Model

Number ofneighbors

Autoregressive model

PCA 1 and 2 PCA 1

Ordinary least squares analysis

Spatial lags

PCA 1 and 2 PCA 1

No spatial lags

PCA 1 and 2 PCA 1

2468

469.4438.0458.8466.2

473.8443.4464.2465.8

596.4576.6576.4564.4

599.6579.6584.0573.8

620.8620.8620.8620.8

617.2617.2617.2617.2

Documents

HABITAT PREFERENCES OF FERAL AMERICAN MINK IN THE …of rabbit warrens seemed to be strongly affected by riverside farmland management, this might eventually determine the distribution