Environmental Economics
Industrial Organization
Applied Econometrics

Industrial Organization
Environmental Economics
Applied Econometrics
Microeconomic Theory

May, 2000 (Oral) Environmental Economics and Industrial Organization
September, 1996 (Written) Microeconomic and Macroeconomic Theory

Valuing Recreational Fishing Opportunities:  Unobserved Characteristics and Sequential Decisions in a Discrete Choice Model of Demand

Professor Steven Berry
Professor Christopher Timmins
Professor Martin Pesendorfer

Summer 2002

M.Phil. (2000), M.A. (1997), Department of Economics, Yale University
B.A. (1995), Summa Cum Laude, Mathematics and Economics, State University of New York at Geneseo

Yale Dissertation Fellowship, 2001
Yale University Graduate Fellowship, 1995-1997, 1999-2000

Instructor, Yale University, Microeconomic Theory and Policy (Summer 2001)
Head Teaching Assistant, Yale University, Introductory Microeconomics (Fall 2000)
Teaching Assistant, Yale University, Intermediate Microeconomics (Spring 2001), Introductory Microeconomics (Spring 2000), and Microeconomics (Graduate) (Fall 1999)

Research Assistant to Martin Pesendorfer, Yale University, 2000-2001
Collected and analyzed data tracking projects financed by the Department of Defense. Wrote flexible code to estimate a wide range of fixed effects regressions to identify patterns in quantity, procurement costs, and research expenditures over time and across categories of defense projects.

Senior Economist (1999), Economist (1997-1998), Triangle Economic Research, Durham, North Carolina
Conducted analysis for litigation support and settlement negotiations in natural resource damage cases involving PCB contamination, including the Fox River in Wisconsin, and oil spills including the 1990 American Trader spill in California and the 1995 Skaubay spill in Texas.

• For the Fox River case, assisted in drafting an assessment plan outlining proposed data collection and estimation strategies, worked on survey design, managed and conducted analysis of collected data, and prepared client presentations and draft reports. Contributed to survey instrument design by developing a sampling plan, selecting the language for the telephone screener, and conducting focus groups, verbal protocol interviews, and pre-testing of both written and telephone surveys. Managed associates in data preparation tasks, analysis, data validation, and report writing. Developed flexible programs in both GAUSS and Stata for estimation of discrete choice models and evaluation of potential restoration projects.
• For the American Trader spill, conducted analyses of existing data, drafted critiques of opposing side’s analysis, replicated their results, and prepared technical summaries in support of an expert witness.
• For the Skaubay spill, collected on-site data and conducted interviews, critiqued opposing side’s analysis, produced preliminary estimates using transfer techniques, and prepared client presentations.

"Valuing Recreational Fishing Opportunities: Accounting for Unobserved Characteristics in a Discrete Choice Model of Demand," manuscript, Yale University, 2001.

"Welfare Implications of Site Aggregation: A Comparison of Conditional Logit and Random Parameters Logit Estimates," Annual W-133 Western Regional Project Technical Meeting, Tucson Arizona, February 1999.

"An Analysis of Household Beach and Recreation Demand and Values," Camp Resources VI, Wilmington, North Carolina, August 1998.

Recreation demand models provide valuable insights and information that supports both management decisions and litigation in cases of damage to natural resources.  While discrete choice models have been applied extensively to recreation demand models, nearly all researchers assume that they observe all relevant characteristics of the recreation locations and that individuals faced with repeated opportunities to choose among locations make each decision independently.  My dissertation presents a model that relaxes both of these assumptions.  The models and techniques developed to estimate demand for recreation, a non-marketed good, extend to other fields using discrete choice models to estimate demand for ordinary marketed goods when not all product characteristics are observed and consumers make repeated choices.  The importance of relaxing these assumptions is evaluated by comparing welfare estimates across model specifications for a wide array of proposed policy changes. 

Recreation demand models in general provide valuable insights and information that supports both management decisions and litigation in cases of damage to natural resources. The models and techniques developed in this dissertation to estimate demand for recreation, a non-marketed good, extend to other fields using discrete choice models to estimate demand for ordinary marketed goods when not all product characteristics are observed and consumers make repeated choices.

Anglers constitute 30 percent of Wisconsin's adult populations and spend 15 million days fishing the state's vast water resources each year.  Using a unique data set that I assisted in collecting, this research examines recreational fishing in Wisconsin.   The data includes a panel survey of 788 Wisconsin anglers during the summer of 1998, data on the characteristics of fishing locations, and daily weather observations across the stae.  The survey carefully tracks anglers and provides detailed information about each fishing trip including the date, exact location, and fishing success.

The large number of recreational locations, changes in characteristics from year to year, and the costs of collecting detailed information make unobservable site characteristics an undeniable reality in recreation demand modeling. However, researchers to date either ignore this fact or deal with it in a very restrictive way by including an ad hoc set of alternative or group specific constants. While the data used in this analysis contains more detailed and extensive information than most existing recreation demand data, it cannot claim to include all of the important characteristics that affect anglers’ choices. Finding substantial effects of unobserved site characteristics in this detailed data underscores the importance of controlling for them in recreation demand modeling in general.

The first chapter of this thesis develops a discrete choice model that recognizes the presence of unobserved site characteristics and individual taste heterogeneity. A two-stage estimation approach based on Berry, Levinsohn, and Pakes (1995), which accounts for unobserved site characteristics by including a full set of alternative specific constants, is presented. Simulation techniques are used to recover the distribution of individuals' tastes and a contraction mapping computes the alternative specific constants, making the estimation tractable.  The two-stage approach can recover parameter estimates for those characteristics that vary only across sites, which are important for policy experiments.  To illustrate the importance of controlling for unobserved site characteristics and explore the ramifications of ignoring them, this chapter presents Monte Carlo simulations.  Simulations show that a conditional logit model with the common assumption that all site characteristics are observed produces biased standard errors that dramatically overstate precision and can produce unreliable parameter estimates, even if these characteristics are not correlated with included variables.   Further, ignoring unobserved site characteristics biases estimates of welfare gains from proposed improvements, causing estimates at unattractive sites to be overstated and estimates at attractive sites to be understated.

The second chapter explores the potential link between successive decisions made by an individual over a relatively short time period measured in days, weeks, or months. The actual catch of fish serves as the link. The model allows the fishing success on previous trips to affect the probability of fishing and the probability of choosing a particular site. Once we control for the angler's tastes, this allows for the possibility that anglers may chose to visit a site that has a high potential catch of a species different from what they caught on their last trip.

The estimated models allow examination of a wide range of policies. These include changes in land-use and shoreland development, fishing regulations and management policies, and fish stocking programs. The welfare implications of the policy experiments are compared across models to determine their sensitivity to the controls for unobserved site characteristics, heterogeneous tastes, and state dependence. Results show that model specification substantially affects the magnitude of benefits from proposed improvements and can even alter the ranking of projects. Given the potentially serious biases that can result from ignoring unobserved site characteristics, the proposed model provides a valuable new tool for recreation demand modeling and other applications of discrete choice models of demand that allows researchers to account for unobserved characteristics, obtain reliable parameter estimates, and perform relevant policy experiments.