Identifying and analysing dominant preferences in discrete choice experiments: An application in health care
Introduction
A methodological issue that has received little attention to date in the literature on discrete choice experiments is whether respondents trade-off attribute levels or have lexicographic preference orderings (where only one attribute matters and individuals do not trade). Assuming that individuals trade-off means that individuals are prepared to accept more of a specific good or characteristic in compensation for less of another. This is also known as compensatory decision making. In the case of a lexicographic ordering of goods or characterisitcs, an individual is not prepared to trade-off and so goods or characterisitics cannot be substituted for one another (non-compensatory decision making). In the case of a lexicographic ordering of a bundle of goods, there are no other bundles (apart from itself) to which it is indifferent. Indifference curves cannot be formulated as no trading takes place. Marginal rates of substitution are meaningless, and a lexicographic ordering cannot be represented by a utility function (Deaton & Muellbauer, 1980). These concepts underpin many methods used to elicit preferences based on economic theory.
Choice situations where individuals do not trade-off (non-compensatory decision making) are ignored by economic theory, yet have been the subject of much research in psychology where their existence is well established empirically (Gigerenzer & Todd, 1999; Payne, Bettman, & Johnson, 1993). If there is evidence of non-compensatory decision making then this may influence the effectiveness of policies to improve welfare (Earl, 1983). For example, if a policy is introduced to reduce waiting times then it may not improve individuals' utility if individuals have a minimum target reduction in waiting time, and the policy does not achieve this. A study that assumes that individuals trade-off may therefore overestimate the effectiveness of such policies since it assumes any reductions in waiting time, including those below the target, will increase utility. This therefore has important consequences when attempting to interpret the results of empirical work that elicits preferences using choice-based techniques.
The aim of this paper is to examine the extent to which respondents trade-off attribute levels in a discrete choice experiment. In particular, we examine the existence of dominant preferences, a special case of a lexicographic ordering. The paper uses data from a study that elicited the users' and non-users' preferences for different models of out of hours care provided by General Practitioners (GPs). After defining hierarchical choice, and discussing how it can be identified empirically, the paper discusses the reasons why such preferences exist. The empirical analysis illustrates these issues and tests for the existence of one form of hierarchical preferences. In particular, it uses a more strict empirical definition for hierarchical preferences, and examines potential reasons for the existence of such preferences. Discussion is focussed on the implications for future research on eliciting preferences using discrete choice experiments.
Section snippets
Defining hierarchical choice
The situation when preferences are not perfectly substitutable has been examined by several authors including Drakopoulos (1994) and Fishburn (1974). At one extreme of this limited substitutability are lexicographic orderings. This implies that there is an absolute order of preferences and precludes any degree of substitution between attributes. Following Drakopoulos (1994), let x and x′ be two alternatives defined as two bundles of attributes:
Identifying dominant preferences
Underlying the discrete choice method of conjoint analysis (where respondents are presented with a series of pairs of scenarios and, for each pair, asked to choose which they prefer), is random utility theory (Manski, 1977). This is the discrete choice analogue of consumer theory, and so assumes that individuals trade-off characteristics when making their discrete choice. The results of this form of conjoint analysis are usually interpreted as utility and marginal rates of substitution between
Reasons for the existence of hierarchical preferences
A main theme in the psychology and behavioural literature on decision making is that preferences are often constructed at the time they are being elicitied, and that decision makers have a variety of strategies that are used when constructing preferences (Payne, Bettman, & Johnson, 1992). This is especially the case for complex tasks and those that involve `conflict' among values, such as tasks involving trade-offs. Such conflicts may be confronted and resolved, or avoided, reduced or modified
Background to the out of hours study
The empirical analysis of those with dominant preferences is based on a study of the preferences of individuals (both users and non-users) for different models of out of hours care provided by GPs (i.e. primary health care delivered during evenings, nights or weekends) (Scott, Watson, & Ross, 2002). The study used a discrete choice experiment to analyse the preferences of parents/guardians of children under 13 years old for out of hours care (frequent users of out of hours care). 6304
Results
Using the method to identify individuals with a dominant preference in Section 5.2, 58% (2255/3891) of respondents had a dominant preference for `whether the doctor seemed to listen'; 10% (402/3891) had a dominant preference for waiting time, and; 7.3% (284/3891) of respondents had a dominant preference for being able to see a GP from their own practice. However, a proportion of these did not rank the attribute as the most important. This introduces doubts as to whether a dominant preference
Discussion
This study has examined the existence and reasons for dominant preferences in the context of a discrete choice experiment examining individuals' preferences for GP out of hours care. Forty five per cent of respondents had a dominant preference. Their responses were analysed separately, providing richer information about the nature of preferences. There is evidence that both past experiences and the complexity of the decision task influenced dominant preferences. Those who had never experienced
Acknowledgements
Many of the ideas in this paper arose from work on conjoint analysis in HERU and so the author is indebted to discussions with Mandy Ryan, Andy Lloyd, Shelley Farrar, Emma McIntosh, Fernando San Miguel and Sandra Vick. The out of hours study was funded by the NHS R&D Primary Secondary Interface Programme. This study would not have possible without the contributions of Stuart Watson and Sue Ross. Thanks also to Nicola Torrance for her contribution to this study. The Health Economics Research
References (30)
Choice by elimination
Journal of Mathematical Psychology
(1972)- Bryan, S., Buxton, M., Sheldon, R., & Grant, A. (1997). The use of magnetic resonance imaging (MRI) for the...
- et al.
Understanding consumers' preferences for dental service
Journal of Health Care Marketing
(1993) - et al.
Economics and consumer behaviour
(1980) The nature of individual preferences: a prologue to Johannesson, Jonsson and Karlsson
Health Economics
(1997)Hierarchical choice in economics
Journal of Economic Surveys
(1994)Ethical preferences and the assessment of existence values: does the neoclassical model fit?
Northeastern Journal of Agricultural and Resource Economics
(1986)Lexicographic orders, utilities and decision rules: a survey
Management Science
(1974)- et al.
Fast and frugal heuristics: the adaptive toolbox
Decisions with multiple objectives:preferences and value trade-offs
A new approach to consumer theory
Journal of Political Economy
Operationally relevant characteristics in the theory of consumer behaviour
Cited by (143)
Financing electricity access in Africa: A choice experiment study of household investor preferences for renewable energy investments in Ghana
2021, Renewable and Sustainable Energy ReviewsA study of the willingness of Spanish drivers to pay a premium for ZEVs
2021, Energy PolicyThe Dynamics of Refugee Return: Syrian Refugees and Their Migration Intentions
2023, British Journal of Political ScienceWhat are the job attribute preferences of physicians and nurses in Türkiye? Evidence from a discrete choice experiment
2023, Human Resources for Health