Today's entry in Abstract Monday is a little technical, but it's a problem I've been interested in for a long time, and it looks like my and John's former colleague, Ju-Chin Huang*, and co-author have found an improvement (if not solution). If I ever get around to writing a second edition of my nonmarket valuation econometrics book, this is definitely going to be in there.
*I'm sure John will agree, Ju-Chin was the smartest of the three of us.
Title: Correcting On-Site Sampling Bias: A New Method with Application to Recreation Demand Analysis
Author(s): Wei Shi and Ju-Chin Huang
Journal: Land Economics, 2018, 94(3): 459-474
Objective: A new method for dealing with truncation and endogenous stratification in on-site sampling of recreational site users.
Background: Assessing recreational site use can involve random samples of the general population, random sample of subsamples of users (e.g. boat owners), or on-site sample of current visitors. While convenient, and often less expensive, on-site samples of current visitors introduce modeling challenges due to truncation (no non-users are sampled), and endogenous stratification (the probability of interviewing an avid visitor is higher--sometimes called avidity bias). Both biases have puzzled recreation demand models for decades and the literature is still fraught with examples of misunderstanding and confusion about the correct way to handle on-site sampling. Most past studies have tried to simultaneously solve both problems, but such solutions depend critically on the assumed statistical distribution of trips. Separating the issue of stratification from truncation may provide recreation demand estimates that are less sensitive to distributional assumptions.
Methods: Shi and Huang first correct endogenous stratification by weighting the assumed expected number of trips taken, after truncation at zero, by the inverse of the observed number of trips. This weighting effectively 'scales down' the probability of interviewing those who take large number of trips, creating a sample that more closely mimics the likelihood of a sample drawn from the general population. The weighted sample is then used to estimate a truncated count data model, thereby separating the issue of truncation from the issue of endogenous stratification.
Results and Conclusions: Using simulations and a case study, Shi and Huang demonstrate that the distribution free re-weighting approach to on-site sampling reduces the sensitivity of results to the researcher-assumed distribution of trips.
