Collision counts data has shown that significant proportion of collisions occur frequently on the boundary of Traffic analysis zones (TAZs). Consequently, the way in which collisions, and also other geocoded data, along TAZs’ boundaries are assigned into adjacent zones is of interest because it has direct impact on the prediction ability of macro-level CPMs. In this study, data for 422 TAZs from the City of Ottawa was used to develop macro-level CPMs. Geocoded data on TAZ’s boundary were assigned between adjacent TAZs using ten different assignment methods. Models to predict total, non-fatal
injury, fatal, property damage only (PDO), bike-involved, and pedestrian-involved collisions were developed. The collisions were related to four categories of independent variables. Results of the developed models show that different geocoded boundary data assignment methods do affect the accuracy of developed CPMs results significantly. It was found that allocating boundary data to TAZs evenly improved model results significantly.