Farm Heterogeneity in Wildlife Production - AgEcon Search

Paper AAEA Annual Meeting August 8-11, 1999 • Nashville, Tennessee

Farm Heterogeneity in Wildlife Production

Jaap van Wenum1, Alfons Oude Lansink1 and Ada Wossink1,2

1

Department of Economics and Management, Wageningen Agricultural University Hollandseweg 1, 6706 KN Wageningen, Netherlands

2

Department of Agricultural and Resource Economics North Carolina State University, Raleigh, NC 27695-8109, USA

Farm Heterogeneity in Wildlife Production

Abstract

The paper discusses the appropriate functional form and estimation technique for a wildlife production function on the farm level. A random effect model was developed to capture the relationship between wildlife output, management practices, natural conditions and non-observed factors that cause a farm specific management bias. Two wildlife measures were compared: species richness and a wildlife yardstick that enables different species to be weighed on their ecological value. The results show that the farm specific factor has a significant impact on the production of wildlife both in terms of species numbers and in terms of ecological values. Parameter estimates that were similar for the species richness specification were found to vary with the yardstick specification, which indicates the added value of the wildlife yardstick. The results have implications for efficient payment schemes for wildlife production.

1. Introduction

World wide, there is growing concern about decreasing biodiversity, caused by the degradation of living conditions of plants and animals. Increasingly, modern society values the wildlife biodiversity benefits that arise as joint outputs with agricultural land use (Edwards and Abivardi, 1998). To the extent that the production of these public goods is not adjusted to social optima, a role for public policy is called for both in the short run and the long run. The US Endangered Species Act and the EU Wild Birds Directive and Habitats Directive are examples of such public policies (Rosso Grossman, 1997; Brown and Shogren, 1998). In the short-run, the public interest is in ensuring socially optimal use of privately controlled inputs that influence the supply of wildlife as a public good. In the long-run public policy can play an important role by provision of research on new technologies that enhance wildlife productivity, for example organic farming systems. The focus of this paper is on the social interest of wildlife preservation specifically in agriculture in the short-run. In agricultural areas where biodiversity has already seriously been damaged, ways have to be found for restoration and enhancement and this requires active wildlife management on

2

farms. Identification of cost efficient wildlife policies depends on the relationships between current land use activities and wildlife values, and assessment of opportunity costs of foregone uses. In this task, ecological economics has an important role to play. A first step includes the estimation of a wildlife production function, which captures the relationship between land use practices, natural conditions and wildlife at the farm level. Effects of options for wildlife conservation management should be determined by using a production function that includes a definition and measurement of wildlife in a tangible way that enables different land use practices and site specific conditions to be compared. A farm specific management factor in wildlife production is expected caused by non-observed factors such as past-activities on the farm. The wildlife production function needs to account for this accordingly. Little work has been done on modeling the relation between agricultural practices and wildlife production at the farm and field level based on non-experimental data. Usually field experiments are done comparing species richness between different crops or fallow alternatives or across soil types (see for example the work on unsprayed field margins by De Snoo (1995)). Data and relationships derived from such experimentally controlled settings do not include the complex, heterogeneous ecosystem processes which may vary across real life observations with input mixes and practices. This causes an expected heterogeneity or management factor that is unobservable but has significant implications for effective policy design. The first objective of the paper is to contribute to a better understanding of the appropriate functional form and estimation technique for the wildlife production function based on agronomic and ecological insights. We also give specific attention to the selection of an indicator that correctly measures the wildlife output in response to wildlife management activities. A second objective is to investigate the importance of farm specific factors in wildlife production. The third objective is to show how the estimates of a wildlife production function based on a panel data of Dutch field crop farmers can be used for management and policy support. The outline of the paper is as follows: Section 2 presents the theoretical framework of the research. Next Section 3 discusses the data and the wildlife indicators used for the estimation. Section 4 describes the estimation procedure and section 5 presents results with special reference to the management factor in wildlife production. The paper finishes with discussion and conclusions.

3

2. Theoretical model

Similar to yields of agricultural crops, wildlife output varies among regions, among farms and from year to year. Production ecologists have developed a conceptual framework that can be used for a better understanding of the economics of crop management (see Wossink and Rossing, 1998). In this section this framework is adapted for wildlife production. Our conceptual framework distinguishes production environments, growth factors, management inputs and production levels (Figure 1). The production environment, E, at a specific site represents the setting for wildlife production and is characterized by physical factors that include climate and aspects of the soil (groundwater table, type of soil). Variation in wildlife production per unit acreage in similar production environments is attributable to differences in three categories of growth factors (output defining, output limiting and output reducing) that vary within a year and between years. Wildlife output defining factors include: (a) weather during the growing season (solar radiation, temperature), (b) factors due to management in the past such as the level of eutrophication and desiccation of the soil, presence of vegetation remnants and extent of the flora seed bank. The output defining factors together determine the 'potential wildlife yield' for a specific production environment. The extent to which this potential level is achieved in practice depends on operational management (control) of output limiting, L, and reducing factors, R. Output limiting factors depend on crops selected (including fallow), rotation and size and spatial pattern of fields and field margins. Together the limiting factors determine ‘attainable wildlife yield’ for a given production environment E. The output reducing factors depend on nutrient management, water management and pest control and determine ‘actual wildlife yield’.

4

Figure 1. Conceptual framework of wildlife management Yield level Management opportunities Defining factors (D) Potential

None

Potential

Crop selection and rotations Limiting factors (L)

Size and spatial pattern of fields and of field margins

Attainable

Nutrient management Reducing factors (R) Attainable

Actual

Water management Pest management

Actual

Production

Production

Environment (E1)

Environment (E2)

Source: Adapted from De Koeijer et al. (1999).

To integrate the insights depicted in Figure 1 we define the wildlife production frontier as: F (Y , X , D ; E) = 0

(1)

where Y is the vector of wildlife output (actual yield), X is a vector of effective levels of management inputs, D is a vector denoting the variable output defining factors that are not controllable by management and E denotes the production environment. The origin of the limiting and reducing factors justifies an indirect role of the management inputs X. Let XL and XR denote the inputs that determine the limiting and reducing factors, respectively. Equation (1) may now be rewritten as F[Y , Φ((X R , Y D )( X L , Y D ))D ; E] = 0

(2)

5

Where YL is the attainable yield level and YD is the yield level as resulting from the defining factors D and given the production environment E. Further Φ is a vector of functions representing the effective result of the management inputs XL and XR affecting production. There is no prior information from non-experimental data that could be directive on the appropriate functional form for Φ. Besides, given that statistical approached allow for indirect effects and interactions, the specification as in (1) must be viewed as unnecessary. That is, equation (1) may be written in composite form as: F (Y , X

R

, X L , D ; E) = 0

(3)

For a specific farm i (i = i, …, I) and year t (t=1,…, T) the wildlife production function may now be denoted as: Y it = f it ( X itR , X itL ; D it ; E i )

(4)

Assuming a common underlying technology of wildlife production over farms i (i = i, …, I) and years t (t=1,…, T) equation (4) may be reformulated as: Y it = φ i + f ( X itR , X itL , D it ; E i )

(5)

where the parameter φi represents the ith farm-specific effects. Note that with nonexperimental data, observations might be available on output Y it , inputs X itR , X itL and on Ei but not on Dit. Hence, when estimating (5) the management factor will include both φi and Dit.

3. Data indicators and empirical model

Data

Data on wildlife conservation in crop farming were available from different wildlife projects in the Netherlands varying in natural conditions and parcel lay out. This study only used data on vascular plants collected on farms in the northern sand area, the northern clay area and the central clay area (young polders). Commonly used census methods for vascular plants are the methods of Braun-Blanquet, Londo and Tansley (Den Held and Den Held, 1992). The projects considered in this paper either used the Londo-scale or Braun-Blanquet method to assess the vegetation. Both methods use small representative plots of standard size within which an inventory of all plant species is made.

6

Data were used from field margin experiments carried out by the provincial authorities of Groningen and Drenthe. Vegetation assessments were carried out on sandy soil in both 6meter wide unsprayed and conventional field margins of winter and spring cereals, potatoes and sugar beet. Furthermore data were used from a nationwide project on alternative fallow field management. Assessments were carried out for different management options including phacelia, grass-clover, nature fallow mixtures (mixtures of more than three legumes/catch crops etc) and natural vegetation (spontaneously developed vegetation). Assessments were carried out in both margins and center of the field (Canters et al. 1996). Finally data were obtained from two projects in the central clay area on grass-clover strips along crop fields (Remmelzwaal and Voslamber, 1995; Kuiper, 1995). Vegetation assessments were made for the strips and for the field edge (ditch banks, verges).

Indicators

Many attempts have been made to indicate wildlife and biodiversity by construction of species diversity indexes. Ecologists often construct biodiversity indexes as a function of species counts and the relative abundance of species (Magurran, 1988), whereas economists construct biodiversity indexes as a function of genetic distances among members of a species set (Solow et al., 1993). Species richness is the simplest form of these indexes, neglecting differences in abundance or genetic distance. In this study two indicators for wildlife production are used. The from the literature well known species richness indicator is compared to an extensive species based indicator specifically developed for agriculture: the wildlife yardstick (Buys, 1995; van Wenum et al., 1998). Both indicators are applied to the vascular plant species group only. The wildlife yardstick consists of a representative set of species from the following species groups: vascular plants, mammals, birds, butterflies, amphibians and reptiles. To each species, a rating (0-100 points) has been assigned based on its protection need as determined by rarity, population tendency and international importance. Rarity of species (Rj) was calculated by dividing the total number of topographical grid cells in the Netherlands (1677, each grid cell at 25 km2) by the number of grid cells in which a species is found (Buys, 1995):

7

Rj = 1677 / number of topographical grid cells with species j found

(6)

Tendency of species (Tj) was calculated as the change in (national) population size in terms of percentage:

 (current population size j − population size j in the past )  ∗ 100 Tj =    population size in the past j  

(7)

The period considered when assessing population changes differs between species group because of data availability. For species with constant or increased population sizes, Tj equals 0. For different species groups, different criteria are used to assess international importance (Bink et al., 1994). The number of criteria ranges from 1 (mammals, amphibians and reptiles) to 4 (nesting birds). International importance of species (Ij) was incorporated in the yardstick rating by determining the relative number of criteria a species meets:

Ij =

number of criteria for international importance species j meets number of criteria for international importance of respective species

(8)

The yardstick rating of a species (Ej) now is composed as follows (Buys, 1995):

(

)

Ej = 18,5∗ log Rj∗ Tj(1+ ( Ij∗0,5))

(9)

The logaritmical transformation was carried out to achieve a rating range from 1 to 100 and for creating distinction between (general) species ratings. For this study plant species not part of the representative set were rated similarly. A yardstick score per area measure now is calculated as the product of species ratings and number of species found for the respective area measure.

8

Empirical model

Estimation of a wildlife production function ideally requires information about limiting factors, restricting factors, defining factors and the production environment. However, the available data contain a limited amount of information. Data on wildlife production and distance of the sampling spot to the field edge are available as continuous variables. Information about agricultural activities takes the form of discrete variables, indicating the specific crop that is grown on the field. Furthermore, the available data contain information about the soil type. A functional specification of the wildlife production function that incorporates this information is the following:

8

2

j =1

i =1

YitW = M i + ∑ α j Aijt + ∑ β j S ij + γ 1 Dit + γ 2 Dit2

(10)

where Mi is the unobservable management variable for the ith farm, Aijt denotes a dummy variable for agricultural activities of the ith farm at time t with j=1 (grass-clover), 2 (nature mix fallow), 3 (natural vegetation), 4 (unsprayed winter cereals), 5 (unsprayed spring cereals), 6 (potatoes), 7 (sugar beet) and 8 (phacelia fallow). The dummy variables Aijt take the value 1 if activity j is present at time t at farm i and 0 otherwise. Sij are regional dummy variables with j=1 (northern clay area) and 2 (central clay area) that take the value 1 if the j-th region applies and zero otherwise. The northern sand area is the reference area in this regression, i.e. Sij is zero for all i,j in the northern sand area. Dit represents the distance in meters from the sampling spot to the edge of the field. The quadratic specification allows for both increasing and decreasing marginal effect of distance on wildlife production.

4. Estimation

This section discusses the estimation procedure that is used in order to estimate the unobserved management variable and the structural parameters of (10). Prior to this discussion, a few remarks are made about the available data. First it should be noted that the available panel data set of Dutch farms is unbalanced, since the length of the time series differs by farm. Second, it should be noted that some

9

variables differ across farms only, i.e. they are constant for individual farms. Therefore, the often applied fixed effects estimator cannot be used, since parameters associated with variables that vary across farms only are incorporated in the fixed effect (Baltagi, 1995). Estimation of the structural parameters and the unobserved management variable is achieved by applying a random effects estimator. The empirical model (10) is rewritten by imposing a specific error structure:

YitW =

8

2

j =1

i =1

∑α j Aijt + ∑ β j S ij + γ 1 Dit + γ 2 Dit2 + uit

(11)

where the composite error term uit has the following structure :

uit = M i + eit

(12)

Mi is assumed to be a random variable representing an unobservable management factor which is i.i.d. (0,σM) and eit is i.i.d. (0,σe) and Mi is independent of eit. In addition, Aijt, Sij and Dit are assumed to be independent of Mi and eit (Baltagi, 1995). Estimation of the random effects model is done by transforming the data prior to estimation. The transformation gives the following equation:

8

2

j =1

j =1

YitW − θ iYi W = ∑α j ( Aijt − θ i Aij ) + ∑ β j Sij +

(13)

γ 1 ( Dit − θ i Di. ) + γ 2 ( Dit2 − θ i DDi ) where θi is defined as 1 − σ e (Tiσ M2 + σ e2 ) 0.5 and all barred variables indicate farm-specific means. From (11), it can be seen that estimating the random effects model requires a consistent estimate of σM and σe. An estimate for σe is obtained from the residuals of a fixed effects ('within') estimation of the model in (13):

σ e2 =

eˆ' eˆ ` n− N − K

(14)

Where n, N and K are the total number of observations, the number of farms and the number of parameters to be estimated, respectively. A consistent estimate of σM is obtained from a regression on the individual means: N

2

j =1

j =1

Yi W = ∑α j Ai + ∑ β j Sij +γ 1 Di . + γ 2 Di Di + M i + ei

(15)

10

This regression uses N observations (one observation for each farm) and it can be shown that the variance of the composite disturbance term M i + ei gives a consistent estimate of

σ M2 + σ e2T , where T is defined as 1

N

1 (Greene, 1998:337). N∑ Ti i =1

Estimates of the structural parameters of the wildlife production function are obtained by calculating the value of θi (using the estimates for σM and σe ) and performing OLS on the transformed equation (13).

5. Results

Equation (15) was estimated using both the species richness and the wildlife yardstick value. Results of the random effects estimation can be found in table 1. Parameters are significant at the critical 5% level for 66% of the species richness specification and 58% of the nature yardstick value specification. The R2 of the species richness specification is 0.73 and 0.68 for the yardstick value specification.

11

Table 1:

Random Effects estimation results of species richness and yardstick value specification Species richness specification

Parameter

Yardstick value specification

Estimate

t-value

Parameter

Estimate

t-value

α1

7.95

5.17*

α1

27.35

2.23*

α2

14.17

10.74*

α2

25.85

2.52*

α3

12.70

4.94*

α3

35.17

1.52

α4

14.72

7.07*

α4

43.75

2.68*

α5

16.92

6.01*

α5

50.35

2.14*

α6

3.99

1.47

α6

-22.51

-0.99

α7

0.10

0.04

α7

-12.93

-0.53

α8

7.09

3.72*

α8

-15.89

-0.96

β1

-3.36

-1.18

β1

-20.60

-1.03

β2

0.29

0.18

β2

26.89

2.24*

γ1

2.07

5.97*

γ1

29.93

9.04*

γ2

-0.16

-6.53*

γ2

-2.27

-9.42*

α1-α8) parameters associated with dummy variables of grass-clover (1), nature mix fallow (2), natural vegetation (3), unsprayed winter cereals (4), unsprayed spring cereals (5), potatoes (6), sugar beet (7) and phacelia fallow (8). β1-β2) parameters associated with regional dummy variables. γ1-γ2) parameters associated with distance form the field edge. *)

Significant at the critical 5% level.

12

Table 2 : Differences between parameters αi for species richness and yardstick value specifications (t-ratios in parentheses). Yardstick value specification α1 α1 S P E C I E S

I C H N E S S

α4

α5

α6

α7

α8

1.50

-7.82

-16.40

-23.00

49.86

40.29

43.25

(0.18)

(-0.36)

(-0.89)

(-0.92)

(2.06)*

(1.56)

(2.88)*

-9.32

-17.90

-24.50

48.36

38.79

41.75

(-0.45)

(-1.06)

(-1.03)

(2.10)*

(1.57)

(3.00)*

-1.47

-8.58

-15.18

57.68

48.11

51.07

(-0.67)

(-0.32)

(-0.48)

(1.87)

(1.49)

(2.06)*

(6.88)* α3 4.75 (2.02)* α4 6.77 α5 8.98 (2.92)* α6 -3.95 (-1.32) α7 -7.84 (-2.54)* α8 -0.86 (-0.53)

*)

α3

α2 6.22

(2.81)* R

α2

0.54

2.02

-6.60

66.26

56.69

59.65

(0.24)

(0.64)

(-0.30)

(3.11)*

(2.64)*

(2.81)*

2.75

4.23

2.21

72.86

63.30

66.25

(0.93)

(1.15)

(0.90)

(2.52)*

(2.13)*

(2.44)*

-10.18

-8.70

-10.72

-12.92

-9.57

-6.61

(3.56)*

(-2.41)*

(-4.56)*

(-3.94)*

(-0.37)

(-0.25)

-14.07

-12.59

-14.62

-16.82

-3.90

2.96

(-4.74)*

(-3.41)*

(-6.40)*

(-5.13)*

(-1.39)

(0.11)

-7.08

-5.61

-7.63

-9.83

3.09

6.99

(-4.78)*

(-2.12)*

(-2.88)*

(-3.02)*

(0.97)

(2.13)*

Significant at the critical 5% level

The results show that in both the species richness and yardstick value specification, unsprayed cereals (α4 and α5) are more beneficial to wildlife production than the fallow alternatives and other crops. Table 2 shows differences between parameters for species richness and yardstick value specifications. The difference between cereals and all other alternatives is significant in most cases, except for nature mix and natural vegetation. It can also be seen that in both specifications, sugar beet and potatoes are least beneficial to wildlife production. Within the group of fallow alternatives (α1 - α3 and α8) phacelia fallow gives the lowest contribution to wildlife production for both specifications. Also, within the group of fallow alternatives, natural vegetation gives a higher contribution to wildlife production in terms of yardstick value than the all other fallow alternatives, even though vegetation has been sown in there for the

13

sake of wildlife production. Furthermore, the nature mix fallow parameter is highest of the fallow alternatives for the species richness specification, whereas it is lower than the grassclover and natural vegetation parameter in the yardstick value specification. This indicates that wildlife produced by nature mix fallow have a lower protection need than wildlife produced by grass-clover and natural vegetation. The soil type parameters (β1 and β2) are not significant for the species richness specification whereas for the yardstick value specification β2 (central clay parameter) is significant and more positive, indicating that species found in the central clay area have a higher protection need than species found on sandy and northern clay soils. The distance parameters (γ1 and γ2) are similar for both specifications. A Lagrange multiplier test (Baltagi, 1995 : 163) is used to test the joint significance of all farm specific effects. Under the null hypothesis: H0 :all µi = 0, or equivalently σµ2 =0, this test is asymptotically distributed as χ2(1). The test statistic is 48.33 for the species richness specification and 145.66 for the yardstick value specification. Therefore, the null hypothesis all µi = 0 is clearly rejected for both specifications at the critical 5% level, which takes the value 3.84. The results of these tests imply that farm specific conditions have a significant impact on the production of wildlife in terms of species richness and in terms of yardstick value.

6. Discussion and Conclusions

The study clearly shows the advantages of the yardstick approach over the species richness indicator in estimating wildlife production functions. Species richness estimates do not give information on the importance for wildlife conservation of the species found, whereas the yardstick incorporates the protection need of the species therefore presenting a more reliable estimate of wildlife production. In European policy there is a trend in agricultural wildlife conservation from effortbased payments to incentives and payment schemes linked to wildlife production. The main advantage of these wildlife output based systems for farmers is that it allows them to choose their own management strategy to obtain a certain wildlife level. For policy makers the system guarantees that wildlife targets are met. However, the conclusion from this study that nonobserved farm specific conditions are having a significant impact on wildlife production, may result in adverse selection. Farmers with already favorable conditions for the presence of

14

wildlife will be attracted to such payment schemes and possibly overcompensated, considering their efforts. Regional or local differentiation in payment levels will be required to control this. Further research on wildlife production functions in agriculture should focus on dynamics in particular. Incorporating the development of wildlife production over time will provide more insights into the complex factors that cause variations in wildlife production in agriculture. Furthermore for a complete view of wildlife in farming other species groups than vascular plants need to be considered as well.

15

References

Baltagi, B.H. (1995). Econometric Analysis of Panel Data. Chichester: Wiley. Bink, R.J., D. Bal, V.M. van den Berk and L.J. Draaijer (1994) Toestand van de natuur 2. IKC Natuur, Bos, Landschap en Fauna; Ministerie van Landbouw Natuurbeheer en Visserij, Wageningen Buys, J.C. (1995). Towards a yardstick for biodiversity on farms (in Dutch with English summary). Publication number CLM 169. Utrecht: Centrum voor Landbouw en Milieu. Brown, G.M. and J.F. Shogren (1998). Economics of the Endangered Species Act, Journal of Economic Perspectives 12(3): 3-20. Canters, K.J., J. De Leeuw, R.J. Van Der Poll, E.P.M. Van Dongen, H.Heemsbergen and E.M.P. Hermens (1996). Natuur in Natuurbraak, werkdocument IKC Natuurbeheer nr. W-101. IKC-L, CML, Wageningen/Leiden De Koeijer, T.J., G.A.A. Wossink, M.K. van Ittersum, P.C. Struik, J.A. Renkema (1999). A conceptual model for analysing input-output coefficients in arable farming systems: from diagnosis towards design (in press Agricultural Systems). Den Held, J.J. and A.J. Den Held (1992) Beknopte handleiding voor vegetatiekundig onderzoek.

Wetenschappelijke

mededelingen

nr.

97.

Koninklijke

Nederlandse

Natuurhistorische Vereniging, Utrecht De Snoo, G.R. (1995). Unsprayed field margins: implications for environment, biodiversity and agricultural practice, Ph.D.-thesis Leiden University, Leiden Edwards, P.J. and C. Abivardi (1998). The value of biodiversity: where ecology and economy blend. Biological Conservation 83(3): 239-246. Greene, W.H. (1998). LIMDEP version 7.0. Users manual. New York. Kuiper, R. (1995) Project akkerrandenbeheer Wieringermeer. In: G.R. De Snoo, A.J.W. Rotteveel and H. Heemsbergen (Eds.) Akkerranden in Nederland. Werkgroep Akkerranden, Wageningen/Leiden Magurran, A.E, (1988) Ecological diversity and its measurement. Croom Helm Limited., London Remmelzwaal, A.J. and B. Voslamber (1996). In de marge, een onderzoek naar ruimte voor de natuur op landbouwbedrijven. Flevobericht nr. 390. Ministry of Transport, Public Works and Water Management, Lelystad.

16

Rosso Grossman, M. (1997). Habitat and Species Conservation in the European Union and the United States, Drake Law Review 45: 19-49. Solow, A., S. Polasky and J. Broadus (1993) On the measurement of biological diversity. Journal of Environmental Economics and Management 24, 60-68 Van Wenum, J.H., J.C. Buys and G.A.A. Wossink (1998). Nature quality indicators in agriculture, In: F. Brouwer and R. Crabtree (Eds.), Environmental Indicators and Agricultural policy, Wallingford UK: CABI, pp. 105-120 Wossink, G.A.A. and W.A.H. Rossing, (1998). On increasing returns and discrete choice: integrating production ecological principles in economic analysis of crop management, Journal of Environmental Management 54 (3): 233-247.

17