Search

An issue constantly raising concerns is the complexity of choice sets in stated choice experiments. Generally, designing choice sets requires trade-offs between the amount of information the researcher can derive, primarily through the number of attributes, and the measurement error caused by increased complexity. However, leaving specific attributes out might misrepresent the good in question and result in less information for decision-makers. An opportunity to reduce this trade-off is to apply partial profile designs. The basic idea is that not all attributes vary on all choice sets and that the overlap among attribute levels is controlled by the experimental design. Partial profile designs have rarely been used in environmental valuation, and if so, no split-sample approach was used to investigate performance differences. In this study, we aim to fill this gap by utilising both design types to value the good environmental status of the German North and Baltic Sea. We find that the partial profile design leads to significantly lower willingness-to-pay estimates, increased statistical efficiency as well as higher attribute attendance and less random choices. We thus recommend that researchers should consider partial profile designs more often as they promise clear benefits. • Two experimental designs – full vs partial profile – are compared. • Partial profile designs reduce complexity of choices. • The designs perform significantly different, for example, regarding attribute non-attendance. • WTP estimates are considerably larger in the full profile treatment. • Respondents do seem to care less about cost levels in the lower range of the price vector. [ABSTRACT FROM AUTHOR]

Attribute non-attendance in stated choice experiments (CE) has gained attention in literature, with some studies finding that not all respondents attend to all attributes. While the current studies show that taking non-attendance into account can significantly influence survey results, it is not yet clear what motivates respondents to ignore or pay less attention to some of the attributes. In the present study, we use 16 different split samples designed according to a design of designs plan, varying different aspects of dimensionality, i.e., the number of choice sets, the number of alternatives, or the number of attributes. Firstly, to analyse the relationship between stated attribute non-attendance and the design dimensions we test whether both are significantly associated. Secondly, we estimate equality-constrained latent class models with classes based on pre-defined rules to infer attribute non-attendance and analyse the influence of the design dimensions. Overall, the results indicate a rather weak relationship between stated or inferred attribute non-attendance and design dimensions. However, an interesting finding is that there is an association with the number of alternatives and with the number of sets. [ABSTRACT FROM AUTHOR]