- Email: [email protected]
Exploiting decision heuristics and IT in the design of a DSS for voluntary agri-environmental programs
Ecological Economics 49 (2004) 303 – 315 www.elsevier.com/locate/ecolecon
METHODS
Exploiting decision heuristics and IT in the design of a DSS for voluntary agri-environmental programs Dennis Collentine a,*, Martin Larsson b, Nils Hannerz c a b
Department of Economics, Swedish University of Agricultural Sciences, Box 7013, S-750 07 Uppsala, Sweden Department of Soil Sciences, Swedish University of Agricultural Sciences, Box 7072, S-750 07 Uppsala, Sweden c Swedish Institute of Agricultural and Environmental Engineering-JTI Box 7033, SE-750 07 Uppsala, Sweden Received 15 October 2002; received in revised form 28 October 2003; accepted 1 February 2004 Available online 19 June 2004
Abstract Low participation rates by farmers in voluntary agri-environmental programs may depend on rationally bounded ex ante estimates of the negative effect of program enrollment on farm income. Uncertainty and the presence of information transaction costs may lead to the use of heuristics by farmers to reduce adoption decision costs. This paper describes how LENNART, a netbased decision support system (DSS), has been designed to exploit the use of heuristics and provide low cost access to information. The model has been developed to evaluate the effects of agronomic measures on farm income and on the leaching of nutrients from cultivated fields. A subsidy program for catch crop cultivation in Southern Sweden served as the basis for development of the DSS and is used throughout the paper for purposes of illustration. D 2004 Elsevier B.V. All rights reserved. Keywords: Agri-environmental policy; BMP implementation; Farm management; Heuristics; SOILNDB
1. Introduction Agronomic practices that contribute to nitrogen leaching are often connected with field cultivation practices. Changes in agronomic practices, so-called best management practices (BMPs), have been identified which if implemented could reduce the level of nitrogen leaching (Carpenter et al., 1998; Gustafson et al., 1998; Feather and Amacher, 1994; Sharp and Bromley, 1979). Implementation of BMPs by farmers * Corresponding author. Tel.: +46-18-67-17-39; fax: +46-1867-35-02. E-mail addresses: [email protected] (D. Collentine), [email protected] (M. Larsson), [email protected] (N. Hannerz). 0921-8009/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolecon.2004.02.002
is generally voluntary and encouraged by support from extension services or other government programs. Unfortunately, these programs have not achieved expected results (Collentine, 2002a; Gustafson et al., 1998; Wolf, 1995; Just et al., 1991; Setia and Magleby, 1987). The success of agri-environmental policy dedicated to the adoption of BMPs by farmers and the cost effectiveness of these policies can be enhanced through an understanding of the factors that determine how farmers make choices with regard to implementation (i.e. when to adopt and which measures to adopt). While farmers attitudes to adoption have been studied, these studies have primarily been surveys to analyze the socioeconomic factors which characterize the adopters and nonadopters (Tucker and Napier,
304
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
2002; Napier and Tucker, 2001; Drake et al., 1999; Buller, 1999; Cooper and Keim, 1996; Amacher and Feather, 1997; Supalla et al., 1995; Feather and Amacher, 1994). Participation in agri-environmental programs that support the adoption of BMPs is primarily an economic decision. If the perceived effect on farm income of a BMP is income neutral or positive then the measure is more likely to be adopted. Buller (1999) observed in a report on the implementation and effectiveness of agri-environmental schemes, that because ‘‘Agri-environmental policy occupies an illdefined middle ground between regulatory approaches to environmental management . . .and more classic generalised market instruments . . . [that] agri-environmental policy critically needs to be placed at the level of the farmer and the farm’’ (pp.105 – 106). That is, successful programs begin with an understanding of how management choices are made by farmers on their farms. Uncertainty with respect to costs associated with the implementation of new field management techniques leads to a need for estimation of the economic effect of adoption on farm income. However, ‘‘since costs and effects of a given practice will differ from farm to farm, the amount of information needed is substantial’’ (Leathers, 1991, p.308). The act of gathering relevant information and then using it for the purpose of decision making, involves costs to the decision maker. These can be classified as a type of transaction cost. That is, costs that are necessary to complete a transaction or in the case of agri-environmental policy to participate in the program. Whitby (2000) notes that with respect to participants in agrienvironmental programs ‘‘they have to incur some level of transaction cost to participate in schemes and these are not generally reimbursed specifically . . . [and that] altruism apart, society must also bear the private transaction costs of contracts’’ (p.322). If the decision making process itself was better understood, then information flows in the form of a decision support system (DSS) could be developed which could reduce the transaction costs of the decisions by farmers to adopt specific measures as well as support authorities in the design, implementation and evaluation of agri-environmental policy. Simon (1987) observed that since decision makers are limited by their capacity to store and process information, they operate in a realm of bounded
rationality when faced with choices. Patrick and DeVuyst (1995) succinctly conclude in their survey on risk research in farm management, ‘‘producers are unwilling to pay more for the information than it is perceived to be worth ...[therefore] efforts should focus on both reducing the cost of risk-related information and increasing its perceived value’’ (p.10). Heuristics are a type of simple decision rule that lowers the cost of accessing and processing information. Information technology (IT) also substantially lowers the cost of computing operations and the storage of data. In addition, the Internet offers the possibility of low cost access to digitalized information. The DSS described in this paper was designed to exploit the use of heuristics by farmers and use the capabilities of IT and the Internet to provide support for the evaluation of the economic and environmental effects of BMPs on individual fields. The first section below analyzes how an individual decision maker, a farmer, searches for and processes information when faced with complex decisions. This section concludes with a description of how heuristics may be used by farmers to reduce the costs of adoption decisions. The next section describes the net-based interactive DSS called LENNART. The paper ends with conclusions and a description of the planned extensions of the model.
2. Farm level implementation decisions 2.1. A simple model of choice The choice by a farmer to implement a new BMP is driven by the flows of returns and benefits associated with the practice relative to the flows and benefits of present practices. The condition for adoption is represented in Eq. (1). The LHS 1 represents the net return (R C) from adopting the BMP of an insown catch crop (cc) on a particular field (a) as a flow over time (t) at a particular expected rate of return (i) and any noneconomic values (M) associated with adoption or nonadoption of the practice. The RHS of Eq. (1) represents the returns from current use (n) of the same field without the BMP. It should be noted that the farmer is already assumed to be profit maximizing using current practices and that the net returns are sufficient to continue with current practices indefi-
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
305
nitely. In the absence of transaction costs, the choice facing the farmer in this case is simple, if the net returns from implementing the BMP are greater than those from current practices, then the practice is adopted.
associated with implementing catch crops as a cultivation practice. If the LHS in Eq. (5) is positive then the decision maker will adopt the measure.
T X
t¼0
cc cc ð1 þ iÞt f ðRcc at Cat ; Mat Þ
t¼0
zð1 þ iÞt f ðRnat Catn ; Matn Þ
ð1Þ
If all of the benefits and costs to farmers are subsumed in the economic returns then there are no externalities considered in the decision to adopt a particular agronomic management practice. In this case, the M term disappears and the decision is simply an evaluation of the difference between the returns (R) and the costs (C) of the two alternatives. In addition, since the timing of flows associated with an insown catch crop corresponds to the timing of flows from sowing the same main crop without a catch crop, then changes in this parameter are highly correlated. Combining both of these assumptions allows Eq. (1) to be rewritten as just the incremental difference between the two practices as in Eq. (2).
T X ð1 þ iÞt ðRdat Catd Þz0
ð5Þ
d ) is composed of two The change in revenues (Rat parts; the areal subsidy for implementing the BMP and the change in income due to changes in the yield of the main crop. The second term can be included as a cost effect, where a fall in yield is included as a production cost and an increase in yield is treated as a production gain. While long-term soil fertility effects are not taken into account explicitly, these could be incorporated in the same manner as the effect on yield. The choice represented in Eq. (5) then is reduced to the present value of the difference between the payment received for planting the catch crop and a linear sum of the field production costs associated with the practice. Following Collentine (2002a), these costs can be expressed as:
Catd ¼
n X
ð6Þ
Abat
b¼1 T X n cc n ð1 þ iÞt ððRcc at Rat Þ ðCat Cat ÞÞz0
ð2Þ
t¼0
Furthermore, if the decision is based only on a comparison of the returns from the two alternative field management practices then farm management choices such as main crop choice and other whole farm decisions are taken as givens based on fixed assets. Thus, the revenue from choosing to plant a catch crop (Rcc) and the costs associated with this practice (Ccc) can be expressed as linear sums of the current practice and the additions in revenue (Eq. (3)) and costs (Eq. (4)) associated with the catch crop. d n Rcc at ¼ Rat þ Rat
ð3Þ
Catcc ¼ Catd þ Catn
ð4Þ
Where each agronomic activity (Ab) is assumed to be linear and independent, these describe a set of values (costs) at the farm and field levels. Substituting this linear cost function into Eq. (5) results in Eq. (7), where the costs and returns associated with the planting of a catch crop are those that each farmer needs to evaluate for each field under consideration. ! T n X X t d ð1 þ iÞ Rat Abat z0 ð7Þ t¼0
Moving from this framework into one where uncertainty enters into the evaluation is possible to model by using the expected value of the costs and returns as seen in Eq. (8). ! T n X X t d ð1 þ iÞ E Rat Abat z0 ð8Þ t¼0
Substituting Eqs. (3) and (4) into Eq. (2) allows the decision to be written as simply an evaluation of the additional revenues (Ratd) and additional costs (Catd)
b¼1
b¼1
Since the returns are a constant subsidy paid out for every hectare enrolled in the program and any yield
306
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
effects are subsumed into production costs, the expected value of the returns is the same as the deterministic value and the only uncertainty that remains is in the cost term, Eq. (9).
for each of the variables must be made.1 In addition, the costs of accessing the information needed to make these estimates also need to be included in the choice model.
! T n X X t d ð1 þ iÞ Rat EðAbat Þ z0
2.2. Choice model with information costs
t¼0
ð9Þ
b¼1
Costs evaluated at the field level for each agronomic activity or effect (Ab) become the determining factors for the decision by individual farmers to enroll acreage in the catch crop program. For catch crops, Collentine (2002a) identifies four agronomic activities and effects with a total of 14 choice variables at the farm and field levels. If the individual costs represented by the choice variables were known, then performing the calculation in Eq. (9) is a relatively simple procedure which could be done on a handheld calculator without a great deal of effort. However, since the choice to implement the use of catch crops on the farm is based on the assumption that the technique is not already being used, it is reasonable to assume that there is uncertainty with respect to the expected values for each type of cost. A risk adverse farmer evaluating the decision would be expected to perform some type of ‘what if’ (sensitivity) analysis. What if the expected cost variable was lower/higher than average, how might this affect the cost of the agronomic activity and in turn, the decision to adopt the measure? If a farmer were to choose only three values for each of these (low, normal and high) to use in calculating the variance of the expected costs, the total number of cost combinations to evaluate in any one time period in Eq. (9) are equal to 314, for a total of 4,782,969 possible cost combinations! It would also be possible for the farmer to choose to analyze costs from the perspective of a worst, average and best case scenario. Even though in this type of analysis there are only three calculations to perform, an assumption needs to be that all the variables are perfectly positively correlated. An assumption that if not made ad hoc would require an even greater number of calculations than for the first type of analysis above. Furthermore, before solving the total costs for each combination of costs during one time period is even possible for the first method, high and low estimates
As described above, there is a need to make estimates of the cost distribution for each of the 14 choice variables in order to make the adoption decision. In some cases, cost estimation may only entail transferring a cost from another activity already practiced on the farm to the practice currently being evaluated; for example, a discount rate or tractor costs. In other cases, gathering the information may require more effort, such as calling around to seed dealers to establish the price of the seed to be used and then adding the cost of transportation to this. The underlying assumption is that irrespective of the effort required accessing the information needed to make an estimation has a positive opportunity cost for the gatherer of the information, an information cost. Although Buchanan (1969) noted that at the moment of the decision these types of costs should rightly be regarded by the decision maker as sunk costs, the problem still remains of how to incorporate these costs into the ex ante choice represented in Eq. (9). To take into account the transaction costs of information, Eq. (9) can be expanded with an additional cost term. In theory, since the net returns from any decision to adopt a particular agronomic measure must be positive, the returns need to cover not only the effect on farm income of adopting the measure (Eq. (9)) but also the cost of accessing and processing the information needed to make the decision to adopt, represented as an uncertain, nondiscounted cost (T) in Eq. (10). ! T n X X t ð1 þ iÞ Rdat EðAbat Þ EðTacc Þz0 t¼0
b¼1
ð10Þ The transaction cost (T acc) in Eq. (10) is the opportunity cost of obtaining and using relevant 1
Assuming a triangular distribution for each variable makes the calculation of the average a simple combination of these two estimates.
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
information for making a choice of whether or not to adopt the practice of sowing catch crops on field a. This cost can vary a great deal. What gives rise to the variation in costs for individual farmers and the need for explicitly recognizing these costs is that the conditions surrounding the choice to adopt a new agronomic practice are different not only between farmers but between fields as well. Easter (1993) writes for farmers ‘‘the differences in asset specificity across farm types mean that the transaction costs of responding to changes in policies or institutional arrangements will be quite different among farms’’ (p. 46). Asset specificity differs between fields as well. The access cost of information may be high or low, it depends on the demands placed on the information. There is a trade-off that is possible to make between reliability and access cost. For example, if information with regard to fertilization schedules for a farmer were available through a phone call to a farm adviser then this could be obtained at a low access cost but may not be reliable because of site-specific information demands. A site-specific fertilizing schedule could also be obtained by first performing soil tests on the field and then studying and comparing research results for the particular crop and soil type using alternative fertilization schedules. An information gathering method that would, in all likelihood, be associated with high access costs. However, in certain circumstances it may be possible to gain access to locally specific information at a low cost. Bingham et al. (1995) write, ‘‘it is important to identify in advance what information is relevant. . .identifying ‘decisive’ information in advance could substantially reduce information costs of public decisions (p. 84)’’. However, this identification also has a cost and implies a tradeoff between the expected value of the additional information and the cost to access it. Another method for accessing information with respect to a site specific fertilization schedule would be to assign a high degree of reliability to another farmer in the same area with what might be known to be a similar type of soil and then follow the application schedule observed on this field or to follow the schedule recommended by that particular farmer, a method which could result in a low access cost and a relatively high degree of site specificity. This method is a type of decision heuristic, a cognitive method for reducing the decision fields of complex problems to
307
make them tractable, a technique associated with rationally bounded decision makers. 2.3. Bounded rationality and the choice model with information costs In the 1950s, Simon (1956) introduced the concept of bounded rationality to describe the process by which decision makers are limited by cognitive constraints in the search for, and evaluation of, the information used in making decisions. In the neoclassical model of economic optimization, a rationally economic actor would be expected to evaluate all the consequences of the decision under consideration and to choose the alternative that maximized returns (utility). However, what Simon introduced was the idea that each actor is limited by both knowledge and computational capacity and consequently, limits the alternatives and consequences considered to make the decision tractable. This then results in a bounded rational choice rather than an optimal choice being made, a choice which is described as ‘satisficing’ rather than ‘optimizing’ from the decision maker’s perspective (Simon, 1987; Hogarth, 1987). Optimization theory disregards decision processes and cognitive limitations (Van den Bergh et al., 2000; Laville, 2000). Patrick and DeVuyst (1995) write that farmers ‘‘are not concerned with the ‘optimal’ decision in an abstract situation such as is typically assumed in research, but rather what is the best decision for them in their specific situation’’ (p. 9). Application of the theory of bounded rationality explicitly concentrates on the role of the decision maker with respect to the problem at hand, the processes that are used for decision making as well as the limitations imposed by the computational capacity of individuals.2 In the choice model described above, it was noted that a rational farmer could perform as many as 4,782,969 calculations in order to estimate the economic consequences of cultivating a catch crop on a particular field. Even if an impulse to perform this heroic task was exhibited, the computational demands placed on the decision maker would be next to 2 See Van den Bergh et al. (2000) for a discussion of the implications for environmental policy of models that assume that satisficing models based on bounded rationality rather than optimization models better explain observed individual behavior.
308
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
Fig. 1. Series of decision nodes.
impossible to satisfy without some type of decision support. In addition, even if it were feasible to somehow perform all of these operations, the associated opportunity cost (the information cost in Eq. (10)) would in all likelihood dominate all possible gains. Leading the rational farmer to conclude that the measure should either not be adopted or that some other lower cost method, with a lower expected information cost E(T) for making the decision, should be found. Heuristics are used as a method by rational actors for reducing information costs in decision making. Following Hogarth (1987), a decision may be regarded as a series of decision nodes rather than as a single decision. At each node, there are three types of decisions which are possible; to accept, to unconditionally reject or to reject but gather more information (Fig. 1). This series of nodes can be illustrated by the decision to purchase a book. The presumptive purchaser of a book would gain the most complete information to make a determination of value with respect to the contents of the book, by first reading the entire book. However, because of the high expected opportunity costs this would presumably lower the subjective value of the book to the would-be purchaser. Fortunately, the cover of the book can provide signals about the content at a low opportunity cost. The title, the name of the author, the cover illustration or graphic design may all be interpreted as signals about the content of the book.3 All of these signals 3 See Collentine (2002b) for a more detailed description of the economic rationality for ‘judging a book by its cover’.
may be regarded as decision nodes leading to one of three decisions; purchase the book, unconditionally reject the idea of purchasing the book or reject the idea but gather more information (look for a review or summary of the contents, for example). Returning to the example of the agri-environmental program described above, the decision of a farmer to adopt cultivation of catch crops, the first decision node is when the farmer becomes aware that there is a financial return associated with the measure, for example, a subsidy available for adopters. This initial information sets a series of decision nodes in motion, a series which ends only when there is an unconditional rejection of the measure or the measure is adopted, a relationship that is illustrated in Fig. 1. The number of times the decision to reject but gather more information is made determines the number of nodes in the series. Each node is associated with positive information costs, so the greater the number of nodes the higher the information costs for the decision. A quick decision, one with very few decision nodes, lowers the transaction costs (in Eq. (10)) that may, in turn, be decisive in the decision to adopt. Thus a farmer, operating within a context of bounded rationality at each decision node is interested in decision rules that are satisfactory. In this case, satisficing means finding a decision rule and the necessary information for determining whether the LHS of Eq. (10) is greater then zero. The two uncertain cost elements in Eq. (10) are the direct costs of adopting the measure, cultivation of catch crops, based on the agronomic activities (Aabt) and the transaction, or information costs (T acc) used to
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
make the decision. As a part of the Swedish catch crop subsidy program, the Swedish Board of Agriculture provides information about the program. This represents an information cost that is in part a sunk cost (low cost) to the receiver, for the purpose of making a decision at the first node. The information provided by the Board about the program includes two pieces of information that address the direct costs of adopting the measure. The measure is anticipated to be income neutral for the farmer, and the expected direct costs are estimated to be 310 –675 SEK/ha (SJV, 2001).4 Since the subsidy is 900 SEK/ha, the rational farmer knows that both pieces of information cannot be true. If the program has the aim of being truly income neutral then the expected costs are underestimated. Thus, although a quick decision to adopt may be made on the basis of the low cost information available (a sunk cost at that particular decision node), the reliability of that information may be judged to be low. Therefore, it is probable that the farmer would reject and gather more information, that is, move to the next decision node and accept the need for increasing information costs. 2.4. Decision heuristics Decision heuristics are a collective term for behaviour rules that serve to simplify choice. Because these rules apply to historic behaviour patterns, the information cost in applying them is low as the information used to generate the rule is already a sunk cost. Actual application is also possible at a relatively low cost since ‘‘one reason that heuristics work is that they can exploit structures of information in the environment’’ (Girgerenzer and Selten, 2001, p. 9). This in turn means that instead of processing information, the user of a heuristic needs only to look for a recognizable pattern in the flow of information. In the extensive work on heuristics pioneered by the psychologist team of Daniel Kahneman and Amos Tversky, three general types of decision heuristics were identified; representativeness, anchoring and availability (Kahneman et al., 1982). Persons faced with complex decisions where there is uncertainty involved with respect to the outcome of the decision intuitively use these 4 SEK, Swedish crowns. On 7/10/2003 the exchange rate was 7.75 SEK/USD.
309
decision heuristics. They serve as a method of structuring subjective probabilities associated with the possible outcomes. Representativeness is used in making judgments about the stationarity of observations, that is, whether the events associated by the user for decision support belong to the same class of events. If the person facing a decision believes that the decision at hand represents a sample of a larger population, then prior beliefs about the larger population can be applied to make a decision. For example, assume that a farmer believes based on prior experience that information from government agencies underestimates the costs to adopters of new agronomic management practices. The use of the representative heuristic may lead the farmer to choose not to adopt a new practice in the absence of other sources of information, in spite of the fact that this prior belief is based on a small sample. In this case, the heuristic provides low cost support for deciding to reject unconditionally, or to reject and gather more information, but it also reveals a bias in judgement. Unfortunately, this bias may also lead to an error in judgement. It may be that due to a belief ‘in the law of small numbers’ the farmer may have an exaggerated confidence in the conclusions based on small samples (Kahneman and Tversky, 1982). Anchoring is often used as a technique for numerical prediction when there is a value available. Estimations of uncertain values need to be based on an information set of values; this set of values then becomes the ‘anchor’ for the quantitative estimate. For example, assuming a farmer knows exactly how much time it takes to sow a particular field with a conventional sowing machine, and then this information may be used as the starting point for the same farmer’s estimate of the time it will take to sow the same field using a different sowing technique and/or equipment. The estimate is adjusted upwards, or downwards, from the known value. The estimate made in this case is sensitive to the level of this starting value (Tversky and Kahneman, 1974; Lichtenstein et al., 1978). This is the use of the anchoring heuristic. Availability refers to access to information for making estimates of the frequency of events. ‘‘There are situations in which people assess the frequency of a class or the probability of an event by the ease with which instances or occurrences can be brought to
310
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
mind’’ (Tversky and Kahneman, 1982, p. 11). For example, if a farmer has experienced one spring plowing over a 10-year period that was delayed because of the weather, the memory of the delay may be so accentuated because of the problems associated with it that the event dominates and is recalled by the farmer as occurring often. In this case, the frequency of the event, delayed spring plowing is overestimated due to the availability heuristic. These three types of heuristics are used to reduce the information costs for making decisions in the presence of uncertainty. However, use of these heuristics can also lead to errors in judgement. Errors that may arise to the use of biases to determine the reliability of the information used to support the heuristic. For example, when the representative heuristic is used, the user has determined that the sample observed is sufficiently similar to the larger population that it represents for decision making purposes. It is in the making of this determination that biases may be found which in turn lead to errors in judgement. Much of the work on the use of biases for support of heuristics focuses on experiments to study to what extent the use of these biases lead to errors in judgement (see for example Kahneman et al., 1982). Alternatively, if support is provided to increase the reliability of information used, then perhaps the importance of the bias will decrease and be replaced by more reliable means for applying the heuristic for making decisions. Providing support to be used in conjunction with the use of heuristics for decision making was one of the starting points in the design of the DSS, LENNART.
3. LENNART; a net-based dynamic DSS LENNART has been designed to support field management decisions in particular, the choice of whether or not to implement a specific field based BMP which is expected to lead to reductions in nutrient leaching. LENNART is designed to be used by individual farmers or farm advisers to explore results of modifications of farming practices, both the effect on the income of the farmer(s) and the effect on the leaching of nutrients. As noted above, there are a series of factors which affect the decision
of the farmer to implement a specific measure. These decision factors include:
field specific qualities: soil type, previous crop, drainage; farm specific qualities: crop rotation, agronomic practices, access to capital, access to information; regional specific qualities: local weather; the producer’s perception of the costs and benefits of the alternatives (subjective probabilities); the individual risk profile of the producer and sectoral risk; the dissipation of information; the rate of adoption by other producers.
The principle idea is that each user can adjust the model to reflect local conditions based on user information. This allows for flexibility in use of the model and ensures that the user is in control of the results generated by the model by ensuring that the user has control over model inputs with respect to the first three points. LENNART also provides a unique opportunity through the use of modern information techniques, to explore how decision heuristics can be actively incorporated into a decision support tool. 3.1. Exploiting heuristics Through the above analysis of how choices may be framed by farmers with respect to participation in agri-environmental programs, the model development team studied the use of three types of heuristics as decision support: representativeness, anchoring and availability. The model LENNART was designed to support the use of these heuristics by decision makers. It does this by increasing the reliability of the information used to minimize the judgement errors due to bias that may be associated with the use of these heuristics. To support the use of the representativeness heuristic, LENNART has been designed to provide access to classes of users in the database and to support the user in determining whether the chosen classes are representative for the decision being considered. Each user logging on to LENNART provides basic information about the size of their farming operation, the type of farming operation and the geographical location of the farm. In addition, for each field entered for
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
calculating the effect of a management option, information is entered on soil type and crop rotation. Furthermore, in the performance of the program calculations, the user enters additional economic and agronomic data such as the selected discount rate, estimated tractor operation costs, etc. (see Collentine, 2002a). The unique construction of LENNART makes it possible for the user to search the database based on the specific class of data designated. For example, the user who wishes to compare their own estimate of labor costs with other users’ estimates will be able to search the database and reproduce a summary of this information (see Fig. 2). If the user believed that a more narrowly defined portion of the reference group more closely represented their own operation, say farms with more than 50 ha in crops, then the database in LENNART could be restricted and this limited database is made available to the user. The possibility of defining a specific reference group to use for comparative purposes allows the user to search for subjectively defined similarities from a broadly defined population to use in comparing value estimates. Nowak et al. (1997) in an evaluation of USDA water quality demonstration projects conclude ‘‘local validation of the profitability and practicality of recommended management practices generally is needed to convince producers to try using them to a significant degree’’. In the absence other alternatives, the decision maker may be reduced to ‘looking over the neighbor’s fence’ as support for the use of the representative heuristic. The model expands the horizons of the user in a structured manner. The default values provided by LENNART are designed to serve as values for instances where anchoring is a factor. Each BMP evaluated with the
Fig. 2. Farm user estimation of labor cost per hour based on data from test-runs of LENNART.
311
model is decomposed into the field level agronomic activities and effects associated with the measure. This permits independent estimation for each activity by the user. For example, the model decomposes the cultivation of catch crops into four separate activities/effects: seed costs, sowing method, harvest effect and weed control. Each of these in turn is broken down into the choice variables that need to be estimated by the user. To estimate the cost for sowing the catch crop, the user evaluates and compares two different sowing methods. In the dialogue box for estimating the cost of a separate seeder, the user selects values from two separate scrollable lists. The first is for the purchase price of the seeder that displays a default value and a scroll arrow. By clicking on the enter key the user accepts this default value for the cost estimation. In the second prototype version of LENNART, a standard value for depreciation and capital costs are used to calculate the entered purchase price as a present value. Clicking on the scroll arrow opens the list and the user can scroll up or down from the default value to select a different value which is then used to perform the estimate. The same process is followed for all of the 14 choice values. The default values displayed in LENNART serve as anchors for the user. When moving through the dialogue boxes in LENNART, the individual user first sees the suggested default values (the anchor) in the area designated for choice values. The displayed default value gives a signal to the user of an appropriate choice for this variable. If this value is not acceptable then the user moves from this value to a new estimate that reflects additional user-based information. The netbased platform of LENNART (see below) allows the default levels to be adjusted on the server. This may be done in response to user driven information (farm size for example) or for research purposes. The model supports the use of the anchoring heuristic by recognizing that the default values displayed are a low cost signal of information to the user. Availability refers to access to information for making estimates of the frequency of events. Patrick and DeVuyst (1995) describe decision making as a learning process: ‘‘as farmers learn, they may reevaluate their situation and, when they become convinced that one alternative tends to dominate the others, the decision is made’’ (p. 8). LENNART is designed to
312
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
provide access to other users’ frequency estimates as well as expert estimates through the use of links to other sources of information. This heuristic shares similarities with the representative heuristic as they both refer to the reliability of a sample as representative of a larger population. However, availability refers primarily to the ability of the decision maker to access similar events. Users of the DSS LENNART may return to their own previous runs for comparative purposes. New estimates made by the farm user of the costs of implementation are easily compared to previous estimates as all the information is saved on the server in a database. The Internet platform also will enable the user to access other sources of information (research results, advising services, etc.) as these become available that may support frequency estimates. Lowering the cost of access to information has driven development of the Internet. By making LENNART available through the net, the cost of access is lowered and the availability of information to the user is increased. 3.2. The model: a net-based dynamic DSS The model is built on a relational database that is located on a web-server. Access to the system is performed via normal Internet browsers using plain HTML-code. The HTML-code is dynamically generated through server-side scripts. Both the system and sub-models of LENNART are maintained inside these scripts. When using the system, the user sends a request to the web-server, which processes the request and sends the result back as an HTML-page to the browser of the user. On the server, LENNART computes the economic costs for adopting catch crops on each field and generates the resulting nitrogen loss reduction on that particular field. The economic model driving these cost estimations does not need any substantial amount of computational power. Therefore, the model may be run directly on the server when the user sends a request. Model responses are produced within seconds. This short response time is, however, not the case for the calculation of nitrogen reduction. The basis for calculation of reductions in leaching losses of nitrogen is the physically based SOILNDB model (Johnsson et al., 2002). This model was developed to quantify nitrogen leaching losses from arable
land for various agri-environmental conditions. SOILNDB is an administrative program which links input data to automatic parameterization procedures for two widely used research models; the water and heat model SOIL (Jansson and Halldin, 1979) and the nitrogen model SOILN (Johnsson et al., 1987). Even though the requirements for input data for SOILNDB are more general, less detailed and less extensive than those required for the direct use of the SOIL –SOILN models, the development team decided that these demands would be too cumbersome in the DSS LENNART. Instead, an extensive number of standardized runs from the SOILNDB model were stored in a separate database.5 Nitrogen leaching for different soils, crop combinations and areas (climates) are kept in the database. Thereby the nitrogen leaching data can be sent back to the LENNART user within seconds. There are three primary factors that led to the choice of a server based web site accessed through the Internet for LENNART; access factors, development factors and data base factors. A server-based program promotes access for a wide group of intended users. Multiple users from individual computers may access the site, with the only personal computer software requirement being a standard web-navigating program (Netscape or Explorer). Enabling access to the program through individual computer connections also allows the program to be demonstrated in a variety of environments. Farm advisers can demonstrate use of the program in consultations with farmers during farm visits. The program can also be demonstrated and used by groups in seminars. Development of the model can be continuous over time as control over the version being used is determined through the server. This quality also means that no problems arise with versions being used that are out of date. Each time a user logs on the version that becomes available is determined through commands on the server. This also allows for partitioning over time to test development of model components. For example, inclusion of a wizard format or tutorial can be tested by incorporating that component into the 5 There are 1168 runs from the SOILNDB model in the prototype. These runs cover two geographic regions in Sweden, 4 – 6 soil types, 10 – 13 crops and with or without an insown catch crop where this is agronomically feasible.
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
model made available to users on the server over for a specific period of time or a specified number of runs. Results from this partitioned model can be compared and choices made by model developers with respect to incorporation or development of the most favourable components. The net-based format also allows for incorporation of changes in development of the independent natural science process based sub-model, SOILNDB. The server platform of LENNART allows changes to be made in the user available model as soon as new information becomes available which affects the results of the sub-model. The entire model does not need to be replaced, only those changes which are made to the model. This ensures that LENNART is able to make use of the best information available. The location of the model on a server also means that the database is developed as the model is used. This represents the dynamic aspect of the model. All of the data is located in one place and can be accessed from anywhere by designated users. As new data becomes available, i.e. every time the model is used, this data is directly available on the server. The immediacy of availability will be able to provide support for users that are interested in comparative data and for users that are interested in aggregate data for policy evaluation and design. Fig. 3 illustrates one of the comparative summary pages in the second
313
prototype of LENNART. This page compares the farm user’s imputed cost estimates with the cost estimates used by the Swedish Agricultural Board for calculating the economic effect of cultivation of catch crops. It is also possible to use partitioning with respect to the database. Open access to the entire database through the Internet will make it possible for those users that are interested in the model to actively work with the database for this purpose. Fig. 2 reproduces a diagram of LENNART user estimates of labor costs per hour. Statistical analysis of this kind of data is of interest for policy analysis and program evaluation. The technical platform for LENNART is a Windows environment using an Access 2000 database, Active Server Pages (ASP) with server-side Visual Basic Script and a few client-side Java Scripts. The second prototype of the model (in Swedish) is available at: http://neptunus.md.slu.se/VASTRA/BAK/index.html.
4. Conclusions and further development Agri-environmental policies will continue to include programs based on voluntary participation by farmers. Since the success of these programs depends ultimately on the rate of implementation of the policy measure by farmers, it is important to understand how
Fig. 3. Comparative summary page in LENNART for farm users (in Swedish).
314
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
farmers make management choices. The uncertainty with respect to costs that necessarily accompanies the adoption of new techniques leads to a need for supporting information to analyze the economic effect of adoption on farm income. Farmers, like the rest of us, are limited by their capacity to store and process information. As decision makers, they operate in a realm of bounded rationality when faced with choices and rather than optimizing over the set of decision alternatives may instead satisfice. Decisions may be perceived as a series of sequential decisions, a set of decision nodes. At each node, one of three actions is possible; accept (enroll in the program), reject unconditionally or reject and gather more information. The last of these three choices includes the expected costs of the additional information. Since these costs are positive and the revenues may be constant, this results in a paradox. If the value of the information gathered is lower than the cost of accessing and processing this information, the likelihood of participation is lower the more well informed the decision is. Thus, quick decisions based on simple decision rules may be effective. The decision support system LENNART was designed to assist farmers with the evaluation of the effects of implementing agronomic measures to reduce the leaching of nutrients from cultivated land. The model currently allows users to evaluate the economic effect and the expected reduction in nitrogen leaching from a set of crop rotations in a specific area of Southern Sweden. Expansion is planned to cover both a larger geographic area as well as to include a greater number of soil types and crop rotations. In addition, the development team plans to improve the graphic interface through the use of focus group tests. Preliminary work is also underway to allow the evaluation of other field management measures in LENNART. New measures planned for development in extensions of the model include the reduction of fertilization intensity on fields and measures where timeliness is a factor such as the timing of cultivation in combination with other practices and the timing of fertilizer applications. Since measures where timeliness is a factor may have an effect on other farm activities beyond the individual field, a preliminary study of these types of measures is necessary before they may be included in LENNART.
Acknowledgements This study was performed within the Swedish Water Management Research Program (VASTRA) and financed by the Swedish Foundation for Strategic Environmental Research (MISTRA). References Amacher, G.S., Feather, P.M., 1997. Testing producer perceptions of jointly beneficial best management practices for improved water quality. Applied Economics 29 (2), 153 – 159. Bingham, G., Bishop, R., Brody, M., Bromley, D., Clark, E., Cooper, W., Costanza, R., Hale, T., Hayden, G., Kellert, S., Norgaard, R., Norton, B., Payne, J., Russell, C., Suter, G., 1995. Issues in ecosystem valuation: improving information for decision making. Ecological Economics 14, 73 – 90. Buchanan, J.M., 1969. Cost and Choice: An Inquiry in Economic Theory. The University of Chicago Press, Chicago, IL. Buller, H., 1999. Evaluating participation in agri-environmental schemes. In: Beihl, D., Schramek, J., Buller, H. Wilson, G. (Eds.), The Implementation and Effectiveness of Agri-Environmental Schemes after Reluation 2078. Final Report to the DGVI of the European Commission (Contract FAIR CT95274, Frankfurt). Carpenter, S.R., Caraco, N.F., Correll, D.L., Howarth, R.W., Sharpley, A.N., Smith, V.H., 1998. Nonpoint pollution of surface waters with phosphorus and nitrogen. Ecological Applications 8 (3), 559 – 568. Collentine, D., 2002a. Economic modelling of best management practices (BMPs) at the farm level. In: Steenvoorden, J., Claessen, F., Willems, J. (Eds.), Agricultural Effects on Ground and Surface Waters, vol. 273. IAHS Press, Oxfordshire, UK, pp. 17 – 22. Collentine, D., 2002b. Book Review. In: Tietenburg, T., Folmer, H. (Eds.), The International Yearbook of Environmental and Resource Economics 2000/2001: A Survey of Current Issues, Edward Elgar Publishing, UK. Environmental and Resource Economics, vol. 23, pp. 467 – 472. Cooper, J.C., Keim, R.W., 1996. Incentive payments to encourage farmer adoption of water quality protection practices. American Journal of Agricultural Economics 78 (4), 54 – 64. Drake, L., Bergstro¨m, P., Svedsa¨ter, H., 1999. Farmers’ attitudes and uptake. In: Van Huylenbroeck, G., Whitby, M. (Eds.), Countryside Stewardship: Farmers, Policies and Markets. Elsevier, Oxford, UK, pp. 89 – 112. Easter, K., 1993. Differences in the transaction costs of strategies to control agricultural offsite and undersite damages. In: Russell, C., Shogren, J. (Eds.), Theory, Modeling and Experience in the Management of Nopoint-Source Pollution. Kluwer Academic Publishing, London, pp. 37 – 68. Feather, P.M., Amacher, G.S., 1994. Role of information in the adoption of best management practices for water quality improvement. Agricultural Economics 11, 159 – 170. Girgerenzer, G., Selten, R., 2001. Rethinking rationality. In: Gir-
D. Collentine et al. / Ecological Economics 49 (2004) 303–315 gerenzer, G., Selten, R. (Eds.), Bounded Rationality: The Adaptive Toolbox. MIT Press, Cambridge, MA, pp. 1 – 12. Gustafson, A., Fleischer, S., Joelsson, A., 1998. Decreased leaching and increased retention potential co-operative measures to reduce diffuse nitrogen load on a watershed level. Water Science and Technology 38 (10), 181 – 189. Hogarth, R., 1987. Judgement and Choice. Wiley, New York. 311 pp. Jansson, P.-E., Halldin, S., 1979. Model for annual water and energy flow in a layered soil. In: Halldin, S. (Ed.), Comparison of Forest Water and Energy Exchange Models. International Society for Ecological Modelling, Copenhagen, pp. 145 – 163. Johnsson, H., Bergstro¨m, L., Jansson, P.-E., Paustian, K., 1987. Simulated nitrogen dynamics and losses in a layered agricultural soil. Agriculture, Ecosystems and Environment 18, 333 – 356. Johnsson, H., Larsson, M., Ma˚rtensson, K., Hoffman, M., 2002. SOILNDB: a decision support tool for assessing nitrogen leaching losses from arable land. Environmental Modelling & Software 17, 505 – 517. Just, R.E., Buss, A., Donoso, G., 1991. The significance of the interface of agricultural and resource policy: conclusions and directions for further research. In: Just, R.E., Bockstael, N. (Eds.), Commodity and Resource Policies in Agricultural Systems. Springer Verlag, Berlin, pp. 293 – 310. Kahneman, D., Tversky, A., 1982. Subjective probability: a judgment of representativeness. In: Kahneman, D., Slovic, P., Tversky, A. (Eds.), Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK, pp. 32 – 47. Kahneman, D., Slovic, P., Tversky, A., 1982. Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK. Laville, F., 2000. Foundations of procedural rationality: cognitive limits and decision processes. Economics and Philosophy 16, 117 – 138. Leathers, H.D., 1991. Best management practices versus socially optimal practices. In: Just, R.E., Bockstael, N. (Eds.), Commodity and Resource Policies in Agricultural Systems. SpringerVerlag, Berlin, pp. 293 – 310. Lichtenstein, S., Slovic, P., Fischoff, B., Layman, M., Combs, B., 1978. Judged frequency of lethal events. Journal of Experimental Psychology. Human Learning and Memory 4, 551 – 578. Napier, T., Tucker, M., 2001. Use of soil and water protection practices among farmers in three midwest watersheds. Environmental Management 27 (2), 269 – 279.
315
Nowak, P., O’Keefe, G., Bennett, C., Anderson, S., Trumbo, C., 1997. Executive summary: communication and adoption evaluation of USDA water quality demonstration projects. Plant and Animal Systems. CSREES/USDA, Washington, DC. Patrick, G.F., DeVuyst, E.A., 1995. Whence and Whither in farm management risk research and extension delivery? Canadian Journal of Agricultural Economics 43, 1 – 14. Setia, P., Magleby, R., 1987. An economic analysis of agricultural nonpoint pollution control alternatives. Journal of Soil and Water Conservation, November-December, 427 – 431. Sharp, B.M.H., Bromley, D.W., 1979. Agricultural pollution: the economics of coordination. American Journal of Agricultural Economics 61 (4), 591 – 600. Simon, H.A., 1956. Rational choice and the structure of the environment. Psychological Review 63, 129 – 138. Simon, H.A., 1987. Bounded rationality. In: Eatwall, J., Milgate, M., Newman, P. (Eds.), The New Palgrave A Dictionary of Economics. MacMillan, London, pp. 266 – 268. SJV, 2001. Tips for Sowing Catch Crops, Greppa Na¨ringen, Brochure (Tips info¨r fa˚nggro¨desa˚dden (in Swedish)). Swedish Board of Agriculture, Stockholm, Sweden. Supalla, R.J., Selley, R.A., Bredeweg, S., Watts, D., 1995. Adoption of nitrogen and water management practices to improve water quality. Journal of Soil and Water Conservation 50 (1), 77 – 82. Tucker, M., Napier, T.L., 2002. Preferred sources and channels of soil and water conservation information in three midwestern US watersheds. Agriculture, Ecosystems & Environment 92 (2 – 3), 297 – 313. Tversky, A., Kahneman, D., 1974. Judgment under uncertainty: heuristics and biases. Science 185, 1124 – 1131. Tversky, A., Kahneman, D., 1982. Judgment under uncertainty: heuristics and biases. In: Kahneman, D., Slovic, P., Tversky, A. (Eds.), Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK, pp. 3 – 22. van den Bergh, C.J.M., Ferrer-i-Carbonell, A., Munda, G., 2000. Alternative models of individual behaviour and implications for environmental policy. Ecological Economics 32, 43 – 61. Whitby, M., 2000. Challenges and options for the UK Agri-Environment: presidential address. Journal of Agricultural Economics 51 (3), 317 – 332. Wolf, A.T., 1995. Rural nonpoint source pollution control in Wisconsin: the limits of a voluntary program. Water Resources Bulletin 31 (6), 1009 – 1022.
METHODS
Exploiting decision heuristics and IT in the design of a DSS for voluntary agri-environmental programs Dennis Collentine a,*, Martin Larsson b, Nils Hannerz c a b
Department of Economics, Swedish University of Agricultural Sciences, Box 7013, S-750 07 Uppsala, Sweden Department of Soil Sciences, Swedish University of Agricultural Sciences, Box 7072, S-750 07 Uppsala, Sweden c Swedish Institute of Agricultural and Environmental Engineering-JTI Box 7033, SE-750 07 Uppsala, Sweden Received 15 October 2002; received in revised form 28 October 2003; accepted 1 February 2004 Available online 19 June 2004
Abstract Low participation rates by farmers in voluntary agri-environmental programs may depend on rationally bounded ex ante estimates of the negative effect of program enrollment on farm income. Uncertainty and the presence of information transaction costs may lead to the use of heuristics by farmers to reduce adoption decision costs. This paper describes how LENNART, a netbased decision support system (DSS), has been designed to exploit the use of heuristics and provide low cost access to information. The model has been developed to evaluate the effects of agronomic measures on farm income and on the leaching of nutrients from cultivated fields. A subsidy program for catch crop cultivation in Southern Sweden served as the basis for development of the DSS and is used throughout the paper for purposes of illustration. D 2004 Elsevier B.V. All rights reserved. Keywords: Agri-environmental policy; BMP implementation; Farm management; Heuristics; SOILNDB
1. Introduction Agronomic practices that contribute to nitrogen leaching are often connected with field cultivation practices. Changes in agronomic practices, so-called best management practices (BMPs), have been identified which if implemented could reduce the level of nitrogen leaching (Carpenter et al., 1998; Gustafson et al., 1998; Feather and Amacher, 1994; Sharp and Bromley, 1979). Implementation of BMPs by farmers * Corresponding author. Tel.: +46-18-67-17-39; fax: +46-1867-35-02. E-mail addresses: [email protected] (D. Collentine), [email protected] (M. Larsson), [email protected] (N. Hannerz). 0921-8009/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolecon.2004.02.002
is generally voluntary and encouraged by support from extension services or other government programs. Unfortunately, these programs have not achieved expected results (Collentine, 2002a; Gustafson et al., 1998; Wolf, 1995; Just et al., 1991; Setia and Magleby, 1987). The success of agri-environmental policy dedicated to the adoption of BMPs by farmers and the cost effectiveness of these policies can be enhanced through an understanding of the factors that determine how farmers make choices with regard to implementation (i.e. when to adopt and which measures to adopt). While farmers attitudes to adoption have been studied, these studies have primarily been surveys to analyze the socioeconomic factors which characterize the adopters and nonadopters (Tucker and Napier,
304
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
2002; Napier and Tucker, 2001; Drake et al., 1999; Buller, 1999; Cooper and Keim, 1996; Amacher and Feather, 1997; Supalla et al., 1995; Feather and Amacher, 1994). Participation in agri-environmental programs that support the adoption of BMPs is primarily an economic decision. If the perceived effect on farm income of a BMP is income neutral or positive then the measure is more likely to be adopted. Buller (1999) observed in a report on the implementation and effectiveness of agri-environmental schemes, that because ‘‘Agri-environmental policy occupies an illdefined middle ground between regulatory approaches to environmental management . . .and more classic generalised market instruments . . . [that] agri-environmental policy critically needs to be placed at the level of the farmer and the farm’’ (pp.105 – 106). That is, successful programs begin with an understanding of how management choices are made by farmers on their farms. Uncertainty with respect to costs associated with the implementation of new field management techniques leads to a need for estimation of the economic effect of adoption on farm income. However, ‘‘since costs and effects of a given practice will differ from farm to farm, the amount of information needed is substantial’’ (Leathers, 1991, p.308). The act of gathering relevant information and then using it for the purpose of decision making, involves costs to the decision maker. These can be classified as a type of transaction cost. That is, costs that are necessary to complete a transaction or in the case of agri-environmental policy to participate in the program. Whitby (2000) notes that with respect to participants in agrienvironmental programs ‘‘they have to incur some level of transaction cost to participate in schemes and these are not generally reimbursed specifically . . . [and that] altruism apart, society must also bear the private transaction costs of contracts’’ (p.322). If the decision making process itself was better understood, then information flows in the form of a decision support system (DSS) could be developed which could reduce the transaction costs of the decisions by farmers to adopt specific measures as well as support authorities in the design, implementation and evaluation of agri-environmental policy. Simon (1987) observed that since decision makers are limited by their capacity to store and process information, they operate in a realm of bounded
rationality when faced with choices. Patrick and DeVuyst (1995) succinctly conclude in their survey on risk research in farm management, ‘‘producers are unwilling to pay more for the information than it is perceived to be worth ...[therefore] efforts should focus on both reducing the cost of risk-related information and increasing its perceived value’’ (p.10). Heuristics are a type of simple decision rule that lowers the cost of accessing and processing information. Information technology (IT) also substantially lowers the cost of computing operations and the storage of data. In addition, the Internet offers the possibility of low cost access to digitalized information. The DSS described in this paper was designed to exploit the use of heuristics by farmers and use the capabilities of IT and the Internet to provide support for the evaluation of the economic and environmental effects of BMPs on individual fields. The first section below analyzes how an individual decision maker, a farmer, searches for and processes information when faced with complex decisions. This section concludes with a description of how heuristics may be used by farmers to reduce the costs of adoption decisions. The next section describes the net-based interactive DSS called LENNART. The paper ends with conclusions and a description of the planned extensions of the model.
2. Farm level implementation decisions 2.1. A simple model of choice The choice by a farmer to implement a new BMP is driven by the flows of returns and benefits associated with the practice relative to the flows and benefits of present practices. The condition for adoption is represented in Eq. (1). The LHS 1 represents the net return (R C) from adopting the BMP of an insown catch crop (cc) on a particular field (a) as a flow over time (t) at a particular expected rate of return (i) and any noneconomic values (M) associated with adoption or nonadoption of the practice. The RHS of Eq. (1) represents the returns from current use (n) of the same field without the BMP. It should be noted that the farmer is already assumed to be profit maximizing using current practices and that the net returns are sufficient to continue with current practices indefi-
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
305
nitely. In the absence of transaction costs, the choice facing the farmer in this case is simple, if the net returns from implementing the BMP are greater than those from current practices, then the practice is adopted.
associated with implementing catch crops as a cultivation practice. If the LHS in Eq. (5) is positive then the decision maker will adopt the measure.
T X
t¼0
cc cc ð1 þ iÞt f ðRcc at Cat ; Mat Þ
t¼0
zð1 þ iÞt f ðRnat Catn ; Matn Þ
ð1Þ
If all of the benefits and costs to farmers are subsumed in the economic returns then there are no externalities considered in the decision to adopt a particular agronomic management practice. In this case, the M term disappears and the decision is simply an evaluation of the difference between the returns (R) and the costs (C) of the two alternatives. In addition, since the timing of flows associated with an insown catch crop corresponds to the timing of flows from sowing the same main crop without a catch crop, then changes in this parameter are highly correlated. Combining both of these assumptions allows Eq. (1) to be rewritten as just the incremental difference between the two practices as in Eq. (2).
T X ð1 þ iÞt ðRdat Catd Þz0
ð5Þ
d ) is composed of two The change in revenues (Rat parts; the areal subsidy for implementing the BMP and the change in income due to changes in the yield of the main crop. The second term can be included as a cost effect, where a fall in yield is included as a production cost and an increase in yield is treated as a production gain. While long-term soil fertility effects are not taken into account explicitly, these could be incorporated in the same manner as the effect on yield. The choice represented in Eq. (5) then is reduced to the present value of the difference between the payment received for planting the catch crop and a linear sum of the field production costs associated with the practice. Following Collentine (2002a), these costs can be expressed as:
Catd ¼
n X
ð6Þ
Abat
b¼1 T X n cc n ð1 þ iÞt ððRcc at Rat Þ ðCat Cat ÞÞz0
ð2Þ
t¼0
Furthermore, if the decision is based only on a comparison of the returns from the two alternative field management practices then farm management choices such as main crop choice and other whole farm decisions are taken as givens based on fixed assets. Thus, the revenue from choosing to plant a catch crop (Rcc) and the costs associated with this practice (Ccc) can be expressed as linear sums of the current practice and the additions in revenue (Eq. (3)) and costs (Eq. (4)) associated with the catch crop. d n Rcc at ¼ Rat þ Rat
ð3Þ
Catcc ¼ Catd þ Catn
ð4Þ
Where each agronomic activity (Ab) is assumed to be linear and independent, these describe a set of values (costs) at the farm and field levels. Substituting this linear cost function into Eq. (5) results in Eq. (7), where the costs and returns associated with the planting of a catch crop are those that each farmer needs to evaluate for each field under consideration. ! T n X X t d ð1 þ iÞ Rat Abat z0 ð7Þ t¼0
Moving from this framework into one where uncertainty enters into the evaluation is possible to model by using the expected value of the costs and returns as seen in Eq. (8). ! T n X X t d ð1 þ iÞ E Rat Abat z0 ð8Þ t¼0
Substituting Eqs. (3) and (4) into Eq. (2) allows the decision to be written as simply an evaluation of the additional revenues (Ratd) and additional costs (Catd)
b¼1
b¼1
Since the returns are a constant subsidy paid out for every hectare enrolled in the program and any yield
306
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
effects are subsumed into production costs, the expected value of the returns is the same as the deterministic value and the only uncertainty that remains is in the cost term, Eq. (9).
for each of the variables must be made.1 In addition, the costs of accessing the information needed to make these estimates also need to be included in the choice model.
! T n X X t d ð1 þ iÞ Rat EðAbat Þ z0
2.2. Choice model with information costs
t¼0
ð9Þ
b¼1
Costs evaluated at the field level for each agronomic activity or effect (Ab) become the determining factors for the decision by individual farmers to enroll acreage in the catch crop program. For catch crops, Collentine (2002a) identifies four agronomic activities and effects with a total of 14 choice variables at the farm and field levels. If the individual costs represented by the choice variables were known, then performing the calculation in Eq. (9) is a relatively simple procedure which could be done on a handheld calculator without a great deal of effort. However, since the choice to implement the use of catch crops on the farm is based on the assumption that the technique is not already being used, it is reasonable to assume that there is uncertainty with respect to the expected values for each type of cost. A risk adverse farmer evaluating the decision would be expected to perform some type of ‘what if’ (sensitivity) analysis. What if the expected cost variable was lower/higher than average, how might this affect the cost of the agronomic activity and in turn, the decision to adopt the measure? If a farmer were to choose only three values for each of these (low, normal and high) to use in calculating the variance of the expected costs, the total number of cost combinations to evaluate in any one time period in Eq. (9) are equal to 314, for a total of 4,782,969 possible cost combinations! It would also be possible for the farmer to choose to analyze costs from the perspective of a worst, average and best case scenario. Even though in this type of analysis there are only three calculations to perform, an assumption needs to be that all the variables are perfectly positively correlated. An assumption that if not made ad hoc would require an even greater number of calculations than for the first type of analysis above. Furthermore, before solving the total costs for each combination of costs during one time period is even possible for the first method, high and low estimates
As described above, there is a need to make estimates of the cost distribution for each of the 14 choice variables in order to make the adoption decision. In some cases, cost estimation may only entail transferring a cost from another activity already practiced on the farm to the practice currently being evaluated; for example, a discount rate or tractor costs. In other cases, gathering the information may require more effort, such as calling around to seed dealers to establish the price of the seed to be used and then adding the cost of transportation to this. The underlying assumption is that irrespective of the effort required accessing the information needed to make an estimation has a positive opportunity cost for the gatherer of the information, an information cost. Although Buchanan (1969) noted that at the moment of the decision these types of costs should rightly be regarded by the decision maker as sunk costs, the problem still remains of how to incorporate these costs into the ex ante choice represented in Eq. (9). To take into account the transaction costs of information, Eq. (9) can be expanded with an additional cost term. In theory, since the net returns from any decision to adopt a particular agronomic measure must be positive, the returns need to cover not only the effect on farm income of adopting the measure (Eq. (9)) but also the cost of accessing and processing the information needed to make the decision to adopt, represented as an uncertain, nondiscounted cost (T) in Eq. (10). ! T n X X t ð1 þ iÞ Rdat EðAbat Þ EðTacc Þz0 t¼0
b¼1
ð10Þ The transaction cost (T acc) in Eq. (10) is the opportunity cost of obtaining and using relevant 1
Assuming a triangular distribution for each variable makes the calculation of the average a simple combination of these two estimates.
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
information for making a choice of whether or not to adopt the practice of sowing catch crops on field a. This cost can vary a great deal. What gives rise to the variation in costs for individual farmers and the need for explicitly recognizing these costs is that the conditions surrounding the choice to adopt a new agronomic practice are different not only between farmers but between fields as well. Easter (1993) writes for farmers ‘‘the differences in asset specificity across farm types mean that the transaction costs of responding to changes in policies or institutional arrangements will be quite different among farms’’ (p. 46). Asset specificity differs between fields as well. The access cost of information may be high or low, it depends on the demands placed on the information. There is a trade-off that is possible to make between reliability and access cost. For example, if information with regard to fertilization schedules for a farmer were available through a phone call to a farm adviser then this could be obtained at a low access cost but may not be reliable because of site-specific information demands. A site-specific fertilizing schedule could also be obtained by first performing soil tests on the field and then studying and comparing research results for the particular crop and soil type using alternative fertilization schedules. An information gathering method that would, in all likelihood, be associated with high access costs. However, in certain circumstances it may be possible to gain access to locally specific information at a low cost. Bingham et al. (1995) write, ‘‘it is important to identify in advance what information is relevant. . .identifying ‘decisive’ information in advance could substantially reduce information costs of public decisions (p. 84)’’. However, this identification also has a cost and implies a tradeoff between the expected value of the additional information and the cost to access it. Another method for accessing information with respect to a site specific fertilization schedule would be to assign a high degree of reliability to another farmer in the same area with what might be known to be a similar type of soil and then follow the application schedule observed on this field or to follow the schedule recommended by that particular farmer, a method which could result in a low access cost and a relatively high degree of site specificity. This method is a type of decision heuristic, a cognitive method for reducing the decision fields of complex problems to
307
make them tractable, a technique associated with rationally bounded decision makers. 2.3. Bounded rationality and the choice model with information costs In the 1950s, Simon (1956) introduced the concept of bounded rationality to describe the process by which decision makers are limited by cognitive constraints in the search for, and evaluation of, the information used in making decisions. In the neoclassical model of economic optimization, a rationally economic actor would be expected to evaluate all the consequences of the decision under consideration and to choose the alternative that maximized returns (utility). However, what Simon introduced was the idea that each actor is limited by both knowledge and computational capacity and consequently, limits the alternatives and consequences considered to make the decision tractable. This then results in a bounded rational choice rather than an optimal choice being made, a choice which is described as ‘satisficing’ rather than ‘optimizing’ from the decision maker’s perspective (Simon, 1987; Hogarth, 1987). Optimization theory disregards decision processes and cognitive limitations (Van den Bergh et al., 2000; Laville, 2000). Patrick and DeVuyst (1995) write that farmers ‘‘are not concerned with the ‘optimal’ decision in an abstract situation such as is typically assumed in research, but rather what is the best decision for them in their specific situation’’ (p. 9). Application of the theory of bounded rationality explicitly concentrates on the role of the decision maker with respect to the problem at hand, the processes that are used for decision making as well as the limitations imposed by the computational capacity of individuals.2 In the choice model described above, it was noted that a rational farmer could perform as many as 4,782,969 calculations in order to estimate the economic consequences of cultivating a catch crop on a particular field. Even if an impulse to perform this heroic task was exhibited, the computational demands placed on the decision maker would be next to 2 See Van den Bergh et al. (2000) for a discussion of the implications for environmental policy of models that assume that satisficing models based on bounded rationality rather than optimization models better explain observed individual behavior.
308
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
Fig. 1. Series of decision nodes.
impossible to satisfy without some type of decision support. In addition, even if it were feasible to somehow perform all of these operations, the associated opportunity cost (the information cost in Eq. (10)) would in all likelihood dominate all possible gains. Leading the rational farmer to conclude that the measure should either not be adopted or that some other lower cost method, with a lower expected information cost E(T) for making the decision, should be found. Heuristics are used as a method by rational actors for reducing information costs in decision making. Following Hogarth (1987), a decision may be regarded as a series of decision nodes rather than as a single decision. At each node, there are three types of decisions which are possible; to accept, to unconditionally reject or to reject but gather more information (Fig. 1). This series of nodes can be illustrated by the decision to purchase a book. The presumptive purchaser of a book would gain the most complete information to make a determination of value with respect to the contents of the book, by first reading the entire book. However, because of the high expected opportunity costs this would presumably lower the subjective value of the book to the would-be purchaser. Fortunately, the cover of the book can provide signals about the content at a low opportunity cost. The title, the name of the author, the cover illustration or graphic design may all be interpreted as signals about the content of the book.3 All of these signals 3 See Collentine (2002b) for a more detailed description of the economic rationality for ‘judging a book by its cover’.
may be regarded as decision nodes leading to one of three decisions; purchase the book, unconditionally reject the idea of purchasing the book or reject the idea but gather more information (look for a review or summary of the contents, for example). Returning to the example of the agri-environmental program described above, the decision of a farmer to adopt cultivation of catch crops, the first decision node is when the farmer becomes aware that there is a financial return associated with the measure, for example, a subsidy available for adopters. This initial information sets a series of decision nodes in motion, a series which ends only when there is an unconditional rejection of the measure or the measure is adopted, a relationship that is illustrated in Fig. 1. The number of times the decision to reject but gather more information is made determines the number of nodes in the series. Each node is associated with positive information costs, so the greater the number of nodes the higher the information costs for the decision. A quick decision, one with very few decision nodes, lowers the transaction costs (in Eq. (10)) that may, in turn, be decisive in the decision to adopt. Thus a farmer, operating within a context of bounded rationality at each decision node is interested in decision rules that are satisfactory. In this case, satisficing means finding a decision rule and the necessary information for determining whether the LHS of Eq. (10) is greater then zero. The two uncertain cost elements in Eq. (10) are the direct costs of adopting the measure, cultivation of catch crops, based on the agronomic activities (Aabt) and the transaction, or information costs (T acc) used to
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
make the decision. As a part of the Swedish catch crop subsidy program, the Swedish Board of Agriculture provides information about the program. This represents an information cost that is in part a sunk cost (low cost) to the receiver, for the purpose of making a decision at the first node. The information provided by the Board about the program includes two pieces of information that address the direct costs of adopting the measure. The measure is anticipated to be income neutral for the farmer, and the expected direct costs are estimated to be 310 –675 SEK/ha (SJV, 2001).4 Since the subsidy is 900 SEK/ha, the rational farmer knows that both pieces of information cannot be true. If the program has the aim of being truly income neutral then the expected costs are underestimated. Thus, although a quick decision to adopt may be made on the basis of the low cost information available (a sunk cost at that particular decision node), the reliability of that information may be judged to be low. Therefore, it is probable that the farmer would reject and gather more information, that is, move to the next decision node and accept the need for increasing information costs. 2.4. Decision heuristics Decision heuristics are a collective term for behaviour rules that serve to simplify choice. Because these rules apply to historic behaviour patterns, the information cost in applying them is low as the information used to generate the rule is already a sunk cost. Actual application is also possible at a relatively low cost since ‘‘one reason that heuristics work is that they can exploit structures of information in the environment’’ (Girgerenzer and Selten, 2001, p. 9). This in turn means that instead of processing information, the user of a heuristic needs only to look for a recognizable pattern in the flow of information. In the extensive work on heuristics pioneered by the psychologist team of Daniel Kahneman and Amos Tversky, three general types of decision heuristics were identified; representativeness, anchoring and availability (Kahneman et al., 1982). Persons faced with complex decisions where there is uncertainty involved with respect to the outcome of the decision intuitively use these 4 SEK, Swedish crowns. On 7/10/2003 the exchange rate was 7.75 SEK/USD.
309
decision heuristics. They serve as a method of structuring subjective probabilities associated with the possible outcomes. Representativeness is used in making judgments about the stationarity of observations, that is, whether the events associated by the user for decision support belong to the same class of events. If the person facing a decision believes that the decision at hand represents a sample of a larger population, then prior beliefs about the larger population can be applied to make a decision. For example, assume that a farmer believes based on prior experience that information from government agencies underestimates the costs to adopters of new agronomic management practices. The use of the representative heuristic may lead the farmer to choose not to adopt a new practice in the absence of other sources of information, in spite of the fact that this prior belief is based on a small sample. In this case, the heuristic provides low cost support for deciding to reject unconditionally, or to reject and gather more information, but it also reveals a bias in judgement. Unfortunately, this bias may also lead to an error in judgement. It may be that due to a belief ‘in the law of small numbers’ the farmer may have an exaggerated confidence in the conclusions based on small samples (Kahneman and Tversky, 1982). Anchoring is often used as a technique for numerical prediction when there is a value available. Estimations of uncertain values need to be based on an information set of values; this set of values then becomes the ‘anchor’ for the quantitative estimate. For example, assuming a farmer knows exactly how much time it takes to sow a particular field with a conventional sowing machine, and then this information may be used as the starting point for the same farmer’s estimate of the time it will take to sow the same field using a different sowing technique and/or equipment. The estimate is adjusted upwards, or downwards, from the known value. The estimate made in this case is sensitive to the level of this starting value (Tversky and Kahneman, 1974; Lichtenstein et al., 1978). This is the use of the anchoring heuristic. Availability refers to access to information for making estimates of the frequency of events. ‘‘There are situations in which people assess the frequency of a class or the probability of an event by the ease with which instances or occurrences can be brought to
310
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
mind’’ (Tversky and Kahneman, 1982, p. 11). For example, if a farmer has experienced one spring plowing over a 10-year period that was delayed because of the weather, the memory of the delay may be so accentuated because of the problems associated with it that the event dominates and is recalled by the farmer as occurring often. In this case, the frequency of the event, delayed spring plowing is overestimated due to the availability heuristic. These three types of heuristics are used to reduce the information costs for making decisions in the presence of uncertainty. However, use of these heuristics can also lead to errors in judgement. Errors that may arise to the use of biases to determine the reliability of the information used to support the heuristic. For example, when the representative heuristic is used, the user has determined that the sample observed is sufficiently similar to the larger population that it represents for decision making purposes. It is in the making of this determination that biases may be found which in turn lead to errors in judgement. Much of the work on the use of biases for support of heuristics focuses on experiments to study to what extent the use of these biases lead to errors in judgement (see for example Kahneman et al., 1982). Alternatively, if support is provided to increase the reliability of information used, then perhaps the importance of the bias will decrease and be replaced by more reliable means for applying the heuristic for making decisions. Providing support to be used in conjunction with the use of heuristics for decision making was one of the starting points in the design of the DSS, LENNART.
3. LENNART; a net-based dynamic DSS LENNART has been designed to support field management decisions in particular, the choice of whether or not to implement a specific field based BMP which is expected to lead to reductions in nutrient leaching. LENNART is designed to be used by individual farmers or farm advisers to explore results of modifications of farming practices, both the effect on the income of the farmer(s) and the effect on the leaching of nutrients. As noted above, there are a series of factors which affect the decision
of the farmer to implement a specific measure. These decision factors include:
field specific qualities: soil type, previous crop, drainage; farm specific qualities: crop rotation, agronomic practices, access to capital, access to information; regional specific qualities: local weather; the producer’s perception of the costs and benefits of the alternatives (subjective probabilities); the individual risk profile of the producer and sectoral risk; the dissipation of information; the rate of adoption by other producers.
The principle idea is that each user can adjust the model to reflect local conditions based on user information. This allows for flexibility in use of the model and ensures that the user is in control of the results generated by the model by ensuring that the user has control over model inputs with respect to the first three points. LENNART also provides a unique opportunity through the use of modern information techniques, to explore how decision heuristics can be actively incorporated into a decision support tool. 3.1. Exploiting heuristics Through the above analysis of how choices may be framed by farmers with respect to participation in agri-environmental programs, the model development team studied the use of three types of heuristics as decision support: representativeness, anchoring and availability. The model LENNART was designed to support the use of these heuristics by decision makers. It does this by increasing the reliability of the information used to minimize the judgement errors due to bias that may be associated with the use of these heuristics. To support the use of the representativeness heuristic, LENNART has been designed to provide access to classes of users in the database and to support the user in determining whether the chosen classes are representative for the decision being considered. Each user logging on to LENNART provides basic information about the size of their farming operation, the type of farming operation and the geographical location of the farm. In addition, for each field entered for
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
calculating the effect of a management option, information is entered on soil type and crop rotation. Furthermore, in the performance of the program calculations, the user enters additional economic and agronomic data such as the selected discount rate, estimated tractor operation costs, etc. (see Collentine, 2002a). The unique construction of LENNART makes it possible for the user to search the database based on the specific class of data designated. For example, the user who wishes to compare their own estimate of labor costs with other users’ estimates will be able to search the database and reproduce a summary of this information (see Fig. 2). If the user believed that a more narrowly defined portion of the reference group more closely represented their own operation, say farms with more than 50 ha in crops, then the database in LENNART could be restricted and this limited database is made available to the user. The possibility of defining a specific reference group to use for comparative purposes allows the user to search for subjectively defined similarities from a broadly defined population to use in comparing value estimates. Nowak et al. (1997) in an evaluation of USDA water quality demonstration projects conclude ‘‘local validation of the profitability and practicality of recommended management practices generally is needed to convince producers to try using them to a significant degree’’. In the absence other alternatives, the decision maker may be reduced to ‘looking over the neighbor’s fence’ as support for the use of the representative heuristic. The model expands the horizons of the user in a structured manner. The default values provided by LENNART are designed to serve as values for instances where anchoring is a factor. Each BMP evaluated with the
Fig. 2. Farm user estimation of labor cost per hour based on data from test-runs of LENNART.
311
model is decomposed into the field level agronomic activities and effects associated with the measure. This permits independent estimation for each activity by the user. For example, the model decomposes the cultivation of catch crops into four separate activities/effects: seed costs, sowing method, harvest effect and weed control. Each of these in turn is broken down into the choice variables that need to be estimated by the user. To estimate the cost for sowing the catch crop, the user evaluates and compares two different sowing methods. In the dialogue box for estimating the cost of a separate seeder, the user selects values from two separate scrollable lists. The first is for the purchase price of the seeder that displays a default value and a scroll arrow. By clicking on the enter key the user accepts this default value for the cost estimation. In the second prototype version of LENNART, a standard value for depreciation and capital costs are used to calculate the entered purchase price as a present value. Clicking on the scroll arrow opens the list and the user can scroll up or down from the default value to select a different value which is then used to perform the estimate. The same process is followed for all of the 14 choice values. The default values displayed in LENNART serve as anchors for the user. When moving through the dialogue boxes in LENNART, the individual user first sees the suggested default values (the anchor) in the area designated for choice values. The displayed default value gives a signal to the user of an appropriate choice for this variable. If this value is not acceptable then the user moves from this value to a new estimate that reflects additional user-based information. The netbased platform of LENNART (see below) allows the default levels to be adjusted on the server. This may be done in response to user driven information (farm size for example) or for research purposes. The model supports the use of the anchoring heuristic by recognizing that the default values displayed are a low cost signal of information to the user. Availability refers to access to information for making estimates of the frequency of events. Patrick and DeVuyst (1995) describe decision making as a learning process: ‘‘as farmers learn, they may reevaluate their situation and, when they become convinced that one alternative tends to dominate the others, the decision is made’’ (p. 8). LENNART is designed to
312
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
provide access to other users’ frequency estimates as well as expert estimates through the use of links to other sources of information. This heuristic shares similarities with the representative heuristic as they both refer to the reliability of a sample as representative of a larger population. However, availability refers primarily to the ability of the decision maker to access similar events. Users of the DSS LENNART may return to their own previous runs for comparative purposes. New estimates made by the farm user of the costs of implementation are easily compared to previous estimates as all the information is saved on the server in a database. The Internet platform also will enable the user to access other sources of information (research results, advising services, etc.) as these become available that may support frequency estimates. Lowering the cost of access to information has driven development of the Internet. By making LENNART available through the net, the cost of access is lowered and the availability of information to the user is increased. 3.2. The model: a net-based dynamic DSS The model is built on a relational database that is located on a web-server. Access to the system is performed via normal Internet browsers using plain HTML-code. The HTML-code is dynamically generated through server-side scripts. Both the system and sub-models of LENNART are maintained inside these scripts. When using the system, the user sends a request to the web-server, which processes the request and sends the result back as an HTML-page to the browser of the user. On the server, LENNART computes the economic costs for adopting catch crops on each field and generates the resulting nitrogen loss reduction on that particular field. The economic model driving these cost estimations does not need any substantial amount of computational power. Therefore, the model may be run directly on the server when the user sends a request. Model responses are produced within seconds. This short response time is, however, not the case for the calculation of nitrogen reduction. The basis for calculation of reductions in leaching losses of nitrogen is the physically based SOILNDB model (Johnsson et al., 2002). This model was developed to quantify nitrogen leaching losses from arable
land for various agri-environmental conditions. SOILNDB is an administrative program which links input data to automatic parameterization procedures for two widely used research models; the water and heat model SOIL (Jansson and Halldin, 1979) and the nitrogen model SOILN (Johnsson et al., 1987). Even though the requirements for input data for SOILNDB are more general, less detailed and less extensive than those required for the direct use of the SOIL –SOILN models, the development team decided that these demands would be too cumbersome in the DSS LENNART. Instead, an extensive number of standardized runs from the SOILNDB model were stored in a separate database.5 Nitrogen leaching for different soils, crop combinations and areas (climates) are kept in the database. Thereby the nitrogen leaching data can be sent back to the LENNART user within seconds. There are three primary factors that led to the choice of a server based web site accessed through the Internet for LENNART; access factors, development factors and data base factors. A server-based program promotes access for a wide group of intended users. Multiple users from individual computers may access the site, with the only personal computer software requirement being a standard web-navigating program (Netscape or Explorer). Enabling access to the program through individual computer connections also allows the program to be demonstrated in a variety of environments. Farm advisers can demonstrate use of the program in consultations with farmers during farm visits. The program can also be demonstrated and used by groups in seminars. Development of the model can be continuous over time as control over the version being used is determined through the server. This quality also means that no problems arise with versions being used that are out of date. Each time a user logs on the version that becomes available is determined through commands on the server. This also allows for partitioning over time to test development of model components. For example, inclusion of a wizard format or tutorial can be tested by incorporating that component into the 5 There are 1168 runs from the SOILNDB model in the prototype. These runs cover two geographic regions in Sweden, 4 – 6 soil types, 10 – 13 crops and with or without an insown catch crop where this is agronomically feasible.
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
model made available to users on the server over for a specific period of time or a specified number of runs. Results from this partitioned model can be compared and choices made by model developers with respect to incorporation or development of the most favourable components. The net-based format also allows for incorporation of changes in development of the independent natural science process based sub-model, SOILNDB. The server platform of LENNART allows changes to be made in the user available model as soon as new information becomes available which affects the results of the sub-model. The entire model does not need to be replaced, only those changes which are made to the model. This ensures that LENNART is able to make use of the best information available. The location of the model on a server also means that the database is developed as the model is used. This represents the dynamic aspect of the model. All of the data is located in one place and can be accessed from anywhere by designated users. As new data becomes available, i.e. every time the model is used, this data is directly available on the server. The immediacy of availability will be able to provide support for users that are interested in comparative data and for users that are interested in aggregate data for policy evaluation and design. Fig. 3 illustrates one of the comparative summary pages in the second
313
prototype of LENNART. This page compares the farm user’s imputed cost estimates with the cost estimates used by the Swedish Agricultural Board for calculating the economic effect of cultivation of catch crops. It is also possible to use partitioning with respect to the database. Open access to the entire database through the Internet will make it possible for those users that are interested in the model to actively work with the database for this purpose. Fig. 2 reproduces a diagram of LENNART user estimates of labor costs per hour. Statistical analysis of this kind of data is of interest for policy analysis and program evaluation. The technical platform for LENNART is a Windows environment using an Access 2000 database, Active Server Pages (ASP) with server-side Visual Basic Script and a few client-side Java Scripts. The second prototype of the model (in Swedish) is available at: http://neptunus.md.slu.se/VASTRA/BAK/index.html.
4. Conclusions and further development Agri-environmental policies will continue to include programs based on voluntary participation by farmers. Since the success of these programs depends ultimately on the rate of implementation of the policy measure by farmers, it is important to understand how
Fig. 3. Comparative summary page in LENNART for farm users (in Swedish).
314
D. Collentine et al. / Ecological Economics 49 (2004) 303–315
farmers make management choices. The uncertainty with respect to costs that necessarily accompanies the adoption of new techniques leads to a need for supporting information to analyze the economic effect of adoption on farm income. Farmers, like the rest of us, are limited by their capacity to store and process information. As decision makers, they operate in a realm of bounded rationality when faced with choices and rather than optimizing over the set of decision alternatives may instead satisfice. Decisions may be perceived as a series of sequential decisions, a set of decision nodes. At each node, one of three actions is possible; accept (enroll in the program), reject unconditionally or reject and gather more information. The last of these three choices includes the expected costs of the additional information. Since these costs are positive and the revenues may be constant, this results in a paradox. If the value of the information gathered is lower than the cost of accessing and processing this information, the likelihood of participation is lower the more well informed the decision is. Thus, quick decisions based on simple decision rules may be effective. The decision support system LENNART was designed to assist farmers with the evaluation of the effects of implementing agronomic measures to reduce the leaching of nutrients from cultivated land. The model currently allows users to evaluate the economic effect and the expected reduction in nitrogen leaching from a set of crop rotations in a specific area of Southern Sweden. Expansion is planned to cover both a larger geographic area as well as to include a greater number of soil types and crop rotations. In addition, the development team plans to improve the graphic interface through the use of focus group tests. Preliminary work is also underway to allow the evaluation of other field management measures in LENNART. New measures planned for development in extensions of the model include the reduction of fertilization intensity on fields and measures where timeliness is a factor such as the timing of cultivation in combination with other practices and the timing of fertilizer applications. Since measures where timeliness is a factor may have an effect on other farm activities beyond the individual field, a preliminary study of these types of measures is necessary before they may be included in LENNART.
Acknowledgements This study was performed within the Swedish Water Management Research Program (VASTRA) and financed by the Swedish Foundation for Strategic Environmental Research (MISTRA). References Amacher, G.S., Feather, P.M., 1997. Testing producer perceptions of jointly beneficial best management practices for improved water quality. Applied Economics 29 (2), 153 – 159. Bingham, G., Bishop, R., Brody, M., Bromley, D., Clark, E., Cooper, W., Costanza, R., Hale, T., Hayden, G., Kellert, S., Norgaard, R., Norton, B., Payne, J., Russell, C., Suter, G., 1995. Issues in ecosystem valuation: improving information for decision making. Ecological Economics 14, 73 – 90. Buchanan, J.M., 1969. Cost and Choice: An Inquiry in Economic Theory. The University of Chicago Press, Chicago, IL. Buller, H., 1999. Evaluating participation in agri-environmental schemes. In: Beihl, D., Schramek, J., Buller, H. Wilson, G. (Eds.), The Implementation and Effectiveness of Agri-Environmental Schemes after Reluation 2078. Final Report to the DGVI of the European Commission (Contract FAIR CT95274, Frankfurt). Carpenter, S.R., Caraco, N.F., Correll, D.L., Howarth, R.W., Sharpley, A.N., Smith, V.H., 1998. Nonpoint pollution of surface waters with phosphorus and nitrogen. Ecological Applications 8 (3), 559 – 568. Collentine, D., 2002a. Economic modelling of best management practices (BMPs) at the farm level. In: Steenvoorden, J., Claessen, F., Willems, J. (Eds.), Agricultural Effects on Ground and Surface Waters, vol. 273. IAHS Press, Oxfordshire, UK, pp. 17 – 22. Collentine, D., 2002b. Book Review. In: Tietenburg, T., Folmer, H. (Eds.), The International Yearbook of Environmental and Resource Economics 2000/2001: A Survey of Current Issues, Edward Elgar Publishing, UK. Environmental and Resource Economics, vol. 23, pp. 467 – 472. Cooper, J.C., Keim, R.W., 1996. Incentive payments to encourage farmer adoption of water quality protection practices. American Journal of Agricultural Economics 78 (4), 54 – 64. Drake, L., Bergstro¨m, P., Svedsa¨ter, H., 1999. Farmers’ attitudes and uptake. In: Van Huylenbroeck, G., Whitby, M. (Eds.), Countryside Stewardship: Farmers, Policies and Markets. Elsevier, Oxford, UK, pp. 89 – 112. Easter, K., 1993. Differences in the transaction costs of strategies to control agricultural offsite and undersite damages. In: Russell, C., Shogren, J. (Eds.), Theory, Modeling and Experience in the Management of Nopoint-Source Pollution. Kluwer Academic Publishing, London, pp. 37 – 68. Feather, P.M., Amacher, G.S., 1994. Role of information in the adoption of best management practices for water quality improvement. Agricultural Economics 11, 159 – 170. Girgerenzer, G., Selten, R., 2001. Rethinking rationality. In: Gir-
D. Collentine et al. / Ecological Economics 49 (2004) 303–315 gerenzer, G., Selten, R. (Eds.), Bounded Rationality: The Adaptive Toolbox. MIT Press, Cambridge, MA, pp. 1 – 12. Gustafson, A., Fleischer, S., Joelsson, A., 1998. Decreased leaching and increased retention potential co-operative measures to reduce diffuse nitrogen load on a watershed level. Water Science and Technology 38 (10), 181 – 189. Hogarth, R., 1987. Judgement and Choice. Wiley, New York. 311 pp. Jansson, P.-E., Halldin, S., 1979. Model for annual water and energy flow in a layered soil. In: Halldin, S. (Ed.), Comparison of Forest Water and Energy Exchange Models. International Society for Ecological Modelling, Copenhagen, pp. 145 – 163. Johnsson, H., Bergstro¨m, L., Jansson, P.-E., Paustian, K., 1987. Simulated nitrogen dynamics and losses in a layered agricultural soil. Agriculture, Ecosystems and Environment 18, 333 – 356. Johnsson, H., Larsson, M., Ma˚rtensson, K., Hoffman, M., 2002. SOILNDB: a decision support tool for assessing nitrogen leaching losses from arable land. Environmental Modelling & Software 17, 505 – 517. Just, R.E., Buss, A., Donoso, G., 1991. The significance of the interface of agricultural and resource policy: conclusions and directions for further research. In: Just, R.E., Bockstael, N. (Eds.), Commodity and Resource Policies in Agricultural Systems. Springer Verlag, Berlin, pp. 293 – 310. Kahneman, D., Tversky, A., 1982. Subjective probability: a judgment of representativeness. In: Kahneman, D., Slovic, P., Tversky, A. (Eds.), Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK, pp. 32 – 47. Kahneman, D., Slovic, P., Tversky, A., 1982. Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK. Laville, F., 2000. Foundations of procedural rationality: cognitive limits and decision processes. Economics and Philosophy 16, 117 – 138. Leathers, H.D., 1991. Best management practices versus socially optimal practices. In: Just, R.E., Bockstael, N. (Eds.), Commodity and Resource Policies in Agricultural Systems. SpringerVerlag, Berlin, pp. 293 – 310. Lichtenstein, S., Slovic, P., Fischoff, B., Layman, M., Combs, B., 1978. Judged frequency of lethal events. Journal of Experimental Psychology. Human Learning and Memory 4, 551 – 578. Napier, T., Tucker, M., 2001. Use of soil and water protection practices among farmers in three midwest watersheds. Environmental Management 27 (2), 269 – 279.
315
Nowak, P., O’Keefe, G., Bennett, C., Anderson, S., Trumbo, C., 1997. Executive summary: communication and adoption evaluation of USDA water quality demonstration projects. Plant and Animal Systems. CSREES/USDA, Washington, DC. Patrick, G.F., DeVuyst, E.A., 1995. Whence and Whither in farm management risk research and extension delivery? Canadian Journal of Agricultural Economics 43, 1 – 14. Setia, P., Magleby, R., 1987. An economic analysis of agricultural nonpoint pollution control alternatives. Journal of Soil and Water Conservation, November-December, 427 – 431. Sharp, B.M.H., Bromley, D.W., 1979. Agricultural pollution: the economics of coordination. American Journal of Agricultural Economics 61 (4), 591 – 600. Simon, H.A., 1956. Rational choice and the structure of the environment. Psychological Review 63, 129 – 138. Simon, H.A., 1987. Bounded rationality. In: Eatwall, J., Milgate, M., Newman, P. (Eds.), The New Palgrave A Dictionary of Economics. MacMillan, London, pp. 266 – 268. SJV, 2001. Tips for Sowing Catch Crops, Greppa Na¨ringen, Brochure (Tips info¨r fa˚nggro¨desa˚dden (in Swedish)). Swedish Board of Agriculture, Stockholm, Sweden. Supalla, R.J., Selley, R.A., Bredeweg, S., Watts, D., 1995. Adoption of nitrogen and water management practices to improve water quality. Journal of Soil and Water Conservation 50 (1), 77 – 82. Tucker, M., Napier, T.L., 2002. Preferred sources and channels of soil and water conservation information in three midwestern US watersheds. Agriculture, Ecosystems & Environment 92 (2 – 3), 297 – 313. Tversky, A., Kahneman, D., 1974. Judgment under uncertainty: heuristics and biases. Science 185, 1124 – 1131. Tversky, A., Kahneman, D., 1982. Judgment under uncertainty: heuristics and biases. In: Kahneman, D., Slovic, P., Tversky, A. (Eds.), Judgment Under Uncertainty: Heuristics and Biases. Cambridge Univ. Press, UK, pp. 3 – 22. van den Bergh, C.J.M., Ferrer-i-Carbonell, A., Munda, G., 2000. Alternative models of individual behaviour and implications for environmental policy. Ecological Economics 32, 43 – 61. Whitby, M., 2000. Challenges and options for the UK Agri-Environment: presidential address. Journal of Agricultural Economics 51 (3), 317 – 332. Wolf, A.T., 1995. Rural nonpoint source pollution control in Wisconsin: the limits of a voluntary program. Water Resources Bulletin 31 (6), 1009 – 1022.