Presented by: 
Debbie Street, University of Technology, Sydney
Date: 
Tue 14 Aug, 3:00 pm - 4:00 pm
Venue: 
Room 442, Building 67

Experiments which show people a number of options, be they meals, computers or medical treatments, and ask them to choose the option they think is best, are called choice experiments. Choice experiments provide a cost-effective way to predict the likely market share of a new product or  the acceptability of a proposed government policy.

In most choice experiments, the options to be considered are described  by a number of attributes, and each attribute has a number of possible levels. Each option is described by a level of each of the attributes and each choice set has a fixed number of options in it, typically between 2 and 6. Each respondent will be shown a number of choice sets in turn and asked to choose their preferred option from each of the choice sets.

The design problem, then, is two-fold. Which attribute-level combinations should be used as the options in a choice experiment? How is it best to arrange these attribute-level combinations into choice sets so that good predictions can be made using a reasonable number of choice sets?
 
In this talk we start by describing some choice experiments used in areas of health economics.  We then show how orthogonal arrays, BIBDs and Latin squares can be used to design choice experiments which have both good statistical efficiency and good respondent efficiency.