Choice experiment

from Wikipedia, the free encyclopedia

Choice experiment or discrete choice experiment (German: Discrete decision experiment or discrete choice experiment ) is a selection or decision based method for the analysis of economic preferences . Social science survey data on hypothetical selections ( expressed preferences ; choice experiment in the narrower sense) or data on real selections ( revealed preferences ) can be analyzed. The statistical ( econometric ) approaches are called discrete decision models .

In a choice experiment, the respondents are asked to select one of at least two different scenarios (alternatives). Each of the scenarios is described by a number of properties (attributes). Through the econometric analysis of many selection decisions, the relative influence of the properties on the selection behavior can be determined. If a property is an amount of money to be paid, marginal willingness to pay can be assessed.

Fields of application

Choice experiments are mainly used in the fields of marketing , transport, environmental economics and health economics . In the field of environmental economics, the choice experiment is used to assess environmental economics to determine the overall economic value .

Theoretical foundations

Choice experiments are based on two economic theories - the theory of consumption by Kelvin Lancaster and the random utility theory ( English random utility theory , short RUT ) by Daniel McFadden . Lancaster's consumption theory states that people do not derive their benefit from the consumption of abstract goods, but from the properties of these goods ( attributes ). According to the random utility theory, people always select the good from the goods available to them, to which they ascribe the greatest benefit (utility maximization calculus). The utility ascribed to the good is thought of in the random utility theory as consisting of two components: a systematic component, which depends on the properties, and a random component. The systematic benefit component can be deduced from the observation of selections. The random benefit component cannot be observed in this way in the good and is an error term in the statistical sense . Depending on the assumption about the distribution of the error terms, different statistical models are used in the evaluation of choice experiments.

Central terms

Alternative , option : a good or a scenario, described by a series of attributes. A distinction is made between generic and labeled alternatives (the former have no name, but are only called "Alternative A", "Alternative B" etc. or "Scenario A", "Scenario B", the latter have a specific name, such as for product brands or transport modes). As a rule, one of the alternatives is a status quo or opt-out alternative that appears in all choice cards.

Attribute : Property of the considered good / scenario, the levels of which vary across the goods / scenarios to be compared.

Choice Card : material embodiment of a selection option in the form of a card or the like that shows the attributes. Sometimes a choice set is also referred to as a choice card if the options are summarized on a card (or similar).

Choice Set : Comparison of two or more options (choice cards) from which a choice can be made. Since the respondents are often presented with more than one choice set in choice experiments, a distinction must be made between the number of selection decisions and the number of respondents.

Binary discrete decision models

Various models are used to analyze the results of discrete choice experiments. The choice of the model is linked to certain assumptions, especially about the distribution of the error terms, but also e.g. B. the properties of the sample. Common models include probit models , multinomial logit models , conditional logit models, hierarchical logit models, and mixed logit models. Since the dependent variable in discrete decision models is always binary (yes-no), the maximum likelihood method is usually used to estimate the model . Various authors have shown that discrete decision models can also be evaluated using PLS path modeling ( PLS for Partial Least Squares , see Partial Least Squares Estimation ).

execution

There are several steps to be taken when performing a choice experiment. The following have been highlighted in the literature:

  • Analysis of the economic good to be considered;
  • Identification of relevant attributes;
  • Test planning:
    • if necessary, decision on the model context to be assumed;
    • Identification of a statistically efficient design from combinations of alternatives;
    • Division of the combinations into choice sets;
  • Implementation of the choice experiment;
  • Statistical evaluation.

Similar methods

Choice Experiment is a further development of the conjoint analysis known from marketing . A functionally similar method is the contingent valuation method .

literature

  • Jordan J. Louviere, David A. Henscher, and Joffre D. Swait: Stated Choice Methods: Analysis and Application . Cambridge University Press, Cambridge, New York 2000, ISBN 978-0-521-78830-4 .
  • Kelvin Lancaster: Consumer Demand: A New Approach . Columbia University Press, New York 1971, ISBN 978-0-231-03357-2 .
  • Emily Lancsar and Jordan Louviere: Conducting Discrete Choice Experiments to Inform Healthcare Decision Making . In: PharmacoEconomics . tape 26 , no. 8 , 2008, p. 661-677 , doi : 10.2165 / 00019053-200826080-00004 .
  • Joseph F. Hair, Christian M. Ringle, Siegfried P. Gudergan, Andreas Fischer, Christian Nitzl and Con Menictas: Partial least squares structural equation modeling-based discrete choice modeling: an illustration in modeling retailer choice . In: Business Research . tape 12 , no. 1 , 2019, p. 115-142 , doi : 10.1007 / s40685-018-0072-4 .