How effective are marker variables at imputing attitudes? An external evaluation using vehicle ownership models
Author(s)
Soria, Jason
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Information obtained from travel behavior surveys is used to inform major policy decisions affecting many people and industries. Such surveys are generally designed with minimizing respondent burden in mind, thereby reducing the number of topics a single survey can address. Attitudinal variables, although inarguably important to explaining behavior, are generally excluded from such surveys for that reason among others. We demonstrate the possibility of using a small set of attitudinal statements to adequately represent factor scores obtained from a much larger set. Specifically, we utilize a completed factor analysis to identify “marker” (important) attitudinal items, apply a Random Forest approach to those items to estimate (“impute”) the factor scores assuming that the other items are unavailable, develop vehicle ownership models without and with the imputed attitudes, then compare the model estimation results and their predictive power. Attitudes provide more information than traditional modeling variables by themselves do, and particularly help predict zero-vehicle households better – likely because owning vehicles is the norm in the study area (Georgia, USA), and the attitudes help identify the exceptions. Comparing models with imputed attitudes to one using the original factor scores, those with imputed attitudes predicted vehicle ownership with less accuracy, but better than the model without any attitudes. This study shows that marker variables can successfully be utilized to add valuable attitudinal information to otherwise conventional travel behavior models. We discuss the implications for future survey development – specifically, the prospect that future surveys can include more attitudinal topics without worsening respondent fatigue.
Sponsor
Date
2024-01
Extent
Resource Type
Text
Resource Subtype
Post-print
Rights Statement
Unless otherwise noted, all materials are protected under U.S. Copyright Law and all rights are reserved