Constructing an Optimal Orthogonal Choice Design with Alternative-Specific Attributes for Stated Choice Experiments

The stated choice (SC) experiment is generally regarded as an effective way to obtain data for discrete choice analysis. The SC experimental design method, which determines the rule for allocating different levels to each attribute in choice situations, has a great impact on parameter estimation. The optimal orthogonal choice (OOC) design is one of the most efficient SC designs, and its use can achieve more reliable parameter estimates with an equal or lower sample size than other methods. However, the OOC design can be applied only to utility models with generic attributes, as using it to assign alternative-specific attribute levels is not discussed fully in literature. This paper provides a method for extending the use of the OOC design to alternative-specific attributes. Column vectors for alternative-specific attributes were introduced, and the value of each vector was forced to be orthogonal with other generic attributes in the same alternative. In this way, orthogonality of the OOC design was kept within an individual alternative but not necessarily across alternatives. The proposed method was compared with traditional orthogonal design and with D-efficient design (another state-of-the-art efficient design method). Three experiments using field data on mode choice were conducted. The results showed that both the proposed method and the D-efficient design had a higher efficiency than the orthogonal design. In addition, under the complex experimental setting in the real world, the proposed method outperformed the D-efficient design in the sense that almost the same efficiency could be obtained while multiple iterations for an optimal solution were avoided.