Accuracy of Geoimputation

Abstract

The role of the built environment in travel behavior has seen increased interest by strategic transportation planners. To capture relationships between travel behavior and the built environment, microenvironment variables representing infrastructure and land uses surrounding trip origins and destinations are being used as explanatory variables in travel demand models. Buffers of various sizes can be created around origins or destinations to capture the microenvironments. A key requirement is knowledge of the exact coordinates (latitude and longitude) of trip locations. However, such information is commonly removed from public use data because of privacy concerns. To assess whether synthetic geo-imputed residences can overcome the removal of exact location information, two data sets from activity-based travel surveys in North Carolina (Research Triangle survey, N = 4,724, and Charlotte survey, N = 3,310) were analyzed. The fundamental question was whether the geoimputed microenvironmental measurements could be used to model travel behavior sufficiently and accurately. Residences, geoimputed residences, and residences assumed to be located at centroids of census blocks (as is current practice) were compared. The data indicate that (a) the assignment of census block centroids results in statistically significant systematic errors when the accessibility measures are measured; (b) geoimputation based on the level of the traffic analysis zone can provide reasonably accurate accessibility measures in larger buffer sizes of 0.75 mi but not in smaller buffers of 0.25 mi; and (c) geoimputation based on census block level provides accessibility measures that are sufficiently accurate for specifying travel behavior models.