Numerical Analysis of the Validity of Uniform Design in Stated Choice Modeling

This paper examines the statistical properties of uniform design for stated choice modeling. Estimation efficiency, prediction efficiency and test power (i.e., the ability to pick up significant effects or exclude insignificant effects) are selected as measures of statistical properties. Both uniform design and orthogonal design are used to generate profiles. The major experimental design strategies for multinomial logit models including shifted pairs, k L , J 2 block, all pairs and McFadden’s sampling rule, are used to construct choice sets from the profiles generated by both orthogonal design and uniform design. Monte Carlo experiments are used to generate models, whose parameters vary in scale. The results show that the performance of uniform design in stated choice modeling is comparable to that of orthogonal design.