The objective of this project is to evaluate omitted variable bias in crash data analysis. Specifically, whether the omission of spatial determinants associated with road crashes leads to incorrect conclusions in models that use road links as the basis of the analysis. Previous research conducted in New Jersey has identified various area-based measures associated with both pedestrian and motor-vehicle crashes. These include area-based income measures, vehicle ownership levels, and population and employment patterns. Many features of the road environment, such as curvature, medians, lane and shoulder widths can influence the probability of vehicle crashes, and are the basis for most link-based crash analysis, since these can be easily measured for different links in the road network. The link-level data can therefore more accurately reflect the features of the road that are associated with crash probabilities, while spatial features can capture the socio-economic conditions of the area in which the road is located. Studies using spatial data as well as those using link-based data typically find statistically significant associations, however, only one study, to our knowledge, has combined these approaches to evaluate whether omitted variable bias occurs; in this case, a link-based study that tested the omission of spatial data in an intersection-based analysis (Mitra and Washington, 2012). This proposed study will seek to evaluate how the omission of key spatial variables from a link-based analysis may bias results of crash studies. Data from New Jersey will be used and different types of crashes will be evaluated (total crashes, fatal only, injury only, single vehicle, multi-vehicle, and pedestrian involved). Results will provide useful guidance not only on key factors associated with road crashes, but also will provide valuable input for developing more detailed road safety models, similar to those in AASHTOs Highway Safety Manual (AASHTO, 2010).