TY - JOUR
T1 - Does uniform design really work in stated choice modeling? A simulation study
AU - WANG, Donggen
AU - Li, Pengfei
N1 - Copyright:
Copyright 2017 Elsevier B.V., All rights reserved.
PY - 2005
Y1 - 2005
N2 - Stated preference/choice methods have become established modeling tools for transportation studies. At the core of these methods is experimental design, which usually employs orthogonal arrays. This conventional design method, however, imposes constraints on the number of attributes and/or levels that may be included in a study, for the number of profiles it provides may be too large to handle, particularly when the number of attributes and/or levels is large and sometimes it cannot provide a reasonable number of profiles. Recently, Wang and Li (2002, Geographical Analysis, 34, 350–362) introduce a new design method that may overcome the shortcomings of the conventional approach: uniform design. The number of profiles generated by uniform design is substantially less than that by orthogonal design, particularly for uneven numbers of levels. Further, uniform design can provide solutions to cases where the orthogonal design cannot. This paper presents a simulation study to analyze the statistical properties of uniform design and to compare uniform design and orthogonal design on these properties. Specifically, we study the ability to pick up the significant variables and the accuracy of parameter estimation and model prediction by uniform design and compare them with that by orthogonal design. The simulation results show that parameters estimated from uniform design are unbiased. The efficiency of the parameter estimations of uniform design is comparable to that of orthogonal design.
AB - Stated preference/choice methods have become established modeling tools for transportation studies. At the core of these methods is experimental design, which usually employs orthogonal arrays. This conventional design method, however, imposes constraints on the number of attributes and/or levels that may be included in a study, for the number of profiles it provides may be too large to handle, particularly when the number of attributes and/or levels is large and sometimes it cannot provide a reasonable number of profiles. Recently, Wang and Li (2002, Geographical Analysis, 34, 350–362) introduce a new design method that may overcome the shortcomings of the conventional approach: uniform design. The number of profiles generated by uniform design is substantially less than that by orthogonal design, particularly for uneven numbers of levels. Further, uniform design can provide solutions to cases where the orthogonal design cannot. This paper presents a simulation study to analyze the statistical properties of uniform design and to compare uniform design and orthogonal design on these properties. Specifically, we study the ability to pick up the significant variables and the accuracy of parameter estimation and model prediction by uniform design and compare them with that by orthogonal design. The simulation results show that parameters estimated from uniform design are unbiased. The efficiency of the parameter estimations of uniform design is comparable to that of orthogonal design.
KW - Experimental design
KW - Orthogonal design
KW - Stated choice method
KW - Stated preference method
KW - Uniform design
UR - http://www.scopus.com/inward/record.url?scp=85009569227&partnerID=8YFLogxK
U2 - 10.1080/18128600508685652
DO - 10.1080/18128600508685652
M3 - Journal article
AN - SCOPUS:85009569227
SN - 1812-8602
VL - 1
SP - 209
EP - 221
JO - Transportmetrica
JF - Transportmetrica
IS - 3
ER -