Original Article
Using Fecal Progestagens and Logistic
Regression to Enhance Pregnancy Detection
in Wild Ungulates: A Bison Case Study
STEVEN L. CAIN,1 National Park Service, Grand Teton National Park, P.O. Box 170, Moose, WY 83012, USA
MEGAN D. HIGGS, Department of Mathematical Sciences, Montana State University, 2-242 Wilson Hall, Bozeman, MT 59717-2400, USA
THOMAS J. ROFFE, United States Fish and Wildlife Service, 1400 S 19th Avenue, Bozeman, MT 59718, USA
STEVEN L. MONFORT, Smithsonian Conservation Biology Institute, National Zoological Park, 1500 Remount Road, Front Royal, VA 22630,
USA
JOEL BERGER, Division of Biological Sciences/Wildlife Conservation Society, Northern Rockies Field Of?ce, University of Montana, Missoula, MT
59812, USA
ABSTRACT Ungulate ecological studies often include components of reproduction because of its demo-
graphic importance and the ecological factors affecting it. Pregnancy status, in particular, is key because it
represents a starting point for succeeding measurements of vital rates. Here, we present a case study using
wild bison (Bison bison), in which we developed a non-invasive method for assessing pregnancy in unmarked,
non-handled animals that improves upon existing approaches for wild ungulates. Speci?cally, we employed a
model-based binary logistic-regression approach to estimate the probability of pregnancy predicted by fecal
progestagen concentrations quanti?ed from a single, late-gestation scat sample. For 155 observations of 42
marked bison from the Jackson herd in northwestWyoming, USA during 1997?2005, we used combinations
of transrectal uterine palpation and calf status as independent measures of pregnancy to reduce the potential
for error inherent in using either measure alone. We evaluated predictive success by calculating mis-
prediction rates from leave-one-out cross-validation, and by calculating the percentage of 95% con?dence
intervals that crossed a pregnant?not-pregnant threshold. Correct predictions, with high con?dence, were
obtained from a model using year-centered, natural-log-transformed progestagen concentrations, resulting
in an overall successful cross-validation pregnancy prediction rate of 93.5%. Our approach will allow
practitioners to consider the uncertainty associated with each prediction, thereby improving prediction
interpretations. The approach should appeal to practitioners because fecal samples are easily collected and
preserved, laboratory procedures are well-documented, and logistic-regression statistical software is readily
available. Furthermore, samples can be obtained non-invasively, which reduces cost and potential bias and
increases animal safety, human safety, and social acceptability. 
 2012 The Wildlife Society.
KEY WORDS bison, fecal, hormones, logistic, non-invasive, pregnancy, pregnancy-specific protein B, regression,
Wyoming, Yellowstone.
Studies of ungulates often include components of reproduc-
tion because of its importance to demography (e.g., Testa
2004, Coulson et al. 2005, Fuller et al. 2007) and the
ecological factors affecting it (Berger et al. 1999). Among
parameters commonly used to assess reproductive perfor-
mance, pregnancy status is key because, following ovulation,
it represents a starting point for succeeding measurements of
vital rates, including birth rates, neonatal survival, and re-
cruitment of breeding-aged individuals into a population
(Ramsay and Sadleir 1979, Messier et al. 1990,
Kirkpatrick et al. 1993, Russell et al. 1998, Stoops et al.
1999). Knowledge of pregnancy status can be useful at the
population level, where it may be expressed as an overall rate
(Messier et al. 1990), or at the individual level, where the
status of a marked individual and its subsequent reproductive
outcome is of interest (Messier et al. 1990,White et al. 1995,
Garrott et al. 1998).
Determining pregnancy of wild ungulates is inherently
dif?cult and presents a serious challenge to researchers in-
terested in reproductive ecology. Handling animals, either
through capture and physical restraint or chemical immobi-
lization, is required for direct evaluations of pregnancy status,
including transrectal palpation (Hein et al. 1991) and ultra-
sound (Barrett 1981, Stephenson et al. 1995), as well as for
indirect assessments using serum pregnancy-speci?c protein
B (Houston et al. 1986, Sasser et al. 1986, Wood et al. 1986,
Haigh et al. 1988). An obvious disadvantage with these
methods is that there are hazards associated with handling
Received: 8 February 2012; Accepted: 22 May 2012
1E-mail: steve_cain@nps.gov
Wildlife Society Bulletin; DOI: 10.1002/wsb.178
Cain et al. 
 Bison Pregnancy Prediction 1
animals, both to the research subjects (e.g., Ballard and
Tobey 1981, Valkenburg et al. 1983, Larsen and Gauthier
1989, Peterson et al. 2003) and/or to the handlers (Jessup
et al. 1988, Haigh 1990, Dein et al. 2005, Roffe et al. 2005,
and references therein), some of which may bias results of the
intended research (Welsh and Johnson 1981). The potential
for capture-related mortality is of biological concern as well,
particularly when handling rare or endangered species
(Association for the Study of Animal Behaviour 2003,
Berger et al. 2010). Likewise, society increasingly questions
(on ethical grounds) the need to manipulate animals or
employ invasive, potentially harmful handling techniques
(Farnsworth and Rosovsky 1993, see Bekoff and Jamieson
1996, Peterson et al. 2003).
In general, non-invasive pregnancy assessment methods
(Lasley and Kirkpatrick 1991, Monfort 2003) are more
limited in inferential capability but have several inherent
advantages, including 1) safety, because capture and handling
are not required; 2) reduced bias from potential effects on
behavior, activity, distribution, or physiology; 3) lower ex-
pense because immobilizing drugs or specially trained per-
sonnel to administer them or to perform uterine transrectal
palpation are not necessary; and 4) social acceptability. In
ungulates, these techniques have focused on the measure-
ment of reproductive steroid hormone metabolites excreted
in urine (Poole et al. 1984, Kirkpatrick et al. 1990, Monfort
et al. 1990) and/or feces (Kirkpatrick et al. 1990, Monfort
et al. 1993, Garrott et al. 1998, Berger et al. 1999,
Schoenecker et al. 2004). Application of these techniques
to accurately estimate pregnancy status for individual animals
from single fecal samples, which is more cost-ef?cient and
less invasive than collecting multiple samples per individual,
has been proposed only for elk (Cervus elaphus; Garrott et al.
1998), bighorn sheep (Ovis canadensis; Schoenecker et al.
2004), and feral horses (Linklater et al. 2000).
Bison (Bison bison) are polygynous seasonal breeders with a
9-month gestation period (Berger and Cunningham 1994).
In our study area, breeding and calving peaked in August and
May, respectively, with most births occurring over a 2-month
period (Berger and Cain 1999).
As part of broader research on bison reproductive ecology,
we sought to evaluate fecundity and prevalence of fetal and
neonatal loss in a marked sample of reproductive-aged ani-
mals. We directly assessed pregnancy status in a subset of our
bison population that was handled each year, with the pri-
mary intent of determining the incidence of brucellosis and
its effect on reproduction. Our goal in the current study was
to develop a non-invasive method for assessing pregnancy
status for the remainder of our study animals that were not
handled each year. Reproductive-aged females were observed
directly at least once weekly during the presumptive parturi-
tion interval to estimate date of birth. Our study population
was wide-ranging and this was a labor-intensive task, so our
objective was to improve ef?ciency by removing animals that
we were con?dent were not pregnant from the group checked
for calves each week.
Our study area did not have continuous snow cover in
winter, so urine collection was not feasible. Therefore, we
focused on developing model-based estimates of pregnancy
probability based on quantitative assessments of fecal pro-
gestagens in a single scat collected during winter. We rec-
ognized the limitations of using single samples, but, similar
to many studies, cost, logistical considerations, and the po-
tential for additional disturbance to our study population
precluded additional fecal-sample collections. Based on past
research (Kirkpatrick et al. 1992, 1993), we expected con-
centrations of fecal progestagens to differ between pregnant
and non-pregnant bison. However, in contrast to similar
ungulate studies, our goal was not to simply compare
mean concentrations of fecal progestagens between pregnant
and non-pregnant groups, but instead to estimate the prob-
ability of pregnancy for individual females based on hormone
concentration, and then to use this probability to predict
pregnancy status. By quantifying the uncertainty inherent in
our probability estimates, we were also able gauge con?dence
in our predictive measures.
STUDY AREA
Our study area in the southern Greater Yellowstone
Ecosystem, USA, focused on Grand Teton National Park
and the National Elk Refuge, which comprised the primary
range of the free-ranging Jackson bison population. This area
included the upper Snake River drainage in a high-elevation
valley commonly referred to as Jackson Hole, bounded by the
Teton Range to the west, the Gros Ventre and Absaroka
Mountains to the east, the Yellowstone Plateau to the north,
and the town of Jackson, Wyoming to the south. During our
1997?2005 study, the population varied from 320 to 950 (S.
L. Cain, unpublished data). Pelleted alfalfa was provided to
elk and bison on the National Elk Refuge for an average of
57 days (range ? 28?85) each winter (E. Cole, National Elk
Refuge, unpublished data).
Brucellosis was discovered in the Jackson bison herd in
1989 (Williams et al. 1993). Its source has been attributed to
the original Yellowstone reintroduction because these bison
were known to be infected (Meagher 1973), or to sympatric
elk from the Jackson elk herd, which were also known to be
infected (Thorne 1982). Based on a criterion that required a
minimum of 2 out of 6 serologic tests to be positive by bovine
Uniform Methods and Rules (USDA APHIS 2003; at least
one of which had to be a quantitative test), brucellosis annual
seroprevalence in non-calf, predominantly female, bison av-
eraged 69% during our study (T. J. Roffe, unpublished data).
We considered bison sampled more than once during the
year as positive if they met the criteria for positive on any
sample.
METHODS
Sample Collection and Processing
Our study protocol required marking, taking samples from,
and monitoring the reproductive performance of a sample of
female bison each year. We immobilized bison on or near
National Elk Refuge feed-lines mid-winter (late Jan?mid-
Mar), during the 2nd or early 3rd pregnancy trimester, from a
tracked snow-vehicle using a carfentanil?xylazine drug com-
2 Wildlife Society Bulletin
bination administered from a projectile dart (Roffe et al.
2001). We deployed Telonics very-high-frequency radiocol-
lars (Telonics, Inc., Mesa, AZ) and numbered 5.8-cm 
7.3-cm bangle ear tags on each individual; individuals
were initially selected randomly from groups of several hun-
dred bison on feed-lines. We did this by generating random
numbers from 1 to 20 to determine the nth adult female we
would choose as we approached or worked our way through a
group of bison. After marking our original sample in 1997,
during 1998?2002 we attempted to re-capture each individ-
ual annually to determine pregnancy status, take blood and
fecal samples, and replace aged radiocollars. During 2003?
2005 we re-captured only half of our marked sample annually
to address a separate objective of evaluating the effects of
handling on reproduction.We enhanced our sample annually
using randomly selected females of appropriate age to replace
those that died or whose radiocollars needed replacement.
For each anesthetized bison, transrectal palpation was per-
formed by a veterinarian to assess pregnancy status, 20 cm3
of whole blood was taken to evaluate brucellosis exposure and
pregnancy-speci?c protein B, approximately 20?50 g of feces
was obtained (unless the animal?s colon was void), fat depth
over the posterior-most rib was measured as an index of body
condition (e.g., Richards et al. 1986, McLaren et al. 1991),
and tooth eruption patterns were recorded for aging (e.g.,
Frison and Reher 1970). For animals not immobilized, we
obtained fecal samples by locating individuals with radio-
telemetry, verifying their identity with ear tags when multi-
ple radioed animals were in a group, watching them until
they defecated, and then collecting the target sample. We
minimized potential for sampling non-target fecal matter by
approaching bison closely, working in pairs whenever possi-
ble (allowing one person to remain focused on the target
sample while the other approached and collected), using a
stationary spotting scope focused on the target sample to
corroborate location, sampling only fresh fecal matter, and
rejecting samples that we could not con?dently attribute to
the target animal. We stored fecal samples frozen in sealed
plastic bags until they were thawed for laboratory analyses.
Endocrine assessments were conducted at the Smithsonian
Conservation Biology Institute (Front Royal, VA). All fecal
samples were processed and extracted using fecal hormone
procedures that have been previously described (Garrott
et al. 1998). Fecal progestagens were quanti?ed with a
monoclonal antibody (CL425; C. Munro, University of
California?Davis, Davis, CA) used in a radioimmunoassay
(1997?2000; Wasser et al. 1994) or enzyme-immunoassay
(2001?2005; Graham et al. 2001). For both assays, serially
diluted fecal extracts demonstrated displacement curves that
were parallel to those of standard hormone preparations. All
samples for each year were evaluated in a single assay, and
intra-assay variation was <5%.
We centrifuged blood, harvested serum, and froze all sam-
ples until processing. Sera were submitted to the American
Association of Veterinary Diagnostic Laboratories certi?ed
Montana Veterinary Diagnostic Laboratory, Bozeman,
Montana (USA), for 8 Brucella serologic tests: card (Rose
Bengal), buffered antigen plate agglutination, standard plate
agglutination, standard tube agglutination, rivanol, and com-
plement ?xation. Samples from 2002 to 2005 were also
processed using ?uorescent polarization. We interpreted
tests as positive or negative based on U.S. Department of
Agriculture bovine brucellosis uniform methods and rules
(USDA APHIS 2003). We classi?ed high-titered animals
based on complement ?xation of a 4? reaction at 1:80
dilution or greater (Roffe et al. 1999).
We also submitted sera to Biotracking, Inc. (Moscow, ID;
biotracking.com) to determine concentrations of pregnancy-
speci?c protein B (Sasser et al. 1986). Pregnancy-speci?c
protein B was determined by radioimmunoassay (earlier
samples) and enzyme-linked immunoassay (later samples)
with a reported 95% accuracy at 40 days pregnant in bison
(Biotracking, Inc.).
Reproductive Status
To evaluate the relationship between fecal progestagens and
pregnancy, we needed an independent measure of pregnancy
against which to assess fecal progestagen levels. We de?ned
??pregnant?? as all adult females observed with a calf in the
spring or summer following winter fecal sampling. We de-
termined ??females with calves?? by locating each marked
reproductive-aged bison at least once weekly during the
calving season (Berger and Cain 1999), and assigning calves
to speci?c females only after a suckling was observed (Berger
and Cunningham 1994).
We de?ned as ??non-pregnant?? those females that were
both negative by palpation and were without a calf, because
??absence of a calf?? alone had a greater margin of error for
labeling pregnant females that lost calves through fetal or
neonatal mortality as non-pregnant. We eliminated females
of uncertain pregnancy status (positive palpation, absence of
calf) from our analysis because of the small potential for error
by palpation. We recognized that using these criteria in-
curred a small potential to include females palpated falsely as
not-pregnant that subsequently lost their fetus or neonate
(and, therefore, were absent calf) in the ??not-pregnant??
group.
Modeling Procedures and Considerations
We modeled the relationship between pregnancy status
and winter-collected (late gestation) fecal progestagens
(WP) using binary logistic regression (e.g., Hosmer and
Lemeshow 2000). Logistic regression modeled the natural
log of the odds [ln(odds) ? logit(P)] of being pregnant as a
linear function of the explanatory variable WP. The odds of
being pregnant were de?ned as the probability of being
pregnant, divided by the probability of not being pregnant.
For each female, the logistic-regression model provided
an estimated logit of the probability of pregnancy, ln[P/
(1  P)], along with standard errors. We then obtained
estimated probabilities of pregnancy simply by inverting
the logit transformation. Likewise, we calculated con?dence
intervals (CI) for probabilities by applying the inverse trans-
formation to the endpoints of the CI calculated on the logit
scale (Agresti 2002). R statistical software was used for all
analyses (R Development Core Team 2012).
Cain et al. 
 Bison Pregnancy Prediction 3
Because our goal was to predict pregnancy status for non-
handled females, we focused our model-assessment efforts
on quantifying predictive success. Classi?cation of a female
as pregnant or not-pregnant based on the magnitude of her
WP required speci?cation of a threshold probability. We
chose the classi?cation threshold minimizing the mis-pre-
diction rate, as calculated using leave-one-out cross valida-
tion. That is, we removed one observation from the data set,
?t the model, and then predicted pregnancy status for that
observation, which allowed for comparison of predicted
status to true status. Repeating this for all observations
allowed us to calculate mis-prediction rate as the number
of incorrect predictions divided by the total number of
observations. We also calculated an average estimation error
using cross-validation by taking the absolute value of the
difference between the estimated probability of pregnancy
and the true response (1 for pregnant, 0 for not pregnant),
averaged over all observations. For example, an estimated
probability of 0.78 gave an estimation error of 0.22 if the
female was pregnant and 0.78 if the female was not pregnant.
The use of mis-prediction rate and average estimation error
relied only on the point estimates of the probabilities and the
chosen classi?cation threshold, and failed to account for
widths of associated CIs for the probabilities. To assess
this uncertainty in the context of prediction, we evaluated
whether CIs crossed the classi?cation threshold, coupled
with whether the pregnancy status was correctly predicted.
For correct predictions, it is undesirable for the CI to cross
the threshold, which indicates a lack of con?dence in a
correct prediction. On the other hand, for incorrect predic-
tions, it is undesirable for the CI to not cross the threshold,
which indicates high con?dence in a wrong prediction. In
these 2 cases, simply relying on the mis-prediction rate would
yield a falsely optimistic or pessimistic view, respectively. We
considered these bad predictive behaviors for a model, and
calculated the proportion of 95% CIs in each case for a
particular model, which allowed comparison of these pro-
portions across models with different covariates. Thus, we
evaluated predictive ability by comparing predictions to
known values, and utilizing the information in the CI re-
garding uncertainty in the estimate used to make the
prediction.
We considered several issues related to assumptions in our
statistical approach. First, the binary logistic-regression
model assumed 1) independent observations, and 2) a linear
relationship between the log odds of pregnancy and the
explanatory variables. Data were collected across years; there-
fore, we assessed the appropriateness of presumptive homo-
geneity in the relationship between WP and the probability
of pregnancy across years. Initially, we found clear year-to-
year differences in the range of WP (Fig. 1A). Factors that
can modulate year-to-year variability in fecal steroid excre-
tion include age (Morden et al. 2011) and nutritional plane
(Cook et al. 2001, 2002), but we found no relationship
between age or body condition and WP for reproductive-
aged animals. Nearly all of our bison received supplemental
feed (on the National Elk Refuge) at the time of sampling, so
large variations in dietary ?ber (which may also affect hor-
mone secretion; Wasser et al. 1993), were unlikely. A more
likely source of between-year variation in progestagen con-
centrations related to sampling females at different stages of
gestation, which resulted from sampling periods that varied
by several weeks among years, or from natural variation in
timing of reproductive season onset (Monfort et al. 1993,
Garrott et al. 1998), as evidenced by a wide range of within-
year calf birth dates in our population (Berger and Cain
1999). However, in addition to sampling strategies and
physiology, intra- and inter-assay variation in hormone tests
can contribute to the year-to-year variation (see reviews,
Monfort 2003, Schwartz and Monfort 2008).
We corrected for year-to-year variation, regardless of its
source, by ?rst applying a natural-log transformation to WP,
which served to make the variance more constant across
years. To adjust for remaining differences in the centers of
the distributions across years, we performed a year-centering
transformation by subtracting each observation from its
Figure 1. Winter-collected (late gestation) fecal progestagen levels (WP) (A), and year-centered ln(WP) (B) by year and assigned pregnancy status for Jackson
bison, northwest Wyoming, USA, 1997?2005 (dark horizontal lines in each box represent medians, boxes span the interquartile range [IQR], whiskers extend
to min. and max. observations within 1.5 the IQR, and points denote observations >1.5 the IQR).
4 Wildlife Society Bulletin
associated year average (Fig. 1B). This approach removed
most of the dependence of WP on year (Figs. 1 and 2).
Second, because most females were accompanied by mul-
tiple observations over time (repeated measures), we assessed
the assumption of independence among all observations.
Ignoring females in the analysis assumed the relationship
between the probability of pregnancy and (transformed) WP
was the same for each female (homogeneity among F), and
that each observation was independent, even if it came from
the same female in a different year. However, we found no
clear indication of heterogeneity among females. For exam-
ple, a visual cut-off of approximately zero for year-centered
ln(WP) was effective at separating pregnant from non-preg-
nant observations, regardless of the female or year (Fig. 2).
We decided against using random effects for females because
of the small number of repeated measures for most individu-
als and our ultimate interest in predicting for females outside
the sample. However, we did use generalized estimating
equations to compare an ??independent?? model to one ac-
counting for within-female dependence (e.g., Zurr et al.
2009) and detected no practical difference in the estimates
or standard errors between the models.
Third, we assessed the potential for serial dependence
among the binary responses from the same female, which
would manifest as longer successive sequences (runs) of 1s or
0s than would be expected by chance. With our short time
series, this was an inherently dif?cult assumption to assess
(Fig. 2).Most runs that contained more than one observation
were for pregnant status, which is to be expected for females
of reproductive age, where the probability of pregnancy is
much >0.5. A small number of females could have had
chronic or acute health issues that led to successive years
of non-pregnant status, but it did not appear that observa-
tions within a run of non-pregnant observations behaved
differently than non-pregnant observations occurring amidst
2 runs of pregnant observations (Fig. 2). Furthermore, we
were not overly concerned with potential serial dependence;
we maintain that the information in each observation is
worth that of an independent measurement regardless of
whether the animal was not-pregnant for 2 successive
years.
In our considerations of data independence, we kept our
foremost goal to develop a technique for non-invasively
predicting the probability of pregnancy from WP at the
forefront. Coupling our ultimate goal for the model with
the lack of clear evidence for heterogeneity, we considered all
observations independent and are con?dent standard errors
accompanying our estimates are not misleadingly small. An
important and often overlooked component to our method
is the use of uncertainty, and we believe the advantages of
incorporating the use of CIs into the process of prediction far
outweigh any disadvantages that could arise from a lack of
independence in this application.
We evaluated the assumption of linearity between the log
odds of pregnancy and the explanatory variable (transformed
WP [YC ln(WP)]), by calculating empirical logits for arti?-
cially created bins of WP values for data collapsed over
females and years. Plots of the empirical logits versus the
explanatory variable revealed the linearity assumption was
adequately met.
Finally, because our study population was brucellosis-
infected, and brucellosis is known to heavily infect the pla-
centa of bison (Cheville et al. 1998, Rhyan et al. 2001; an
important producer of progesterone), we were concerned
with the possibility that disease status could affect our fecal
progestagen measurements. Our question focused on the
possibility that fecal progestagen values were not only indic-
ative of pregnancy status, but also of disease status, and that
our model could misclassify pregnant animals as non-preg-
nant due to active brucellosis infections and reduced produc-
tion of progesterone. Quantitative serology is known to be
related to infection in greater Yellowstone area Brucella-
infected bison (Roffe et al. 1999). To investigate the rela-
tionship between infection status and fecal progestagen, we
ranked the Brucella antibody response of bison as either
??high?? or ??not high?? (low or negative antibody response)
based on the complement ?xation test (Roffe et al. 1999),
and compared mean (transformed) WP by serological status,
age, and pregnancy status using 128 cases for which we had
all 3 variables. We report here comparisons of ??high?? versus
Figure 2. Year-centered, log-transformed values of winter-collected (late
gestation) fecal progestagens (WP) by year and assigned pregnancy status
(solid circle ? pregnant, X ? not pregnant) for 42 female bison from the
Jackson bison herd, northwest Wyoming, USA, 1997?2005.
Cain et al. 
 Bison Pregnancy Prediction 5
??not high?? antibody response because these are most likely to
represent animals with active Brucella infections (high) versus
those without infections (not high). We found no evidence
of an interaction between age and serological status
(P ? 0.468), and we found inconclusive evidence of a dif-
ference in mean WP between animals with ??high?? and ??not
high?? titers after pregnancy status was accounted for (2-sided
P ? 0.092; Fig. 3). We estimated the mean WP for animals
with ??high?? titers to be 0.23 units smaller than for those with
??not-high?? titers (95% CI ? 0.49?0.037). Although we
did ?nd evidence of a slight decrease in WP for high-titered
animals, the magnitude was not large enough to negatively
affect predictions from our model. When simply comparing
Brucella antibody test-positive versus test-negative individu-
als, we found no evidence of a difference in mean (trans-
formed) WP.
Procedures were approved by the Animal Care and Use
Committee of Montana State University (proposal clearance
no. 97-779) and the Biological Resources Division?U.S.
Geological Survey, and conformed to the Animal Welfare
Act and U.S. Government principles for the use and care of
vertebrate animals used in testing, research, and training.
Our work was conducted under approved permits by the
National Park Service and Wyoming Game and Fish
Department.
RESULTS
Reproductive Status
From 1997 to 2005, we assigned pregnancy status using our
palpation and calf status criteria and obtained WP values for
155 observations of 42 unique bison; 114 were classi?ed as
pregnant and 41 as not pregnant. Years of data for individual
females ranged from 1 to 8 (x ? 3.7). Of the 42 unique
females, 34 were pregnant in 1 year and 25 were not
pregnant in 1 year. Only 15 females had 1 observation
for each status. Twelve of the 41 not-pregnant observations
came from females below reproductive age (<2 yr), partially
explaining the number of females with only non-pregnant
observations (Fig. 2). We collected winter fecal samples and
determined fecal progestagen levels for all samples. Of the
155 fecal samples, we collected 129 from immobilized ani-
mals and 26 from free-ranging individuals.
We obtained blood and conducted pregnancy-speci?c pro-
tein B pregnancy analyses for 129/155 observations for which
females were assigned a pregnancy status. Twenty-?ve of the
129 comparisons between our palpation-calf assigned preg-
nancy status and pregnancy-speci?c protein B-derived preg-
nancy status were in disagreement (Table 1).
Rectal palpation was performed for 123 of the 155 obser-
vations that were assigned a pregnancy status (32 non-pal-
pated F were assigned solely based on presence of a calf). Of
41 animals determined not pregnant from palpation, 2
(4.9%) were subsequently documented with calves. The 2
incorrect palpations occurred for observations with the sec-
ond- and third-longest palpation?birth intervals (186 and
197 days; median of all intervals ? 93 days, n ? 81), which
likely rendered fetal detection more dif?cult. Regardless,
based on these data, we estimated a false-negative palpation
rate across all palpation?birth intervals as about 5% in our
study. In addition to 88 females palpated as pregnant, sub-
sequently observed with a calf, and assigned as pregnant in
our analyses, we recorded 24 animals palpated as pregnant
that were never observed with a calf. These observations were
not used in ?tting the model because they did not meet our
criteria for assignment as pregnant or not pregnant.
Although these cases represented 22.6% (24/106) of obser-
vations recorded as pregnant from palpation, the potential
for fetal or neonatal loss subsequent to palpations precluded
us from estimating a false-positive palpation error rate.
Model Prediction
We investigated the prediction success of multiple models,
focusing on including predictors that would be useful for
predicting in new, unhandled, unmarked females. For ex-
ample, age will not be accurately known in these cases;
therefore, we had little interest in coming up with a ?nal
model for prediction that included it. Mis-prediction rates
Figure 3. Year-centered, log-transformed values of winter-collected (late
gestation) fecal progestagens [ln(WP)] grouped by disease status (Sero
Code) and coded by pregnancy status (circle ? pregnant, triangle ? not
pregnant) from the Jackson bison herd, northwest Wyoming, USA, 1997?
2005. The lines connect the average for each disease status within pregnancy
status.
Table 1. Comparison of pregnancy determination by pregnancy-specific
protein B (PSPB) from blood samples and a combination of rectal palpation
and annual calf status (assigned status) in Jackson bison, northwest
Wyoming, USA, 1997?2005 (shaded cells show disagreement between the
2 methods).
PSPB results
Assigned status
TotalNot pregnant Pregnant
Not pregnant 26 10 36
Pregnant 15 78 93
Total 41 88 129
6 Wildlife Society Bulletin
obtained using cross-validation varied from 6.45% to 14.19%
among the 4 logistic-regression models we investigated
(Table 2). The models including only year-centered
ln(WP) had the lowest mis-prediction rate (6.45%) and
the lowest overall percentage of CIs accompanying correct
predictions crossing the threshold value (Table 2). Correct
predictions were obtained with high con?dence using this
model and resulted in an overall successful pregnancy pre-
diction rate of 93.5%.
The model incorporating date of fecal-sample collection
(WDAT) slightly reduced the average estimation error and
the percent of CIs accompanying mis-predictions that did
not cross the threshold (Table 2). Fewer wrong predictions
were made with this model, but many of these were made
with false con?dence. On the other hand, the inclusion of
WDAT increased the mis-prediction rate from 6.45% to
7.10%. However, the small differences among the results
from the models are probably not practically meaningful and
either could be used to make predictions. For the following
results, we use the model with only YC ln(WP).
The estimated logistic-regression equation using year-cen-
tered ln(WP) was
Logit?P? ? 1:685 ? 3:940  ?YC ln ?WP?
Standard errors for the regression coef?cients were 0.338
(intercept) and 0.696 (slope). Performing the inverse logit
transformation to obtain the equation for estimating the
probability of pregnancy, we obtained
^P ?
expf1:69 ? 3:94  ?YC ln ?WP?g
1 ? expf1:69 ? 3:94  ?YC ln ?WP?g
The threshold for determining pregnant versus not-pregnant
status that resulted in the best predictive success for this
model was 0.60, and this was used to make all predictions.
Using the year-centered ln(WP) model, 110 of the 114
observations assigned pregnant status were correctly pre-
dicted as pregnant (Table 3). Of the 110, only 3 of the
95% CIs crossed the threshold, which indicated a lack of
con?dence in the correct prediction. The model correctly
predicted 35 of the 41 not-pregnant observations, with 3 CIs
crossing the threshold. Of the 10 total mis-predictions, 3 of
the associated CIs overlapped the threshold, which indicated
an appropriate lack of con?dence in the prediction. For
those not overlapping the threshold, which indicated false
con?dence in the predictions, most came very close to the
threshold and, thus, in practice would be ?agged as uncertain
predictions (Fig. 4).
In addition to estimating the probability of pregnancy for a
particular female given their fecal progestagen levels, the
logistic-regression model allowed interpretation of the rela-
tionship between the progestagen level and the odds of
pregnancy. As suspected we found convincing evidence
Table 2. Leave-one-out cross-validation measures of pregnancy status predictive success using winter-collected (late gestation) fecal progestins (WP) in
4 logistic-regression models for Jackson bison, northwest Wyoming, USA, 1997?2005 (YC ? yr-centered, WDAT ? sample collection date).
Model
% Mis-
predicted
Average
estimation error
% Mis-prediction
CIs not overlapping
cut-off
% Correct prediction
CIs covering
cut-off
Pregnancy P
cut-off value
WP 14.19 0.283 4.52 27.74 0.5
ln(WP) 12.26 0.223 5.81 13.55 0.5
YC ln(WP) 6.45 0.123 4.52 3.87 0.6
YC ln(WP) ? WDAT 7.10 0.117 3.23 5.81 0.5
Table 3. Pregnancy predictions using logistic regression and year-centered,
log-transformed values of winter collected (late gestation) fecal progestins
(WP) from Jackson bison, northwest Wyoming, USA, 1997?2005.
Model predictions
Assigned pregnancy status
TotalPregnant Not pregnant
CIs do not
overlap cut-off
Pregnant 107 5 112
Not pregnant 2 32 34
CIs overlap cut-off
Pregnant 3 1 4
Not pregnant 2 3 5
Total 114 41 155
Figure 4. Fitted assigned pregnancy probabilities and 95% confidence inter-
vals for the logistic regression model using year-centered, log-transformed
values of winter-collected (late gestation) fecal progestagens (WP) for
Jackson bison, northwest Wyoming, USA, 1997?2005 (dashed line repre-
sents proposed cut-off for model-predicted pregnancy status).
Cain et al. 
 Bison Pregnancy Prediction 7
that year-centered fecal WP level was associated with preg-
nancy status (ZWald?s ? 5.66, P < 0.001). A doubling of
year-centered WP was associated with an estimated 15.4
(?23.94)-fold increase in the odds of a female being pregnant
(95% pro?le likelihood CI from 6.7 to 45.6 times).
DISCUSSION
Our logistic-regression model provided predictions of mid-
to late-gestation bison pregnancy status based on one fecal
sample with high accuracy (93.5%). Other studies have
reported successful pregnancy-prediction rates based on fecal
hormones, of 100% in a small sample of bison (14 pregnant, 4
non-pregnant based on palpation; Kirkpatrick et al. 1992),
85% in captive moose (Alces alces; Monfort et al. 1993),
approximately 97?100% in elk (Garrott et al. 1998,
Stoops et al. 1999), and 93?100% in bighorn sheep
(Borjesson et al. 1996, Schoenecker et al. 2004). However,
unlike these studies, using logistic regression we derived a
measure of con?dence for each individual prediction, which
allows practitioners to further evaluate estimates for individ-
ual animals, and thereby improve prediction interpretations.
The Importance of Independent Pregnancy Measures
Our study underscores the importance of independently
validating fecal progestagens with other objective measures
of pregnancy when developing pregnancy?fecal hormone
relationships. Using calf status alone as an indicator of
pregnancy, for example, is not suf?cient, because fetal and
neonatal losses from predation, disease, or other environ-
mental factors are possible. In our study area, where brucel-
losis and predators capable of taking bison?wolves (Canis
lupus) and grizzly bears (Ursus arctos)?were common, we
documented a 22.6% reduction between bison females pal-
pated as pregnant and those observed with a calf within about
a week of birth. Fetal losses due to brucellosis infections in
our population could partially explain this reduction, but
abortion rates are generally low in chronically infected herds,
such as the Jackson and Yellowstone bison herds (Cheville
et al. 1998). We also considered the possibility that our
palpations could have contributed to fetal losses. To evaluate
this, we compared fetal?neonatal loss rates in our bison that
were and were not palpated, using the model presented
herein to retrospectively determine pregnancy status of ani-
mals that were not palpated and not observed with calves.We
found that 11/41 (26.8%) pregnant animals fell into this
category, which was similar to but slightly higher than the
fetal?neonatal loss rate in our palpated sample (22.6%),
which suggests that palpation had no effect. Neonatal pre-
dation could also have contributed, but our study was not
designed to measure this. Nevertheless, our data make it clear
that pregnancy estimates based on calf status alone can be
prone to underestimating bias.
Transrectal uterine palpation is generally regarded as an
accurate indicator of pregnancy status, but its reliability is
highly dependent on the experience of the practitioner and
the extent of fetal development (Weber et al. 1982). Using
experienced veterinarians with mid?late-gestation bison, we
documented a 5% false-negative palpation rate, which may
have been related to expectations of a larger fetal size at this
time in gestation during palpation. The 2 false-negative
palpations were in earlier stages of development than
May-parturient bison. We were unable to estimate false-
positive rates because of confounding factors discussed
above. Although the accuracy of palpation is probably suf?-
cient for most studies, we recommend that it be coupled with
or substituted by ultrasonography whenever possible to
improve our understanding of fecal progestagens as indica-
tors of pregnancy.
Interestingly, based on our pregnancy assignment protocol
using both palpation and calf status, we found pregnancy-
speci?c protein B to be an unreliable indicator of pregnancy
status. Pregnancy-speci?c protein B, which Haigh et al.
(1991) showed was 93% accurate in wood bison (Bison bison
athabascae), and which has been used in other studies as an
independent indicator of pregnancy status (Garrott et al.
1998, Russell et al. 1998, Joly and Messier 2005), disagreed
with 19% (25/129) of our pregnancy assignments based on
?eld measurements, including 10/36 (28%) assigned ??not
pregnant?? by pregnancy-speci?c protein B and con?rmed
to have calves later in the season (Table 1). Of these, blood-
sample?calf birth-date intervals were 66?78 days for 8 sam-
ples, 131 days, and 244 days, which suggests that only the
latter case could be suspected of early (gestation) sample-
related error. The laboratory that conducted our analyses
speci?ed 95% accuracy in bison.
Considerations for Model Refinement
Detailed knowledge of the normal endocrine excretion dy-
namics of bison was essential for determining the optimal
time for sampling (i.e., early- vs. mid- or late-pregnancy).
In our study, we used information derived from monthly
assessments of urinary and fecal progestagen in bison by
Kirkpatrick et al. (1992). However, these data were limited
to monthly mean progestagens assessed in 14 females sam-
pled before conception through the ?rst 4.5 months of
gestation (Sep, Nov, and Jan). Sampling strategies for preg-
nancy detection would likely bene?t from more comprehen-
sive longitudinal assessments of fecal progestagens excretion
in pregnant and non-pregnant females sampled before and
throughout the entire period of gestation. Aligning sampling
effort with the gestational stage when females excrete maxi-
mal fecal progestagens would increase the likelihood of
discriminating pregnant from non-pregnant or cycling
females (Monfort et al. 1993). Furthermore, more intensive
sampling, perhaps in a captive herd, would help to elucidate
the extent of among-individual variation, including age-
related effects on fecal hormone production. Nevertheless,
our model produced a useful predictor of pregnancy in bison
with a single mid?late-gestation fecal sample.
Fecal progestagens vary in non-pregnant, reproductively
active female bison, reaching a peak during the luteal phase
when the corpus luteum is producing progesterone. Though
bison are seasonally polyestrus in the late summer/early fall,
we occasionally observed young (<3 months old) calves, and
males tending apparently cycling females, during our winter
captures. We made no attempt to ascertain the reproductive
8 Wildlife Society Bulletin
status of non-pregnant bison (anestrus vs. actively cycling),
but this suggests the possibility of non-pregnant bison with
high WP in our sample. Some of the bison designated not
pregnant by our ?eld criteria but model-predicted to be
pregnant because of high WP levels may have been repro-
ductively active. Had we been able to control for this variable,
our model may have been improved.
MANAGEMENT IMPLICATIONS
Knowledge of pregnancy rates is often lacking in demo-
graphic models used to manage ungulate populations, but
their inclusion would greatly increase the informative power,
problem-solving attributes, and overall utility of such mod-
els. While our study results are directly applicable to bison,
the approach will facilitate the work of wildlife biologists,
researchers, reserve managers, and others by providing an
effective and ef?cient method for estimating pregnancy?
and assessing the associated uncertainty in the estimates?in
free-ranging ungulates at scales ranging from individual
animals to populations. The approach is useful to practi-
tioners because fecal samples are easily collected and pre-
served, laboratory procedures are well-documented, and
logistic regression is readily available in statistical computer
software. Furthermore, samples can be obtained non-inva-
sively, without the need to handle animals and incur draw-
backs associated with immobilization such as cost, human
and animal safety, potential data bias, and lack of social
acceptability. The bene?ts of the latter should not be under-
estimated in contemporary wildlife management or research
settings, given the increasing concern among the general
public and special interests for manipulation of wild animals
(see Peterson et al. 2003 and Farnsworth and Rosovsky
1993).
ACKNOWLEDGMENTS
Funding was provided by the National Park Service, U.S.
Fish and Wildlife Service, U.S. Geological Survey, and
Wildlife Conservation Society. C. Cunningham, K.
McFarland, K. Dhillon, K. Gasaway, and R. Leshan collect-
ed fecal samples. K. Mashburn, D. Kersey, and N. Presley
assisted with hormone analyses. C. Cunningham, K.
McFarland, M. Reid, K. Cannon, S. Sweeney, and K.
Cof?n assisted with bison immobilizations. B. Smith, B.
Reiswig, and F. Escobedo helped with ?eld logistics. The
manuscript bene?ted from the comments of 2 anonymous
reviewers. Any use of trade, product, or ?rm names is for
descriptive purposes only and does not imply endorsement by
the U.S. Government.
LITERATURE CITED
Agresti, A. 2002. Categorical data analysis. Second edition. John Wiley &
Sons, Hoboken, New Jersey, USA.
Association for the Study of Animal Behaviour. 2003. Guidelines for the
treatment of animals in behavioural research and teaching. Animal
Behaviour 65:249?255.
Ballard,W. B., and R.W. Tobey. 1981. Decreased calf production of moose
immobilized with anectine administered from helicopter. Wildlife Society
Bulletin 9:207?209.
Barrett, R. H. 1981. Pregnancy diagnosis with Doppler ultrasonic fetal pulse
detectors. Wildlife Society Bulletin 9:60?63.
Bekoff,M., andD. Jamieson. 1996. Ethics and the study of carnivores: doing
science while respecting animals. Pages 15?45 in J. L. Gittleman, editor.
Carnivore behavior, ecology, and evolution, II. Cornell University Press,
Ithaca, New York, USA.
Berger, J., and S. L. Cain. 1999. Reproductive synchrony in brucellosis-
exposed bison in the southernGreater Yellowstone Ecosystem and in non-
infected populations. Conservation Biology 13:357?366.
Berger, J., and C. Cunningham. 1994. Bison: mating and conservation in
small populations. Columbia University Press, New York, New York,
USA.
Berger, J., K. M. Murray, B. Buuveibaatar, M. R. Dunbar, and B.
Lkhagvasuren. 2010. Capture of ungulates in central Asia using drive
nets: advantages and pitfalls as illustrated by endangeredMongolian saiga.
Oryx 44:512?515. DOI: 10.1017/S003060531000058X
Berger, J., W. Testa, T. Roffe, and S. L. Monfort. 1999. Conservation
endocrinology: a noninvasive tool to understand relationships between
carnivore colonization and ecological carrying capacity. Conservation
Biology 13:980?989.
Borjesson, D. L., W. M. Boyce, I. A. Gardner, J. DeForge, and B. Lasley.
1996. Pregnancy detection in bighorn sheep (Ovis canadensis) using
a fecal-based enzyme immunoassay. Journal of Wildlife Diseases 32:
67?74.
Cheville, N. F., D. R. McCullough, and L. R. Paulson. 1998. Brucellosis in
the Greater Yellowstone Ecosystem. National Research Council, National
Academy Press, Washington, D.C., USA.
Cook, R. C., J. G. Cook, R. A. Garrott, L. L. Irwin, and S. L. Monfort.
2002. Effects of diet and body condition on fecal progestagen in elk.
Journal of Wildlife Diseases 38:558?565.
Cook, R. C., D. L. Murray, J. G. Cook, P. Zager, and S. L. Monfort. 2001.
Nutritional in?uences on breeding dynamics in elk. Canadian Journal of
Zoology 79:845?853.
Coulson, T., J.-M. Gaillard, and M. Festa-Bianchet. 2005. Decomposing
the variation in population growth into contributions from multiple
demographic rates. Journal of Animal Ecology 74:789?801.
Dein, F. J., D. E. Toweill, and K. P. Kenow. 2005. Care and use of wildlife
in ?eld research. Pages 185?196 in C. E. Braun, editor. Techniques for
wildlife investigations and management. Sixth edition. The Wildlife
Society, Bethesda, Maryland, USA.
Farnsworth, E. J., and J. Rosovsky. 1993. The ethics of ecological ?eld
experimentation. Conservation Biology 7:463?472.
Frison, G. C., and C. A. Reher. 1970. Age determination of buffalo by teeth
eruption and wear. Plains Anthropologist 15:46?50.
Fuller, J. A., R. A. Garrott, P. J. White, K. E. Aune, T. J. Roffe, and J. C.
Rhyan. 2007. Reproduction and survival of Yellowstone bison. Journal of
Wildlife Management 71:2365?2372.
Garrott, R. A., S. L. Monfort, K. L. Mashburn, P. J. White, and J. G. Cook.
1998. One-sample pregnancy diagnosis in elk using fecal steroid meta-
bolites. Journal of Wildlife Diseases 34:126?131.
Graham, L., F. Schwarzenberger, E. Mostl, W. Galama, and A. Savage.
2001. A versatile enzyme immunoassay for the determination of proges-
togens in feces and serum. Zoo Biology 20:227?236.
Haigh, J. C. 1990. Opioids in zoological medicine. Journal of Zoo and
Wildlife Medicine 21:391?413.
Haigh, J. C., M. Cran?eld, and R. G. Sasser. 1988. Estrus synchronization
and pregnancy diagnosis in red deer. Journal of Zoo Animal Medicine
19:202?207.
Haigh, J. C., C. Gates, A. Ruder, and R. Sasser. 1991. Diagnosis of
pregnancy in wood bison using a bovine assay for pregnancy-speci?c
protein B. Theriogenology 36:749?754.
Hein, R. G., J. L. Musser, and E. F. Bracken. 1991. Serologic, parasitic and
pregnancy survey of the Colockum elk herd in Washington. Northwest
Science 65:217?222.
Hosmer, D. W., and S. Lemeshow. 2000. Applied logistic regression.
Second edition. John Wiley & Sons, New York, New York, USA.
Houston, D. B., C. T. Robbins, C. A. Ruder, and R. G. Sasser. 1986.
Pregnancy detection in mountain goats by assay for pregnancy-speci?c
protein B. Journal of Wildlife Management 50:740?742.
Jessup, D. A., R. K. Clark, R. A. Weaver, and M. D. Kock. 1988. The
safety and cost effectiveness of net-gun capture of desert bighorn sheep
(Ovis canadensis nelson). Journal of Zoo Animal Medicine 19:208?213.
Cain et al. 
 Bison Pregnancy Prediction 9
Joly, D. O., and F. Messier. 2005. The effect of bovine tuberculosis and
brucellosis on reproduction and survival of wood bison in Wood Buffalo
National Park. Journal of Animal Ecology 74:543?551.
Kirkpatrick, J. F., K. Bancroft, and V. Kincy. 1992. Pregnancy and ovulation
detection in bison (Bison bison) assessed by means of urinary and fecal
steroids. Journal of Wildlife Diseases 28:590?597.
Kirkpatrick, J. F., D. F. Gudermuth, R. L. Flagan, J. C.McCarthy, and B. L.
Lasley. 1993. Remote monitoring of ovulation and pregnancy of
Yellowstone bison. Journal of Wildlife Management 57:407?412.
Kirkpatrick, J. F., S. E. Shideler, and J. W. Turner, Jr. 1990. Pregnancy
determination in uncaptured feral horses based on free steroids in feces and
steroid metabolites in urine-soaked snow. Canadian Journal of Zoology
68:2576?2579.
Larsen, D. G., and D. A. Gauthier. 1989. Effects of capturing pregnant
moose and calves on calf survivorship. Journal of Wildlife Management
53:564?567.
Lasley, B. L., and J. F. Kirkpatrick. 1991. Monitoring ovarian function in
captive and free-roaming wildlife by means of urinary and fecal steroids.
Journal of Zoo and Wildlife Medicine 22:23?31.
Linklater, W. L., K. M. Henderson, E. Z. Cameron, K. J. Stafford, and E.
O. Minot. 2000. The robustness of faecal steroid determination for
pregnancy testing Kaimanawa feral mares under ?eld conditions. New
Zealand Veterinary Journal 48:93?98.
McLaren, D. G., J. Novakofski, K. F. Parrett, L. L. Lo, S. D. Singh, K. R.
Neumann, and F. K. McKeith. 1991. A study of operator effects on
ultrasonic measures of fat depth and longissimus muscle area in cattle,
sheep and pigs. Journal of Animal Science 69:54?66.
Meagher, M. M. 1973. The bison of Yellowstone National Park. National
Park Service Scienti?cMonograph Series no. 1, Washington, D.C., USA.
Messier, F., D. M. Desaulniers, A. K. Goff, R. Nault, R. Patenaude, and M.
Crete. 1990. Caribou pregnancy diagnosis from immunoreactive proges-
tins and estrogens excreted in feces. Journal of Wildlife Management
54:279?283.
Monfort, S. L. 2003. Non-invasive endocrine measures of reproduction and
stress in wild populations. Pages 147?165 inD. E.Wildt,W. Holt, and A.
Pickard, editors. Reproduction and integrated conservation science.
Cambridge University Press, Cambridge, England, United Kingdom.
Monfort, S. L., N. P. Arthur, and D. E. Wildt. 1990. Monitoring ovarian
function and pregnancy by evaluating excretion of urinary oestrogen
conjugates in semi-free-ranging Przewalski?s horses (Equus przewalskii).
Journal of Reproduction and Fertility 91:155?164.
Monfort, S. L., C. C. Schwartz, and S. K. Wasser. 1993. Monitoring
reproduction in captive moose using urinary and fecal steroid metabolites.
Journal of Wildlife Management 57:400?407.
Morden, C. C., R. B. Weladji, E. Ropstad, E. Dahl, O. Holand, G.
Mastromonaco, and M. Nieminen. 2011. Fecal hormones as a non-inva-
sive population monitoring method for reindeer. Journal of Wildlife
Management 75:1426?1435.
Peterson, M. N., R. R. Lopez, P. A. Frank, M. J. Peterson, and N. J. Silvy.
2003. Evaluating capture methods for urban white-tailed deer. Wildlife
Society Bulletin 31:1176?1187.
Poole, J. H., L. H. Kasman, E. C. Ramsay, and B. L. Lasley. 1984. Musth
and urinary testosterone in the African elephant (Loxodonta africana).
Journal of Reproduction and Fertility 70:255?260.
R Development Core Team. 2012. R: a language and environment for
statistical computing. R Foundation for Statistical Computing, Vienna,
Austria: ISBN 3-900051-07-0, http://www.R-project.org
Ramsay, M. A., and R. M. F. S. Sadleir. 1979. Detection of pregnancy in
living bighorn sheep by progestin determination. Journal of Wildlife
Management 43:970?973.
Rhyan, J. C., T. Gidlewski, T. J. Roffe, K. Aune, L. M. Philo, and D. R.
Ewalt. 2001. Pathology of brucellosis in bison from Yellowstone National
Park. Journal of Wildlife Diseases 37:101?109.
Richards, M. W., J. C. Spitzer, and M. B. Warner. 1986. Effect of varying
levels of postpartum nutrition and body condition at calving on subsequent
reproductive performance in beef cattle. Journal of Animal Science
62:300?306.
Roffe, T. J., K. Cof?n, and J. Berger. 2001. Survival and immobilizingmoose
with carfentanil and xylazine. Wildlife Society Bulletin 29:1140?1146.
Roffe, T., J. C. Rhyan, K. Aune, L. M. Philo, D. R. Ewalt, T. Gidlewski,
and S. G. Hennager. 1999. Brucellosis in Yellowstone National Park
bison: quantitative serology and infection. Journal of Wildlife
Management 63:1132?1137.
Roffe, T. J., S. J. Sweeney, and K. Aune. 2005. Chemical immobilization of
North American mammals. Pages 286?302 in C. E. Braun, editor.
Wildlife investigational techniques. Sixth edition. The Wildlife
Society, Bethesda, Maryland, USA.
Russell, D. E., K. L. Gerhart, R. G.White, and D. Van DeWetering. 1998.
Detection of early pregnancy in caribou: evidence for embryonic mortality.
Journal of Wildlife Management 62:1066?1075.
Sasser, R. G., C. A. Ruder, K. A. Ivani, J. E. Butler, and W. C. Hamilton.
1986. Detection of pregnancy by radioimmunoassay of a novel pregnancy-
speci?c protein in serum of cows and a pro?le of serum concentrations
during gestation. Biology of Reproduction 35:936?942.
Schoenecker, K. A., R. O. Lyda, and J. Kirkpatrick. 2004. Comparison of
three fecal steroid metabolites for pregnancy detection used with single
sampling in bighorn sheep (Ovis canadensis). Journal of Wildlife Disease
40:273?281.
Schwartz, M. K., and S. L. Monfort. 2008. Genetic and endocrine tools
for carnivore surveys. Pages 228?250 in R. A. Long, P. MacKay,
J. C. Ray, and W. J. Zielinski, editors. Noninvasive survey methods
for North American carnivores. Island Press, Washington, D.C.,
USA.
Stephenson, R. R., J. W. Testa, G. P. Adams, R. G. Sasser, C. C. Schwartz,
and K. J. Hundertmark. 1995. Diagnosis of pregnancy and twinning in
moose by ultrasonography and serum assay. Alces 31:167?172.
Stoops, M. A., G. B. Anderson, B. L. Lasley, and S. E. Shideler. 1999.
Use of fecal steroid metabolites to estimate the pregnancy rate of a
free-ranging herd of Tule elk. Journal of Wildlife Management
63:561?569.
Testa, J. W. 2004. Population dynamics and life history trade-offs of moose
(Alces alces) in south-central Alaska. Ecology 85:1439?1452.
Thorne, E. T. 1982. Brucellosis. Pages 54?63 in E. T. Thorne, N. Kingston,
W. R. Jolley, and R. C. Bergstrom, editors. Diseases of wildlife in
Wyoming. Second edition. Wyoming Game and Fish Department,
Cheyenne, USA.
U.S. Department of Agriculture, Animal and Plant Health Inspection
Service [USDA APHIS]. 2003. Brucellosis eradication: uniform methods
and rules, effective October 1, 2003. APHIS Publication 91-45-013,
Washington, D.C., USA.
Valkenburg, P., R. O. Boertje, and J. L. Davis. 1983. Effects of darting and
netting on caribou in Alaska. Journal of Wildlife Management 47:1233?
1237.
Wasser, S. K., S. L. Monfort, J. Southers, and D. E. Wildt. 1994. Excretion
rates and metabolites of oestradiol and progesterone in baboon (Papio
cynocephalus cynocephalus) faeces. Journal of Reproduction and Fertility
101:213?220.
Wasser, S. K., R. Thomas, P. P. Nair, C. Guidry, J. Southers, J. Lucas, D. E.
Wildt, and S. L. Monfort. 1993. Effects of dietary ?ber on faecal steroid
measurements in baboons (Papio cynocephalus cynocephalus). Journal of
Reproduction and Fertility 97:569?574.
Weber, B. J., M. L. Wolfe, and G. C. White. 1982. Use of serum proges-
terone levels to detect pregnancy in elk. Journal of Wildlife Management
46:835?838.
Welsh, T. H., and B. H. Johnson. 1981. Stress-induced alterations in
secretion of corticosteroids, progestins, luteinizing hormone and testos-
terone in bulls. Endocrinology 109:185?190.
White, P. J., R. A. Garrott, J. F. Kirkpatrick, and E. V. Berkeley. 1995.
Diagnosing pregnancy in free-ranging elk using fecal steroid metabolites.
Journal of Wildlife Diseases 31:514?522.
Williams, E. S., E. T. Thorne, S. L. Anderson, and J. D. Herriges, Jr. 1993.
Brucellosis in free-ranging bison (Bison bison) from Teton County,
Wyoming. Journal of Wildlife Diseases 29:118?122.
Wood, A. K., R. E. Short, A. Darling, G. L. Dusek, R. G. Sassar, and C. A.
Ruder. 1986. Serum assays for detecting pregnancy in mule and white-
tailed deer. Journal of Wildlife Management 50:684?687.
Zurr, A. F., E. N. Ieno, N. J. Walker, A. A. Saveliev, and G. M. Smith.
2009. Mixed effects models and extensions in ecology with R. Springer,
New York, New York, USA.
Associate Editor: White.
10 Wildlife Society Bulletin