Use of a genetic algorithm for building efficient choice designs

Choice design building based on D-error minimisation can be accomplished either by using predefined β values or by assuming probabilistic distributions for them. Several mathematical techniques have been used for both approaches in the past, resulting in algorithms that obtain efficient designs, which guarantee the high quality of the information that will be provided by the respondents. This paper proposes the use of a genetic algorithm for dealing with the problem of building designs with minimum D-error, describing the technique and applying it successfully to several benchmark cases. Design matrices, D-error values, percentages of level overlap and computation times are provided for each case.