Enhancing Transportation Data Accuracy and Integration: Techniques for Small Area Estimation and Data Linkage

Abstract

Reliable transportation data is essential for the development, planning, and management of transportation systems. Traditional data collection methods, such as travel surveys, often lack the necessary granularity and context, making it difficult to fully understand the underlying reasons for travel behaviors and patterns. This dissertation proposes methods for enhancing transportation data integration and reliability, including small area estimation and mobile device data linkage. The research consists of four chapters, each addressing different aspects of data integration and analysis. The first chapter applies small area estimation techniques to combine multiple traditional surveys and census data to estimate the total number of household trips at the census tract level. This chapter provides a comprehensive SAE modeling framework within the transportation data context by integrating direct and synthetic estimations to produce accurate statistics. It includes various small area estimation techniques, such as regression-based models and population synthesis for areas with zero samples, as well as the Fay-Herriot model for areas with small samples. The second chapter focuses on integrating traditional surveys with modern GPS- derived surveys. The feasibility of using data record linkage to link similar travel behavior/trips to identify similar persons is explored. Two techniques are proposed: probabilistic record linkage and similarity-based approach. The objectives are to explore the feasibility of using them in transportation contexts and comparing the effectiveness of these methods. The third chapter extends the second chapter by focusing on optimization techniques for the methods proposed, particularly in the context of high-uncertainty datasets where the ground truth is unknown. This chapter offers solution techniques to optimize overall accuracy, including graphical solutions to identify discrepancies between the two datasets in the absence of ground truth dataset. The fourth chapter presents a practical application that integrates the optimization techniques introduced in Chapter 3 with the linking approaches developed in Chapter 2. Specifically, the optimization techniques are employed to estimate the discrepancy between two datasets: a self-reported trip survey and a GPS-derived trip survey. The estimated discrepancy is subsequently employed to determine the optimal matching parameters, which include the offset, scale, and threshold value, for implementing the linking approaches. In addition to the linkage application, this chapter conducts a statistical comparison of the two linking approaches by analyzing the results of the linked groups. The comparison focuses on aggregated trip characteristics, such as trip duration, trip distance, and the total number of trips per individual. Statistical metrics, including the t-test and distribution overlap values, are applied to evaluate the differences between the matched groups.