You are on page 1of 15

THE ECONOMICS OF MANAGING INFECTIOUS WILDLIFE DISEASE

RICHARD D. HORAN AND CHRISTOPHER A. WOLF
We use a two-state linear control model to examine the socially optimal management of disease in a valuable wildlife population when diseased animals cannot be harvested selectively. The two control variables are nonselective harvests and supplemental feeding of wildlife, where feeding increases both in situ productivity and disease prevalence. We derive a double singular solution which depends on the initial state and does not require bang-bang controls. The case of bovine tuberculosis among Michigan white-tailed deer is analyzed. In the base model, the disease is optimally maintained at low levels, with intermittent investments (via feeding) in deer productivity. Key words: bovine tuberculosis, double singular solution, linear control, nonselective harvests, whitetailed deer.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

The spread of infectious disease among and between wild and domesticated animals has become a major problem worldwide, with examples involving wildlife as the primary vector of disease transmission almost too numerous to count (see the Journal of Wildlife Diseases). The economic costs of wildlife-to-livestock disease transmission can be substantial. For instance, a foot and mouth disease outbreak in the U.S., which could easily spread through a wildlife vector such as deer, could cost billions of dollars the first year alone (USDA-APHIS 2002). For this and other diseases, farmers incur losses when infected livestock die or become less productive due to the disease, because the demand for livestock products is diminished, and because of strict regulations imposed when a herd becomes infected (e.g., depopulating a herd). But the costs are not limited to infected farms, as trade sanctions are often imposed on entire counties, states, or even countries where the disease is present. Human

Richard D. Horan and Christopher A. Wolf are associate professors, Department of Agricultural Economics, Michigan State University. The authors gratefully acknowledge funding provided by the Economic Research Service-USDA cooperative agreement number 43-3AEM-3-80105 through ERS’ Program of Research on the Economics of Invasive Species Management (PREISM). The views expressed here are those of the authors and should not be attributed to ERS or USDA. The authors thank Stephen Swallow, three anonymous reviewers, Erwin Bulte, Eli Fenichel, Larry Karp, Ken Matthews, and Aart de Zeeuw for helpful comments and suggestions. Any remaining errors are the authors’.

health is also an important concern and is often the primary justification for eradication programs and controls. A disease outbreak among wildlife may also impose costs on those who value wildlife products or services. These costs would arise if hunters place a premium on healthy wildlife, and if infected or even healthy populations in close proximity to an outbreak are culled to prevent additional spread. The costs could be greater for threatened or endangered species, particularly those protected in parks not large enough to support a viable population. As population members wander outside protected areas, the risk of infection increases—both for wandering wildlife and for those in protected areas. Conservation measures must therefore be taken with disease control in mind (Simonetti). There has been relatively little research in the area of the economics of disease control among wildlife populations. Most animal disease work has estimated the costs to farmers and consumers under alternative control strategies, with little regard given to the wildlife dimension (e.g., Mahul and Gohin; Kuchler and Hamm; McInerney; Ebel, Hornbaker, and Nelson; Dietrich, Amosson, and Crawford; Liu). An exception is Bicknell, Wilen, and Howitt, who developed a bioeconomic model to analyze a bovine tuberculosis (TB) problem in New Zealand in which TB is spread by Australian brush-tailed possums to dairy herds. Specifically, they explored optimal disease control strategies (e.g., testing at the

Amer. J. Agr. Econ. 87(3) (August 2005): 537–551 Copyright 2005 American Agricultural Economics Association

538

August 2005

Amer. J. Agr. Econ.

farm level and hunting possums off the farm) from the perspective of a single farmer. Our wildlife disease model differs from prior literature in three important ways. First, we consider the socially optimal management of an infected wildlife population as opposed to controls taken by a single farmer. Wildlife disease problems affect many people (e.g., many landowners and hunters), and so an individual farmer would tend to under-invest in disease control investments that provide external benefits. Second, we assume infected wildlife cannot be identified until after they are killed and examined (Williams et al.), rendering it impossible to selectively harvest only infected animals. Accordingly, any off-take of infected animals is likely to be accompanied by healthy animals that may be economically valuable. Third, healthy but susceptible wildlife such as deer and lions (which are infected with TB in South Africa) may hold considerable economic value. Thus, exterminating wildlife as a way of eradicating a disease outbreak, as is often proposed, might be a comparatively costly approach. The model is applied to the case of bovine TB among white-tailed deer in Michigan, the only known area in North America where bovine TB has become established in a wild deer population. In the early to mid-1990s, signs of bovine TB started to re-emerge both in the wild deer population and also among some farms. Michigan lost bovine TB accreditedfree status in June 2000 and was required to adopt a testing program for all Michigan cattle, goats, bison, and captive cervids. In addition, other states could place movement restrictions on Michigan livestock at their discretion (MDA, USDA-APHIS 1999). On 1 June 2004, Michigan received “split-state” status for bovine TB, resulting in two disease management zones having separate requirements for animal movement, identification, and testing. This status came about because extensive testing found the disease confined to the northeast corner of Michigan’s lower peninsula, and so regulatory costs are now primarily confined to this area. Michigan agriculture is obviously concerned about disease-related costs and supports culling the deer population to eradicate the disease. However, such extreme measures could be very costly, particularly since deer hunting is arguably the highest-valued use of the land in the infected region. A final contribution of this article is a methodological one. Because our bioeconomic model involves two state variables, we adopt

a linear control model to make the problem tractable in two dimensions. In contrast to the bang-bang solutions that arise in conventional linear control models, the solution is a nonlinear feedback law along a singular arc, where the particular arc pursued will vary according to the initial state and where bang-bang controls are not required to initially move to this arc.
Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

A Model of Infectious Wildlife Disease Consider a wildlife population that grows unexploited on a fixed land area. The aggregate wildlife population, N, consists of two subpopulations: a healthy but susceptible stock, s, and an infected stock, z. In the absence of exploitation or disease, the susceptible stock grows according to the logistic growth function, rs(1 − N/k), where r is the intrinsic growth rate and k is the carrying capacity. The logistic function is not required for the general model, but the use of a specific functional form facilitates understanding of the population dynamics and the logistic is consistent with our numerical model of Michigan white-tailed deer. Following Barlow (1991a), the densitydependent component of the logistic equation, (1 − N/k), depends on the aggregate population because susceptible and infected wildlife compete for the same habitat. Population growth is reduced as members of the susceptible stock become infected, which occurs when a susceptible animal comes into contact with an infected animal. Disease transmission can be modeled in a number of ways. Historically, ecologists have used a mass action or density-dependent transmission function, zs, where is the contact rate per infectious deer. McCallum, Barlow, and Hone note, however, that the mass action model often does not hold up empirically. A competing transmission function, which often fits data better than mass action for diseases such as cowpox in bank voles and wood mice (Begon et al. 1998, 1999) and brucellosis in Yellowstone bison (Dobson and Meagher), is frequencydependent or density-independent transmission. The frequency-dependent model, which we adopt, assumes that the z infected animals make on average z contacts in each time period, with s/N contacts being with susceptible animals so that total transmission is zs/N.1
1 McCallum, Barlow, and Hone propose a functional form which reduces to mass action at one extreme parameter value and

Horan and Wolf

Managing Infectious Wildlife Disease

539

Finally, consider the potential controls for the infected stock. Vaccination is not considered because for many diseases, such as bovine TB in wild deer, there are no effective vaccines (MDA). The disease could be expediently controlled or even eradicated (albeit at possibly high cost) if the manager could selectively harvest from the infected stock. However, we assume selective harvesting is not an option because infected animals cannot be identified prior to the kill; outward signs of an illness often take a long period to manifest (MDA; Williams et al.). Harvesting will therefore include both healthy and infected individuals, which could be costly for species that are highly valued for recreational purposes (such as deer) or that are endangered. Nonselectivity is not unique to the current situation. For instance, hunters and fishermen cannot selectively harvest from different cohorts within exploitable populations of many species (Reed, Clark), and by-catch of nontargeted species is often a problem in fisheries. Given nonselective harvesting, a manager can only choose the aggregate harvest, h, with the harvest from each stock depending on the proportion of animals in that stock relative to the aggregate population. That is, hs = hs/N and hz = hz/N, where hi denotes the harvest from population i. Given this specification, the equation of motion for the healthy stock is (1) s = r s(1 − N /k) − zs/N − hs/N . ˙

It will be more intuitive and mathematically more convenient to work in terms of the variable N instead of s, and the variable = z/N instead of z, where represents the disease prevalence rate. The relations s = N − z and z = N can be used to substitute for s and z in equations (1) and (2), and without loss we focus on the following equations of motion (3) (4) ˙ N = r N [1 − (N /k)] − ˙ = [ − ](1 − ) . N −h
Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

The infected stock also grows according to the logistic growth function (assuming infected mothers pass the disease to their young, either in utero or shortly after birth through contact), although the disease increases mortality by a rate of . We assume < r, which is the most realistic case for our white-tailed deer application and also the most interesting case in general; otherwise, the diseased stock would not be sustainable. Given this specification and that newly infected animals add to the infected stock, the equation of motion for the infected stock is (2) z = r z(1 − N /k) − z + zs/N − hz/N . ˙

Neither h nor N appears in equation (4), so clearly harvesting does not affect disease prevalence in this model. An interior solution arises in the unlikely event that = , and all wildlife become healthy after a disease outbreak if < . There is a saddle equilibrium with z = 0 when > and r > , although the likelihood of achieving this outcome is essentially zero because the initial state would have to lie on the saddle path since (nonselective) harvesting policies cannot be used to move to this path. More generally, all wildlife become infected when > .2 Hence, the only effective harvest-based option for disease eradication in this model when > is extermination of all wildlife. Harvesting is not the only control option, however. Other potentially effective control strategies would be those that have the effect of increasing or decreasing . For instance, some hunters bait wildlife with small amounts of feed (baiting) and others actually provide large amounts of feed to wildlife (feeding) throughout the year. Such practices might effectively decrease (because infected animals are better nourished) and increase (because individual animals come into close contact as they feed). So, a prohibition on feeding or baiting could facilitate disease eradication. We now turn to the case of Michigan white-tailed deer to examine management possibilities and the economic question of how much control is warranted.

frequency-dependent transmission at the other extreme, although the form is generally too complex to apply analytically. Still, there is no generally accepted modeling approach and the data is too sparse in most cases to determine which form is appropriate (McCallum, Barlow, and Hone). For simplicity and because frequencydependence fits the data better than mass action in many cases, we apply the frequency-dependent form. But, we do make note of where this assumption could have important management implications.

2 The local stability properties of each equilibrium are given by the eigenvalues associated with the linearized forms of equations (3) and (4) (Conrad and Clark, pp. 45–46). Also, note the result that harvesting does not affect prevalence stems from two underlying assumptions: frequency-dependence and nonselective harvesting (McCallum, Barlow, and Hone). Under mass action transmission, equation (4) would be written as ˙ =[ N − ](1 − ) , and harvests, by impacting N, would affect the possibility of an interior equilibrium (McCallum, Barlow, and Hone; Barlow 1991b).

540

August 2005

Amer. J. Agr. Econ.

Wildlife Management and Disease Control for Michigan White-Tailed Deer Bovine tuberculosis among Michigan whitetailed deer is primarily concentrated in a fourcounty area in the northeastern part of the lower peninsula, formally designated as deer management unit (DMU) 452 or less-formally as the “core” (see Hickling). A few cases of infection have been found beyond this area but the disease does not appear to be sustainable outside the core. This has led many to speculate that unique, core-specific features such as human–environment interactions— particularly feeding programs—have enabled the disease to become endemic (Hickling). Indeed, prior to 1995, only eight cases of bovine TB had ever been reported in wild deer from North America (Schmitt et al.) and conventional wisdom held that the disease was not self-sustaining in wildlife populations (Hickling). Several hunt clubs in the core sponsored feeding programs to increase deer density. Originating in the late 1800s, these clubs purchased large amounts of core area land on which its members could hunt. This land was desirable because it was easily accessible from highways and, as it consisted of generally poor soil for agronomic purposes, the land was inexpensive (Hickling). The historic density of deer in the area is estimated to have been 7–9 deer/km2 (O’Brien et al.). The hunt clubs, desiring greater density, began aggressive deer feeding programs to encourage herd growth. These programs have been known to dump tractor–trailer loads of food in the woods and fringe areas, with the resulting massive food piles being visible from the air along with the tracks of many congregating deer. The result is that deer density increased to an estimated 25 deer/km2 by the mid-1990s. Of course, while increasing the carrying capacity of core deer generates hunting benefits, while supplemental feeding also generates external costs by leading to increased disease transmission as deer congregate, and by supporting sick animals, thereby reducing disease-related mortality. Denote f as feed provided by feeding programs. Increased feed availability reduces the density-dependent component of growth by a factor (1 − f ), increases the disease transmission coefficient by a factor (1 + f ), and decreases mortality due to the disease by a factor (1 − f ), where , , and are parameters. The equations of motion for the deer population and disease prevalence rate, (3) and (4), are

modified as follows to account for these impacts of feeding (5) ˙ N = r N [1 − (N /k)(1 − f )] − (1 − f ) N − h ˙ = [ (1 + f ) − (1 − f )](1 − ) .

(6)

With r > (as is widely believed), the disease would be endemic in the core if (1 + f ) > (1 − f ).3 If > , then the disease will persist regardless of feeding or hunting choices (apart from wildlife eradication). In that case, some other effort to reduce disease transmission will be required to eradicate the disease. But if < , then the disease would be eliminated by setting f < [ − ]/[ + ] for some time. A smaller f means the disease is eliminated earlier but at an interim cost of lost deer productivity. Of course, it is important to consider whether eradication is optimal. Economic Specification Hunters gain utility from the actual process of shooting deer or consuming meat and other deer products. The (constant) marginal utility from harvesting healthy deer is denoted by p, which is not less than the (constant) marginal utility from harvesting infected deer, pz , that is, p ≥ pz . For simplicity, we set pz = 0 so that harvests of infected animals yield no benefits.4 The benefits from hunting are therefore phs/N = p(1 − )h. Assume harvests occur according to the Schaefer harvest function (although in general this specification is not required), and that the unit cost of effort, c, is constant. Then, total harvesting costs, restricted on the in situ stocks, are (c/q)h/N, where q is the catchability coefficient. The unit cost of food is w. Finally, the costs of the disease, particularly to farmers and related agribusinesses, must also be considered. Denote the variable economic damages caused by infected deer by D(z) (with D(0) = 0, D > 0, D ≥ 0). These variable damages are due to infections in the cattle herd that result in lost stock, increased testing, and business interruption loss. The imposition of trade restrictions and federally mandated testing requirements in response to
The disease would not be sustainable outside the core if < 0 , where 0 and 0 represent parameter values outside the core. These parameters may differ from and due to human– environment interactions apart from feeding. 4 This assumption should not affect the qualitative nature of the results, but it may affect the trajectories in the numerical exercise.
0 3

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Horan and Wolf

Managing Infectious Wildlife Disease

541

the disease may also result in a significant lump sum damage component. Such lump sum damages are primarily policy-induced and, if large enough, could affect the optimal plan. We begin by investigating an optimal plan without these lump sum costs, as the solution is efficient from Michigan’s point of view in the absence of exogenous regulatory impositions. Later, in the numerical section, we analyze the effect of the regulatory-based lump sum costs. We find that these additional costs might sufficiently alter Michigan’s optimal management plan so that it is more in line with the federal government’s stated objectives.5 Optimality Conditions and the Double Singular Solution Wildlife managers have two objectives when dealing with the disease: reduce the number of diseased animals and control the spread of the disease. To accomplish these goals, managers have focused on harvest levels and the amount of food provided by feeding programs as the primary choice variables (Hickling). Given the discount rate , an economically optimal allocation of harvests and feeding maximizes social net benefits, SNB, that is, (7) Max SNB
h, f

The marginal impact of harvests on the Hamiltonian is given by (9) ∂ H/∂h = p(1 − ) − c/(q N ) − .

=
0

[ p(1 − )h − (c/q)(h/N )

− w f − D( N )] e− t dt subject to the equations of motion (5) and (6) and the feasibility conditions h ≥ 0 and 0 ≤ f ≤ f max .6 The current value Hamiltonian is (8) H = p(1 − )h − (c/q)(h/N ) − w f − D( N ) + [r N [1 − (N /k)(1 − f )] − (1 − f ) N − h] + [( [1 + f ] − [1 − f ])(1 − ) ] where and are the co-state variables associated with N and , respectively.
5 Deer are also important causes of automobile accidents and damage to agricultural crops (Rondeau, Rondeau and Conrad). We ignore these other damages in order to focus on the impacts of disease, but we note that these other damages could be important. 6 We do not place an upper bound on h. Rather, our primary concern is that f ≤ min(1/ , 1/ ). A value of f > 1/ would result in a negative mortality rate due to the disease, which is not possible. A value of f > 1/ would result in a negative density-dependence factor, which also does not seem realistic.

If this expression is positive so that marginal rents exceed the marginal user cost, then harvests should be set at their maximum levels. If this expression is negative, then no harvesting should occur. The singular solution is pursued when marginal rents and the marginal user cost are equated. This is the standard condition for linear control problems involving renewable resources (e.g., Clark), except for two important differences. First, marginal rents are reduced by p because not all harvested animals are valued (as some are infected). Second, because harvests of N are nonselective, the marginal user cost of N can be positive or negative, that is, the sign of is ambiguous. It is easy to show that the following relation must hold: = (∂SNB/∂s)(1 − ) + (∂SNB/∂z) , where ∂SNB/∂z < 0 and ∂SNB/∂s > 0 or <0 depending on whether increases in s only end up fueling the growth of a larger infected population z. If = 1, then all additions to N only add to the infected stock so that < 0, and vice versa when = 0. There must exist a set of values for the state variables such that = 0, other sets such that > 0, and still others such that < 0. A potential nonconvexity therefore emerges, with the possibility of multiple optimality candidates (Rondeau; see also Tahvonen and Salo; Huffaker and Wilen; Maler, Xepapadeas, and de Zeeuw). The po¨ tential for nonconvexities does not arise when harvests can be made selectively. Now consider the marginal impacts of feeding on the Hamiltonian (10) ∂ H/∂ f = −w + [r (N 2 /k) + + [ + ](1 − ) . N]

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Feeding can be thought of as an investment in the productivity of the resource, although it has the unwanted side-effect of increasing disease prevalence. As we show below, the solution has similarities to, but also important differences from, Clark, Clarke, and Munro’s problem of investing in harvesting capital. The singular solution should be followed whenever the unit cost of feeding equals the in situ net marginal value of feeding on the two state variables. The in situ net marginal value is the difference between the marginal benefits of feeding on the overall stock (which includes

542

August 2005

Amer. J. Agr. Econ.

increased productivity and decreased mortality, and which may be negative when < 0) and the marginal costs of feeding in terms of an increased proportion of infected animals (due to increased transmission and decreased mortality among the infected stock). If the marginal in situ values exceed the unit cost, then feeding should proceed at the maximum rate. If the unit cost exceeds the in situ value, then feeding should optimally cease. It must be the case that < 0, for disease is never beneficial. Equation (10) therefore implies that > 0 must hold along a singular path, and so nonconvexities can only emerge along a nonsingular feeding path. Clearly f = f max is not optimal if < 0 as feeding would only create costs in this case. The only nonsingular feeding path consistent with < 0 is f = 0, but < 0 would not be sustainable in this case because → 0 as long as f = 0 is maintained (see also footnote 10). (13) =r−

period of time during which one of the controls is “blocked” or constrained from following the double singular path (Arrow, Clark). Blocked intervals will be shown to introduce some interesting complexities into the model. However, to understand how the solution transitions from the double singular solution to a blocked interval, it is first necessary to understand the unconstrained solution, for “the optimal path must always lie as close as possible to the [double] singular path” (Clark, p. 56). Indeed, it will be helpful to first graph the double singular solution in the context of a numerical example before examining the complete solution of double and partial singular paths. We therefore delay our discussion of partial singular solutions until the numerical section below. Differentiating condition (9) with respect to time, substituting the right-hand side (RHS) of condition (12) in for ˙ , and using (9) to substitute for the co-state variable , we have

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

2r N (1 − f ) − (1 − f ) k c/(q N 2 )[r N [1 − (N /k)(1 − f )] − (1 − f ) N ] − D + p(1 − ) − c/(q N ) p[ (1 + f ) − (1 − f )](1 − ) − . p(1 − ) − c/(q N ) Equation (13) is a variant of the conventional “golden rule” for renewable resource management: the rate of return for holding the healthy stock in situ equals the marginal productivity of the stock, plus net marginal stock effects (i.e., the marginal cost savings that accrue as harvests come from a larger stock minus the marginal damages, normalized by marginal user cost), minus the (normalized) value of foregone revenues as some of the remaining healthy in situ stock will become infected and result in a larger proportion of infected deer in future harvests. Equation (13) must hold at all times along the double singular path (or even a partial singular path involving only the harvest; see below). In conventional autonomous renewable resource models involving a single state variable, the singular solution is a point, N ∗ , because the golden rule is only a function of the stock and can be solved for a unique value of N. In contrast, condition (13) is a (linear) function of one of the control variables, f . If we solve equation (13) for f as a function of the current state variables, N and , the result is a nonlinear feedback law (Bryson and Ho). This

The necessary arbitrage conditions for an optimal solution are given by: (11) ˙ = − ∂ H/∂ N = +D (12) ˙ = + − (1 − f ) ] − ∂ H/∂ = + ph + D N (1 − f )N − [ (1 + f ) − ch/(q N 2 ) − [r − 2r (N /k)(1 − f )

− (1 − f )](1 − 2 ). Conditions (11) and (12) reflect intertemporal changes in optimal marginal resource values. The remainder of this section is devoted to deriving the conditions that characterize the double singular path—that is, the solution path involving singular controls for both harvesting and feeding, so that conditions (9) and (10) both vanish. As we show in the numerical section to follow, partial singular solutions, in which the solution is singular for only one control variable, will also be part of the overall solution. Partial singular solutions in this context arise as part of a blocked interval, a

Horan and Wolf

Managing Infectious Wildlife Disease

543

feedback rule implies that the double singular solution will be a path. Now differentiate condition (10) with respect to time, substitute the right-hand side (RHS) of conditions (11) and (12) for ˙ and ˙ , respectively, and use conditions (9) and (10) to substitute for and (the procedure is analogous to the one described for equation (13), and is outlined by Conrad and Clark, and others). The result is a golden rule expression for managing disease prevalence (14) = −{[−2 pr 2 (1 − f )N 3 /k 2 −c +p +rp −p
2

the need for bang-bang controls to move to the arc. In contrast, more conventional singular arcs are independent of the initial state— they are defined by a single curve (or a point), with bang-bang controls required to move to the curve if it does not pass through the initial state.8 Numerical Example We now examine the optimal solution numerically because the feedback rules and the differential equations that define the solution are too complex to analyze analytically. The data used to parameterize the model are described in the Appendix. While we have made every effort to calibrate the model realistically, research on the Michigan bovine TB problem is still evolving and at a fairly early stage so knowledge of many parameters is somewhat limited. The following analysis is therefore best viewed as a numerical example rather than a true reflection of reality. Nonetheless, the results shed light on the economics of wildlife disease management in general and specifically on TB in Michigan deer. The numerical solution, arrived at using the software Mathematica 5.0 (WolframResearch), is presented in figure 1 for the case of = 0.1 (we explore the sensitivity of this choice of below). Although not presented, an interior but unstable steady state arises at the point (N = 5,561, = 0.013), just northeast of the point d. This steady state is an unstable focus, which means that it is only optimal to be at this point if the system begins at this point. Otherwise, it is optimal to spiral away from this point (see Clark), which explains the cyclical nature of paths in its vicinity. Indeed, we find an interior cycle, involving a portion of path 3 along with paths 4 and 5, is optimal. The optimal solution begins at the initial state values N 0 and 0 , represented by point a in figure 1. From this point, the double singular path 1 is followed. This path spirals away from the interior equilibrium, initially increasing and N. The result that feeding should be initially encouraged runs contrary to Michigan’s current policy approach of banning
8 Linear control problems having nonlinear feedback laws as the solution seem to be more common in noneconomic applications, at least for the class of autonomous problems. For instance, Bryson and Ho provide a famous example of optimal thrust programs for rockets. Swallow provides the only example of which we are aware in the resource economics literature. He derives a nonlinear feedback rule for the harvest of a resource stock; however, his singular solution lies along a single curve, with bang-bang controls required when the system is initially off the curve.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

( + ) ](1 − ) ](1 − ) + N [D ( +p (2 + [ + ] + f ) ] (1 − ) −2 ] f ] + )

+ h[cr /k − 2 pr (N /k)

+ (r N /k)[2 pr − p − p [ + ] − p[ × (1 − f ) − cr + c × {− pr (N /k)(1 − )
2

× (1 − ) + (r N /k)[(r N /k)c (1 − f ) ]}

+ (w + c − Np

) + N cr /k (1 − )}−1 .

Equation (14), which depends linearly on both controls, h and f , must also hold along the double singular path. If we plug the feedback law for f into this expression, it is possible to solve equation (14) for a feedback law for the harvest, h(N, ). The feedback laws h(N, ) and f (N, ) can be plugged into the differential equations (5) and (6) to (numerically) solve for the double singular path, given the initial states, N 0 and 0 , and assuming that the feedback laws satisfy feasibility conditions at these initial states.7 Note that the double singular path—the solution to the equations of motion—depends on the values of the initial state variables. Hence, different singular arcs will be defined for different initial states, and so the system can begin on a singular arc (an interior solution) without
7 That equations (9) and (10) both vanish when the feedback rules are followed, for any state variable combination such that the nonnegativity constraints are satisfied, is verified by setting equations (9) and (10) equal to zero and noticing that the coefficient matrix for the vector [ ] for this system is not singular—thus, a unique value of both and satisfy the singular conditions for all relevant combinations of N and .

544

August 2005

Amer. J. Agr. Econ.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Figure 1. Solution of the benchmark numerical example (N 0 = 13,298; 0 = 0.023; r = 0.5703; k = 14,049; = 0.339; = 0.356; = 0.00008; = 2.64 × 10−6 ; = 0.3623; p = 1270.8; c/q = 231,192; w = 36.53; = 5491; = 0.1)

feeding (although it is consistent with the voluntary actions of the hunt clubs in recent years). Feeding represents an investment in stock productivity, initially increasing the stock while enabling large harvests along the singular path 1. The disease prevalence rate goes up along the singular path, but the increased damages are offset by the rewards of larger near-term harvests. Feeding and also prevalence rates continue to grow along the path 1. Eventually f (N, ) = f max = 10,000, represented by the boundary f = f max in figure 1. This boundary creates a blocked interval that prevents the state variables from following the double singular path (Arrow; Clark, p. 56).9 The farsighted planner knows the boundary is approaching. Therefore, at some point prior to the singular path becoming constrained at the f = f max boundary, the planner will set f equal to an extremal value and abandon the myopic double singular path, as is required by the “premature switching principle” (Clark, p. 57). Arrow (p. 16) points out that actually finding the optimal path “is a process of trial and error,” first finding the eligible intervals and then numerically finding when it becomes optimal to jump from one interval to another.
9 The continuation of the singular path 1 is not illustrated, but after the boundary it changes direction and moves northwest: with increasing prevalence, damages are reduced by reducing the stock since prevalence is increasing.

The first interval consists of path 1, and the second interval begins by setting f = f max prior to having reached the f = f max boundary. Define T as the time at which this extremal value of f is chosen and the double singular path 1 is abandoned. At T, it becomes optimal to pursue the partial singular solution for N conditional on f = f max —at least provided that f = f max will be maintained for more than an instant. Suppose this was the case. Then, the resulting partial singular solution is characterized by equation (13), holding f fixed at its constrained value. Specifically, equation (13) can be solved for N( , f max ), a unique deer population given the value of at time T and also given f = f max . This value is optimally approached along a most rapid approach path (MRAP). However, N( , f max ) lies at point c in figure 1, which is to the left of the f = 0 curve. At this point, it is actually optimal to set f = 0 and move towards the f = 0 curve along a MRAP that also involves zero harvests (so that N might increase), resulting in damages accruing along this path without any offsetting benefits. Since we know that f = 0 is optimal immediately after the cull at T, it is better to set f = f max for an instant at point b, and then immediately jump to point c, where f = 0 is optimal. Path 2 in figure 1 illustrates this jump. The instantaneous move from the f = f max boundary to the f = 0 boundary may at first glance seem excessive, but it makes sense from the perspective of the premature switching

Horan and Wolf

Managing Infectious Wildlife Disease

545

principle. For instance, suppose we had not jumped prematurely at point b but rather had followed the double singular solution to the f = f max boundary, and then set f = f max and moved northwesterly along the boundary. In this case, the f = f max constraint would not remain binding indefinitely, for it eventually leads to very high prevalence rates and low stocks. Rather, at some point, it would become optimal to follow another double singular solution: moving northwesterly along the f = f max boundary, a double singular path (not illustrated) does eventually emerge from the f = f max boundary and moves towards the f = 0 boundary. The f = 0 boundary would then be followed, reducing prevalence and allowing the stock to recover. This long process would eventually lead to point c, but the jump along path 2 gets there sooner. The premature switching principle says that if we foresee constraints on the horizon, then an early adjustment will produce larger net benefits. Our numerical simulations confirm this. We now move on to define the third interval in our solution. Once on the f = 0 boundary, we must check to see if it becomes optimal to pursue a double singular solution given the current stock and prevalence levels. At point c, the unconstrained double singular solution would move to the left of the f = 0 curve, which implies f < 0 and hence is infeasible. Hence, it is optimal to maintain zero feeding. Given that f is constrained exogenously in this fashion, equation (13) can be solved for N( , 0), with moving exogenously through time. Specifically, N( , 0) is derived from the following modified form of equation (13) with f =0 (15) =r− 2r N − k c/(q N 2 )(r N [1 − N /k] − N ) − D + p(1 − ) − c/(q N ) p( − )(1 − ) − . p(1 − ) − c/(q N )

cides with the f = 0 curve in the present model.10 Disease prevalence diminishes while wildlife stocks increase along the partial singular path 3. Continuation along this path would eventually lead to an outcome with a diseasefree wildlife stock, after which time feeding could be reintroduced without creating any disease problems. But that outcome is not pursued. When point d is attained on path 3, f = 0 is no longer a binding constraint but rather the solution to the feedback function f (N, ) along a double singular path, labeled 4, that moves northeasterly away from the f = 0 boundary. At this point, small amounts of supplemental feeding can significantly boost productivity while adding little to disease prevalence, as is reflected by the relatively flat slope of path 4 in the vicinity of d. The opportunity cost of waiting for the disease to die out therefore becomes too high relative to the gains that can be made from re-investing in deer productivity at d. So, the planner abandons the partial singular path 3 for the double singular path 4, which is the fourth solution interval. Path 4 eventually turns around and heads back to the f = 0 boundary, as the marginal costs of feeding in terms of increased disease prevalence approach the marginal benefits of feeding on increased deer productivity. But knowing that this boundary is imminent, the premature switching principle requires the planner to optimally cull the stock some time prior to reaching the boundary, jumping to a point such as e by way of path 5, the fifth interval in the solution. Point e lies on path 3 and so, once at e, the remaining portion of path 3 is pursued and the cycle 3-4-5-3 repeats. The disease is never eradicated because the deer are highly valuable and feeding intermittently becomes a good investment to boost productivity of the stock. Once the eligible intervals have been identified, the solution algorithm proceeds as follows. Starting at point d (which is easily calculated numerically as N = 5,379, = 0.009) and moving along path 4, determine the

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

The solution N( , 0) is a partial singular path for N, which is essentially a nonautonomous singular path given the exogenous movement of (e.g., see Clark or Conrad and Clark) and is optimally approached along a most rapid approach path (MRAP). This singular path, which we label as path 3 in figure 1, coin-

10 Note that > 0 everywhere along path 3, and so nonconvexities do not emerge. Nonconvexities could emerge on the f = 0 curve where p(1 − ) < c/(qN), or 1 − c/(pqN) < , but this only occurs at very large values of . Such an outcome could optimally arise as part of a solution for a problem involving much larger initial prevalence rates than the one in this example. But even in such instances would diminish as f = 0 is maintained, and eventually would become positive. The remainder of the solution would then converge with the one being described for the present example.

546

August 2005

Amer. J. Agr. Econ.

stock-prevalence rate combination on this path for which switching from path 4 to path 5 (and then continuing in the equilibrium cycle in perpetuity) results in maximum discounted net benefits. Once the equilibrium cycle is calculated, it is possible to start at point a and determine the stock-prevalence rate combination for which switching from path 1 to path 2 (and then moving on path 3 and into the equilibrium cycle) results in maximum discounted net benefits. The solution is (1) starting at point a, follow path 1, which involves increasing supplemental feeding and harvesting rates, until N = 14,123 and = 0.05; (2) an instantaneous cull is required to bring the deer population to N = 2,970; (3) follow path 3, which involves zero feeding and increasing harvests, until N = 5,379 and = 0.009; (4) follow path 4, which involves increasing supplemental feeding and harvesting rates, until N = 7,681 and = 0.018; (5) cull to a population of N = 4,874, and continue the equilibrium cycle. In many respects, the optimal path is similar to that of Clark, Clarke, and Munro, who analyze irreversible investments in harvesting capacity for renewable resources. They find it is optimal to temporarily over-capitalize (relative to the steady state) prior to a stockdepletion phase. The reason is that the larger capital levels allow more harvesting early on, which generate greater near-term benefits prior to advancing to the steady state. Somewhat analogously in our model, we find that initial and intermittent future investments in resource productivity create opportunities for near-term gains. But, an important difference is that a steady state is not optimal in our model. Unlike Clark, Clarke, and Munro, investment in our model (via feeding) produces adverse effects on resource dynamics: along with the productivity enhancing investments comes the unwanted side-effect of the disease, and sustained investment (feeding) would only lead to increasing disease prevalence. If allowed to continue unabated, this increasing prevalence eventually causes damages to swamp benefits. Therefore, intermittent dis-investment in the disease is warranted. Incorporating Fixed Damage Costs Consider now the imposition of trade restrictions and federally mandated testing requirements in response to the disease that result in a significant lump sum damage component. Wolf and Ferris estimated $4 million annually

in such costs under the current split-state TBfree status. Such costs do not affect the optimal interior cycle that we just described, as the marginal incentives are unaffected. But, of course, the lump sum costs never vanish under an interior strategy, and so these costs could affect whether the interior cycle is in fact optimal over a strategy to eradicate the disease and hence eliminate the lump sum costs. Given the magnitude of the lump sum damage costs, we find the optimal eradication strategy is to immediately stop feeding and immediately cull to the f = 0 curve. Once on that curve, the solution is of the partially singular type as harvesting is set at levels to achieve the stock levels that solve equation (15). The f = 0 curve is followed until = 0, at which point feeding can resume. It can easily be verified that the singular solution in this case involves equation (9) being satisfied as a strict equality and equation (10) as a strict inequality, so that feeding should be set at its maximum level. Equation (13) then uniquely determines the singular stock, N ∗=0, f = f max , which should be approached along a most rapid approach path. This implies zero harvests until the stock has increased to the steady-state value N ∗=0, f = f max , which equals 30,942 deer (compared to N 0 = 13,298). Simulated discounted net benefits under this scenario appear to be about $0.5 million larger under the eradication strategy.11 Hence, the eradication strategy appears optimal in the presence of lump sum damage costs. Because these costs are exogenously imposed, primarily as a result of federal policies, and because the federal government has clearly stated eradication as a policy goal (USDAAPHIS 1999), the results suggest that federal policies ensure that eradication is an incentive compatible strategy for Michigan to follow. Sensitivity Analysis We conclude our numerical example with a sensitivity analysis (of the model with no fixed costs) designed to shed light on how certain key parameters influence the optimal interior solution. First, consider figure 2, which illustrates
11 It is difficult to report a comparison of social net benefits, as eradication is an asymptotic process that does not occur in finite time in the simulation. Our estimate of a $0.5 million advantage for the eradication strategy arises for eradication time periods for which the combination of prevalence rates and stock levels suggest about 25–50 infected deer. While not extremely close to zero, the prevalence rate seemed sufficiently low and the time intervals sufficiently long to believe the disease would be wiped out in reality.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Horan and Wolf

Managing Infectious Wildlife Disease

547

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Figure 2. Solution of the numerical example when as in figure 1) the solution when = 0.05. There are several important differences between figures 2 and 1. First, the double singular path 1 is negatively sloped in figure 2 but positively sloped in figure 1. This difference arises because a smaller discount rate results in a more equal weighting of the near-term benefits that accrue from investing in in situ deer productivity (via feeding) and the costs that stem from the greater associated disease prevalence. Accordingly, feeding occurs at lower levels under smaller discount rates and hence provides a smaller productivity boost. Moreover, the jump to the f = 0 curve occurs at a greater distance from the f = f max curve (when N = 13,298 and = 0.028). Another difference is that the double singular path 4 in figure 2 has shifted down and exhibits a shorter upward path relative to its analogue in figure 1 (with path 4 starting at N = 6,163, = 0.006, and path 5 starting at N = 7,681, = 0.018). The shorter upward path results from the same sort of tradeoff influencing singular path 1. The downward shift results because the smaller discount rate reduces the opportunity cost of waiting for smaller disease prevalence levels relative to the benefits of feeding-induced productivity enhancements. In sum, the equilibrium cycle is shorter and occurs at lower disease prevalence rates for smaller discount rates. Solutions analogous to that in figure 2 could also arise for larger marginal damages, feeding costs, or disease mortality rate. For instance, upon increasing from 0.356 to 0.373 (so that / increases from 1.05 to 1.1), holding all other parameters from the basic model

= 0.05 (with all other parameter values

constant, the solution looks almost identical to figure 2 but for different reasons.12 With larger disease mortality, infected deer die more rapidly. This means the productivity boost to the aggregate deer stock is not as great and the stock declines more rapidly for a given harvest level—hence the negatively sloped double singular path 1 and the short upward portion of the double singular path 4. Greater disease-related mortality also decreases the opportunity cost of waiting for reduced disease prevalence. Hence, the downward shift in singular path 4. Similar results occur for larger marginal damages, which reduces the opportunity cost of waiting for reduced prevalence, and for larger feeding costs, which reduces the marginal benefits of productivity investment via feeding and hence reduces the cost of waiting for reduced prevalence. If feeding costs, marginal damages, or / is increased enough, we find that the opportunity cost of eradication of the disease becomes optimal in the long run. This process is presented in figure 3 for the case of increased feeding costs. After an initial productivity investment, there is a jump to the partial singular path 3 along the f = 0 curve. This partial singular path is optimally followed until = 0 and N = N ∗=0,f =0 , as there is no double singular path that moves easterly out of the
12 The solution arising under a mass action transmission function is also qualitatively similar to figure 2. This is because densitydependent disease transmission implies larger opportunity costs associated with holding the deer stock in situ, so that there are incentives to start depleting the stock sooner and to maintain it at lower levels.

548

August 2005

Amer. J. Agr. Econ.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Figure 3. Solution of the numerical example when feeding costs are increased tenfold (with all other parameter values as in figure 1) f = 0 curve (coinciding with the fact that there is no interior focus point in this case). Once = 0, feeding is set at its maximum level, provided the feeding costs are not too great (otherwise, the system remains at N ∗=0,f =0 = 5,921 deer). Equation (14) then uniquely determines the singular stock, N ∗=0, f = f max , which should be approached along a most rapid approach path 4. This implies zero harvests until the stock has increased to the steady-state value N ∗=0, f = f max = 30,942 deer. Discussion and Conclusions This article represents a first step in understanding the economics of disease control in wildlife populations. A general model of wildlife growth and disease transmission was used to illustrate limitations of harvesting strategies when harvests cannot be made selectively from the diseased population. Strategies to address disease prevalence must therefore focus on more than just the harvest, and can be particularly effective if they address disease transmission and mortality. The state of Michigan announced a goal of eradicating bovine tuberculosis in Michigan deer populations by 2010. To that end, the wild white-tailed deer population in the core disease area was to be decreased through hunting programs that sold increased licenses. In addition, the practice of legally feeding deer in the infected area was ended and the practice of baiting was temporarily ended. Both of these disease control instruments were examined in our model. With deer being a highly valued resource, we find that eradicating the disease is not likely to be economically optimal—at least when Michigan does not bear fixed policyrelated costs due to the exogenous imposition of trade restrictions and federally mandated testing requirements. It takes too long for the disease to dissipate naturally once supplemental feeding is halted, which is not surprising considering that it took sixty-two years to previously eliminate the disease in cattle herds under much more controlled conditions. It is also too difficult and costly to kill all the deer in the infected area, as policy-makers in Michigan are currently discovering. Instead, it is optimal for the disease to remain endemic in the area at very low levels, with intermittent investments (via supplemental feeding) in in situ deer productivity. Of course, an endemic disease is not always optimal. If marginal damages, feeding costs or disease mortality are large enough, we find that it may be optimal to delay feeding-induced productivity enhancements in favor of disease eradication. Perhaps more importantly, we find that eradicating the disease may be optimal when Michigan bears exogenously imposed, fixed policy-related costs. Specifically, the results suggest that the restrictions imposed by states and the federal government ensure that disease eradication is an incentive compatible strategy for Michigan to follow. Although the model was applied to the specific case of bovine TB in deer herds, the model

Horan and Wolf

Managing Infectious Wildlife Disease

549

and results are likely to be applicable to other wildlife disease problems—even those problems where supplemental feeding is not an issue. Supplemental feeding decisions in our model represent the easiest method of affecting disease transmission for the Michigan case, and control of disease transmissions would likely be a part of any wildlife disease management strategy. For other diseases, alternative environmental variables could be manipulated in ways that reduce disease transmission, and it is reasonable to believe that such actions might result in tradeoffs in in situ productivity (e.g., if contact is somehow reduced then fertility might also be expected to decline). Hence, the current model provides a foundation for analyzing a range of wildlife disease problems. Finally, two caveats are worth mentioning. First, the disease was assumed to be unsustainable beyond the core area. This appears reasonable for the Michigan bovine TB problem, but it may not be the case for some other diseases. Rather, it might be possible for some other diseases to spread among additional populations. Such a situation might imply greater marginal damages due to the disease and hence more incentives to contain the disease. Additional tradeoffs may also arise for spatially differentiated populations that possibly interact through migratory processes. A spatially explicit analysis would be required in such instances to fully assess the implications of spatial disease transmission. A second caveat is that livestock sector management responses (e.g., biosecurity investments) were not included in our model. The economics of jointly managing valuable wildlife and livestock (and even human) populations to control zoonotic diseases is left to future research. [Received January 2004; accepted September 2004.] References
Arrow, K.J. “Optimal Capital Policy with Irreversible Investment.” In J.N. Wolfe, ed. Value, Capital, and Growth: Papers in Honour of Sir John Hicks. Edinburgh: University Press Edinburgh, 1968, pp. 1–19. Barlow, N. “A Spatially Aggregated Disease/Host Model for Bovine Tb in New Zealand Possum Populations.” Journal of Applied Ecology 28(1991a):777–93. ——. “Control of Endemic Bovine Tb in New Zealand Possum Populations: Results from a

Simple Model.” Journal of Applied Ecology 28(1991b):794–809. Begon, M., S.M. Feore, K. Brown, J. Chantrey, T. Jones, and M. Bennett. “Population and Transmission Dynamics of Cowpox in Bank Voles: Testing Fundamental Assumptions.” Ecology Letters 1(1998):82–6. Begon, M., S.M. Hazel, D. Baxby, K. Bown, R. Cavanagh, J. Chantrey, T. Jones, and M. Bennett. “Transmission Dynamics of a Zoonotic Pathogen within and between WildLife Host Species.” Proceedings of the Royal Society London (Biology) 266(1999):1939–45. Bicknell, K.B., J.E. Wilen, and R.E. Howitt. “Public Policy and Private Incentives for Livestock Disease Control.” Australian Journal of Agricultural and Resource Economics 43(1999):501– 21. Boyle, K.J., B. Roach, and D. Waddington. 1996 Net Economic Values for Bass, Trout, and Walleye Fishing, Deer, Elk and Moose Hunting, and Wildlife Watching: Addendum to the 1996 National Survey of Fishing, Hunting and Wildlife Associated Recreation. Washington DC: U.S. Fish and Wildlife Service, 1998. Bryson, A.E. Jr., and Y.C. Ho. Applied Optimal Control: Optimization, Estimation, and Control. New York: Hemisphere Publishing, 1975. Clark, C.W. Mathematical Bioeconomics. New York: Wiley, 1976. Clark, C.W., F.H. Clarke, and G.R. Munro. “The Optimal Exploitation of Renewable Resource Stocks: Problems of Irreversible Investment.” Econometrica 47(1979):25–47. Conrad, J.W., and C.W. Clark. Natural Resource Economics: Notes and Problems. New York: Cambridge University Press, 1987. Dietrich, R.A., S.H. Amosson, and R.P. Crawford. “Bovine Brucellosis Programs: An Economic/Epidemiological Analysis.” Canadian Journal of Agricultural Economics 35 (1987):127–40. Dobson, A.P., and M. Meagher. “The Population Dynamics of Brucellosis in the Yellowstone National Park.” Ecology 77(1996):1026–36. Ebel, E.D., R.H. Hornbaker, and C.H. Nelson. “Welfare Effects of the National Pseudorabies Eradication Program.” American Journal of Agricultural Economics 74(1992):638–45. Frawley, B.J. “1999 Deer Harvest for Early Archery, Early Firearm, and Regular Firearm Deer Seasons in the Northeast Lower Penninsula.” Michigan Department of Natural Resources, Wildlife Bureau Report No. 3304, 1999. Heesterbeek, J.A.P., and M.G. Roberts. “Mathematical Models for Microparasites of Wildlife.” In B.T. Grenfell and A.P. Dobson, eds. Ecology

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

550

August 2005

Amer. J. Agr. Econ.

of Infectious Diseases in Natural Populations. Cambridge: Cambridge University Press, 1995, pp. 90–122. Hickling, G.J. “Dynamics of Bovine Tuberculosis in Wild White-Tailed Deer in Michigan.” Michigan Department of Natural Resources, Wildlife Division Report No. 3363, 2002. Hill, H.R. “The 2002 Deer Pellet Group Surveys.” Michigan Department of Natural Resources, Wildlife Report No. 3376, 2002. Huffaker, R.G., and J.E. Wilen. “Dynamics of Optimal Stocking in Plant/Herbivore Systems.” Natural Resource Modeling 3(1989):553–73. Kuchler, F., and S. Hamm. “Animal Disease Incidence and Indemnity Eradication Programs.” Agricultural Economics 22(2000):299–308. Liu, C. “An Economic Impact Evaluation of Government Programs: The Case of Brucellosis Control in the United States.” Southern Journal of Agricultural Economics 11(1979):163–8. Mahul, O., and A. Gohin. “Irreversible Decision Making in Contagious Animal Disease Control Under Uncertainty: An Illustration Using FMD in Brittany.” European Review of Agricultural Economics 26(1999):39–58. Maler, K.-G., A. Xepapadeas, and A. de Zeeuw. ¨ “The Economics of Shallow Lakes.” Environmental and Resource Economics 26(2003):603– 24. McCallum, H., N. Barlow, and J. Hone. “How Should Pathogen Transmission Be Modeled?” Trends in Ecology and Evolution 16(2001):295– 300. McCarty, C.W., and M.W. Miller. “A Versatile Model of Disease Transmission Applied to Forecasting Bovine Tuberculosis Dynamics in White-Tailed Deer Populations.” Journal of Wildlife Diseases 34(1998):722–30. McGehee, R., and R. Armstrong. “Some Mathematical Problems Concerning the Ecological Principle of Competitive Exclusion.” Journal of Differential Equations 23(1977):30–52. McInerney, J. “Old Economics for New Problems—Livestock Disease: Presidential Address.” Journal of Agricultural Economics 47(1996):295–314. Meagher, M., and M.E. Meyer. “On the Origin of Brucelossis in Bison of Yellowstone National Park: A Review.” Conservation Biology 8(1994):645–53. Michigan Department of Agriculture (MDA). Michigan Bovine TB. Available at http://www. bovinetb.com (downloaded 2002). Miller, M.W., and B. Corso. Risks Associated with M. bovis in Michigan Free-Ranging White-Tailed Deer: An Update to the 1996 Report. Fort Collins, CO: U.S. Department of Agriculture,

Animal and Plant Health Inspection Service, Centers for Epidemiology and Animal Health, 1999. Miller, R., J.B. Kaneene, S.D. Fitzgerald, and S.M. Schmitt. “Evaluation of the Influence of Supplemental Feeding of White-Tailed Deer (Odocoileus Virginianus) on the Prevalence of Bovine Tuberculosis in the Michigan Wild Deer Population.” Journal of Wildlife Diseases 39(2003):84–95. O’Brien, D.J., S.M. Schmitt, J.S. Fierke, S.A. Hogle, S.R. Winterstein, T.M. Cooley, W.E. Moritz, K.L. Diegel, S.D. Fitzgerald, D.E. Berry, and J.B. Kaneene. “Epidemiology of Mycobacterium bovis in Free-Ranging WhiteTailed Deer in Michigan.” Preventive Veterinary Medicine 54(2002):47–63. Reed, W.J. “Optimum Age-Specific Harvesting in a Nonlinear Population Model.” Biometrics 36(1980):579–93. Rondeau, D. “Along the Way Back from the Brink.” Journal of Environmental Economics and Management 42(2001):156–82. Rondeau, D., and J. Conrad. “Managing Urban Deer.” American Journal of Agricultural Economics 85(2003):266–81. Schmitt, S.M., S.D. Fitzgerald, T.M. Cooley, C.S. Bruning-Fann, L. Sullivan, D.E. Berry, T. Carlson, R.B. Minnis, J.B. Payeur, and J. Sikarskie. “Bovine Tuberculosis in FreeRanging White-Tailed Deer in Michigan.” Journal of Wildlife Diseases 33(1997):749–58. Simonetti, J.A. “Wildlife Conservation Outside Parks Is a Disease-Mediated Task.” Conservation Biology 9(1995):454–6. Swallow, S.K. “Depletion of the Environmental Basis for Renewable Resources: The Economics of Interdependent Renewable and Nonrenewable Resources.” Journal of Environmental Economics and Management 19(1990):264–80. Tahvonen, O., and S. Salo. “Nonconvexities in Optimal Pollution Accumulation.” Journal of Environmental Economics and Management 31(1996):160–77. U.S. Department of Agriculture, Animal and Plant Health Inspection Services, Veterinary Services. Bovine Tuberculosis Eradication Uniform Methods and Rules, Effective January 22, 1999. Available at www.ncagr.com/vet/ FedTBUMRTxt.htm (downloaded 2004). U.S. Department of Agriculture, Animal and Plant Health Inspection Service (USDA-APHIS). Foot-and-Mouth Disease. Available at http:// www.aphis.usda.gov/oa/pubs/fsfmd301.htm, 2002 (downloaded 2002). U.S. Department of the Interior, Fisheries and Wildlife Service, U.S. Department of

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010

Horan and Wolf

Managing Infectious Wildlife Disease

551

Commerce, Bureau of the Census. 1996 National Survey of Fishing, Hunting, and Wildlife-Associated Recreation, Washington DC, 1996. Williams, E.S., M.W. Miller, T.J. Kreeger, R.H. Kahn, and E.T. Thorne. “Chronic Wasting Disease of Deer and Elk: A Review with Recommendations for Management.” Journal of Wildlife Management 66(2002):551–63. Wolf, C.A., and J.N. Ferris. Economic Consequences of Bovine Tuberculosis for Michigan Livestock Agriculture. A report to the Michigan Department of Agriculture, 2000.

Appendix
The model is calibrated using parameters obtained from a variety of sources. The initial core-area deer population, N 0 , was estimated to be 13,298 in Spring 2002 (Hill). Since 1995, core prevalence rates have fluctuated between 2.2% and 4.8%, averaging 2.3% during the period 1998–2000 (O’Brien et al.). We adopt a value of 0 = 0.023, as it is believed this has been fairly constant over the past few years (Hickling, O’Brien et al.). Core carrying capacity is taken to be the upper bound of estimates from the 1960s prior to extensive hunt club feeding activities. This value is 9 deer/km2 (Miller et al., O’Brien et al.), which implies k = 14,049 for the 1,561 km2 core area. Recent populations, which benefit from extensive feeding, have peaked around 19–23 deer/km2 (O’Brien et al., Hickling). Since these populations have been subject to significant exploitation, we use a slightly higher effective carrying capacity value of 27 deer/km2 , which translates to k/(1 − f ) = 42,147. Miller et al. report approximately 8,212 kg/km2 of fruits, vegetables, and grains being fed to deer in the core area (probably an underestimate). We adopt this value and set f = 8,212, solving for = 0.00008. Note that, in our analysis, we set the maximum value of f equal to f max = 10,000. This choice is somewhat arbitrary but it has little bearing on our qualitative graphical results.

For disease transmission, we derive (1 + f ) = 0.346 using reported rates of infected contact by sex along with survival rates from the time of contact to that of infection (Miller and Corso), and also reported deer sex ratios (McCarty and Miller). Miller et al.’s results on the relation between increased feeding and increased transmission are used to calibrate = 2.64 × 10−6 . Given this value and f = 8,212, we can solve for = 0.339. The intrinsic growth rate for white-tailed deer is taken to be r = 0.5703 (Rondeau and Conrad). Then, using results from Hill on nonhunting mortality, we adopt an effective disease-related mortality of (1 − f ) = 0.2. This rate does not imply that 20% of all infected deer die as a direct result of the disease, as few deer actually die from tuberculosis. Rather, the deer are weakened by their infection and ultimately die from something else. More information is needed to calibrate and . The unsustainable nature of the disease outside the core suggests that > . We assume / = 1.05, as this value produces reasonable results relative to historical changes in disease prevalence when recent deer populations, disease prevalence rates, harvests, and feeding choices are plugged into the model. Using our previously defined values, we can solve for = 0.356 and = 0.3623. The price per harvested deer is p = $1,270.80, which is derived from various estimates of consumer’s surplus, hunting effort, and expenditures provided by Boyle, Roach, and Waddington; Frawley; and U.S. DOI-FWS. Scaled harvesting costs, c/q, are taken from Rondeau and Conrad to be $231,192. This cost might be smaller in the Michigan case, but we find the results are insensitive to even fairly significant changes in this value (Rondeau and Conrad report a similar finding). The price of feed is set at w = 36.53, which is imputed from Miller et al.’s feed density rates for the core along with anecdotal evidence about feed expenditures. Finally, using the results of Wolf and Ferris (given “split-state” status), we take variable damages in the core area to be about $1.6 million per year and, using a linear damage function of the form z, we derive = 5,491. Fixed damage costs are taken to be $4 million per year.

Downloaded from ajae.oxfordjournals.org at Virginia Polytechnic Institute and State University on June 27, 2010