Adaptation of a Canadian Culpability Scoring Tool to Alberta Police Traffic Collision Report Data
Author(s): Pitt, Aucoin, Graff, Howard, Nettel-Aguirre, McCormack, Owens, Anderson, Rowe, Hagel
Slidedeck Presentation Only (no paper submitted):
Abstract:
Traffic research techniques often require the ability to assign responsibility in motor vehicle collisions (MVCs). In doing so, researchers can select control drivers from the non-responsible group (represent the driving source population) and identify risk factors from the responsible. These traffic research techniques are predicated on the ability to consistently score driver fault. A Canadian Culpability Scoring Tool (CCST) uses police collision report (PCR) data and is automated to score driver fault in MVCs while accounting for external factors (e.g., weather conditions). This tool was previously validated using British Columbia PCR data; however, PCRs vary among provinces. To use the CCST for traffic research in Alberta, the tool required adaptation to Alberta PCR cohort.
To adapt the Canadian Culpability Scoring Tool to Alberta police report data.
Calgary and Edmonton PCRs from 2010-2014 were used. Adaptation of the CCST was completed through collaboration with Alberta Transportation, contributing to face validity. The CCST is scored on seven categories, each receiving a score from 1 to 5. Motorists with scores ≤13 are considered culpable and ≥16 non-culpable; scores 14&15 are indeterminate. Two research assistants (RAs), given only the information necessary for scoring, manually evaluated 175 randomly selected MVCs. Discussion of disagreements between the two RAs, and consultation from experts in Alberta Transportation, informed the rules used in the automation of the Alberta adapted CCST. The automation was applied to 656,934 motorists in the study sample. Binary logistic regression was used to examine differences between culpable (n=396,248) and non-culpable drivers (n=244,442) for those characteristics that were not included in the tool.
The kappa value of inter-rater reliability for the random sample (n=175) was nearly perfect (0.95; 95% CI: 0.92-0.99). From the original sample, 244,442 (37.2%) were deemed not-culpable, 396,248 (60.3%) were deemed culpable and 16,244 (2.5%) were indeterminate. The culpable group had higher crude odds than the non-culpable group of each: being male (OR=1.12; 95% CI: 1.11-1.14), being impaired by drugs (OR=11.57; 95% CI: 7.82-17.11), and being impaired by alcohol (OR=32.59; 95% CI: 27.23-39.02). The odds of being 54 years old were higher in the culpable group (OR=1.47, 95% CI: 1.40-1.50); whereas, the odds of being in the age group 40-54 years (OR=0.85 95% CI: 0.84-0.86) were lower in the culpable. Culpable drivers had higher odds of driving between 12 am and 6 am than any other 6 hour time-block.
The culpable group, as determined by the Alberta-specific culpability tool, exhibited characteristics expected of drivers who are at-fault in collisions. The age groups 25-39 and 40-54 demonstrated slightly different results than the CCST results. However, this is the only difference that exists in the findings of our tool compared with the CCST and could exist due to difference in the datasets, rather than the adaptation approach. It is possible to adapt the CCST to Alberta PCR data and assign fault on large data sets with high reliability and validity. In doing so, we can identify risk factors for collision contribution and not-at-fault drivers who represent the driving source population.