Professional Documents
Culture Documents
Estimating Survival Rates With Time Series
Estimating Survival Rates With Time Series
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
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
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
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.
(=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).
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
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
/ (1 — Q_i,/_i) , (1 - Co,)
Var(.v,_i,_i) = t= l . n
Xot a = ——-— i = o, ..., w.
1=0 (=1
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
—
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
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
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