Ecology, 83(11), 2002, pp.

3003–3013
� 2002 by the Ecological Society of America

THE EFFECT OF PREY AND PREDATOR DENSITIES ON

WOLF PREDATION
JOHN A. VUCETICH,1 ROLF O. PETERSON, AND CARRIE L. SCHAEFER
School of Forestry, Michigan Technological University, Houghton, Michigan 49931 USA

Abstract. Predator kills rate (i.e., kills per predator per time) is routinely presupposed
to depend exclusively on prey density. However, per capita rates of killing may typically
depend on the density of both prey and predator. Unfortunately, our perception of many
ecological phenomena may be limited by the inappropriate assumption that kill rates do
not depend on predator density. One of many ways to represent the influence of predator
density is ratio-dependent predation, where kill rate depends on the ratio of prey to predator
rather than the actual numbers of prey and predator. Determining the role of ratio dependency
in predation theory has been contentious. Assessments of the influence of predator density
on kill rate have been primarily limited to theoretical considerations, indirect evidence, and
simplified laboratory demonstrations. We directly observed the influence of both prey and
predator density on kill rates in an unmanipulated terrestrial system of large mammals—
wolves (Canis lupus) and moose (Alces alces). Predator density explained more variation
in kill rate than did prey density (R2 � 0.36 vs. R2 � 0.17, respectively). Moreover, the
ratio-dependent model greatly outperformed the prey-dependent model. Nevertheless, the
ratio-dependent model failed to explain most of the variation in kill rate (i.e., R2 � 0.34).
The ratio-dependent–prey-dependent controversy may dissipate with greater appreciation
and acknowledgment that both models may be overly simplistic, both have value, and
neither deserves primacy.
Key words: Alces alces; Canis lupus; functional response; Isle Royale National Park (Michigan,
USA); kill ratio, per capita; moose–wolf system; predation, prey-dependent vs. ratio-dependent; pred-
ator kill rates; predator–prey dynamics.

INTRODUCTION pendence between prey equilibrium density and prey

Fundamental conceptions of predation evolved slow- growth rate (Oksanen et al. 1981); (3) the impossibility
ly during the 20th century. In the 1920s, predation was of prey equilibria that are both low and stable, i.e., the
modeled according to the chemical law of mass action, paradox of biological control (Arditi and Berryman
where per capita kill rate increases linearly with prey 1991); and (4) the indirect positive effect of increased
density (Lotka 1925, Volterra 1926). Although this the- primary productivity on predators prevents herbivores
ory may capture essential dynamics of some popula- from increasing in food chains with even numbers of
tions (e.g., Elton and Nicholson 1942, Huffaker 1958), trophic levels, but not in food chains with odd numbers
inadequacies were quickly noted (e.g., Thompson of trophic levels (Hairston et al. 1960; Arditi and Ginz-
1937, Smith 1952). In the 1950s, theoretical consid- burg 1989, the situation is more complex for food webs
erations and behavioral experiments gave rise to other [Polis and Strong 1996]).
chemical models where per capita kill rate increases Whether these properties are really all that unusual
asymptotically with prey density (prey-dependent pre- can be rightly contested (Abrams 1994, 1997). For ex-
dation; Holling 1966). This theory forms the founda- ample, the paradox of enrichment has been reported to
tion of five decades of predation research (Taylor 1984, occur in microcosm experiments (Holyoak 2000).
Kerfoot and Sih 1987, Boyce 2000) and underlies par- Moreover, prey density is an important determinant of
adigms of population regulation (Hairston et al. 1960, per capita kill rate, and prey-dependent models have
Fretwell 1977, Oksanen et al. 1981). However, prey- been and continue to be useful. However, like all simple
dependent models have been criticized for being char- models, prey-dependent models are limited, and these
acterized by several allegedly unusual properties (Ar- limitations have motivated assessments of the notion
diti and Ginzburg 1989, Ginzburg and Akçakaya 1992, that additional factor(s) are also generally important
Akçakaya et al. 1995): (1) overexploitation of prey and determinants of the per capita kill rate.
destabilization of predator–prey interactions in re- During the past 25 yr, models have been developed
sponse to increased prey carrying capacity (i.e., the to account for the manner in which both predator den-
paradox of enrichment, Rosenzweig 1971); (2) inde- sity and prey density affect kill rate (Beddington 1975,
DeAngelis et al. 1975, Arditi and Ginzburg 1989, Ak-
Manuscript received 12 July 2001; revised 9 April 2002; ac-
cepted 12 April 2002. çakaya et al. 1995). Factors that may lead to predator-
1 E-mail: javuceti@mtu.edu dependent kill rates include: (1) various types of be-
3003
3004 JOHN A. VUCETICH ET AL. Ecology, Vol. 83, No. 11

havior by prey and predator, such as predator avoidance conceptual progress is hobbled by the relative paucity
(Abrams 1993), group hunting (Cosner et al. 1999), of direct empirical data. Since 1989, when ratio de-
and interference among predators (Beddington 1975); pendency was first formalized, as far as we know there
(2) the limitation of the predator population by re- has been only a single direct assessment of the influ-
sources other than prey (Abrams 1994); and (3) the ence of predator density and prey density on kill rate
allocation of prey among predators (Arditi and Ginz- (i.e., Reeve 1997, see also Arditi and Akçakaya 1990,
burg 1989). Predator dependence via prey allocation is Abrams and Ginzburg 2000). Ratio dependency has
thought to arise from processes that generate variation been supported largely on the basis that such models
in vulnerability among individual prey, such as prey predict positive correlations in abundance among tro-
refugia, age- or size-structured prey populations, or phic levels—a pattern consistent with observation and
spatial heterogeneity (Abrams and Walters 1996). experimentation (Arditi and Ginzburg 1989, Arditi et
Moreover, spatial heterogeneity may represent a ge- al. 1991a, b, Arditi and Saiah 1992, Ginzburg and Ak-
neric generator of predator dependence (Keeling et al. çakaya 1992, Abrams 1994, Akçakaya et al. 1995).
2000). Because many natural systems may be charac- However, such patterns represent indirect evidence of
terized by one or more such factors, predator-dependent ratio dependency and may be explained by other pro-
models accommodate what would seem to be a com- cesses (Abrams 1994). The lack of direct evidence has
mon phenomenon in nature (Abrams and Ginzburg also led to theoretical considerations playing an im-
2000). Predator-dependent models also lead to the res- portant role in evaluating the appropriateness of ratio
olution of many of the purported problems that char- dependency (Arditi and Ginzburg 1989, Freedman and
acterize prey-dependent models, and are consistent Mathsen 1993, Abrams 1994, Yodzis 1994, Akçakaya
with numerous observed patterns of community struc- et al. 1995, Abrams 1997). For example, the plausi-
ture (Hassell 1978, Abrams 1992, 1993, Akçakaya et bility of ratio dependence may be doubted because only
al. 1995). Although many of purported inadequacies of a limited set of plausible parameter values appear to
prey-dependent models can also be rectified within the be associated with persistence in ratio-dependent sys-
framework of prey-dependent theory (Abrams 1994), tems (Freedman and Mathsen 1993, Abrams 1994).
predator dependency remains an important topic for
Advancement in this theory-laden topic requires di-
investigation.
rect empirical assessments of the factors that may affect
Aside from the issue of whether predator density is
per capita kill rates. However, direct assessments of per
generally important is debate concerning how predator
capita kill rates can be limited by bias associated with
dependence should be incorporated into existing the-
prey depletion (Abrams 1994), the artificial setting of
ory. One view seems to be that because predator de-
laboratory experiments, and short time scales that re-
pendence arises from a variety of factors, prey-depen-
veal behavioral, but not population dynamical, aspects
dent models should be tailored for specific situations
of predation (Abrams 1994, Akçakaya et al. 1995).
according to the specific factors thought to cause pred-
While avoiding these pitfalls, we present a direct as-
ator dependence (see Abrams 1994, Abrams and Ginz-
burg 2000). Another view seems to be that ecological sessment of the influence of predator density and prey
thinking would be advanced by blanket replacement of density on the per capita kill rate for an unmanipulated
prey-dependent models with some generic form of system on a time scale relevant for population dynam-
predator dependency, except when specific situations ics. In doing so, we provide evidence required for as-
dictate the need for more specific formulation (Akçak- sessing the influence and nature of prey density and
aya et al. 1995, Abrams and Ginzburg 2000). A generic predator density on per capita kill rates.
form that has been suggested is ratio dependency,
where kill rate depends on the ratio of prey to predator METHODS AND ANALYSIS
rather than the actual numbers of prey and predator Data Collection
(Arditi and Ginzburg 1989). Mechanistically, ratio de-
pendency arises by replacing the constant prey detec- Our observations consist of predator (wolves, Canis
tion rate (a) of the prey-dependent formulation with a lupus) abundance, estimates of prey (moose, Alces al-
rate that decreases monotonically with predator density ces) abundance, and the number of kills made by pred-
(i.e., a/P; see Table 1). ators during winter observation sessions conducted
These alternative views underlie a decade-long de- each January and February from 1971 to 2001. Data
bate over the value of ratio dependency, and antipa- were collected in Isle Royale National Park, (Michigan,
thetic assertions that establishment of ratio dependency USA), an island (544 km2) in Lake Superior where
would represent a ‘‘major clarifying step for the future wolves and moose interact essentially as an isolated
of ecology (Slobodkin 1992:1566)’’, and that ‘‘wide- single-predator–single-prey system (Peterson and Page
spread acceptance of this idea would set predator–prey 1988). Immigration and emigration is likely zero, or at
theory back by decades (Abrams 1994:1842)’’. Al- worst negligible, for wolves and moose. Moose com-
though important philosophical issues underlie these prise �90% of the biomass in wolf diet (Peterson and
unresolved differences (Abrams and Ginzburg 2000), Page 1988); other species capable of preying upon
November 2002 WOLF PREDATION 3005

TABLE 1. The relative performance of instantaneous-kill-rate models fit to empirical data from the wolf (P) and moose (N)
populations in Isle Royale National Park, USA (1971–2001).

Model† Form �‡ Rank§ w� R2

Constant a 36.4 12 4.21 � 10 �9
0.0

Prey dependent
Linear (Lotke 1925, Volterra 1926) aN 36.5 13 4.00 � 10�9 0.0
Asymptotic (Holling 1959) aN/(1 � ahN) 20.9 9 9.91 � 10�6 17.1
Sigmoid (Holling 1959) aN2/(1 � ahN2) 21.6 10 6.69 � 10�6 16.4
Exponential (Rosenzweig 1971) a(1 � exp[�bN]) 20.7 8 1.09 � 10�5 17.3

Ratio dependent
Linear a(N/P) 56.8 14 4.75 � 10�13 0.0
Linear a � b(N/P) 3.7 6 0.052 30.7
Asymptotic (Arditi and Ginzburg 1989) aN/(P � ahN) 0 1 0.335 33.6
Sigmoid aN2/(P � ahN2) 6.1 7 0.015 29.1
Exponential (Gutierrez 1992) a(1 � exp[�bN/P]) 0.8 2 0.221 33.1

Predator dependent
Nonlinear, zero intercept (Hassell and Varley 1969) aN/Pm 33.6 11 1.68 � 10�8 5.0
Nonlinear, zero intercept (Hassell and Varley 1969) aN/[Pm � ahN] 2.2 4 0.112 33.6
Linear, non-zero intercept (Beddington 1975, aN/(bN � P � c) 1.3 3 0.177 34.3
DeAngelis et al. 1975)
Nonlinear, non-zero (Yodzis 1994) aNn/[b(c � P)m � Nn] ¶ ¶ ¶ 0
Nonlinear, zero intercept a(1 � exp[�bN/Pm]) 2.7 5 0.086 33.2
Note: N is prey abundance, P is predator abundance, and a, b, c, h, m, and n are parameters to be estimated from the data.
† Prey- and ratio-dependent models are described according to their shape in relation to kill rate. Predator-dependent models
are described by the shape of the corresponding predator isocline, assuming the numerical response is linearly related to the
functional response (see Yodzis 1994). Some of the ratio-dependent models are expressed in simplified terms, rather than in
terms of the ratio, N/P. Sources are in parentheses.
‡ � � the AICc (Akaike’s information criterion, corrected for small sample size) for the model of interest minus the smallest
AICc for the set of models being considered.
§ The rank order of the model on the basis of �.
� The AICc weight.
¶ The model could not be estimated. Four of the five parameters were unstable and not significantly different from zero
for all sets of initial conditions (see Results).

moose are absent, and hunting is prohibited on the on �44 d of observation (median � 44 d, interquartile
island. range � [38–47 d]), during which travels and kills of
The wolf population was censused annually from wolf packs were determined by aerial observations of
1971 to 2001 by use of fixed-wing aircraft. Confidence tracks in snow (Mech 1966, Peterson 1977). Per capita
in census accuracy is increased by: (1) the frequent kill rates were calculated as the number of kills made
visibility of entire wolf packs at a single location and by a pack divided by the number of wolves in that pack
time and (2) making several complete counts during divided by the number of days during which that pack
each winter survey. Moose abundance was estimated was observed. With 31 yr of winter observations and
annually from 1988 to 2001 by aerial survey using typically three packs in the population at a time (range
fixed-wing aircraft (Peterson and Page 1993). From � 2–5 packs), we obtained 94 measures of kill rate.
1971 to 1991 moose abundance was estimated by co- Additional details on these field methods are reported
hort analysis. The method of cohort analysis used here elsewhere (i.e., Thurber and Peterson 1993, Peterson
is similar to that described by Solberg et al. (1999). et al. 1998).
For complete details see J. A. Vucetich and R. O. Pe-
terson, unpublished manuscript. Between 1988 and Analysis
1991 estimates of moose abundance are available for We compared our kill-rate data to several models
both aerial surveys and cohort analysis. The results each belonging to one of three classes of instantaneous
presented below are based on estimates derived from kill-rate models (Table 1): (1) dependent exclusively
cohort analysis. However, the results are qualitatively on prey, (2) dependent on both prey and predator (here-
identical and quantitatively nearly identical when co- after, predator–prey dependent), and (3) dependent on
hort analysis estimates are replaced with aerial survey prey and predator, but only through the ratio. This set
estimates during the period of overlap. of kill-rate models can produce a predator isocline with
During winter wolves hunt and travel in groups any shape (i.e., linear, concave up, or concave down)
called ‘‘packs’’ (Olson 1938, Murie 1944). Kill rates or y-intercept (i.e., no intercept, zero intercept, and
were calculated for each wolf pack during each winter nonzero intercept) (see Yodzis 1994).
session from 1971 to 2001. Each kill rate was based Estimation of model parameters was based on min-
3006 JOHN A. VUCETICH ET AL. Ecology, Vol. 83, No. 11

imizing the (Gaussian) error sum of squares (i.e., TABLE 2. The relative performance of models that isolate
the influence of predator density on kill rate.
�(Xobserved � Xpredicted)2; see Zar 1984:263). Parameter
estimates were calculated by the nonlinear regression Form† �‡ w§ R2
routine of the software S-PLUS 2000 (MathSoft, Seat-
c � aP �1
0 0.63 35.5
tle, Washington, USA). Results were also confirmed c � a ( b � P ) �1 2.18 0.21 35.5
with the software SigmaStat Version 1.0 (Jandel Cor- a ( b � P ) �1 3.29 0.12 33.2
poration, San Rafaal, California, USA). Selection of aP�1 5.72 0.04 30.0
the most parsimonious model was based on the infor- † P is predator density; a, b, and c are parameters estimated
mation-theoretic approach (Burnham and Anderson from the data.
1998, Anderson et al. 2000). Suitable models with the ‡ � � the AICc (Akaike’s information criterion, corrected
for small sample size) for the model of interest minus the
smallest AICc (Akaike’s information criterion, cor- smallest AICc for the set of models being considered.
rected for small sample size) and � were selected as § The AICc weight.
the best model; � equals the AICc for the model of
interest minus the smallest AICc for the set of models
being considered. AICc relies on principles of parsi- consume prey (Rogers 1972, Abrams and Ginzburg
mony and information theory to estimate the relative 2000). In any event, prey depletion was minimal (i.e.,
distance between a model and the underlying process median � 1.6%; interquartile range � [1.1%, 2.3%]).
that created the observed data. Models with � � 2 are For the integrated models, parameter estimates were
considered worthy of consideration (Burnham and An- calculated by a program written in the programming
derson 1998). From AICc values we also calculated language C�� that considered the error sum of squares
AICc weight of each model i (wi). The ratio wi:wj es- for the entire plausible parameter space. Standard soft-
timates how many times more support the data provide ware packages cannot estimate the integrated models
for model i than model j. because the dependent variable (i.e., number of kills)
We also conducted four follow-up analyses designed cannot be isolated to a single side of the equality (Table
to increase confidence in the initial assessment. First, 3; Reeve 1997). Standard errors and P values for pa-
we examined models that are exclusively predator de- rameters of the instantaneous models were estimated
pendent (Table 2). Second, we addressed the possibility by observing the distribution of parameter estimates
that adequate descriptions of kill rate require account- for 1000 bootstrapped realizations of the original data
ing for a dramatic decline in wolf abundance that oc- set (Efron and Tibshirani 1993).
curred in the early 1980s. Between 1980 and 1982, the
wolf population declined from 50 to 12 individuals. RESULTS
Canine parvovirus was the probable proximate cause Each of the prey-dependent models performed dis-
of the crash, and genetic deterioration may have been mally compared to the other models (�’s � 20; Table
contributing to what may be altered dynamics between 1). The asymptotic form of a ratio-dependent model
wolves and moose since 1980 (Peterson et al. 1998, performed best (� � 0). It had 30, 700 times more
Peterson 1999). To test for this potential change in support than the best (i.e., asymptotic) prey-dependent
dynamics, we estimated and assessed models where model. The asymptotic ratio-dependent model also ex-
indicator variables were used to model differences in plained approximately twice as much variation in kill
parameter values before and after the 1980 population rates as did the asymptotic prey-dependent model (34%
crash. For example, to modify a type II prey-dependent vs. 17%; Fig. 1; Table 4). The similar performance of
functional response to account for time-dependent the exponential and asymptotic ratio-dependent models
predator handling time (h) we used the expression: aN/ (� � 0.8 and 0.0) may be attributable to similarities
(1 � a(h � (Ih�))N), where N is prey density, a is attack in their form.
rate, I � 0 for years prior to 1980, I � 1 for years after The predator–prey-dependent model derived by Bed-
1980, and the term Ih� represents the change in han- dington (1975) ranked third, and its � (�1.3) suggests
dling time for years after 1980. We also estimated and further consideration of this model is warranted. How-
compared models where kill rates were assumed to be ever, this model may be over-parameterized. In this
prey dependent during one time period and ratio de- model, c, which corresponds to the intercept of the
pendent during the other time period. predator isocline, did not differ significantly from zero
Third, we compared models (Table 3) where the de- (c � 6.75 � 5.62 [mean � 1 SE]; P � 0.23). Moreover,
pendent variable was kills per month rather than kills the estimate for c was positive, indicating predator per-
per predator per month (Table 1). This analysis alle- sistence in the absence of prey. The imprecise and, most
viates statistical concerns regarding regression where certainly, biased estimate of c indicates that the esti-
the dependent and independent variables are both func- mate is an unrealistic extrapolation arising from a
tions of predator density. dearth of observations for prey at low densities. Al-
Fourth, we compared instantaneous kill-rate models though the underlying process may be well represented
(Table 3), which account for prey depletion that occurs by a model with a nonzero intercept like that of Bed-
during each season’s period of observation as predators dington (1975), 31 years of data from Isle Royale
November 2002 WOLF PREDATION 3007

TABLE 3. The relative performance among models expressed in terms of kill rate, kills per
month (rather then per capita kill rate) and the relative performance among models of in-
tegrated kill rate (dependent variable � kills).

Description Form† �‡ w§ R2
Dependent variable: kills/mo
Prey dependent aNP/(1 � ahN) 14.4 0.001 12.4
Ratio dependent aNP/(P � ahN) 0 0.729 24.9
Predator and prey aNP/(bN � P � c) 2.0 0.271 25.1
Dependent variable: kills
Asymptotic ratio dependent N[1 � exp(�aT � bKP�1)] 0 0.737 36.1
Predator dependent� N[1 � exp(�aTP1�m � bKP�m)] 2.1 0.262 27.1
Asymptotic prey dependent¶ N[1 � exp(�aTP � bK)] 12.5 0.001 25.9
Notes: Models were fit to empirical data from the wolf and moose populations in Isle Royale
National Park, Michigan, USA (1971–2001). Comparisons based on AICc between, for example,
the instantaneous and integrated forms of the ratio-dependent model are not possible because:
(1) AICc can only be compared among models with exactly the same set of observations for
the dependent variable, and (2) the dependent variable is ‘‘kills per wolf per month’’ for the
instantaneous models and ‘‘kills’’ for the integrated models.
† N � prey (moose) density, P � predator (wolf) density, K � the number of kills observed
in T, and lowercase letters are model parameters.
‡ � � the AICc (Akaike’s information criterion, corrected for small sample size) for the
model of interest minus the smallest AICc for the set of models being considered (in this table).
§ The AICc weight.
� Derived from Hassell-Varley model of Table 1 (see Reeve 1997).
¶ See Rogers (1972).

(Michigan, USA) are inadequate for such an assess- creased predator density tends to reduce kill rates. The
ment. basic structure for expressions relating the effect of
Similarly, two of the predator–prey-dependent mod- predator density to kill rate is 1/P (e.g., the Beddington
els corresponding to nonlinear predator isoclines per- model). Thus, to assess the role of predator density,
formed reasonably well (i.e., � � 2.2 and 2.7; Table we examined four models based on this structure (Table
1). However, these models were characterized by non- 2). The best of these models received �3 times more
significant parameter estimates (a � 0.67 � 0.86 [mean support than the second-ranked model (Table 2). More-
� 1 SE] for the model with � � 2.2; b � �7.2 � 6.8 over, on the basis of proportion of total variation ex-
for the model with � � 2.9). The apparent over-pa- plained, kill rates are more sensitive to changes in pred-
rameterization of these models suggests that either the ator density (R2 � 0.36; Table 2) than prey density (R2
underlying relationship is well approximated by a lin- � 0.17; Table 1).
ear predator isocline, or the Isle Royale data are in- Per capita kill rates appear to be well described with-
adequate for discerning such nonlinearity (see Carpen- out explicitly accounting for potential changes in dy-
ter et al. 1994). Finally, the predator–prey-dependent namics associated with the population crash of 1980.
model presented by Yodzis (1994), which corresponds More specifically, prey-dependent and ratio-dependent
to a nonlinear predator isocline and nonzero intercept models including indicator variables to account for
is apparently so over-parameterized that for any set of time-dependent constants (see Methods and analysis:
initial conditions four of the five model parameters Analysis, above) were less parsimonious (i.e., � � 16.4
were unstable and not significantly different from zero for all prey-dependent models with indicator variables,
(i.e., all P’s � 0.50). and �1.6 for all ratio-dependent models with indicator
Not surprisingly, the constant model (i.e., a), and the variables). Models where dynamics were assumed to
models that are linear with respect to a (i.e., aN, aN/ be prey dependent during one time period and ratio
Pm, a(N/P)) performed poorly overall and within each dependent during the other time period were also less
of their respective classes (�’s � 33). Also not sur- parsimonious (i.e., � � 7.7) than the simple asymptotic
prisingly, the sigmoidal models of each class performed ratio-dependent model.
less well than did the asymptotic models of the cor- The comparisons depicted in Table 1 are similar to
responding class (Table 1). Again, because of the dearth comparisons among models where the dependent var-
of observations for prey at low density, it is unclear iable was kills per month (Table 3) and to comparisons
whether the underlying process is well represented by among integrated models that account for prey deple-
an asymptotic model or whether available data are in- tion (also Table 3). In both sets of comparisons, the
adequate for assessing a sigmoidal relationship (see ratio-dependent model greatly outperformed the prey-
Marshal and Boutin 1999). dependent model (Table 3). Among models depending
The high performance of the ratio-dependent (and on kills per month, and among integrated kill-rate mod-
predator–prey-dependent) model(s) suggest that in- els, the ratio-dependent model received more than 700
3008 JOHN A. VUCETICH ET AL. Ecology, Vol. 83, No. 11

proportions of total variation explained by the instan-

taneous (R2 � 0.34; Table 1) and integrated (R2 � 0.36;
Table 3) ratio-dependent models were similar. How-
ever, the proportion of total variation explained by the
ratio-dependent model with kills per month as a de-
pendent variable was somewhat lower (i.e., R2 � 0.25).

DISCUSSION
Implications for understanding wolf predation
The per capita rates at which Isle Royale (Michigan,
USA) wolves kill moose is poorly described by prey-
dependent models, and much better described by the
asymptotic ratio-dependent model (Table 1, Fig. 1). To
this extent, our results are consistent with the hypoth-
esis that correlations in wolf and prey density across
ecosystems (Arditi et al. 1991a, see also Fuller 1989)
arise from per capita kill rates that are approximated
by the ratio-dependent models.
Moreover, wolf density was a better predictor of kill
rate than was moose density (Fig. 1, Table 2). The
importance of wolf density is consistent with the ob-
servation that wolf–prey interactions are characterized
by many of the factors expected to generate predator
interference and promote predator-dependent kill rates
(see Abrams and Ginzburg 2000). Wolves exhibit group
hunting, aggressive interactions among groups, and so-
cial interactions that are time consuming and density
dependent. Predator dependence may also be promoted
by the limited number of sites where prey can be cap-
tured easily (e.g., winter yarding areas). In moose pop-
ulations, the proportion of prey vulnerable to predation
varies greatly with age structure, because juvenile and
senescent moose are more vulnerable to wolf predation
(Peterson et al. 1984, see also Murie 1944). Finally,
wolves and their prey interact in a spatially heteroge-
neous environment (see Keeling et al. 2000). Despite
intensive study of these populations, it would be nearly
impossible to build a useful mechanistic model capable
of disentangling how each of these factors give rise to
predator dependence. The phenomenological nature of
the ratio-dependent model represents a critical first step
for better understanding the dynamics of wolf–prey
systems. This complexity probably typifies many or
most natural systems.
FIG. 1. The effect of (a) prey density, (b) prey to predator The inadequacy of prey-dependent kill-rate models
ratio, and (c) predator density on the kill rate (i.e., kills per
wolf per month). Each data point represents an estimate for does not, by itself, imply that moose predation by
each pack present in Isle Royale National Park (Lake Su- wolves is well described by ratio-dependent models.
perior, Michigan, USA) during each year from 1971 to 2001. On the contrary, the ratio-dependent model explained
The solid lines represent the best-fitting models (see Tables only 34% of the variation. The rate at which wolves
1 and 2). Solid symbols represent observation prior to 1980,
and open symbols represent data after 1980, when the pop- prey on moose appears to be substantially more com-
ulation crashed from 50 to 12 wolves. plex than: aN/(P � ahN) (Table 4). Frustratingly, more
complex models depending on prey and predator den-
sity do not explain enough additional variation to be
times more support than the prey-dependent model. justified on the basis of parsimony (Table 1). Never-
The predator–prey-dependent models also performed theless, it is significant that a single variable (the ratio
well, but were over-parameterized similar to the Yodzis of prey to predators) explains as much variation as it
model (Table 1) and Beddington model (Table 4). The does when one considers the many factors that could
November 2002 WOLF PREDATION 3009

TABLE 4. Parameter estimates for the best model within each of the four model classes (i.e.,
prey, ratio, predator–prey, and integrated; see Tables 1 and 3).

Model† Parameter � 1 SE P
aN/(1 � ahN) a � 1.48 � 10�3 � 3.35 � 10�4 �0.01
h � 0.42 � 0.13 �0.01
aN/(P � ahN) a � 4.19 � 10�2 � 7.51 � 10�3 �0.01
h � 0.58 � 6.42 � 10�2 �0.01
aN/(bN � P � c) a � 3.39 � 10�2 � 8.18 � 10�3 �0.01
b � 2.25 � 10�2 � 5.63 � 10�3 �0.01
c � 6.75 � 5.62 0.23
N[1 � exp(�at � aKP�1)] a � 3.1 � 10�4 � 4.7 �10�5 �0.01
b � 64 � 7.9 �0.01
aNP/(P � ahN) a � 1.06 � 10�2 � 1.77 � 10�3 �0.01
h � 2.30 � 0.38 �0.01
† In the models, N � prey (moose) abundance, P � predator (wolf) abundance, K � number
of kills observed, and lowercase letters are model parameters.

plausibly have important influences on kill rate (e.g., unpublished data). Also, stochasticity may arise if
climate, social structure, age structure of prey and pred- chance predation events lead to substantial deviations
ator, and spatial heterogeneity in prey and predator hab- from expected rates. Such stochasticity may be im-
itat quality). For example, even though disease has been portant for predation in locally small populations (e.g.,
an important component of the dynamics of the Isle rare or endangered predators or prey, predator–prey
Royale system, wolf and moose density are neverthe- systems in fragmented landscapes). In any event, sto-
less useful predictors of kill rate. chasticity associated with kill rates may have important
Although the lack of stability observed in the Isle and potentially nonintuitive implications for the dy-
Royale wolf–moose system is at odds with the tendency namics of predator–prey systems (May 1976, Markus
for deterministic models of ratio-dependent predation et al. 1987, Boyce 2000), and warrants further study.
to produce highly stable point equilibria, highly vari-
able population abundances might be explained if a The application of models to management
substantial portion of the temporal variation in kill rate and conservation
was inherently stochastic. In fact, interannual vari- Our study highlights a dilemma for the interpretation
ability in winter climate may explain an additional 10– and application of empirically based studies of pre-
15% of the total variation in kill rates for Isle Royale dation (see also Yodzis 1994, Abrams and Ginzburg
wolves and moose (J. A. Vucetich and R. O. Peterson, 2000):
1) The ratio-dependent model predicts a linear pred-
ator isocline with a positive slope that is determined
by empirical observation, but is restricted to having a
zero intercept. A predator isocline that passes through
the origin is problematic because it predicts predator
persistence at vanishingly small numbers of prey (Fig.
2; Abrams 1994), and has been the basis for justifying
misguided management (Yodzis 1994).
2) The prey-dependent model relies on empirical ob-
servation to predict the value at which the predator
isocline crosses the prey axis, but restricts the isocline
to being vertical. A vertical predator isocline excludes
the possibility of potentially important dynamical prop-
erties (e.g., in a simple prey-dependent model the prey
FIG. 2. The effect of prey reduction on predators with equilibrium is independent of increased prey growth
ratio-dependent and predator-dependent kill rates. Equilibria rate and carrying capacity; Arditi and Ginzburg 1989,
occur wherever prey isoclines (lines with circles) intersect but see Abrams 1997).
predator isoclines (lines without circles). Dramatic reduction 3) The predator–prey-dependent model predicts a
in the prey carrying capacity (shifting from line with solid
circles to line with hollow circles) causes dramatic population linear isocline with a positive slope and (nonzero)
decline for ratio-dependent predators (dashed line) and ex- intercept that are both determined by the data, but the
tinction for predator–prey-dependent predators (solid line). intercept cannot be estimated accurately or precisely
For example, as prey carrying capacity shifts from high to due to a dearth of observations at low prey density
low, equilibrium abundance for ratio-dependent predators
shifts from P1 to P2. Contrary to common sense, ratio- (Fig. 2).
dependent predators persist even when prey are extremely Although the ratio-dependent model solves the prey-
rare (see Yodzis 1994). dependent model’s greatest weakness, and vice versa,
3010 JOHN A. VUCETICH ET AL. Ecology, Vol. 83, No. 11

neither model has a completely satisfying structure. In generational overlap, four is a reasonable value for the
contrast, the predator–prey-dependent model possesses smallest number that can be considered a wolf popu-
a satisfying model structure; however, it depends on lation.) Under some circumstances (e.g., Yodzis 1994),
parameters that often will not be reliably estimatable a conservative estimate might be appropriate and thus
from field data (Tables 1 and 4). A useful model would justify a value as high as 82 � 81 � 163 for � (Fig.
seem to require both appropriate model structure and 3). In the absence on this sort of information, even a
reasonable parameter estimates. Empirically based well-reasoned value of � might be more appropriate
models may also require the flexibility to retain what than assuming its value to be zero, as in the ratio-
is assumed to be appropriate structure even if the data dependent model. For example, because a pack of four
are unavailable to support such complexity (see Boyce wolves needs to kill a moose approximately every 15
[2000] for a comparable philosophy; cf. Burnham and d to survive (Thurber and Peterson 1993), and because
Anderson 1998). If common sense is used to veto the a typical recruitment rate for a moose population is
model structure supported by available data, then the �13% (Peterson and Vucetich 2001), � should be �190
properties of the model not explicitly favored by the (�365 d � 15 d per moose � 0.13) moose (Fig. 3).
data should be clearly identified and isolated. Although assigning a value to � may be reasonable if
Fortunately, the ratio-dependent model is easily ad- appropriate data are unavailable, assigning a value of
justed to accommodate this approach. Specifically, a � that conflicts with the data may be less appropriate.
predator isocline with a negative intercept can be To assess this possibility, we calculated � for models
achieved by subtracting from N (prey abundance) the with assumed values of � (� � 0.0 for � � 0 [see Table
threshold number of prey (�) below which a predator 1]; � � 0.6 for � � 82; � � 2.0 for � � 190).
population cannot be supported: This illustration highlights the challenge for science-
kill rate � a(N � �)/[P � ah(N � �)] (1) based management to carefully balance empirical ob-
servation, theoretically derived knowledge, and com-
where P � predator abundance. If the numerical re- mon sense. In some cases, it may be ‘‘well worthwhile
sponse is a linear function of the kill rate (Yodzis 1994, to add one or two parameters to get results that make
Ginzburg 1998), this model predicts a linear predator sense’’ (Abrams 1997:170). Unfortunately, it can be
isocline with a slope of a(q � h) and an intercept of easy to underestimate the difficulty of clearly identi-
�a(h � q), where q is trophic efficiency (e) divided by fying and isolating portions of a model more supported
mortality (m). In a mechanistic manner, this modifi- by ‘‘common-sense’’ than by empirical observation.
cation transforms the ratio-dependent model into a Although these considerations are critical for better
predator–prey-dependent model that can be expressed management and conservation, they may be less useful
in terms of Beddington’s (1975) model. This modifi- for advancing scientific understanding.
cation would be especially pertinent for analyzing how
human harvest might impact persistence. Most impor- Implications for resolving the controversy
tantly, if available data were too sparse to support any
value other than zero for �, other bases might allow A great deal of theoretical weight supports the notion
one to judge some other value of �, and the effect of that predator density can affect per capita kill rates
such subjective modification would be isolated and (Abrams and Ginzburg 2000; but see Rosenzweig and
could be easily assessed. MacArthur 1963). Theory also suggests that the mech-
To illustrate the application of this modification, we anisms underlying this influence are diverse, and that
used the data from Isle Royale. First, we fit the model these mechanisms, collectively, may occur commonly
to the data with � as a free parameter. The data suggest in nature (Abrams and Ginzburg 2000). Although direct
that � is 5.3 � 10�6 � 405.8 (mean � 1 SE; P � 1.00). measurement of the influence of predator density and
This result is analogous to the nonsignificant estimate prey density on kill rate is rare, the only experimental
of c in the predator–prey-dependent model (Table 4). assessment (Reeve 1997) and the only observational
Because wolf predation was not observed for small assessment (this study) support the claim that predator
values of N, these estimates represent an extrapolation. density can be an important influence on per capita kill
The estimate of � can also be judged unreliable because rate. Nevertheless, an assessment of the generality of
� � 0 corresponds to a model structure that predicts predator dependence currently eludes empirical con-
wolf persistence for extremely low moose abundances firmation. The prevalence of various mechanisms that
(i.e., predator isocline with zero intercept). Thus, we might lead to predator dependence is even less certain.
may be justified, or at least aboveboard, in selecting a Future studies should aim for a better empirical un-
value for � based on more reliable information. Infor- derstanding of the conditions (e.g., behavioral, life-
mation about � is contained in a correlative study of history, and environmental characteristics) for which
wolf and moose abundances across North American an accounting of predator dependence is essential. Al-
study sites (Messier 1994). The correlation produced though additional direct measurements will likely in-
by Messier (1994) predicts that four wolves could be crease our understanding, the notions that predator de-
supported by 82 � 81 (95% CI) moose. (Because of pendence is frequently important and that the causal
November 2002 WOLF PREDATION 3011

1), the ratio-dependent model for poor explanatory

power (R2 � 0.34, table 1), and the predator-dependent
models for over-parameterization (Tables 1 and 4). Al-
though each rejection is logically justified, rejecting
one overly simplistic model as a means of gaining pri-
macy for another overly simplistic model may not be
very useful. Controversy may dissipate with less focus
on declarations that one model or the other is wrong,
and more focus on appreciating the circumstances for
which each model is wrong, how wrong, and which
neglected factors contribute most to each model’s
wrongness.
Another source of controversy seems to be differing
views on the value of simple models of predation
(Abrams and Ginzburg 2000:341). One view appears
to be based on the premise that simple predation models
are most appropriately used as conceptual tools for the
development of theory. Thus, prey-dependent models,
with their solid theoretical underpinnings, should be
the foundation to which mathematical terms can be
added to account for specific mechanisms of predator
dependence. The other view is based on the premise
that simple predation models can be useful tools for
prediction and management. Under such circumstanc-
es, the influence of predator dependence may be sus-
pected or detected, but information may be too limited
to support models based on detailed mechanisms of
predator dependence. In such cases ratio dependence
FIG. 3. The effect of imposing a minimum threshold num- may typically be a useful and parsimonious approxi-
ber (�) of prey (moose) below which predators (wolves) can-
not persist on the predicted (a) per capita kill rate and (b) mation. Although our study exemplifies the second
predator isocline. The values for � are 0 moose (solid line), point of view, this aspect of controversy is resolved by
82 moose (dashed line), and 190 moose (circles); see Dis- recognizing that simple predation models are used both
cussion for justification. The curves in (a) were generated for the development of conceptual theory (Rosenzweig
from Eq. 1, and parameters were estimated from the data (see
Fig. 1). The curves in (b) are obtained by assuming that the 1971, Abrams and Walters 1996, Keeling et al. 2000)
numerical response is linearly related to the functional re- and as tools for prediction and management (Rosen-
sponse and that q, the ratio of trophic efficiency to predator zweig 1971, May et al. 1979, Boyce 1993, Flaaten and
mortality rate, is 0.20 (see text below Eq. 1). The curves Stollery 1994, Yodzis 1994, 2001, Spencer 1997).
depicted in (a) occur when the number of predators (P) is 20 Ratio-dependent predation deserves increased atten-
(see Eq. 1).
tion from theorists, empiricists, and educators (Ricklefs
and Miller 2000), because at times it will be the best
mechanisms are likely to be varied do not seem to elicit approximation of nature, and because the proper in-
much controversy (see Abrams and Ginzburg 2000). terpretation of simple models is so challenging. The
Controversy seems to arise over the relative value more general influence of predator density on kill rate
of ratio-dependent and prey-dependent models. Some deserves increased attention from empiricists to better
aspects of the controversy seem to have been resolved. understand its causes, frequency, and importance in
It is no longer generally accepted that primacy should natural systems. The primacy of exclusive prey depen-
be given to a process (e.g., ratio dependence) on the dence may be an artifact of historical development, not
basis of observing a pattern (e.g., correlated abun- its superior depiction of nature. Neither prey depen-
dances among trophic levels) that can be predicted by dency nor ratio dependency deserves primacy. Both are
many processes. sufficiently useful and limited and to serve as co-null
The controversy may persist, in part, from attempts models.
to reject prey-dependent models as a means of gaining
primacy for ratio-dependent models (and vice versa) ACKNOWLEDGMENTS
on the basis of purportedly anomalous predictions (e.g., We are grateful to P. Abrams, R. Arditi, J. Fryxell, and L.
Abrams 1994, 1997, Akçakaya et al. 1995). Our study Ginzburg for valuable discussion, and two anonymous review-
ers for helpful comments. We thank the U.S. National Science
highlights the limitations of this approach. For the Isle Foundation (DEB-9317401, DEB-9903671), EarthWatch, Na-
Royale system, one could reject the prey-dependent tional Geographic Society, and U.S. National Park Service for
model for poor relative performance (� � 20.7, Table financial support.
3012 JOHN A. VUCETICH ET AL. Ecology, Vol. 83, No. 11

LITERATURE CITED the North-east Atlantic minke whale’s (Balaenoptera acu-

torostrata) consumption of fish. Cemare Research Papers
Abrams, P. A. 1992. Predators that benefit prey and prey that
number 67.
harm predators: unusual effects of interacting for aging
Freedman, H. I., and R. M. Mathsen. 1993. Persistence in
adaptations. American Naturalist 140:573–600.
predator–prey systems with ratio-dependent predator influ-
Abrams, P. A. 1993. Why predation rate should not be pro-
portional to predator density. Ecology 74:726–733. ence. Bulletin of Mathematical Biology 55:817–827.
Abrams, P. A. 1994. The fallacies of ‘‘ratio-dependent’’ pre- Fretwell, S. D. 1977. The regulation of plant communities
dation. Ecology 75:1842–1850. by food chains exploiting them. Perspectives in Biology
Abrams, P. A. 1997. Anomalous predictions of ratio-depen- and Medicine 20:169–185.
dent models of predation. Oikos 80:163–171. Fuller, T. K. 1989. Population dynamics of wolves in North-
Abrams, P. A., and L. R. Ginzburg. 2000. The nature of ern Minnesota. Wildlife Monographs Number 105.
predation: prey dependent, ratio dependent or neither? Ginzburg, L. R. 1998. Assuming reproduction to be a func-
Trends in Ecology and Evolution 15:337–341. tion of consumption raises doubts about some popular pred-
Abrams, P. A., and C. I. Walters. 1996. Invulnerable prey ator–prey models. Journal of Animal Ecology 67:325–327.
and the paradox of enrichment. Ecology 77:1125–1133. Ginzburg, L. R., and H. R. Akçakaya. 1992. Consequences
Akçakaya, H. R., R. Arditi, and L. R. Ginzburg. 1995. Ratio- of ratio-dependent predation for steady-state properties of
dependent predation: an abstraction that works. Ecology ecosystems. Ecology 73:1536–1543.
76:995–1004. Gutierrez, A. P. 1992. Physiological basis of ratio-dependent
Anderson, D. R., K. P. Burnham, and W. L. Thompson. 2000. predator–prey theory: the metabolic pool model as a par-
Null hypothesis testing: problems, prevalence, and an al- adigm. Ecology 73:1552–1563.
ternative. Journal of Wildlife Management 64:912–923. Hairston, N. G., F. E. Smith, and L. B. Slobodkin. 1960.
Arditi, R., and H. R. Akçakaya. 1990. Underestimation of Community structure, population control, and competition.
mutual interference of predators. Oecologia 83:358–361. American Naturalist 94:421–425.
Arditi, R., and A. A. Berryman. 1991. The biological control Hassell, M. P. 1978. The dynamics of arthropod predator–
paradox. Trends in Ecology and Evolution 6:32. prey systems. Princeton University Press, Princeton, New
Arditi, R., and L. R. Ginzburg. 1989. Coupling in predator– Jersey, USA.
prey dynamics: ratio dependence. Journal of Theoretical Hassell, M. P., and G. C. Varley. 1969. New inductive pop-
Biology 139:311–326. ulation model for insect parasites and its bearing on bio-
Arditi, R., L. R. Ginzburg, and H. R. Akçakaya. 1991a. Var- logical control. Nature 223:1133–1136.
iation in plankton densities among lakes: a case for ratio- Holling, C. S. 1959. The components of predation as revealed
dependent predation models. American Naturalist 138: by a study of small mammal predation of the European pine
1287–1296. sawfly. Canadian Entomologist 91:293–320.
Arditi, R., N. Perrin, and H. Saı̈ah. 1991b. Functional re- Holling, C. S. 1966. The functional response of invertebrate
sponse and heterogeneities: an experimental test with cla- predators to prey density. Memoirs of the Entomological
docerans. Oikos 60:69–75. Society of Canada 48:1–60.
Arditi, R., and H. Saı̈ah. 1992. Empirical evidence of the Holyoak, M. 2000. Effects of nutrient enrichment on pred-
role of heterogeneity in ratio-dependent consumption. ator–prey metapopulation dynamics. Journal of Animal
Ecology 73:1544–1551. Ecology 69:985–997.
Beddington, J. R. 1975. Mutual interference between para- Huffaker, C. B. 1958. Experimental studies on predation. II.
sites or predators and its effect on searching efficiency. Dispersion factors and predator–prey oscillations. Hilgar-
Journal of Animal Ecology 44:331–340. dia 27:343–383.
Boyce, M. S. 1993. Predicting the consequences of wolf re- Keeling, M. J., H. B. Wilson, and S. W. Pacala. 2000. Re-
covery to ungulates in Yellowstone National Park. Pages interpreting space, time lags, and functional responses in
234–269 in R. S. Cook, editor. Ecological issues on rein- ecological models. Science 290:1758–1760.
troducing wolves into Yellowstone National Park. United Kerfoot, W. C., and A. Sih, editors. 1987. Predation: direct
Stated Department of Interior, National Park Service, Na- and indirect impacts on aquatic communities. University
ture Resource Publications office, Denver, Colorado, USA. Press of New England, Hanover, New Hampshire, USA.
Boyce, M. S. 2000. Modeling predator–prey dynamics. Pages Lotka, A. J. 1925. Elements of physical biology. Williams
253–287 in L. Boitani and T. Fuller, editors. Research tech- and Wilkins, Baltimore, Maryland, USA.
niques in animal ecology. Columbia University Press, New Markus, M., B. Hess, J. Rossler, and K. Kiwi. 1987. Popu-
York, New York, USA. lations under periodically and randomly varying growth
Burnham, K. P., and D. R. Anderson. 1998. Model selection conditions. Pages 267–277 in H. Degn, A. V. Holden, and
and inference. Springer-Verlag, New York, New York, L. F. Olsen, editors. Chaos in biological systems. NATO
USA. ASI Series 138. Plenum, New York, New York, USA.
Carpenter, S. R., K. L. Cottingham, and C. A. Stow. 1994. Marshal, J. P., and S. Boutin. 1999. Power analysis of wolf–
Fitting predator–prey models to time series with observa- moose functional responses. Journal of Wildlife Manage-
tion errors. Ecology 75:1254–1264. ment 63:396–402.
Cosner, C., D. L. DeAngelis, J. A. Ault, and D. B. Olson. May, R. M. 1976. Models for single populations. Pages 4–
1999. Effects of spatial grouping on the functional re- 25 in R. M. May, editor. Theoretical ecology: principles
sponse of predators. Theoretical Population Biology 56: and applications. Blackwell Scientific, Oxford, UK.
65–75. May, R. M., J. R. Beddington, C. W. Clark, S. J. Holt, and
DeAngelis, D. L., R. A. Goldstein, and R. V. O’Neill. 1975. R. M. Laws. 1979. Management of multispecies fisheries.
A model for trophic interaction. Ecology 56:881–892. Science 205:267–277.
Efron, B., and R. J. Tibshirani. 1993. An introduction to the Mech, L. D. 1966. The wolves of Isle Royale. National Park
bootstrap. Chapman and Hall, New York, New York, USA. Service Scientific Monograph Series Number 7. U.S. Gov-
Elton, C., and M. Nicholson. 1942. The ten year cycle in ernment Printing Office, Washington, D.C., USA.
numbers of the lynx in Canada. Journal of Animal Ecology Messier, F. 1994. Ungulate population models with predation:
11:215–244. a case study with North American moose. Ecology 75:478–
Flaaten, O., and K. Stollery. 1994. The economic effects of 488.
biological predation: theory and application to the case of Murie, A. 1944. The wolves of Mount McKinley. National
November 2002 WOLF PREDATION 3013

Park Service Scientific Monograph Series Number 5. U.S. Rogers, D. J. 1972. Random search and insect population
Government Printing Office, Washington, D.C., USA. models. Journal of Animal Ecology 41:369–383.
Oksanen, L. S., S. D. Farewell, J. Arruda, and P. Niemela. Rosenzweig, M. L. 1971. Paradox of enrichment: destabili-
1981. Exploitation ecosystems in gradients of primary pro- zation of exploitation ecosystems in ecological time. Sci-
ductivity. American Naturalist 118:240–261. ence 171:385–387.
Olson, S. F. 1938. Organization and range of the pack. Ecol- Rosenzweig, M., and R. H. MacArthur. 1963. Graphical rep-
ogy 19:168–170. resentation and stability conditions of predator–prey inter-
Peterson, R. O. 1977. Wolf ecology and prey relationships action. American Naturalist 97:217–223.
on Isle Royale. National Park Service Scientific Monograph Slobodkin, L. B. 1992. A summary of the special feature and
Series Number 11. U.S. Government Printing Office, Wash- comments on its theoretical context and importance. Ecol-
ington, D.C., USA. ogy 73:1564–1566.
Peterson, R. O. 1999. Wolf–moose interaction on Isle Royale: Smith, F. E. 1952. Experimental methods in population dy-
the end of natural regulation? Ecological Applications 9: namics: a critique. Ecology 33:441–450.
10–16. Solberg, E. J., B.-E. Saether, O. Strand, and A. Loison. 1999.
Peterson, R. O., and R. E. Page. 1988. The rise and fall of Dynamics of a harvested moose population in a variable
Isle Royale wolves, 1975–1986. Journal of Mammalogy environment. Journal of Animal Ecology 68:186–204.
69:89–99. Spencer, P. D. 1997. Optimal harvesting of fish populations
Peterson, R. O., and R. E. Page. 1993. Detection of moose with nonlinear rates of predation and autocorrelated en-
in midwinter from fixed-wing aircraft over dense forest vironmental variability. Canadian Journal of Fisheries and
cover. Wildlife Society Bulletin 21:80–86. Aquatic Sciences 54:59–74.
Peterson, R. O., N. J. Thomas, J. M. Thurber, J. A. Vucetich, Taylor, R. J. 1984. Predation. Chapman and Hall, New York,
New York, USA.
and T. A. Waite. 1998. Population limitation and the wolves
Thompson, W. R. 1937. Science and common sense: an Ar-
of Isle Royale. Journal of Mammalogy 79:828–841.
istotelian excursion. Longmans, Green, London, UK.
Peterson, R. O., and J. A. Vucetich. 2001. Ecological studies
Thurber, J. M., and R. O. Peterson. 1993. Effects of popu-
of wolves on Isle Royale. Annual report. Michigan Tech-
lation density and pack size on the foraging ecology of
nological University, Houghton, Michigan, USA. gray wolves. Journal of Mammalogy 74:879–889.
Peterson R. O., J. D. Woolington, and T. N. Bailey. 1984. Volterra, V. 1991. Fluctuations in the abundance of a species
Wolves of Kenai Penisula, Alaska. Wildlife Monograph 88: considered mathematically. [Reprint of the translation (Na-
1–52. ture 118:558–560) from the 1926 original (in Italian.)] In
Polis, G. A., and D. R. Strong. 1996. Food web complexity L. A. Real and J. H. Brown, editors. Foundations of ecol-
and community dynamics. American Naturalist 147:813– ogy. University of Chicago Press, Chicago, Illinois, USA.
842. Yodzis, P. 1994. Predator–prey theory and management of
Reeve, J. D. 1997. Predation and bark beetle dynamics. Oec- multispecies fisheries. Ecological Applications 4:51–58.
ologia 112:48–54. Yodzis, P. 2001. Must predators be culled for the sake of
Ricklefs, R. E., and G. L. Miller. 2000. Ecology. Fourth fisheries? Trends in Ecology and Evolution 16:78–84.
edition. W. H. Freeman and Company, New York, New Zar, J. H. 1984. Biostatistical analysis. Second edition. Pren-
York, USA. tice Hall, Engelwood Cliffs, New Jersey, USA.