Statewide comparison of origin-destination matrices between California travel model and twitter

Social media data can potentially be a good source of information for travel behavior and could potentially become a record of longitudinal travel behavior. Before judging their suitability as a source of data, information extensive comparisons with more traditional travel demand models are needed. In this paper we create an origin-destination (OD) matrix for the entire State of California using data from Twitter and compare it to the OD matrix of the calibrated California Statewide Travel Demand Model (CSTDM). The comparison uses a spatial lag Tobit model and latent class regression models to compare OD matrices taking into account the censored distributions of trips and spatial heterogeneity. In the process we also develop a conversion technique between Twitter-based ODs and CSTDM ODs. Our findings include discovery of four different classes of relationships between CTDM ODs and Twitter-based ODs producing different unit-contributions that depend on spatial contexts of the analyzed OD pairs. These four classes represent different types of trips in California. We also find considerable heterogeneity in the mix of these classes within different regions of the State of California.