Previous research shows that various geometric and non-geometric road elements significantly
affect collision occurrence and severity on highways and arterial roads; however, little is known
about how these elements affect the safety performance of urban residential collector roads.
Therefore, this study investigated the impact of these elements on collision occurrence and
collision severity for urban residential collector roads. An extensive data collection effort was
conducted to synthesize collision records, traffic counts, road geometry, traffic control and other
features of residential collector road segments in the city of Edmonton (COE), Alberta, Canada.
Negative binomial safety performance functions (SPFs) were developed for total collision
occurrence and collision severity using four years of data. The proposed models were estimated
using the maximum likelihood estimation technique under a Bayesian context. An outlier test
was performed to improve the models’ goodness-of-fit. Scaled Deviance (SD) and the Pearson 2 χ statistic were used to assess the models’ goodness-of-fit. Results reveal that the exposure
covariates (segment length and traffic volume) are highly significant and positively related to the
predicted collisions in all of the SPFs. The property damage only (PDO) collision model has the
same significant covariates as the total collisions model, indicating that the number of PDO
collisions is predominantly higher than other collisions. For predicted total and PDO collisions,
there is a statistically significant positive relationship between collisions and access-point
density, stop-controlled intersection density, the presence of a horizontal curve, the presence of a
licensed premises, the presence of a seniors’ centre and the presence of on-street parking. In
contrast, there is a significant negative relationship between the presence of median and
predicted total and PDO collisions. For severe (i.e., fatal and injury) collisions, there is a
statistically significant positive relationship between collisions and segment length, traffic
volume, number of lanes, access-point density, stop-controlled intersection density, bus stop
density, the presence of a horizontal curve, the presence of a licensed premises, the presence of a
seniors’ centre and the presence of on-street parking. On the contrary, there is a significant
negative relationship between predicted severe collisions and the presence of a median, the
presence of a centre line and the presence of manned enforcement sites. From a model
application perspective, the city authority could use this information to assess the associated
safety risk of different geometric and non-geometric road elements on residential collector roads
and, hence, prioritize collision prone road segments.
