Discrete Choice Experiments: Constructions and Properties

Abstract

This thesis explores the construction, optimisation, and evaluation of discrete choice experiments (DCEs), a powerful method for understanding individual decision-making across various fields such as healthcare, marketing, transportation, and policy-making. By closely replicating real-world scenarios, DCEs provide critical insights into preferences, allowing researchers, manufacturers, and policymakers to make informed decisions about the features of products and services based on perceived utility. The purpose of this study is to enhance the efficiency and practicality of DCEs by addressing the gaps in their design and proposing novel methodologies that reduce complexity while maintaining robust statistical properties.

The research is guided by four key questions: (1) evaluate current approaches for constructing efficient DCEs, (2) improve existing designs to enhance efficiency, (3) propose novel construction methods, and (4) compare the performance of new and existing designs under various criteria.

The first research question examines optimal (orthogonal) methods for constructing DCEs. These approaches are assessed for their ability to minimise the number of choice sets while preserving essential design properties such as orthogonality and level balance. A detailed review, as published in Heliyon (Alamri, Georgiou, & Stylianou, 2023), highlights the relative strengths and limitations of these methods, offering a practical reference for researchers. This work provides insights into sample sizes needed for reliable experimentation and identifies techniques that achieve high efficiency under specific experimental conditions.

The second research question focuses on improving D-optimal paired choice designs, particularly for main effects models. This work, published in Humanities and Social Sciences Communications (Alamri, Georgiou, & Stylianou, 2023), introduces a novel approach to address cases where the number of attributes equals the number of runs, as well as scenarios that involve attributes with varying levels. The proposed designs achieve up to a 37.5% reduction in the number of choice pairs at the cost of a maximum 10% loss in D-efficiency relative to an optimal design. These methodologies are further extended to accommodate higher-level attributes (l > 2), showcasing their versatility and practical applicability. Simulation studies confirm the robustness of the proposed designs, emphasising their potential to improve experimental precision while reducing respondent burden. By significantly reducing the complexity of experiments without compromising efficiency, these contributions provide practitioners with robust and scalable solutions.

The third research question explores the development of novel three-level choice designs using the Box-Behnken approach. This methodology is applied to estimate both main effects and partial second-order interactions under the utility-neutral multinomial logit model. The introduction of this design offers a significant reduction in the size of the choice sets required, preserving the precision of the experimental results. These designs address a critical gap in the literature by providing compact yet highly efficient frameworks for multi-level attribute studies. Simulation studies confirm the robustness of the proposed designs, emphasising their potential to improve experimental precision while reducing respondent burden.

This thesis advances the field of discrete choice experiments by proposing innovative methods that address long-standing challenges in design efficiency, scalability, and practicality. The contributions offer a systematic framework for researchers to construct efficient and accessible experiments across a wide range of applications. By integrating theoretical insights with practical solutions, this work establishes a robust foundation for future research, enabling the continued evolution of DCE methodologies to meet the demands of modern experimental research.