Estimating Forest Recreation Demand Using Count Data Models

Abstract

Forests, along with related natural areas such as mountains, lakes, and rivers, provide opportunities for a wide variety of recreational activities. Although the recreational services supplied by forested areas produce value for the consumers of those services, the measurement of recreational value is complicated by the fact that access to most natural areas is non-priced. Because outdoor recreation often competes with commodity uses of forests, such as timber harvesting or mineral extraction, failure to account for the recreational use of forest land makes it impossible to determine the efficient use of forest resources. A key insight attributed to Harold Hotelling is that the price of recreational access can be inferred from information on travel costs. Subsequent development of this idea was undertaken by Marion Clawson (1959) and, a few years later, articulated in a general work on the economics of outdoor recreation (Clawson and Knetsch 1966). The basic Hotelling-Clawson-Knetsch (HCK) approach to estimating recreation demand is to statistically regress the number of trips taken to a recreational site on the round-trip cost of travel between trip origins and the site. A set of demand shift variables are also typically included in the specification to control for socio-economic characteristics of visitors, indicators of site quality, and costs associated with visiting substitute sites. Once a travel cost demand curve is estimated, the value of a recreational site can be computed by integrating the area under the demand curve. Two types of data can be used to estimate travel cost models (see, for example, Bockstael et al. 1991 and Freeman 1993). The early studies Sills and Abt (eds.), Forests in u Market Economy, 341-359. typically used aggregate data on origin zones; these are often referred to as zonal travel cost models. Per capita visitation rates for each origin zone (often counties, but also distance zones) were computed, and distances were translated into travel costs using cost per mile multipliers. Socio-economic variables for origin zones were proxy variables for the representative visitor, and prices based on travel costs to substitute sites were included in the specification. The second type of data that can be used to estimate travel cost models is based on individual observations of visitation rates and socio-economic variables (referred to as individual travel cost models). Individual data do not rely on the representative visitor assumption. The added precision in describing individual characteristics and trip decisions has led to the development of a rich array of empirical methods and, in particular, models based on random utility maximization (RUM). The RUM approach models the choice of a recreation site from among a set of alternative sites as a utility-maximizing decision, where utility includes a stochastic component. RUM models emphasize the impact of site quality on recreation demand and are estimated using either multinomial or nested logit models. Forestry examples include Englin et al. (1996) and Pendleton and Shonkwiler (200 1). Another approach that focuses attention on site quality is the hedonic travel cost (HTC) method. The HTC method is used to estimate the demand for site characteristics using a two-step procedure (Brown and Mendelsohn 1984). In the first stage, marginal values (implicit prices) of the site characteristics are estimated for each origin zone. Then, demand functions for characteristics are estimated in the second stage across all origins. Applications to forestry include Englin and Mendelsohn (1991), Holmes et al. (1997), and Pendleton et al. (1998). During the past decade, there has been an explosion of interest in the application of count data models based on the Poisson distribution to estimation of HCK models of recreation demand. In this chapter, we provide an overview of the major developments in count data travel cost modeling and show how they can be applied to forest-based recreation. In contrast to earlier HCK modeling that used ordinary least-squares (OLS) regression, count data models emphasize the non-negative, integer nature of data on the number of trips taken and are most useful when the counts (per person) are small. This is often the case with forest recreation, such as backcountry trips or adventure activities, which most people participate in only a few times a year. Although the normal distribution is a good approximation of the Poisson distribution (which is sometimes called the " law of rare events ") if the mean of the distribution is large, the normal distribution provides a poor approximation of the Poisson for small mean