Multicriteria Analysis for Agricultural Practices

4.2.1 Selecting farming practices by multiattribute models

This section concentrates on using multicriteria analysis to select agricultural prac-tices expressed as discrete alternatives, although various multicriteria analyses have been applied to agricultural and natural resource management (Hayashi, 1999a, 2000b). As shown in Table 4.1, there are two types of methods in the applications.

One is the compensatory approach which aggregates multiple attributes into over-all values by, e.g., multiattribute value (utility) functions in which the concept of tradeoffs plays a crucial role. The other is the non-compensatory or outranking approach which introduces aggregation procedures based on concordance and dis-cordance concepts that are derived from outranking relations, which express that an alternative is at least as good as another one. Although the distance-based ap-proach (e.g., compromise programming), in which the distance between the ideal point and the alternatives is minimized, can also be used for decision making, this table contains no examples of that approach.

One of the main features in these applications is that attention is paid to the tradeoffs between economic objectives and environmental objectives [except for Arondel and Girardin (1998)]. That is, most of the problems can be expressed hierarchically as depicted in Figure 4.1; agricultural practices are evaluated from the viewpoint of profitability and environmental quality of soil and water.

As a representative method for analyzing the tradeoff, we focus on a method based on an additive multiattribute value function used for ranking discrete alterna-tives. This method has been applied to many selection problems of farming prac-tices, and has been used in many papers presented at both the First International Conference on Multiple Objective Decision Support Systems (MODSS) for Land, Water, and Environmental Management in 1995 (El-Swaify and Yakowitz, 1998) and the second such conference in 1999 (MODSS‘99, 1999).

This method is based on an importance order of attributes which is elicited from decision makers or experts without specifying numerical values of attribute weights and has the following procedures. First, single-attribute values are calculated using value functions. Next, the priority order of the attributes is given. Then, the best

82 Kiyotada Hayashi Table 4.1. Criteria used for selecting farming practices.

Environmental

Authors Economic Fertilizer Pesticide Other

Yakowitz et al. Net income N (percolation) Atrazine (surface) Sediment yield

(1993) N (surface) Atrazine (percolation)

P (surface) Serin (surface) Carbofuran (surface)

Foltz et al. Net returns N (surface) Atrazine (surface) Soil loss

(1995) N (percolation) Alachlor (surface) [USLE]

[EPIC] [GLEAMS]

Heilman et al. Net returns N (runoff) Atrazine (runoff) Soil detachment (1997) NO3-N (percolation) Atrazine (sediment) Sediment yield

All other pesticides in surface or groundwater

Lawrence et al. Above ground Range condition

(1997) net primary Channel erosion

production Annual runoff

Annual maximum Peak runoff rate Quail and javalina (NRCS wildlife habitat index)

Arondel and N management Pesticide management Water management

Girardin (1998) (amount, balance, (amount, half-life, (hydric balance, date, splitting up, mobility, toxicity, amount improving location, date)

techniques)

Tiwari et al. Farmers’ NPV Land suitability

(1999) Government Energy output/input

NPV Water requirement

Societal NPV Environmental cost

and worst overall values are determined by solving the following linear programs:

minimize or maximize v(a_i) = Xn j=1

w_jv_j(x_ij)

subject to Xn j=1

wj = 1,

w₁≥w₂ ≥. . .≥w_n≥0,

Applicability of Multicriteria and Risk Analysis 83

Improve management

Maximize

profitability Returns

Minimize environmental impacts

Minimize

soil loss Erosion

Minimize contamination

N

P

Figure 4.1. An example of a value tree for agricultural practices.

where v(ai) is the overall value of alternativeai; wj is the weight for the jth at-tribute;v_j(·)is the value function for thejth attribute;x_ij is thejth attribute level for alternativea_i; andnis the number of attributes.

This method, however, is associated with the following problems (Hayashi, 1999b). (1) The meaning of rank-ordered weights based on the relative importance of attributes is ambiguous; moreover, the weights based on importance judgments may distort rescaling of single-attribute value functions. (2) The importance order of attributes may not be sufficient to determine the final ranking of alternatives, because the intervals of overall values obtained from the importance order are wide and overlap each other. (3) The fact that the structure of a value tree (a hierarchy of criteria) may affect the final results is not taken into account.

4.2.2 Issues on attribute weights

One of the methods to remedy the first problem, the most serious difficulty, is the use of difference value measurement (Dyer and Sarin, 1979) and the application of weight elicitation techniques based on the measurement. It is well-known that using weight elicitation methods without relying on attribute ranges might lead to biased weights (von Nitzsch and Weber, 1993; Fischer, 1995).

This insight is very important because similar procedures have been employed in many fields. For example, multiattribute evaluation techniques are used to inte-grate geographical data into spatial decision making using GIS, and the range prob-lem just mentioned has been recognized as a common source of error (Malczewski, 1999). Moreover, weighting is a subjective and crucial part in LCA (Goedkoop et

84 Kiyotada Hayashi al., 2000) and thus weighting steps are referred to as an “optional element” in ISO 14042 (2000).

However, the weights for attributes expressed as raw data (e.g., the levels of NO3-N in groundwater) are sometimes difficult for decision makers – and even for experts – to understand. Moreover, the difficulty in weighting when the problems have 10 attributes or more is pointed out in LCA (Goedkoop et al., 2000). Indeed, the problems in the real world tend to have a considerable number of attributes.

Therefore, it is necessary to introduce a methodology for transforming the data into the other values to make the meaning easy to grasp and to reduce the number of attributes using a common physical unit. This is especially true for societal decision making because in order to settle the differences in perceptions, it is necessary to introduce an understandable scale into evaluation processes.