Computing travel time when the exact address is unknown: a comparison of point and polygon ZIP code approximation methods

Background: Travel time is an important metric of geographic access to health care. We compared strategies of estimating travel times when only subject ZIP code data were available. Results: Using simulated data from New Hampshire and Arizona, we estimated travel times to nearest cancer centers by using: 1) geometric centroid of ZIP code polygons as origins, 2) population centroids as origin, 3) service area rings around each cancer center, assigning subjects to rings by assuming they are evenly distributed within their ZIP code, 4) service area rings around each center, assuming the subjects follow the population distribution within the ZIP code. We used travel times based on street addresses as true values to validate estimates. Population-based methods have smaller errors than geometry-based methods. Within categories (geometry or population), centroid and service area methods have similar errors. Errors are smaller in urban areas than in rural areas. Conclusion: Population-based methods are superior to the geometry-based methods, with the population centroid method appearing to be the best choice for estimating travel time. Estimates in rural areas are less reliable.