Movement Regularity Analysis using Geo-Located Twitter Data

Time-geographic approaches to human traveling behavior have traditionally used origin-destination data (e.g. Cascetta and Nguyen 1988) or activity-travel data collected via diaries and other forms of survey (e.g. Bowman et al. 2001). Origin-destination data is spatially coarse. It can be used to model interactions among places but is of limited use in understanding movement. Survey data can be spatially detailed, but surveys are repeated infrequently and sample size is typically small due to data collection expense and the need to find participants willing to provide longitudinal data (Handy 1996, Calabrese et al. 2013). As an alternative, researchers have begun to consider location-based and mobile technologies as potential sources of travel activity data. In one example, banknote data was used as a proxy for inter-city mobility in the conterminous U.S. by Brockmann et al. (2006). Cell phone data has been used to provide more detail on individual users’ movements, with behaviors explored at different scales including: urban (e.g. Gonzalez et al. 2008, Calabrese et al. 2010, Kang et al. 2012), region (e.g. Calabrese et al. 2013), country (e.g. Krings et al. 2009). Additionally, social media data serve as a proxy for global-scale movements (e.g. Hawelka et al. 2014) as well as national or urban scales (e.g. Azmandian et al. 2013). The ultimate goal is to enhance understanding of geographic variation in travel behavior in the U.S. and to develop methods for leveraging social media to study spatial behavior. To do so, this paper aims at 1) developing and assessing an algorithm for estimating each Twitter user’s residential county by leveraging a full year of individual-based geo-located tweets (i.e. geo-tweets), 2) investigating relationships between tweeter characteristics and geo-tweeting behaviors, and 3) characterizing counties by weekly, daily, and hourly aggregated number of active residential/non-residential Twitter users.