Addressing onsite sampling in recreation site choice models

Abstract: Independent experts and politicians have criticized statistical analyses of recreation behavior that rely upon onsite samples due to their potential for biased inference. The use of onsite sampling usually reflects data or budgetary constraints but can lead to two primary forms of bias in site choice models. First, the strategy entails sampling site choices rather than sampling individuals– a form of bias called endogenous stratification. Under these conditions, sample choices may not reflect the site choices of the true population. Second, exogenous attributes of the individuals sampled onsite may differ from the attributes of individuals in the population – the most common form in recreation demand is avidity bias. We propose addressing these biases by combining two existing methods, Weighted Exogenous Stratification Maximum Likelihood Estimation and propensity score estimation. We use the National Marine Fisheries Service’s Marine Recreational Fishing Statistics Survey to illustrate methods of bias reduction, employing both simulated and empirical applications. We find that propensity score based weights can significantly reduce bias in estimation. Our results indicate that failure to account for these biases can overstate anglers’ willingness to pay for improvements in fishing catch, but weighted models exhibit higher variance of parameter estimates and willingness to pay.

In a full free version of the paper available at http://www.ecu.edu/cs-educ/econ/upload/ecu1002-20100331-AddressingOnsiteSamplingInRecreationSiteChoiceModels-Landry.pdf the authors point out:
...
As we found in the simulated exercise, the unweighted estimators appear to suffer from upward bias in the WTP measures for one additional caught fish. We find that the weighting strategy leads to a 31% decrease in the WTP for an additional inshore species and a 6% decrease in the WTP for an additional reef species. We do not find statistically significant WTP measures for changes in catch of offshore or pelagic species.
...
Utilizing our simulated dataset with an unweighted estimator, on average, we find point estimates of WTP for changes in fishing catch to be biased upward by 45%. When we only account for endogenous stratification, the point estimates for WTP are still biased upward by 39% on average. In the presence of both endogenous stratification and size-biased sampling, the avidity weight, which only accounts for size-biased sampling, actually increases the bias in point estimates to an average of 89%.

The table below displays key Willingness-to-pay results:


by Paul Hindsley 1* Craig E. Landry 2 and Brad Gentner 3
1. Environmental Studies, Eckerd College, St. Petersburg, FL 33711 hindslpr@eckerd.edu
2. Department of Economics, East Carolina University, Greenville, NC 27858 landryc@ecu.edu
3. Gentner Consulting Group, Silver Spring, MD 20901 brad@gentnergroup.com
* Corresponding author

Journal of Environmental Economics and Management
Article in Press, Accepted Manuscript, Available online March 21, 2011
Keywords: On-site sampling; Propensity score weighting; Recreation demand; Random utility models