Preference instability in stated choice surveys: More evidence

Stated choice surveys are widely used to inform policy making across a broad range of domains, including transport and environmental planning, health, and marketing. It is common practice to present respondents with multiple choice tasks in order to increase the amount of information collected. A central concern in this context is whether the same underlying preferences drive choices across all tasks, as violations of this assumption may affect the reliability of the survey. There are several reasons to expect that this may not hold: preferences may vary due to fatigue, learning effects, anchoring, or other forms of inertia. In particular, the role of the first choice task has attracted considerable attention. Some authors argue that it is the only valid task, while others view it primarily as a learning exercise. In this paper, we deliberately take no position on the validity of these competing viewpoints. Instead, we question the robustness of the commonly reported finding of stable VTT estimates across choice tasks, noting that much of the existing evidence relies on models that do not incorporate random preference heterogeneity. We therefore propose and test a range of more advanced model specifications to assess whether VTT estimates are stable across individual choice tasks and groups of tasks. Using three different datasets, we identify key modelling features that influence whether preference instability is detected. While estimates are stable for some attributes and choice tasks, they can vary by about 30% for others, and by up to 60% when using latent class models. These differences have important implications for data collection and model estimation in stated choice research, which we discuss in the concluding section.