Visitors’ willingness to pay for ecosystem conservation in Grenada

Abstract

AbstractUnderstanding visitors’ willingness to pay (WTP) for natural ecosystems can assist policymakers in managing the environmental assets that form the foundation of nature-based tourism. To earn the highest economic return from limited budgetary resources, tourist-dependent destinations must identify sustainable sources of conservation funding and prioritize spending across ecosystems and initiatives. The purpose of this study is to assess the potential for visitor donations to serve as a source of conservation funding and investigate preferences and willingness to pay for multiple ecosystems concurrently, using data from a stated preference survey administered in Grenada, a small island state in the southern Caribbean that relies on nature-based tourism. Results from a discrete choice experiment and a contingent valuation exercise reveal a reluctance to donate to conservation initiatives by more than half of respondents. Visitors place the highest value on improvements to coral reefs, followed by rainforests, mangroves, and beaches. Average willingness to donate to a conservation trust fund ranged from US $8.00 to US $15.00, and from US $17.00 to US $52.00 for conservation plans targeting improvements to specific ecosystems. Few factors are found to be associated with willingness to pay. Scale heterogeneity may be a more significant source of variation than preference heterogeneity.Keywords: Willingness to paycontingent valuationdiscrete choice experimentecosystem conservationGrenadatourism Disclosure statementThe authors report there are no competing interests to declare.Notes1 Johnston et al. (Citation2017) recommend the use of a single, binary-choice WTP question for the valuation of public goods using CVM to maintain incentive compatibility.2 Survey participation was voluntary. No identifying information was collected from respondents. An introductory informed consent script informed respondents of the anonymous and confidential nature of the survey and their freedom to end the survey or refuse to answer any questions. Written consent was provided by respondents through survey completion.3 The D-efficient choice experiment design was created in SAS using the %choiceff and %mktblock macros (Kuhfeld, Citation2002).4 MXL models were specified with normal distributions and estimated using a panel specification using 500 draws from a Halton sequence. All regression analyses performed using NLOGIT 6.5 The CVM exercise was presented earlier in the survey and is shorter and less burdensome than the DCE.6 The weight command in SAS 9.4 was used to calculate weighted descriptive statistics and CVM response distributions. For logit regression analyses of CVM and DCE response data, sampling weights were applied using the ;wts specification in NLOGIT version 6, where weights were calculated using the subsample of completed responses for each valuation exercise. All calculations and analyses were also conducted using the unweighted sample, with negligeable impact on results. We present the weighted sample results to better reflect characteristics and preferences of the Grenada visitor population.7 These “other” reasons mostly pertained to tourists having already donated to other causes and feeling they had already spent enough money in Grenada.8 Exclusion of protest zero responses increases mean WTP by roughly $1.9 This “pooling back” of responses maintains the monotonicity assumption and dampens the resulting WTP estimates as responses to the pooled back bid values ($2 and $25) are treated as belonging to the next-lowest bid values ($1 and $20). See Haab and McConnel (2002, p. 69).10 Numerous model specifications were estimated, including a base model (fee value only), parsimonious models with select covariates and fee value, and more complex models, including specifications estimated using stepwise regression procedures.11 For example, respondents who are willing to return to Grenada were also generally willing to recommend Grenada to others, higher income respondents tended to be more educated, older, married, visit Grenada for shorter periods and have more Caribbean travel experience. Caribbean residents were more likely to travel to Grenada for business purposes and have more Caribbean travel experience, etc.12 Indicator variables for these countries of origin and return visitation were statistically insignificant in all model runs, hence we do not include them in our results.13 Like constant terms in linear regression models, ASCs capture the effects of non-observable variables on respondent choices. Including two ASCs for the opt-in alternatives provides more a more granular view of choices than a single ASC assigned to the opt-out alternative with no impact on other coefficients.14 All models were estimated with the full sample of DCE respondents, the subsample of respondents excluding the protest zero respondents from the CVM exercise and the subsample of respondents excluding respondents who did not select any of the conservation plans in the choice experiment (those who chose “neither plan” for all three choice panels). Analysis of these subsamples did not produce any meaningful changes in the signs or significance of the associated coefficients, and as expected, resulted in marginally higher WTP estimates.15 Models where error components were assigned to the donation plans (Plan A, Plan B) and those that mimic a nested decision structure - whereby respondents first decide whether to donate and if so, choose between Plan A and Plan B - were also estimated. Coefficient estimates were nearly identical, but these other structures produced moderately lower overall fit. Results available from the authors.16 The RPLCL specification also supports the existence of two preference groups but does not reveal significant preference heterogeneity for attributes within each class. No additional gains in fit were captured by the more complex RPLCL.