23
The first specification (specification I) is our basic model, while specification II is
broader and explores the possible effects of individual characteristics and perceptions
about conservation. In both specifications, mean actual trips are obtained by setting all
the scenario dummies to zero. Both models fit the data well, in the sense that likelihood
ratio tests of the null that all slopes are equal to zero reject the null soundly.
In both models we impose the restriction that the coefficient on price, β2 , to be
equal to -0.1263—the estimate from running the model on actual trip data only. We
believe that this approach is desirable because it “grounds” intended behaviors to
observed behaviors, thus providing reasonable and conservative estimates of the benefits
of conservation of monuments in Armenia.18, 19, 20
18 This approach is in the spirit of Azevedo et al.’s point that revealed preference data (i.e., actual trips)
should be viewed as complementary sources of values and information with stated preference data (i.e.,
hypothetical trips) (Azevedo et al., 2003). Revealed preference methods bring the “discipline of the
market” to stated preference valuations, while the latter can shed light on consumer preferences for price
and quality levels that are currently not observed.
19 Of course, we attempted to estimate the unrestricted model, but were dissatisfied with the fit of the model
and with the implausible value of the unrestricted coefficient on price. All other coefficients, however,
were very close to those of the restricted model. Accordingly, we opted for imposing the restriction and for
reporting only the results of the restricted maximum likelihood estimation in this paper. We also explored
random effects Poisson to allow for the possibility of correlation among the responses provided by the
same person. In the unrestricted model we find some evidence of the presence of random effects, but the
coefficients on all other variables are virtually the same as those of the model where the observations are
independent within respondents. The random effect model does not converge when we impose the
restriction that β2=-0.1263. Finally, we experimented with negative binomial models for the actual trip
data, but encountered convergence difficulties, a problem probably caused by the functional form of the
probability function for the negative binomial model with correction for endogenous truncation (Haab and
McConnell, 2003). Monte Carlo simulations under controlled conditions suggest that these problems with
the negative binomial occur frequently, and that Poisson models are well-behaved even when the true data
generating process is a negative binomial (Alberini and Reppas, 2005).
20 Other researchers have investigated whether the slope of the demand function implied by the responses to
the hypothetical questions is different from that implied by actual travel. Results are mixed. For example,
Rosenberger and Loomis (1999) find that the slope of the demand function (i.e., the coefficient on price per
trip) is the same across actual and hypothetical data, and Alberini et al. (2005) report a similar result in a
travel cost method study that examines fishing trips to the Lagoon of Venice by a sample of anglers in the
Venice area. By contrast, Azevedo et al. (2003) find that individuals appear to be less sensitive to price in
contingent behavior questions than we observe them to be in real life. They are, however, careful to point
out that this could be due to the researcher’s poor measurement of the respondents’ travel costs. Finally,
Grijalva et al. (2002) observe rock climbers on multiple occasions before and after the implementation of a
policy for the management of rock climbing routes in natural parks in Texas, finding that pre-policy