The Non-Market Value of Biodiversity Enhancement in New Zealand's Planted Forests

Abstract

This study investigates the non-market value of biodiversity enhancement in New Zealand’s planted forests using the stated choice experiments (CE) approach. This study focuses on two issues. One issue is policy orientated where we estimate the non-market value of biodiversity enhancement and the determinants of this value. The other issue is about the neutrality of major experimental design criteria used in CE. Specifically, we examine the impact of using different criteria on attribute non-attendance, choice variability, choice determinism and learning.

To estimate the non-market value of biodiversity enhancement, a random parameters logit model with error components is used to analyse choice data collected from 209 respondents across New Zealand. The panel nature of the choice data set is exploited to calculate the marginal willingness-to-pay (WTP) for environmental attributes of each respondent. Panel random-effects regression models are subsequently employed to determine the factors that influence individual-specific WTP values. Results suggest that New Zealand taxpayers would be willing to pay $26.5 million per year for five years for a proposed biodiversity enhancement programme. Random effects regression analysis suggest that respondents living close to large planted forests (i.e., less than 10 kilometres away) would pay more for the programme.

To study whether the selection of experimental design criterion affects attribute non-attendance and choice variability, we analyse a balanced sample with split designs. The balanced sample is composed of 1509 choice observations equally distributed across three experimental designs, namely: orthogonal, Bayesian D-efficient and optimal orthogonal. Results from latent class logit analysis suggest that tasks derived from the Bayesian D-efficient design (BDD) criterion are more attended than those derived from orthogonal and optimal orthogonal designs. Heteroskedastic logit analysis indicates that, unlike the two other designs, higher choice task complexity (as measured by entropy proxies) in the BDD does not increase choice variability of respondents. This is indicated by the absence of a significant increase in the variance of the Gumbel error in the choice data collected using BDD unlike the data collected using the two other criteria.

To study whether the three experimental designs vary in terms of choice determinism and task order effects, a separate analysis of the balanced data set using heteroskedastic logit models is undertaken. Results show that higher levels of choice task complexity (as measured by attribute dispersion proxies) in BDD contribute to increasing choice determinism of respondents but not in the orthogonal design. Choice data collected using BDD choice tasks exhibit a steady learning effect, unlike the other designs which do not exhibit any form of continuous learning.

We conclude that the BDD criterion provides choice tasks that are superior compared to the other two design criteria. Choice data collected using this criterion has a higher quality as indicated by more attended choice tasks, lower choice variability and a pattern of continuous learning. These results point to a higher behavioural efficiency of respondents in evaluating complex choice tasks. However, these results might be specific to the choice data collected in this current study. We suggest that future studies should further investigate the impacts of different experimental designs to verify the findings of this study.