Cats are rare where coyotes roam
Roland Kays,* Robert Costello, Tavis Forrester, Megan C. Baker, Arielle W. Parsons, Elizabeth
L. Kalies, George Hess, Joshua J. Millspaugh, and William McShea
North Carolina Museum of Natural Sciences, 11 W. Jones St., Raleigh, NC 27601, USA (RK, AWP)
Department of Forestry & Environmental Resources, North Carolina State University, Campus Box 7646, Raleigh, NC 27695,
USA (RK, GH)
Smithsonian National Museum of Natural History, 10th and Constitution Ave. NW, Washington, DC 20560, USA (RK, RC)
Smithsonian Conservation Biology Institute, 1500 Remount Rd., Front Royal, VA 22630, USA (TF, MCB, WM)
Department of Fisheries and Wildlife Sciences, 302 Anheuser-Busch Natural Resources Building, University of Missouri,
Columbia, MO 65211, USA (ELK, JJM)
* Correspondent: Roland.Kays@NCSU.edu
Domestic cats (Felis catus) have caused the extinction of many island species and are thought to kill many
billions of birds and mammals in the continental United States each year. However, the spatial distribution and
abundance of cats and their risk to our protected areas remains unknown. We worked with citizen scientists to
survey the mammals at 2,117 sites in 32 protected areas and one urban area across 6 states in the eastern United
States using camera traps. We found that most protected areas had high levels of coyote (Canis latrans) activity,
but few or no domestic cats. The relative abundance of domestic cats in residential yards, where coyotes were
rare, was 300 times higher than in the protected areas. Our spatial models of cat distribution show the amount of
coyote activity and housing density are the best predictors of cat activity, and that coyotes and cats overlap the
most in small urban forests. Coyotes were nocturnal at all sites, while cats were nocturnal in protected areas, but
signiﬁcantly more diurnal in urban sites. We suggest that the ecological impact of free-ranging cats in the region
is concentrated in urban areas or other sites, such as islands, with few coyotes. Our study also shows the value
of citizen science for conducting broadscale mammal surveys using photo-vouchered locations that ensure high
Key words: camera trap, Canis latrans, citizen science, Felis catus, invasive species, protected areas
© 2015 American Society of Mammalogists, www.mammalogy.org
Free ranging domestic cats (Felis catus) are a major conserva-
tion concern because of their predation on native wildlife (Loss
et al. 2013). This situation is worse on oceanic islands, where
prey species typically evolve without mammalian predators
and have little innate ability to avoid cats. Cats have caused the
extinction of 18 small terrestrial island vertebrates and are the
primary extinction risk for another 36 island vertebrates that are
now critically endangered (Medina et al. 2011). In island sys-
tems with simple food web structure, the addition of a predator
species can severely shift community dynamics.
A recent review of cat predation in the United States also
highlighted the conservation problems they pose to continen-
tal ecosystems. By combining typical kill rates and country-
wide cat population estimates, Loss et al. (2013) estimated that
domestic cats kill 1.4–3.7 billion birds and 6.9–20.7 billion
mammals annually. Free-ranging, un-owned cats, as opposed
to pet cats, are thought to cause the majority (69% for birds and
89% for mammals) of this mortality.
However, the spatial extent and ecological signiﬁcance of
this predation on native species remains unknown. Exactly
where the more than 74 million pet cats (Shepherd 2012),
and additional un-owned cats, hunt in the United States is a
critical question. We might expect cats using residential areas
to hunt primarily common prey species that are of lower con-
servation concern. However, if they penetrate public lands
designed to protect native biodiversity, management action
may be needed to reduce their impact. Two tracking studies
of urban cats found them to avoid nature preserves, presum-
ably because of abundant predators populations (Kays and
DeWan 2004; Gehrt et al. 2013). Less is known about cats
outside of developed areas, where most important protected
natural areas are, although one tracking study found that
Journal of Mammalogy, 96(5):981–987, 2015
982 JOURNAL OF MAMMALOGY
un-owned rural cats frequently used natural habitats (Horn
et al. 2011).
We worked with citizen scientists to use camera traps to sur-
vey cats and native wildlife in 32 protected areas across 6 states
in the eastern United States and in residential yards and small
urban forests (some along greenway trails) in Raleigh, North
Carolina. If coyotes in protected areas are negatively inﬂuenc-
ing cats, as seen in urban areas (Kays and DeWan 2004; Gehrt
et al. 2013), then we expect to ﬁnd a negative relationship
between coyote and cat detections. The intensity of this compe-
tition between predators is not only interesting ecologically, but
also important for conservation managers concerned about the
potential negative impacts of cats on native prey.
Materials and Methods
Citizen science camera trap surveys.—From 2012 to 2014, we
recruited and trained 486 volunteer citizen scientists, under-
graduate students, and middle school students to deploy camera
traps across the study area. Most protected area cameras were
run from April to November, while the Raleigh area cameras
were run year-round. Camera traps set in protected areas were
deployed in groups of 3, with one camera placed on a hiking
trail, one 50 m from the trail, and one 100–200 m from the
trail. Camera traps along Raleigh’s greenways were set in pairs
with one camera on the trail and one approximately 25 m off-
trail in nearby wooded areas. Backyard camera traps were set
to minimize pictures of resident humans, typically along the
edge or towards the back of the yard, as described in our ear-
lier work (Kays and Parsons 2014). Volunteers used Reconyx
(RC55, PC800, and PC900; Reconyx, Inc. Holmen, Wisconsin)
and Bushnell (Trophy Cam HD, Bushnell Outdoor Products,
Overland Park, Kansas) camera traps that were equipped with
an infrared ﬂash. These cameras all function similarly in hav-
ing highly sensitive triggers and quick trigger times, allowing
them to record animals passing rapidly in front of the camera
without the addition of bait. Volunteers attached the cameras to
trees at 40 cm above the ground and returned after 3 weeks to
retrieve the images and move the cameras. Cameras were set
on maximum trigger sensitivity and recorded multiple photo-
graphs per trigger, re-triggering immediately if the animal was
still in view.
Volunteers used custom eMammal software to provide ini-
tial identiﬁcation of all wildlife species in camera trap images,
enter camera metadata (e.g., location), and uploaded pictures
to our database. We then reviewed the quality of all data using
the eMammal Expert Review Tool, conﬁrming or correcting all
volunteer species identiﬁcations, and evaluating camera setup
from the view of the camera. After expert review, all data were
downloaded to a Smithsonian data repository for storage.
We grouped consecutive photos into sequences if they were
< 60 s apart and used these sequences as independent records
for subsequent analysis of detection rate and daily activity pat-
terns. Cats photographed from protected areas were identiﬁed
to individuals based on coat color pattern independently by 2
reviewers, who agreed on 100% of identiﬁcations. The Raleigh
area cats were not identiﬁed to individual because most cam-
eras were too widely scattered to obtain recaptures of the same
animals in different cameras.
Environmental variables.—We used ArcMap (ESRI 2012)
to obtain 2 environmental variables for each of our camera
sampling points: housing density and coyote (Canis latrans)
relative abundance. We used the Silvis housing density dataset
(Hammer et al. 2004) to calculate the average housing density
(houses/km2) at 2 spatial scales for each camera using a 250-m
and 5-km radius. We also used a 5-km buffer around each cam-
era and calculated the average coyote detection rate (count/day)
from our cameras within each buffer. On average, these 5-km
buffers included 73 cameras (SE = 1).
Statistical models.—We used spatial statistics (R pack-
age GeoR— Ribeiro and Diggle 2015) to evaluate if detec-
tion rates spatially autocorrelated using a semivariogram to
calculate a minimum distance at which spatial autocorrela-
tion becomes negligible (semivariogram range). We ﬁtted a
semivariogram model to each empirical semivariogram using
weighted least squares and assessed goodness-of-ﬁt by the
minimized sum of squares. We examined detections for cats
and coyotes across sites to consider removing outliers that
could represent a den or feeding station with very high detec-
tion rates. To evaluate the spatial determinants of cat distribu-
tion, we ﬁtted a Poisson count model with a log-link to predict
the count of cat detections, offset by camera deployment
duration (n = 2,117). We used 6 covariates as ﬁxed effects:
habitat type (dummy variable for yard or not yard, small
urban forests, protected areas), latitude and longitude of site,
average housing density (houses/km2) in a 250-m and 5-km
radius of the site, rate of coyote detection at the site (total
count/number of camera days), and average rate of coyote
detection from all cameras within a 5-km radius of the site.
We tested all covariates for multicollinearity using a correla-
tion matrix in Program JMP and considered any correlation
below 30% to be acceptable. We ran our model in a Bayesian
framework using OpenBUGS (Thomas et al. 2006) and R (R
Development Core Team 2011). Our model included a term
for extra-Poisson variation (Breslow 1984) to account for
overdispersion and excess zeros in our dataset. We compared
a suite of 18 covariate combinations which we felt best tested
potential relationships affecting cat distribution. We assessed
relative model deviance using deviance information criterion
(DIC) and ﬁt of the top model using Pearson’s goodness-of-ﬁt
statistic drawn from the posterior distribution (Johnson 2004).
We calculated posterior means and 95% Bayesian credibility
intervals using the most parameterized model within the top
4 DIC points. We separately assessed the signiﬁcance of dif-
ferences in the intensity of use of different habitat types by
cats and coyotes using a Kruskal–Wallis rank-sum test and
Mann–Whitney U-test for pairwise comparisons in Program
JMP (SAS Institute 2012).
To estimate probabilities of occupancy, we used a single sea-
son occupancy model (MacKenzie et al. 2006) and estimated
KAYS ET AL.—DOMESTIC CAT DISTRIBUTION 983
detection probability (P), deﬁned as the probability of detect-
ing an occurring species at a site and occupancy (ψ), deﬁned
as the expected probability that a given site is occupied. For
each species (cat and coyote), we used RMark (Laake 2011)
in R (R Development Core Team 2011) to build and ﬁt models
for each of our covariates, including no covariates (i.e., assum-
ing probabilities were constant across the sites), each combined
with a null model of detection. For each model, we computed
Akaike’s Information Criterion adjusted for small sample size
(AICc), ΔAICc, and Akaike weights (wij, weight of covariate i
for species j) (Burnham and Anderson 2002) and used these
values to assess model ﬁt. We used the most parsimonious
model of occupancy probability for cats (containing the “habi-
tat” variable) to estimate the probability of occupancy in each
We created daily animal activity patterns by ﬁtting density
functions based on circular statistics to independent animal
detections (MacKenzie et al. 2006) using package overlap
(Meredith and Ridout 2014) in R (R Development Core Team
2011). We tested for signiﬁcant differences in activity patterns
using Watson’s 2-sample test for homogeneity in package
CircStats (Lund and Agostinelli 2014) in R (R Development
Core Team 2011).
Mammal surveys.—With 42,874 camera nights of survey effort
across 1,953 locations in 32 protected areas, we obtained 52,863
detections of native wildlife. This same effort returned only 55
detections of cats (0.0012 detections/day). Our semivariogram
showed that autocorrelation for cat rate became negligible after
only 3 m, indicating spatial independence. One camera had very
high detection rates for coyotes, probably representing a den
site or feeding station and so was removed from the analyses.
Cats were detected at 31 camera sites scattered through half of
the protected areas surveyed (Fig. 1; Supporting Information
S1), resulting in an occupancy rate (ψ) of 0.027 (SE = 0.0060)
across the region. Based on coat coloration, we were able to
identify all cats from protected areas to individual; in 14 of
the 32 protected areas, we detected only a single cat (some
Fig. 1.—Distribution of cats detected across 32 protected areas: 50% had no cats (Felis catus) detected, while 44% had just 1 cat, and 2 (6%) had
multiple cats. The 2 map insets are at the same scale and show typical camera arrangement and detection patterns for cats in rural (left, Stone
Mountain State Park) and urban protected areas (right, Rock Creek Park).
984 JOURNAL OF MAMMALOGY
photographed multiple times). In the 2 protected areas with
the highest levels of cat activity, we detected 5 or 6 individual
cats photographed multiple times (Supporting Information S1).
Coyotes (Supporting Information S2) were detected 33 times
more often than cats in protected areas (0.044/day), occurred at
a higher level of occupancy (ψ = 0.49 ± 0.020), and were found
in all but one of the 32 protected areas.
We also used camera traps to survey mammals at 171 loca-
tions in Raleigh, North Carolina for a total of 2,760 camera
nights: including 60 sites in residential yards, and 111 in
small urban forests, 45 of which were along greenway trails.
Domestic cats (Supporting Information S3) were detected
more often on trail than off-trail, but were found at the high-
est rates in yards, followed by small urban forests, which were
both much higher than in larger protected areas (Fig. 2; Tables
1 and 2). Indeed, residential yards (0.44 detections/day) had
300 times more cat activity than protected areas (0.11 detec-
tions/day). Coyotes were detected in all habitats but were rare
in residential areas. Similarly, probability of occupancy for cats
was highest in yards (0.53 ± 0.067), followed by small urban
forests (0.27 ± 0.044), with the lowest rates in protected areas
(0.016 ± 0.0029). Coyotes had the lowest probability of occu-
pancy in yards (0.085 ± 0.041), with higher rates in small urban
forests (0.57 ± 0.061) and protected areas (0.35 ± 0.013).
Spatial and temporal model results.—None of our covari-
ates were highly correlated, all pairwise correlations fell
below 30%. Our top models predicting cat distribution across
all sites ﬁt well (χ2 > 0.4). Parameter estimates from the most
parameterized model within the top 4 DIC points showed that
coyote activity levels and housing densities had the stron-
gest effects on cat detections (Tables 1 and 2). Housing den-
sity was included in only one of the top models (Table 1),
showing it was less important than coyote activity. Although
habitat was not an important covariate in the top multivariate
models of cat distribution, we note that cat detection rate dif-
fered signiﬁcantly across these categories (Kruskal–Wallis test,
χ2 = 502.35, P < 0.0001). Cat rate was signiﬁcantly higher
in yards than small urban forests (mean difference = 24.82,
SE = 6.76, Z = 3.67, P = 0.0002) and protected areas (mean dif-
ference = 508.17, SE = 22.76, Z = 22.33, P < 0.0001). Cat rate
was also signiﬁcantly higher in small urban forests than pro-
tected areas (mean difference = 254.91, SE = 16.89, Z = 15.09,
P < 0.0001; Fig. 2). The intensity of use by coyotes across these
3 habitats was also signiﬁcantly different (Kruskal–Wallis test,
χ2 = 31.1927, P < 0.0001, protected areas versus small urban
forests P < 0.0001, yards versus small urban forests P < 0.0001,
yards versus protected areas P = 0.001).
Our distribution model found a strong negative relationship
between the detection rate of coyotes at the 5-km scale and
cats across all sites (Tables 1 and 2; Supporting Information
S1). Additional support for the hypothesis that coyotes exclude
cats from protected areas comes from the only protected area
in which we found no coyotes, Gambrill State Park, which also
had the highest level of cat detections; the detection rate there
(0.023/day) was 3 times higher than the next most cat-rich
Fig. 2.—Average detection rates of cats and coyotes (Canis latrans)
recorded by camera traps set in different habitats including 32 pro-
tected areas in the eastern United States and 177 urban sites around
Raleigh, North Carolina. Error bars show SE of the mean. Rates
were statistically different across habitats for both coyotes and cats
(Kruskal–Wallis test, P < 0.0001).
Table 1.—Model selection for describing variation in cat (Felis
catus) distribution as ranked by the deviance information criterion
(DIC). The yard variable is a categorical classiﬁcation (yard, not
yard), the house variables are measures of housing density, and the
coyote (Canis latrans) variables are detection rates from camera traps.
LatXLong is a spatial term to account for broadscale geographic trends.
Model DIC Delta DIC
Coyotes 5 km 638 0
Coyotes 5 km + House 5 km 639 1
Coyotes at site 644 6
Null 645 7
Houses 5 km 645 7
LatXLong 650 12
LatXLong + Houses 5 km 651 13
Houses 250 m 651 13
LatXLong + Coyotes 5 km 652 14
Yard + Coyotes 5 km 653 15
LatXLong + Yard + Houses 250 m + Houses
5 km + Coyotes at site + Coyotes 5 km
Yard 655 17
Yard + Houses 5 km 655 17
Yard + Coyotes 5 km + Houses 5 km 656 18
LatXLong + Yard + Coyotes 5 km + Houses 5 km 656 18
LatXLong + Yard + Coyotes at site + Houses 250m 657 19
LatXLong + Yard + Coyotes 5 km 659 21
LatXLong + Yard 660 22
Table 2.—Parameter estimates results for Coyote (Canis latrans) 5
km + Housing Density 5 km model predicting cat (Felis catus) use of
protected and urban areas. Both variables were considered signiﬁcant
because the 95% credible intervals did not include 0.
Variable Posterior mean 95% Credible interval
House 5 km 0.79 0.35, 1.20
Coyote 5 km −1.46 −2.08, −0.86
KAYS ET AL.—DOMESTIC CAT DISTRIBUTION 985
Coyotes and cats in protected areas showed nocturnal activ-
ity patterns that were not different from each other (Watson’s
U2 = 0.077, P = 0.50) but were both different than the more
diurnal activity of urban cats (coyotes-urban cats: Watson’s
1 = 2.242, P < 0.0001; park cats–urban cats: Watson’s
1 = 0.300, P = 0.005; Fig. 3).
Our large-scale survey shows that free-ranging cats are not
widespread in large protected areas in the eastern United States.
We detected no domestic cats in half of the 32 protected areas
we surveyed, and more than 1 individual cat in only 2 of them.
This is the ﬁrst study to address cats in North American pro-
tected areas and suggests that they are not a widespread con-
servation concern for the larger protected areas in this region.
We found evidence that predators may be preventing cats
from colonizing protected areas, as most cat-free areas had
high activity rates of coyotes. Additionally, our measures of
coyote activity were negatively associated with cat activity in
our spatial model. Finally, the only protected area in which we
detected no coyotes had, by far, the highest levels of cat activ-
ity. Coyotes have been shown to prey on cats (Quinn 1997) and
prevent cats from using some urban natural areas (Crooks and
Soule 1999; Gehrt et al. 2013), but this is the ﬁrst study to docu-
ment the partitioning of space between coyotes and cats across
large scales. The virtual absence (one or fewer) of cats in 94%
of the protected areas we surveyed suggests that there is a low
threshold level of coyote activity that effectively prevents cats
from using an area, and that most of the relatively large, pro-
tected areas we surveyed were above that level.
Compared to these protected areas, we detected many more
cats in our surveys of residential yards and small urban forests,
showing the extent to which cat activity is focused in urban
areas. Not surprisingly, cats were most common in yards,
which had 300 times more cat activity than protected areas.
This shows the limited degree to which most urban cats venture
past their neighborhoods, which is similar to what was found
for radio-tracked pet cats (Kays and DeWan 2004). Habitat type
was not featured in our top multivariate distribution models,
presumably because coyote distribution was a better predictor,
and was also correlated with habitat types (Fig. 2).
Why populations of cats and coyotes have virtually no over-
lap in protected areas but substantial spatial overlap in small
urban forests remains an important question. A tracking study
in Chicago also found spatial overlap between coyotes and cats
in small urban forests, although their core areas were separate
(Gehrt et al. 2013). We suspect that the fragmented arrange-
ments of natural and developed areas typical of American cities
may provide cats with sufﬁcient nearby refuges they can access
if they encounter a predator. Our data conﬁrm that residential
yards in Raleigh are safe havens for cats, with coyotes detected
in only 5 of the 64 yards we surveyed. This may be different
from some cities where coyotes are more urbanized (Gehrt
et al. 2009). The increased diurnal activity we found in urban
cats could also be a strategy, by cats or their owners, to avoid
We used our camera trap photos as measures of local cat
relative abundance in 3 different ways, all showing the same
results, with cats rare in larger protected areas, present in small
urban forests at varying levels, and common in residential yards.
Because our camera traps were unbaited and simply recorded
the frequency that cats and coyotes walked by, we could use
this as a measure of relative abundance, as well as a measure
of the ecological impact of these 2 predators (Rowcliffe et al.
2008). Our occupancy models mirrored our results based on
raw detection rate. Finally, for the protected areas, we were
able to identify individual cats based on coat coloration, con-
ﬁrming that few individuals (typically one) were present.
We were not able to evaluate the origin of the cats we photo-
graphed, i.e., whether they were pets, wide-ranging feral cats,
or from a cat colony. Free-ranging cats are thought to prey on
substantially more native prey than pet cats (Loss et al. 2013),
and the establishment of localized cat colonies is of special
conservation concern due to the incredibly high predator den-
sities that can result, in addition to disease concerns (Clarke
and Pacin 2002). We did not speciﬁcally target sampling of cat
colonies and, given the relatively low cat detection rates, appar-
ently did not detect any within the protected areas or small
urban forests we surveyed. We were surprised to detect single
cats in the middle of large protected areas, far from houses or
neighborhoods. We think this detection rate is too low to rep-
resent truly feral populations, and suspect that some of these
could be cats that were abandoned at the parks by their own-
ers. Another alternative is that these were unusual pet cats that
moved much further from their house than is typical, which has
also been observed in one radio-tracked cat (Kays and DeWan
2004). Such wandering cats in protected areas might not have
knowledge of the local coyote populations, which could also
explain why they are temporally overlapping with them in
being primarily nocturnal.
The large scale of our survey, with more than 2,000 sample
points across 6 states, shows that our main result of few to no
cats in protected areas is a consistent pattern across the region.
This also shows the beneﬁt of working with volunteer citizen
Fig. 3.—Daily activity patterns showing high overlap for coyotes
(Canis latrans) and cats (Felis catus) in protected areas and less in
urban areas, where cats are more diurnal. The activity patterns of coy-
otes from protected and urban areas were not different, and thus are
combined into one line.
986 JOURNAL OF MAMMALOGY
scientists to scale-up this mammal survey (Cooper et al. 2007),
allowing us to not only sample more parks, but also efﬁciently
survey urban areas. By reviewing all photographic data col-
lected by citizens, we were able to ensure the high quality of the
data, unlike other citizen science work, which may not collect
vouchers (Cooper et al. 2007).
Because our urban results were from only one city, addi-
tional research will be needed to evaluate how predators and
cats interact in the varied urban landscapes around the coun-
try, including those with more urbanized coyotes (Gehrt et al.
2013), or even larger predators. Additional surveys of cats in
protected areas around the world with varied predator commu-
nities could also shed more light on which situations require
active management to reduce cats (Loyd and Devore 2010) and
which can allow their native predators to keep the cats out.
We thank all of our 486 volunteers for their hard work collect-
ing data for this study, including Master Naturalists, 3 North
Carolina State University undergraduate classes, Prairie
Ridge Ecostation, and students from Exploris Middle School.
For their ﬁeld assistance and volunteer coordination, we thank
A. Rogers, J. Pearson, K. Holliﬁeld, M. Davies, S. Hartley,
B. Davis, G. Schneider, D. Walker, D. Engebretson, J. Hall,
D. Todd, R. Gubler, S. Henry, R. Hughes, B. Yeaman,
M. Milton, A. Landsman, C. DiAntonio, D. Stapleton,
E. Kelley, L. Donaldson, M. Spurrier, D. Nisbet, B. Thompson,
C. Croy, M. Smith, L. Potts, V. Lebsock, J. Palumbo, and
the staff of the National Parks Service, United States Fish
and Wildlife Service, United States Forest Service, North
Carolina State Parks, The Nature Conservancy, North Carolina
Wildlife Resources Commission, Tennessee State Parks,
Tennessee Division of Forestry, South Carolina State Parks,
Virginia State Parks, Virginia Division of Game and Inland
Fisheries, Western Virginia Water Authority, the Wintergreen
Nature Foundation, Maryland State Parks, and Raleigh, Parks,
Recreation and Cultural Resources. For help reviewing photo-
graphs, we thank N. Fuentes, S. Higdon, C. Bland, T. Perkins,
L. Gatens, R. Owens, R. Gayle, C. Backman, K. Clark,
J. Grimes, and J. Simkins. We thank R. Montgomery for
input on study design. This work was conducted with fund-
ing from the National Science Foundation grant #1232442
and #1319293, the VWR Foundation, the US Forest Service,
the North Carolina Museum of Natural Sciences, and the
The Supporting Information documents are linked to this
manuscript and are available at Journal of Mammalogy online
(jmammal.oxfordjournals.org). The materials consist of data
provided by the author that are published to beneﬁt the reader.
The posted materials are not copyedited. The contents of all
supporting data are the sole responsibility of the authors.
Questions or messages regarding errors should be addressed
to the author.
Supporting Information S1.—Summary of the cats and coy-
otes detected by citizen science camera trap surveys of 32 pro-
tected areas across six states and Washington, DC. Rates are
Supporting Information S2.—Camera trap picture of a pack
of coyotes hunting Sand Hills State Forest, South Carolina.
Supporting Information S3.—Camera trap picture of a
domestic cat walking on the Raleigh greenway through a small
urban forest in North Carolina.
Breslow, N. E. 1984. Extra-poisson variation in log-linear models.
Applied Statistics 33:38–44.
Burnham, K. P., and R. D. Anderson. 2002. Model selection and
multi-model inference. Springer, New York.
Clarke, A. L., and T. Pacin. 2002. Domestic cat “colonies” in natu-
ral areas: a growing exotic species threat. Natural Areas Journal
Cooper, C. B., J. Dickinson, T. Phillips, and R. Bonney. 2007.
Citizen science as a tool for conservation in residential ecosystems.
Ecology and Society 12:11.
Crooks, D. R., and M. E. Soule. 1999. Mesopredator release
and avifaunal extinctions in a fragmented system. Nature
ESRI. 2012. ArcMap. Ver. 10.1. Esri Inc., Redlands, California.
Gehrt, S., C. Anchor, and L. White. 2009. Home range and land-
scape use of coyotes in a metropolitan landscape: conﬂict or coex-
istence? Journal of Mammalogy 90:1045–1057.
Gehrt, S. D., E. C. Wilson, J. L. Brown, and C. Anchor. 2013.
Population ecology of free-roaming cats and interference competi-
tion by coyotes in urban parks. PLoS One 8:1–11.
Hammer, R. B., S. I. Stewart, R. Winkler, V. C. Radeloff, and
P. R. Voss. 2004. Characterizing spatial and temporal residential
density patterns across the U.S. Midwest, 1940–1990. Landscape
and Urban Planning 69:183–199.
Horn, J. A., N. Mateus-Pinilla, R. E. Warner, and E.
J. Heske. 2011. Home range, habitat use, and activity patterns
of free-roaming domestic cats. Journal of Wildlife Management
Johnson, V. 2004. A Bayesian χ2 test for goodness-of-ﬁt. The Annals
of Statistics 32:2361–2384.
Kays, R., and A. W. Parsons. 2014. Mammals in and around subur-
ban yards, and the attraction of chicken coops. Urban Ecosystems
Kays, R. W., and A. A. DeWan. 2004. The ecological impact of
inside/outside house cats around a suburban nature preserve.
Animal Conservation 7:1–11.
Laake, J. 2011. RMark: R code for MARK analysis. R package.
Ver. 2.0.7 cran.r-project.org/web/packages/RMark/. Accessed 10
Loss, S. R., T. Will, and P. P. Marra. 2013. The impact of free-
ranging domestic cats on wildlife of the United States. Nature
Loyd, K. A. T., and J. L. Devore. 2010. An evaluation of feral cat
management options using a decision analysis network. Ecology
and Society 15:10. Special Section.
Lund, U., and C. Agostinelli. 2014. CircStats: circular statistics. R
package version 0.2-4. cran.r-project.org/web/packages/CircStats/.
Accessed 10 December 2014.
MacKenzie, D., J. D. Nichols, J. A. Royle, K. H. Pollock,
L. Bailey, and J. E. Hines. 2006. Occupancy estimation and
KAYS ET AL.—DOMESTIC CAT DISTRIBUTION 987
modeling: inferring patterns and dynamics of species occurrence.
Academic Press, New York.
Medina, F. M., et al. 2011. A global review of the impacts of inva-
sive cats on island endangered vertebrates. Global Change Biology
Meredith, M., and M. Ridout. 2014. Overlap: estimates of animal
activity. R package version 0.2.3. cran.r-project.org/web/packages/
overlap/. Accessed 10 December 2014.
Quinn, T. 1997. Coyote (Canis latrans) food habits in three
urban habitat types of western Washington. Northwest Science
Ribeiro, P. J., and P. J. Diggle. 2015. R package GeoR. cran.r-
project.org/web/packages/geoR/. Accessed 10 December
R Development Core Team. 2011. R: a language and environ-
ment for statistical computing, version 2.13.1. R Foundation
for Statistical Computing, Vienna, Austria. www.R-project.org.
Accessed 10 December 2014.
Rowcliffe, M., J. Field, S. Turvey, and C. Carbone. 2008.
Estimating animal density using camera traps without the need for
individual recognition. Journal of Applied Ecology 45:1228–1236.
SAS Institute. 2012. Program JMP. SAS Institute Inc., Cary, North
Shepherd, A. J. 2012. U.S. pet ownership & demographics source-
book. American Veterinary Medical Association, Schaumburg,
Thomas, A., B. O. Hara, U. Ligges, and S. Sturtz. 2006. Making
BUGS open. R News 6: 12–17.
Submitted 13 August 2014. Accepted 26 May 2015.
Associate Editor was Bradley J. Swanson.