Estimating survival rates with time series of standing age-structure data

Author(s): Mark S. Udevitz and Peter J. P. Gogan

Source: Ecology, Vol. 93, No. 4 (April 2012), pp. 726-732
Published by: Wiley on behalf of the Ecological Society of America
Stable URL: https://www.jstor.org/stable/23213721
Accessed: 06-10-2018 21:15 UTC

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/23213721?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Wiley, Ecological Society of America are collaborating with JSTOR to digitize, preserve and
extend access to Ecology

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 Ecology, 93(4), 2012, pp. 726-732
© 2012 by the Ecological Society of America

Estimating survival rates with time series

of standing age-structure data
Mark S. Udevitz1'3 and Peter J. P. Gogan2

1 U.S. Geological Survey, Alaska Science Center, 4210 University Drive, Anchorage, Alaska 99508 USA
2U.S. Geological Survey, Northern Rocky Mountain Science Center, 2327 University Way, Suite 2, Bozeman, Montana 59

Abstract. It has long been recognized that age-structure data contain useful information
for assessing the status and dynamics of wildlife populations. For example, age-specific
survival rates can be estimated with just a single sample from the age distribution of a stable,
stationary population. For a population that is not stable, age-specific survival rates can be
estimated using techniques such as inverse methods that combine time series of age-structure
data with other demographic data. However, estimation of survival rates using these methods
typically requires numerical optimization, a relatively long time series of data, and smoothing
or other constraints to provide useful estimates. We developed general models for possibly
unstable populations that combine time series of age-structure data with other demographic
data to provide explicit maximum likelihood estimators of age-specific survival rates with as
few as two years of data. As an example, we applied these methods to estimate survival rates
for female bison (Bison bison) in Yellowstone National Park, USA. This approach provides a
simple tool for monitoring survival rates based on age-structure data.
Key words: age structure; bison; demographic rates; fecundity; inverse methods; mortality; population
growth rate; population models; survival; Yellowstone National Park.
+-J

Introduction tions that cannot be assumed stable. For example, these

o It has long been recognized that age-structure survival
datarates can be estimated with samples from the
standing
contain useful information for assessing the status and age structure and the age structure of mortality
dynamics of wildlife populations (Caughley 1977). Age year, without assuming stability, but with
from a single
structure data are often used alone or in combination additional information about the population growth
with other types of data (Cooper et al. 2003, Tavecchia rate (Udevitz and Ballachey 1998) or overall mortality
et al. 2009) to estimate demographic parameters such rate (Fryxell
as 1986). Alternatively, time series of standing
survival (Rohwer 2004), productivity (Peery et al. 2007), age-structure data and independent estimates of popu
or changes in population growth rate (Harris et lation al. size can be combined to estimate survival rates
2008). using inverse methods (Gross et al. 2002), which are a
Use of age-structure data alone to estimate demo general class of techniques for estimating demographic
graphic parameters requires relatively restrictive as parameters from time series of age (or stage) class
sumptions. Age-specific survival rates can be estimated abundances (Caswell 2001:142-154). However, estima
with just a single sample from a standing age distribu tion of survival rates using inverse methods typically
tion (i.e., distribution of age frequencies among living requires numerical optimization, a relatively long time
individuals) if the population is stable and stationary series of data, and smoothing or other constraints to
(Caughley 1977). Time series of such samples can be cope with problems of identifiability and sensitivity
used to increase precision of the estimated survival rates (Nelson et al. 2004).
that are assumed constant in stable populations. It is We developed general models that combine time
also possible to use just the time series of standing age
series of age-structure data with information about
structure data for estimating survival rates in popula
population growth rate or fecundity to provide explicit
tions that may not be stable, but the proportion of the
maximum likelihood estimators of age-specific survival
population included in each sample must be constant
rates. Age structures are not required to be stable and
(Hoenig and Gedamke 2007).
the survival rates are estimable with as few as two years
Other types of data may be combined with age
of age-structure data. Standard likelihood-based model
structure data for estimating survival rates in popula
selection procedures can be used to obtain more
parsimonious estimators that assume rates for some
Manuscript received 28 September 2011; revised 9 January
sets of age classes are the same, and we derived explicit
2012; accepted 10 January 2012. Corresponding Editor: B. D.
Inouye. expressions for these estimators, as well. As an example,
3 E-mail: mudevitz@usgs.gov we applied these methods to estimate survival rates for
726

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 ESTIMATING SURVIVAL RATES

female bison (Bison bison) in Yellowstone National

Park, USA (see Plate 1).

Methods
"EK2*^)*)- p)
All of the age class proportions in this likelihood can be
Time series of standing age-structure data can be
estimated directly; maximum likelihood estimators are
combined with information about either population
the same as in Eq. 2. Moreover, maximum likelihood
growth rate or fecundity to estimate age-specific survival
estimates of the survival rates are given by
rates without any assumptions about stability.

Estimates based on population growth rate f,=^-!~

Q-l,i-l
t = 2, ..., T (6)
Consider an age-structured population with w + 1 age
where
classes, at time points t,t=l,...,T, that are separated byX, is an independent maximum likelihood estimate
exactly 1 year. Let cit be the proportion of age class (Appendix
of X, i A). Thus, each successive pair of age
structure
individuals in the population and xit be the number of age samples provides estimates of survival during
the period between the sampling times.
class i individuals in a random sample from the population
The maximum likelihood estimates of all of the
at time t. The xit, i=0, ..., w, are referred to as the standing
parameters, including X„ can also be obtained simulta
age-structure data for year t. Conditioning on the total
neously by maximizing the full likelihood, consisting of
sample sizes, Y17=ox'ti the joint likelihood for these data is
the product of the unconditional likelihood for the
multinomial and is proportional to
population growth rate data and the conditional
likelihood Eq. 5 for the age-structure data. The form

(=1 1=0
nfe (i) of this likelihood (and the estimator for X,) will, of
course, depend on the sampling process used to obtain
the population growth rate data, but the maximum
Maximum likelihood estimates of the age class proportions

$
are given by likelihood estimators of the survival rates will always
have the form of Eq. 6 (Appendix B).
*» = -£- i = 0, w t=l, (2) The variance of c„ is estimated by
X/"
v=(4)-2^
Now, let N, be the total number of individuals in the
i=0
o
population in year t, so that Ntcit is the number of age class i
individuals in the population at time t, and let sit be the Using this with the delta method (Seber 1982) gives
probability of surviving to be in age class i+ 1 at time t+ 1 P"
for individuals in age class i at time I. Because w is the (P/A \ /(I — ci-\,t— l) (1 ~ Cit)\ *2
VarCs,= — + —
maximum age, sK,=0. Udevitz and Ballachey (1998) noted \ %it J
that the number of individuals in age class i at time t is
related to the number in age class i— 1 the previous year by
+ 6^) (7)
\Cl-1,(-1/
NtCu = i = 1, ..., w (3)
as an estimator for the variance of the survival rate
so that age class proportions areestimator
related (Eq.by
6).

Estimates based on fecundity

Cit — ^ 1 = 1, ...,w (4)
Eq. 4 gives expressions for age class proportions as
where X, = N,/N,_i is the population growth rate from
functions of the population growth rates. To derive
equivalent expressions as functions of fecundities, we
time / — 1 to t. Conditioning on X, and reparameterizing
Eq. 1 according to Eq. 4 for a single year of datanote
gives a the number of individuals in age class 0 at time
that
likelihood that depends on survival rates and age t is class
given by
proportions from the previous year, and these parame
ters are not estimable without additional assumptions N,c0, = W,_ i Cjj- if 1 (8)
(Udevitz and Ballachey 1998). However, with a series of 1=0

at least two years of standing age-structure data,

where f), is the expected number of individuals in age
conditional on the t = 2,... ,T, the likelihood
class 0 at time t + 1 produced per individual in age class i
becomes proportional to the following:
at time t. These are the parameters that appear in the
first row of a Leslie matrix, for example (Caswell 2001).
n<i n
Q—1 ,t-1 si-1 ,t— 1
They combine birth rates with survival of either the
h
newborn or the adult, depending on when the popula

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 728 MARK S. UDEVITZ AND PETER J. P. GOGAN Ecology, Vol. 93, No. 4

tion is assessed relative to births. From Eq. 8, it is tion growth rates, maximum likelihood estimates of the
evident that the population growth rates can be fecundities can be obtained as

expressed as /<-1 = coAt'

Estimates for stable populations

h= Jj-l (9) None of the development so far has made any
Cot Cot
assumption about the stability of the population. If
where f, is the overall (non-age specific) fecundity per reproductive and survival rates are constant over a long
individual in the population. Using this expression and enough period of time, a population will eventually
conditioning on the fecundities, rather than the popu attain a constant growth rate and a stable structure in
lation growth rates, the likelihood corresponding to Eq. which the age-class proportions will also be constant
5 becomes proportional to over time (Keyfitz 1985:255-262). Therefore, if we
assume the population is stable, the age class propor
tions at time t will be the same as at time t - 1 and we
IK II M fi-i have

COtCi-\,t-\Si-\,t-\ d, = —i=l, ...,w (12)

*n
/=i f
(10)
ryi
W*
and maximum likelihood estimates of the survival rates
where /,- is the assumed constant pro
are given by
that survive to age i,

Citft /o = l
$i— 1 ,t— 1 i=l, ...,w t = 2, T
coft'i-l.i-i
(11) i-i

u=w
where/, is a maximum likelihood estimate of the overall k=0

fecundity rate for year t (Appendix A). Again, the form

of the full likelihood (and the estimator for f,_\) will
vu ' depend on the sampling process used to obtain the
fecundity data, but the maximum likelihood estimators
of the survival rates will always have the form of Eq. 11
(Appendix B). The delta method (Seber 1982) variance
estimator for the survival rates in Eq. 11 is
X, is the assumed constant population growth rate, and
is the assumed constant survival rate for age class k
(Udevitz and Ballachey 1998). Using Eq. 12 in the
likelihood Eqs. 5 or 10 gives maximum likelihood
estimators of the of the survival rates that are the same
as Eqs. 6 or 11 without the subscripts for time and with
T

/ (1 — Q_i,/_i) , (1 - Co,)
Var(.v,_i,_i) = t= l . n
Xot a = ——-— i = o, ..., w.

1=0 (=1

(1-4) 2 .2 Assumption of stability will increase precision, but th

\-V
Xit resulting estimates will be biased if the assumption is no
J2x« valid. A likelihood-ratio test of this assumption can b
obtained by comparing the maximized likelihood with
the constraints Eq. 12 to the maximized likelihood
Ci,
Var (/,-])■ without these constraints, or likelihood-based model
C0lCi-U-l selection criteria such as AIC (Burnham and Anderson
From Eq. 9, it is also apparent that, given maximum 2002) can be used to select among models that do or d
likelihood estimates of the fecundities, maximum not assume stability.
likelihood estimates of the population growth rates can More parsimonious models
be obtained as
Often, there is interest in modeling age-related
patterns in survival rates or modeling them as function
/ Cj,t-tfi,t-l
of covariates. For example, survival might be modele
X,= M as a smoothly decreasing function of age such as
CQt Cot

or given maximum likelihood estimates of the popula sn = e °"

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 April 2012 ESTIMATING SURVIVAL RATES 729

where a is a constantlikelihood-ratio
hazard testrate (Siler
of this assumption 1979),
can then be or as
function of an environmental covariate such as obtained by comparing this maximized likelihood to the
Sit = I,Si + (1 - I,)s* maximized likelihood without this assumption, or
likelihood-based model selection criteria such as AIC
where /, is an indicator variable for the occurrence of (Burnham
an and Anderson 2002) can be used to select
environmental event such as an El Nino phase of theamong El models with different assumptions about age
Nino/La Nina Southern Oscillation in year 1, and st andrelated patterns in survival rates.
sf are the age class i survival rates associated with theAny number of age classes can be assumed to have the
occurrence or nonoccurrence of the event (Wielgus etsame al. survival rate; so, for example, if survival rates for
2007). It is straightforward to incorporate these types all
of ages < w are assumed to be the same, the maximum
functions into the likelihood Eqs. 5 and 10, butlikelihood
in estimator of this rate would be
general, the maximum likelihood estimators of the
survival rates will not have explicit forms and will have
to be found by numerically maximizing the full
likelihood. However, when there are just two years of (14)
data, there is at least one important class of more
parsimonious age-related patterns in survival rates that
does have relatively simple, explicit maximum likelihood
estimators. This is the class of models in which survival where again cit is as defined in Eq. 2 (Appendix A), and
the maximum likelihood estimates of the age class
rates for two or more age classes are assumed to be the
same. These types of models are often used with avian proportions in Eq. 5 are given by
populations, for example, where it is assumed that first
year survival rates differ from adult rates, but that rates
are the same for all of the adult age classes (Brownie et
al. 1985).
If there are just two years of data, the maximum
likelihood estimator of the survival rate for any set of
age classes that are assumed to have the same rate If is survival rates are assumed to be the same for some age
obtained simply by combining the corresponding age classes and there are more than two years of data, the
class proportions in the maximum likelihood estimators maximum likelihood estimators of survival rates may
of the individual rates. For example, if survival rates for
have explicit forms but, if so, they are more complex and
ages i and j are assumed to be the same, then the we did not attempt to derive them. Also, it does not
maximum likelihood estimate of this rate, conditional appear that maximum likelihood estimators of the age
on an estimate of the population growth rate is given by class proportions have explicit forms. In these cases, all pi
of the estimates can be obtained by numerically
__ (cu + cjt)i, maximizing the full likelihoods incorporating Eq. 5 or
—j (I J)
Cf— 11 + Cj-1](-|
Eq. 10 with the appropriate constraints on the survival
rate parameters.
where c„ is still as defined in Eq. 2, but is no longer the
maximum likelihood estimate of the corresponding
Application to Yellowstone Bison age
class proportion (Appendix A). Maximum likelihood
As an example,
estimators for the other survival we considered
rates are two years of standing
unchanged.
age-structure and
The maximum likelihood estimates ofaerial
the survey data obtained from the
corresponding
central subpopulation
age class proportions that appear (Olexa and Gogan
in likelihood 2007)5of are
Eq. bison
now given by (Bison bison) in Yellowstone National Park, USA. Age
and sex were determined for bison that were non
_ (XjJ-1 + +xj,t-1) selectively culled during management actions during the
''' ' + */+l,<) winters of 2001-2002 and 2002-2003. We considered
data from 42 females culled during April and May 2002
where
and 101 females culled during March 2003 that were
genetically determined to be members of the central
*=£ Xif
subpopulation (Fig. 1). We assumed the ages of these
females represented random samples from the age
These maximum likelihood estimates of the age class structures of the standing population in spring of those
proportions are useful because they can be inserted years. The oldest age in these samples was 14 years, so
along with the maximum likelihood estimates of the we assumed survival rates were 0 for ages >14.
survival rates (Eq. 13) to obtain the value of the We used the ratio of counts from aerial surveys
maximized likelihood Eq. 5 under the assumption ofconducted in spring 2002 and 2003 (Gates et al. 2005) to
equal survival rates for the specified age classes. A
estimate the population growth rate X2oo3 = 2975/3103 =

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 730 MARK S. UDEVITZ AND PETER J. P. GOGAN Ecology, Vol. 93, No. 4

2002
were quite large, ranging from 0.31 to 1.40. Less than
6% of the estimated variance was attributable to the
assumed variance associated with the variance of the

pi survey counts (i.e., the last term in Eq. 7). This suggests
r-j
that the age-structure sample sizes were not sufficient for
pi r~|
obtaining useful estimates of survival rates with this
—

II n n n n I"! n degree of age specificity. At the other extreme, however,

assuming that survival rates were the same for all age
2003 classes and using Eq. 14 provided a single overall
estimate of 0.85, with a standard error of 0.07, or a CV
—
of only 0.08. In this case, 80% of the variance was due to
r-j the assumed variance of the survey counts. This is the
—

n model that would be selected on the basis of AIC (AAIC

—
PI
> 6.0 for the other three models; Supplement 1).
n n No software was required for obtaining any of the
0 2 4 6 8 10 12 14
estimates in this example because all of the estimators
Age (yr) have explicit forms, but it is convenient to organize the
work in spreadsheets. Excel spreadsheets illustrating the
Fig. 1. Numbers of individuals in each age class in samples
of female bison obtained during spring of 2002 and 2003 from
calculations are available in the Supplement.
Yellowstone National Park, USA.
Discussion

Our approach may be viewed as belonging to the

0.96. The delta method (Seber 1982) variance estimator more general class of techniques, known as inverse
for this estimate of X, is methods (Caswell 2001:142-154), that estimate demo
graphic parameters from time series of data on
population size and structure. Inverse methods do not
v£(it) = + (J-^j vlr(/v() require any assumptions about population stability, but
they typically assume that vital rates are either constant
where N, is the population estimate for year t. With just
over time (Dennis et al. 1995, Gross et al. 2002, Wielgus
a single count each year, we had no estimate of variance
et al. 2008), smoothly varying functions of time (Wood
for N2002 or ^2003- For illustration purposes however, we
^ • 1994, Nelson et al. 2004), or functions of covariates that
assumed the coefficient of variation for these population
vary over time (Link et al. 2003, Holmes et al. 2007).
estimates was 0.05. This gave an estimate for Var(X.2oo3)
Inverse methods also generally require relatively long
of 0.0046 to use in Eq. 7. The counts provided no time series of data and numerical methods for estima
information about sex ratio, so we assumed the
tion. Our approach differs in that it makes no
population growth rate for females was the same as
assumption about time-related patterns in vital rates,
the overall population growth rate. requires only two years of data, and provides explicit
Using Eq. 6, it is possible to estimate survival rates for
estimators for age-specific survival rates and overall
all age classes with frequencies > 0 in 2002, but three of
the older age classes had frequencies of 0 and there were
also unreasonably large fluctuations in frequencies for
adjacent age classes in the 2002 data (Fig. 1). These
1.2
fluctuations were most likely due to the small sample size
for 2002, which was less than half that for 2003.
s 10"
Combining frequencies for each pair of adjacent age
CO

classes (i.e., combining frequencies ages 0 and 1, 2 and03± 3,

0.8
>
and so on) eliminated the 0 frequencies and gave a more
£ 0.6
reasonable age structure. This is equivalent to using
estimator Eq. 13 along with the assumption that survival
rates were equal for both age classes in a combined pair,
0.2
and it would allow estimation of survival rates for all age
classes. However, because of the particularly low O.O-I • -
number of 8- and 9-year-old individuals in the 2002 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

sample, the survival rate estimate for ages 8 and 9 would Age (yr)
still be >1. Adding the assumption that rates were the
Fig. 2. Estimated age-specific rates (±SE) of survival from
same for ages 6 through 11 gave reasonable estimates of
spring 2002 to spring 2003 for female bison in Yellowstone
survival rates for all age classes (Fig. 2), but coefficients
National Park. Rates were assumed to be the same within ages
of variation, based on variances estimated with Eq.0-1, 7, 2-3, 4-5, 6-11, and 12-13.

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
April 2012 ESTIMATING SURVIVAL RATES

Plate 1. Adult male plains bison (Bison bison) at Yellowstone National Park, US

fecundity or population growth time

structure methods, the data must
rate.
series
Asofwith
the comprise
sample of the population age-structure.
age class
many
abundances
angiven
linear dynamics unbiased
age for a population with
by Eqs. 3 and 8, there would
be no uncertainty in the annual age structures and
Our estimators of survival rates will be maximum population growth rates or fecundities and the survival
£
likelihood only if they are based on maximum likelihood
rates given by Eqs. 6 or 11 would be exact. o
estimates of population growth rates or fecundities. Much of the motivation for characterizing processes
However, valid estimates of the survival rates and of
their
annually changing vital rates in terms of mean
variances (not necessarily maximum likelihood) functions
can be comes from the problems of identifiability,
obtained using the methods outlined here based sensitivity,
on any and poor precision associated with applica
reasonable estimates (not necessarily maximum likeli
tions where there is significant uncertainty in the
hood) of the population growth rates or fecundities.
observation processes (Nelson et al. 2004). Small
Regardless of the length of the time series, our
samples are likely to produce spurious estimates of age
estimates of survival rates for any given year depend
structures, as was particularly evident with our bison
only on data from that year and the next; they are
data for 2002. Sample sizes will have to be large enough
independent of assumptions about population dynamics
to provide reasonably accurate estimates of annual age
or data from any previous years. Moreover, the process
structures for our approach to provide useful, explicit
of annually changing rates is completely described by this
estimates of survival rates with much age specificity.
sequence of annual estimates. All of the uncertainty in
However, smaller samples may support reasonably
the estimates comes from the observation processes used
accurate estimates of more coarsely defined age struc
to obtain data for estimating the standing age structures
tures, and reasonably precise, explicit estimates of
and either the population growth rates or fecundities.
overall survival rates for combined age classes. Likeli
Other inverse methods generally represent processes
hood-based model selection criteria such as AIC
of annually changing vital rates in terms of mean
(Burnham and Anderson 2002) can be used to determine
functions for the rates. At least part of the uncertainty
the degree of age specificity supported by the data.
associated with estimates of these rates is due to process
Alternatively, other age or time-related patterns of
variation consisting of deviations of the mean functions
survival,
from the actual rates. In some cases, inverse methods are including the types of mean functions that
have
applied to time series of age (or stage) abundances more typically been used in inverse methods, can
that
are completely enumerated (Dennis et al. 1995, Linkeasilyet be incorporated in our likelihood Eqs. 5 or 10.
Maximum likelihood estimators based on these types of
al. 2003, Wielgus et al. 2008) so there is no observation
assumptions will not generally have explicit forms and
error and all of the uncertainty is due to this process
the the
variation. If it was possible to completely enumerate estimates will have to be found numerically.

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms
 732 MARK S. UDEVITZ AND PETER J. P. GOGAN Ecology, Vol. 93, No. 4

Use of more complex age and time-related

Gross, K., B. A.mean
Craig, and W. D. Hutchison. 2002. Bayesian
functions for survival rates will extendestimation of a demographic matrix model from stage
the applicability
frequency data. Ecology 83:3285-3298.
of our approach to cases with time Harris,
series of sparser
N. C., M. J. Kauffman, and L. S. Mills. 2008. Inferences
annual age-structure data sets, but will likely require
about ungulate the
population dynamics derived from age ratios.
time series to be longer. Our explicit estimators will
Journal of Wildlife be
Management 72:1143-1151.
most useful for cases where only short Hoenig,
timeJ. M.,series
and T. Gedamke.
are 2007. A simple method for
available or where there is no basis or desire for estimating survival rate from catch rates from multiple years.
Transactions of the American Fisheries Society 136:1245
assuming more general patterns in survival.
1251.These
estimators provide a simple and convenient method for
Holmes, E. E., L. W. Fritz, A. E. York, and K. Sweeney. 2007.
monitoring survival rates based on annual samples of
Age-structured modeling reveals long-term declines in the
natality of western Steller sea lions. Ecological Applications
standing age-structure data and corresponding estimates
17:2214-2232.
of population growth rate or fecundity.
Keyfitz, N. 1985. Applied mathematical demography. Second
Acknowledgments edition. Springer-Verlag, New York, New York, USA.
Link, W. A., J. A. Royle, and J. S. Hatfield. 2003. Demographic
We are grateful to A. Gertonson and T. Linfield for allowing
analysis from summaries of an age-structured population.
sampling of culled Yellowstone bison. D. Corbet and K. Fogel
Biometrics 59:778-785.
facilitated collection of samples at the abattoirs. L. Bacon, L.
Nelson,
Barecz, M. Biel, D. Christianson, C. Hendrix, M. Lamoreux, E. W. A., E. McCauley, and J. Wimbert. 2004. Capturing
dynamics with the correct rates: inverse problems using
Olexa, k. Podruzny, D. Poinsette, J. Schomburg, and S.
semiparametric approaches. Ecology 85:889-903.
Wallace helped conduct the sampling. J. A. Mack, W. Brewster,
and other staff of the Yellowstone Center for Resources; Olexa, E. M., and P. J. P. Gogan. 2007. Spatial population
structure of Yellowstone bison. Journal of Wildlife Manage
Yellowstone National Park provided administrative and
ment 71:1531-1538.
logistical support. We also appreciate helpful comments by T.
Peery, M. Z., B. H. Becker, and S. R. Beissinger. 2007. Age
Arnold, J. Bromaghin, K. Irvine, and two anonymous reviewers
on earlier versions of the manuscript. Any mention of ratios
trade as estimators of productivity: testing assumptions on a
threatened seabird, the marbled murrelet (Brachyramphus
names is for descriptive purposes only and does not constitute
endorsement by the federal government. marmoratus). Auk 124:224-240.
Rohwer, S. 2004. Using age ratios to infer survival and despotic
Literature Cited breeding dispersal in hybridizing warblers. Ecology 85:423
431.
Brownie, C., D. R. Anderson, K. P. Burnham, and D. S.
Seber, G. A. F. 1982. The estimation of animal abundance and
Robson. 1985. Statistical inference from band recovery data
a handbook, second edition. Resource Publication Number related parameters. Second edition. Oxford University Press,
New York, New York, USA.
156. U.S. Fish and Wildlife Service, Washington, D.C., USA.
Burnham, K. P., and D. R. Anderson. 2002. Model selection Siler, W. 1979. A competing-risk model for animal mortality.
Ecology 60:750-757.
and multimodel inference. Springer-Verlag, New York, New
York, USA. Tavecchia, G., P. Besbeas, T. Coulson, B. J. T. Morgan, and
Caswell, H. 2001. Matrix population models. Sinauer Associ T. H. Clutton-Brock. 2009. Estimating population size and
ates, Sunderland, Massachusetts, USA. hidden demographic parameters with state-space modeling.
American Naturalist 173:722-733.
Caughley, G. 1977. Analysis of vertebrate populations. John
Wiley and Sons, New York, New York, USA. Udevitz, M. S., and B. E. Ballachey. 1998. Estimating survival
Cooper, A. B., R. Hilborn, and J. W. Unsworth. 2003. An rates with age-structure data. Journal of Wildlife Manage
ment 62:779-792.
approach for population assessment in the absence of
abundance indices. Ecological Applications 13:814-828. Wielgus, J., F. Ballantyne IV, E. Sala, and L. R. Gerber. 2007.
Dennis, B., R. A. Desharnais, J. M. Cushing, and R. F. Viability analysis of reef fish populations based on limited
Costantino. 1995. Nonlinear demographic dynamics: math demographic information. Conservation Biology 21:447
ematical models, statistical methods, and biological experi 454.
ments. Ecological Monographs 65:261-281. Wielgus, J., M. Gonzalez-Suarez, D. Aurioles-Gamboa, and
Fryxell, J. M. 1986. Age-specific mortality: an alternative L. R. Gerber. 2008. A noninvasive demographic assessment
approach. Ecology 67:1687-1692. of sea lions based on stage-specific abundances. Ecological
Gates, C. C., B. Stelfox, T. Muhly, T. Chowns, and R. J. Applications 18:1287-1296.
Hudson. 2005. The ecology of bison movements and Wood, S. N. 1994. Obtaining birth and mortality patterns from
distribution in and beyond Yellowstone National Park. structured population trajectories. Ecological Monographs
University of Calgary, Calgary, Alberta, Canada. 64:23-44.

Supplemental Material

Appendix A
Derivation of explicit estimators (Ecological Archives E093-063-A1).

Appendix B
Example illustrating equivalence of survival rate estimates based on full and conditional likelihoods (Ecological Archives E093
063-A2).

Supplement
Excel spreadsheet with example calculations (Ecological Archives E093-063-S1).

This content downloaded from 168.176.5.118 on Sat, 06 Oct 2018 21:15:35 UTC
All use subject to https://about.jstor.org/terms