Environmental Valuation with Discrete Choice Experiments in R

This chapter provides an overview of the key steps of conducting a discrete choice experiment (DCE). We begin with the research questions guiding the study design, move through several design issues related to the DCE and the questionnaire, and conclude with important considerations for the model estimation and post-estimation analysis. The main objective of this chapter is to highlight the decisions involved in conducting a DCE and the potential consequences of these choices for the realisation of the final study and subsequent results. Instead of offering step-by-step guide for the implementation of DCEs, we draw on our research experience in this field to offer insights and practical knowledge for you to consider in a DCE project.

This chapter provides an overview of the random utility maximisation (RUM) model, reviewing its assumptions and delving into its theoretical foundations. We explore the multinomial logit (MNL) model, which is widely used in DCE literature due to its many advantages. These include its robustness, ease of estimation, and straightforward interpretation, with closed-form choice probabilities that simplify calculations. We also review advanced specifications of the mixed logit model, including the random parameters logit (RP-MXL) and latent class (LC-MXL) models, and walk you through the most common goodness of fit models used in DCE research.

This chapter introduces the case study used throughout the book to guide the reader through the steps of conducting a discrete choice experiment in R. The case study is based on the paper published as Meyerhoff et al. (2010), studying the landscape externalities of wind power generation. The original study was a choice experiment surveying preferences on the design of onshore wind farms in two regions in Germany, and the details of the study, particularly the attributes and their levels, demonstrate the real-world applications of discrete choice experiments. In this chapter, we familiarise readers with the wind power case study to prepare them for the estimation, data collection and analysis phase in the following chapters of the book.

This chapter guides you through the process of using R Shiny to collect your own data for your DCE. We provide a sample code on how to implement a bare-bones DCE survey using Shiny, Google Sheets and Shinyapps.io, walking you through the following steps: (1) setting up the data storage location, (2) generating the choice tasks, and (3) creating the Shiny app. To create the Shiny app, we review the user interface, the server function, and the Shiny app function. Finally, we discuss the advantages and disadvantages of using R Shiny to implement DCE surveys.

This chapter highlights the importance of data exploration as a preliminary step in building choice models. By examining choice patterns and uncovering key influencing factors, you can establish a solid basis for the subsequent model development. This initial step involves a series of exploratory data analysis techniques that are simple yet informative. In this chapter, we demonstrate how to use R to effectively conduct this exploration, transforming raw data into meaningful insights. By creating summary tables and examining data visualisations, researchers can gain a deeper understanding of their data, paving the way for a more effective analysis and model building.

This chapter explores Maximum Likelihood (ML) estimation, a statistical method used to estimate parameters of a given probability distribution. We begin with an introduction to the fundamental components of ML estimation, including the likelihood function, the density function, and the process of identifying parameter values that maximise the likelihood of the observed data. This chapter also covers numerical optimisation methods, both gradient-based and non-gradient, for situations where analytical solutions are impractical. We address sample variation in statistical estimation, highlighting the issues that may arise when relying on a single sample to infer population parameters, and review the use of simulation techniques, such as generating artificial datasets, to evaluate the reliability of these estimates.

This chapter covers the estimation of discrete choice models. We begin with a simple Multinomial Logit (MNL) model (with and without unobserved preference heterogeneity) and highlight the importance of optimisation diagnostics, review goodness-of-fit indicators and the identification of outliers, and provide insights into the interpretation of estimates. Progressing to more advanced topics, we continue with the Random Parameters Mixed Logit (RP-MXL) model, addressing both uncorrelated and correlated coefficients, and discuss different parameterisation approaches such as the preference space and the willingness-to-pay space. We conclude with an analysis of Latent Class Mixed Logit (LC-MXL) models and a discussion of extensions of RP-MXL and LC-MXL models.

This chapter highlights the importance of the post-estimation analysis in extracting meaningful insights from discrete choice models. While constructing and estimating a model is an essential step, the true value of a DCE-based model lies in interpreting and applying its results. This chapter demonstrates how to translate model outputs, such as marginal willingness to pay and changes in consumer surplus, into practical, policy-relevant insights that enhance decision-making. We explore issues relevant to these outputs and provide guidance on how to accurately report uncertainty in your results.

This book has provided a foundational overview of designing, collecting, and analysing discrete choice experiments (DCEs) in environmental economics using R. However, it is not intended to be a definitive guide on the subject: mastery of DCEs will require further reading and exploration beyond the content covered in this book. We carefully selected what we believe to be the essential elements of a DCE project to provide a solid foundation, while aiming to spark curiosity and encourage further inquiry. As we conclude this book, we offer some final thoughts and additional pointers to consider as you continue your journey of mastering DCEs.