The role of constants in discrete choice models: It’s not a constant sum game

Authors: John M. Rose, Antonio Borriello, Andrea Pellegrini, Daniel Masters

Released: August 2020

Keywords: Constants; Alternative specific constants; Discrete Choice Experiments; Prediction, forecasting

Abstract: Discrete choice experiments undertaken within the health economics domain involve respondents having to choose from amongst two or more alternatives. An examination of the literature suggests that models applied to unlabelled choice data often do not involve the estimation of a constant term, or when a no choice or status quo alternative is present, a single constant linked to the no-choice alternative. In this paper, we argue that choice models estimated using unlabelled choice data should always involve the estimation of constant terms for all but one alternative, without which significant biases can occur with respect to parameter estimates and willingness to pay outputs.

We argue that constants should be included for all but one alternative, even if they are not statistically significantly different from each other, or from zero. We posit that despite not offering a behavioural interpretation for such data, alternative specific constants can account for different data issues that may go undetected. We also recommend that papers reporting models estimated using discrete choice data provide more detail about the descriptive statistics of the data than currently appears to be the case. Finally, we propose that future efforts exploring the external validity of models estimated using discrete choice experiments focus on real markets in which observed changes are likely to occur, rather than attempt to simply predict the market shares of an existing market at one point in time

Full paper: BIDA-WP-2002 (PDF)

Are health preferences normally distributed? No, not normally ...

Author: John M Rose

Released: 2020

Keywords: Mixed Multinomial Logit Model; Logit Mixed Logit Model; Discrete Choice Experiment; Parametric distributions; Non-parametric distributions.

Abstract: The Mixed Multinomial Logit (MMNL) model has recently become the most dominant discrete choice model (DCM) used within the Health Economics literature. The model relaxes many of the most restrictive assumptions associated with other DCMs, including the allowance of unexplained taste heterogeneity. Nevertheless, estimation of the model requires a large number of decisions be made by the analyst, such as what parameters should be treated as randomly distributed over the population, and what distributional form these different random parameters should take. The most common assumption imposed is that of normally distributed random parameters for non-cost attributes. In this paper, we compare the standard MMNL model to a non-parametric version of the called the Logit Mixed Logit (LML) which makes no assumptions about the densities of the random parameter distributions. We find that preferences, as obtained from the MMNL model, are not normally distributed as is most often assumed.

Full paper: BIDA-WP-2001 (PDF)

The incorporation of budget constraints within stated choice experiments to account for the role of outside goods and preference separability

Author: John M Rose

Released: 26 February 2018

Keywords: Stated choice experiment, multiple discrete continuous extreme value model, household budget, preference separability, incentive compatibility

Abstract: Stated choice (SC) experiments are a popular means of collecting preference data for discrete alternatives. Many SC experiments include a no-choice alternative, either as an opt-out or as a status quo alternative. Even in the presence of a no-choice alternative, it is not clear that respondents understand fully the trade-offs being made between the choice alternatives, and other outside goods. As such, it is possible that many SC experiments are in violation of one of the central tenets underlying the micro-economic theory of demand. In this paper, we report two studies, in which respondents are required to indicate how they would readjust their household budget in light of choices made in an SC experiment. In both case studies, we find significant differences in the results obtained between traditional SC tasks and tasks involving the reallocation of household budgets. We argue that tasks involving the reorganisation of the household budget, are at least in part, more incentive-compatible given that respondents are faced with the financial consequence of their choices, as well as bound by their true budget constraint. The results of the paper reaffirms classical micro-economic demand theory with respect to SC experiments.

Full paper: BIDA-WP-1801 (PDF)

Explaining the components of the pleasure of driving using a partial least square-path model approach

Author: Antonio Borriello

Released: May 2016

Keywords: Driving pleasure; structural equation modelling; attitudes toward driving

Abstract: Instrumental factors, like cost and travel time, play a determinant role in transport mode choice. Irrespective of the purpose of travel, private car use remains the dominate mode of transport in highly developed countries, whilst simultaneously becoming more popular in less developed ones. In order to limit car use, it may be necessary to pay more attention to soft psychological factors, which may help to better understand the transport decision-making process and, as a result, suggest possible corrective actions and policies. There is general agreement that psychological factors, such as environmental awareness, safety, convenience and symbolic aspects of the car, influence the mode choice.

This study seeks to examine the latent construct ‘pleasure of driving’, meant as the act of driving itself. This feeling clearly fosters private car use, clashing with the public interest in reducing it. Using a component-based structural equation model on data collected among young commuters in Lugano (Switzerland), the paper explores the psychological aspects towards driving. Evidence shows that, through a hierarchical model, attitudes related to car performance (such as speed, design, brand), convenience (such as comfort, practicality) and emotion (such as relax, stress, boredom) are strongly connected with the pleasure of driving. A second important latent construct, which influences the private car choice, is related to ‘green’ attitudes, including attitudes towards environmental concerns and sharing vehicles.

Full paper: BIDA-WP-1802 (PDF)