On-Site Sampling in Economic Valuation Studies Termeh Shafie 1 1 Department of Statistics, Stockholm University, Sweden e-mail: termeh.shafie@stat.su.se Abstract An overview of models suggested for analysis of on-site data will be presented, with a special focus on the problems of economic valuation. The purpose is partly to present an overview of models for analysis of on-site data and their relations to models for length biased samples. In addition, a new model for estimation of binary choice data from on-site samples is presented. 1 On-Site Sampling On-site sampling can be an efficient sampling design when it is difficult to construct an effective sampling frame. This sampling design implies that sample inclusion probabilities depend on individual characteristics such as visiting frequency. This in turn leads to an estimation problem when using standard estimation techniques that do not account for the differences in inclusion probabilities. When sampling is done on-site, the researcher surveys respondents on a specific site e.g., visitors to a park - during the surveying time frame. There are two immediate problems connected to this sampling design. First, the sampling inclusion probabilities will depend on the respondent’s visiting frequency to the site. Second, the researcher may have a subjective influence when sampling indivuals at the site. Thus, care needs to be taken when using this sampling method to control for these two issues. Concerning the second issue, it is important to make sure that an objective sample mechanism is used and that randomness is somehow built in the sampling design. An example of how this can be done is by choosing the entrance at the park as a sampling location and using Bernoulli sampling to choose individuals that enter the park. On-site sampling is often used in research fields such as marketing research, research that typically uses on-site data when applying consumer tests. One example is found in Keillor et al. (2001) where on-site samples are used to explore the notion that consumers around the globe are becoming more similar in terms of psychological consumer tendencies. Sudman(1980) identified procedures for shopping-center sampling that can improve the quality and reduce the biases in shopping center samples. In a resource economics study that examined fishing methods, Pollock et al. (1994) proposed sampling anglers at fishing sites to collect data on angler effort and catch. This method intends to estimate the characteristics of the population of fishing trips during a season. The sampling unit is ”fishing trips”; one trip represents one element in the population. Note that when sampling is done this way, we have a frame based sampling procedure and the estimation is not problematic. Shaw (1988) also considers the problem of on-site sampling when surveying visitors at recreational sites. However, Shaw (1988) is interested in the sampled individuals and the underlying factors that