Identifying crash risk factors and high risk locations on an interstate network

Highway safety improvement projects are identified by using either (i) a site-specific or (ii) a systemic approach. In the site-specific approach, locations for improvements are ranked according to different performance measures such as critical crash rate, expected crash rate or equivalent property damage only. Alternatively, in the systemic approach, roadway characteristics such as number of lanes, shoulder width, etc. are flagged as a ‘risk’ (or ‘preventative’) feature that increases (decreases) the risk of negative outcomes. Using the Highway Safety Information System database, we seek to merge the two approaches by, first, identifying roadway factors associated with an increased occurrence of car crashes (features we call ‘risk factors’) and, subsequently, identifying roadway segments with a higher crash risk. Specifically, we model the locations of crashes as a realization from a spatial point process. We then parameterize the associated intensity surface of this spatial point process as the sum of a regression on roadway characteristics and spatially correlated error terms. Thus, through the regression piece, we identify hazardous roadway features and through the spatially correlated error terms, we identify locations of high risk