Joint estimation and consumers’ responses to pesticide risk: Young Sook Eom and V. Kerry Smith
Introduction Sometimes a paper has a long gestation period, but nearly 20 years is probably too long. We started this research as part of the first author’s PhD thesis in the late 1980s, completed the survey providing the data for this analysis in 1990, and then circulated earlier drafts of the paper in the early 1990s. Fortunately for us the problems associated with consumers’ responses to risk have remained a continuing intellectual interest of environmental economists and there have been few efforts to combine revealed and stated preference responses to choices involving risk.1 Equally important, food safety is once again in the news. In 2006 it was e-coli and spinach. Two years later it was salmonella linked to raw tomatoes, and jalapeño peppers.2 The literature in the intervening years has not been able to deal convincingly with food safety. Indeed, a recent session (2007) on the demand for food safety at the American Agricultural Economics Association finds stark contrasts in the authors’ confidence in estimates of consumers’ willingness to pay to reduce risk of food borne illness. For example, in the papers published based on this session, Lusk (2007) notes one study’s estimates (Hammitt and Haninger 2007) are over five times larger than earlier results from Hayes et al. (1995) based on comparable risk changes. Moreover, initial survey research by Shogren and Stamlund (2007) raises concerns with respondents’ ability to understand risk information and thus with analysts’ ability to include models of the perception/adjustment process in those models. While this research was designed and completed before these issues were raised, we believe it nonetheless offers some information that is responsive to both concerns. There are other reasons why we were lucky. At the time our survey was conducted there was no organic produce in major supermarkets in Raleigh, North Carolina (the site of our survey). As a result, the choices available to consumers were more limited than they are today. One final aspect in which our results may have relevance a decade after initial draft of the work was circulated stems from the model. It appears that no one has considered a mixed discrete/continuous demand model interpreting a demand function describing the determinants of the amount consumed with a random utility framework for the choice of the type of
commodity when the choices involve risk. Our application is to risks due to pesticide residues on fresh fruits and vegetables.3 Our results indicate that the stated preference models recover statistically significant estimates for all variables where a priori hypotheses suggested there should be effects. Revealed preference data are incapable of measuring some of these influences. Joint estimation assures compatibility in price and income responses. Of course, to do so we must accept the maintained assumptions about how pesticide residues influence consumers’ food purchasing choices. We are able to estimate separate effects for price, income, and the perceived risk of pesticide residues on the typical consumer’s fresh produce demand. Moreover, our models do not reject the null hypothesis that RP and SP information would be judged consistent. To our knowledge, this is also one of the few applications where the two types of data have been found compatible in the sense that we fail to reject the null hypothesis of consistent shared parameters.4 Of course, this conclusion is conditioned by the shared parameters our models can identify and estimate. The next section outlines the theoretical framework used to combine revealed and stated preference choices. It has three elements: (1) a continuous demand model for a consistently aggregated measure of fresh produce; (2) a model of individual risk perception that recognizes both prior beliefs and new information (see Viscusi 1989); and (3) an expected utility model of the stated choices between existing and “new” produce tested for pesticide residues (based on those subjective risk perceptions). The models used in steps (1) and (3) are derived from a common preference structure. The subsequent section describes the collection of the data used in estimating our model. The last section discusses the prospects for joint estimation in general and the relevance of our early results for current policy questions.