engleski

Methodological issues in choice experiments: Biasedness of willingness-to-pay estimates and attribute multidimensionality

Discrete choice experiments (DCEs) are a popular methodology for eliciting human preferences and willingness-to-pay (WTP) measures. However, there are a number of methodological challenges, such as choice task complexity, hypothetical bias and non- attendance to the choice attributes, which have been examined in the two surveys included in this dissertation. One survey was conducted at the School of Dental Medicine, University of Zagreb, to estimate patients’ WTP for dental care delivery at the school clinic and investigate the validity of derived estimates. Among the attributes included in a DCE, treatment explanation was on average the most valued attribute, followed by dental staff behavior, dental care provider and waiting time in the office. As patients often have diverse expectations and preferences for health care interventions, it is important to consider heterogeneity in their preferences when estimating WTP for health care. Mixed logit and latent class analyses detected a large heterogeneity in patients’ preferences. Four classes of patients with distinct preferences for dental care were identified. Older and/or more educated patients tended to give relatively less importance to treatment explanation, but were more attentive to interpersonal characteristics. Higher education was also associated with a higher propensity to substitute faculty dental care with private care providers. Different survey design features, such as the opt- out option and the selection of attributes and their levels, may affect the choice behavior of respondents and estimated utility weights. However, not giving an opt-out option to respondents did not affect the preferences for dental care delivery and response error variance. Furthermore, adding the cost attribute to a DCE task did not change the rank order of other attributes and their utility weights did not differ significantly, indicating that respondents were willing to trade off cost with other dental care attributes. However, adding the cost attribute increased the response error variance, which could be best explained by a higher cognitive burden from adding an extra attribute to the choice task. We did not find any strong evidence that the effect of including an extra cost attribute is any different from the expected effect of including any other choice attribute ; therefore, its influence may not be as relevant as some of the previous studies may have suggested. Choice studies have shown that respondents often ignore one or more attributes when making choices and may ignore the cost attribute itself, thus precluding the calculation of marginal WTP values. The contingent valuation (CV) method directly estimates respondents’ WTP, thereby avoiding this issue. The DCE method yielded significantly higher welfare estimate for dental care than the CV method. While strategic behavior in direct WTP estimates was not evident, almost every second DCE respondent appears to have ignored the cost attribute, which may explain the discrepancy in welfare estimates between the two methods. The findings suggest that direct elicitation of WTP through the CV method produced more realistic estimates, and raise concerns about the derived WTP measures in DCE studies that did not consider non-attendance to the cost attribute during the estimation process. However, a typical CV study values a good as a whole and offers no information about the value of the individual attributes that comprise a good. To deal with this limitation, the CV task was accompanied by the best-worst choice (BWC) task which provided information used to derive the attribute-specific WTP values from the holistic WTP values estimated through the CV method. The proposed methodology could be a valuable alternative to traditional DCEs when the choice behavior of respondents raises concerns over the validity of estimated welfare measures. Choice contexts often cannot be easily described with a restricted number of attributes, which may lead to the omitted variable bias. On the other hand, asking respondents to evaluate large number of attributes imposes high cognitive burden and stimulates them to simplify the choice task, which may contribute to an increased error variance and affect the validity of utility estimates. Environmental valuation generally involves trade- offs between complex goods and services. How to describe a complex valuation context is an important consideration, and was investigated in a second survey that evaluated the preferences of trail users for different resource, social and managerial conditions in the Medvednica Nature Park. Two different choice experiments that differed in the provision of information about the recreational setting were designed. One DCE used multidimensional attributes to keep the number of choice attributes manageable for respondents, while considering all relevant aspects of visitors’ experience. The other DCE used a greater number of more specific, unidimensional, attributes to describe the same recreational setting, but implemented a partial profile design in which only a subset of attributes appeared in each choice set. Perceived difficulty of the choice task, self-reported choice certainty and choice consistency were similar across the two experiments. Heterogeneity in preferences and scale was detected in both experiments. Indications of non-compensatory behavior, and greater error variance among less experienced trail users were found in the partial profile experiment with more specific trail attributes, but not in the experiment with multidimensional attributes. Environmental conditions are often inter-related and as such cannot be simultaneously included in a DCE. A linked structural choice model was designed to deal with the correlated or causally-related DCE attributes as well as with the ambiguity when using multidimensional choice attributes, since respondents might be focused on different aspects of those attributes when making choices. A structural choice model was used to link the two choice experiments and investigate the amount of variability in multidimensional attributes (crowding on trail and intensity of road traffic, evaluated within the DCE) explained by each of its sub-dimensions (i.e. specific crowding-related conditions, evaluated within object case best- worst scaling (BWS) experiment). The findings indicate that visitors’ preferences for crowding were mostly influenced by perceptions of ‘trail use level’ and ‘noise from road traffic’. Our approach offers richer information to policy makers than other solutions to the attribute correlation/causality issues in complex choice studies, namely the exclusion of some relevant attributes or simply their aggregation into a single composite attribute.

discrete choice experiment ; willingness-to-pay ; choice attributes ; cognitive difficulty ; contingent valuation

nije evidentirano