The Federal Insecticide, Fungicide and Rodenticide Act (FIFRA) and its amendments and related legislation currently require that applicants for registration or reregistration of a pesticide demonstrate to the U.S. Environmental Protection Agency (EPA) that there is not “any unreasonable risk to man or the environment, taking into account the economic, social, and environmental costs and benefits of use of the pesticide.” Present benefit evaluation guidelines call for consideration of effects on users, nonusers, consumers, GNP, and employment. Minimum guidelines call for a partial budgeting analysis and, if output effects of eliminating the pesticide appear large, a more sophisticated analysis using neoclassical multimarket economic-surplus concepts.

In previous work (Cropper et al.), we examined the U.S. Environmental Protection Agency's (EPA's) decision to cancel or continue the registration of pesticides that went through its Special Review process between 1975 and 1989. Our focus in that paper was on the final decision (Notice of Final Determination) issued by the EPA at the end of the rule-making process. Specifically, we asked whether this decision could be explained by the reported risks and benefits associated with pesticide use, and by the comments of special-interest groups that were entered in the public docket.

Pesticide regulation has become a topic of increasing interest in recent years, owing to rising public concerns about residues on foods, in drinking-water wells, and damage to wildlife. Public-opinion polls and political responses to incidents like the controversy over Alar suggest that demand for government intervention to protect public health and the environment from pesticides is high. Pesticides are toxic by design; survey evidence indicates that they are perceived as riskier than other, more common pollutants like auto exhaust (see, for example, Horowitz). Pesticide residues are not easily observable (short of laboratory analysis), making averting strategies by individuals extremely difficult and/or excessively costly to implement.

The paper by Taylor has merit because it is provocative, but like many things, this can be overdone. His self-described diatribe against neoclassical economics is in special need of a response. He appears to dismiss the notion of utility maximization, but in fact there is nothing to dismiss. Although many economists do not realize this fact, the assumption of utility maximization is not testable. Give me any pattern of behavior by an individual (or a society) and I can give you a utility function which, when maximized subject to whatever constraints that person or society may face, yields the observed behavior. An obvious example would be U = 1 for the observed behavior and U = 0 otherwise, although more complicated examples could also be constructed. The power of economics lies not in utility maximization, but in the specification of constraints on choice and in describing how changes in the constraints affect behavior. To illustrate, even a consumer who made decisions randomly would tend to have downward-sloping demand curves because of his or her budget constraint (Becker).

Two issues lurk in the background of the papers by Lichtenberg, Taylor, and Cropper et al. The first is a specific question about the goal of regulatory policy. The second is a more global consideration regarding politics, technological change, and the future.

The concept of nonparametric analysis, estimation, and inference has a long and storied existence in the annals of economic measurement. At least four rather distinct types of analysis are lumped under the broad heading of nonparametrics. The oldest, and perhaps most common, is that associated with distribution-free methods and order statistics. Similar in spirit, but different in emphasis, is nonparametric density estimation, such as the currently popular kernel estimator for regression. Semi-parametric or semi-nonparametric estimation combines parametric analysis of portions of the problem with nonparametric specification for the remainder, such as the specification of a specific functional form for a regression function with a nonparametric representation of the error distribution. The final type of nonparametrics is that associated with data envelopment analysis and revealed preference, although the use of the term nonparametrics for this research is perhaps a misnomer. This paper will briefly review each of the four types of analysis, leaning heavily on other published work for more detailed exposition. The paper will then discuss in more detail the application of the revealed-preference approach to four specific economic problems: efficiency, the structure of technology or preferences, technical or taste change, and risky choice. The paper is not complete, exhaustive, or detailed. The primary purpose is to expose the reader to a variety of techniques and provide ample reference to the relevant literature.

Nonparametric techniques have recently come into vogue in agricultural economics: Applications abound in both consumer and producer models of the agricultural economy. Moreover, several distinct approaches to nonparametric analysis exist. There are nonparametric statistical techniques, semiparametric estimation techniques, nonparametric revealed-preference analysis of consumption data, and nonparametric analysis of production data. Both revealed-preference analysis and nonparametric analysis of production data rely on the basic fact, which provides the foundation for much of modern duality theory, that convex sets can be completely characterized by their supporting hyperplanes. This observation allows one to apply simple mathematical programming (in particular, linear programming) methods to analyze production and consumption data. My task today is to provide an overview of nonparametric programming approaches to production data. Thus, I will not address any of the other topics cited above. However, I would be remiss if I did not mention the close connection between these subject areas and what I intend to survey today. Moreover, one should also recognize that very closely related to the literature on nonparametric programming analysis of production data are the fields of estimation of efficiency frontier via statistical methods. (A useful survey here is Lovell and Schmidt).

When I was invited to be a discussant in this session dealing with nonparametric analysis of production and consumption, my initial response was rather mixed. On one hand, I teach a graduate-level course in applied statistics, and one of the topics I cover is nonparametric statistics and its application to the biological sciences. Hence, I thought it would be interesting to relate what I cover in class with what is used in applied agricultural production economics research. On the other hand, I am not familiar with this approach since I have not used the nonparametric programming method for efficiency analysis in my past research. Thus, together with my graduate students, we started applying the nonparametric production efficiency analysis in order to gain a better understanding of this approach. The deeper we got into the research topic, I realized that there was very little relationship between what is called nonparametric statistics by statisticians and nonparametric methods used to analyze production efficiency.

The nonparametric approach to consumer-demand analysis—based on revealed-preference axioms—is reviewed. Particular attention is paid to questions of size and power of tests for consistency of data with the existence of a stable, well-behaved utility function that could have generated the data. An application to Australian meat demand is used to show how these notions can be quantified and how prior information about elasticities, following Sakong and Hayes, may be used to increase the power of the approach.

This paper explores the issue of the power of nonparametric tests to check for the consistency of data with utility maximization. Alston and Chalfant provide an excellent review of nonparametric approaches to consumer-demand analysis. They test for consistency, separability, and power. The authors address two important questions: First, how does one define power for nonparametric situations, and is that definition comparable to the parametric situation? Second, can the power of nonparametric tests be improved? The authors measure the statistical power of nonparametric methods using a parametric test, though they do not address whether it is legitimate to use parametric tests on nonparametric methods.

Firm-varying production technologies were estimated using random coefficients regression methods for a sample of Massachusetts dairy farms. Results were compared to OLS Cobb-Douglas production function estimates. The random coefficients regression model was found to virtually eliminate conventionally measured firm technical inefficiencies by estimating individual firm technologies and ascribing remaining inefficiencies to specific inputs. Input-specific measures of firm inefficiencies showed hired labor, land, and machinery inputs to be used in excess of efficient levels. Livestock supplies were underutilized by all farms. Efficiencies of feed, crop materials, fuels, and utilities varied, although estimated means were closer to optimal levels.

Seasonal swings in milk production in Florida result in a need to import milk on a seasonal basis. A linear programming analysis is used to analyze alternate freshening-date distributions and project the cost savings to Florida dairy farmers from reduced milk imports.

Protest bids are often excluded during analysis of contingent valuation method data. It is suggested that this procedure might introduce significant bias. Protest bids are often registered by respondents who may actually place a higher- or lower-than-average value on the commodity in question but refuse to pay on the basis of ethical or other reasons. Exclusion of protest bids may therefore bias willingness to pay (WTP) results, but the direction of bias is indeterminate a priori.

This paper empirically tested the three conditions identified by McConnell for equivalence of the linear utility difference model and the valuation function approach to dichotomous choice contingent valuation. Using a contingent valuation survey for deer hunting in California, two of the three conditions were violated. Even though the models are not simple linear transforms of each other for this survey, estimates of mean willingness to pay and their associated 95% confidence intervals around the mean were quite similar for the valuation methods.