Research Papers

Development and Validation of a Canadian Culpability Scale

Filename 3B-1-Mark-Asbridge.pdf
Filesize 117 KB
Version 1
Date added May 8, 2011
Downloaded 7 times/fois
Category 2011 CMRSC XXI Halifax
Tags Session 3B
Author/Auteur Jeff Brubacher, Herbert Chan, Mark Asbridge


Purpose: Several researchers have employed culpability analysis in an effort to assess the contribution of specific risk factors to collisions when working with administrative collision data. Culpability studies attempt to approximate case-control studies by separating drivers into those culpable and non-culpable for their collision, with those non-culpable approximating a non- collision control population. The premise is that if a risk factor increases the chance of a collision, the risk factor should be more likely observed in drivers judged to be culpable for their collision. While a standard scale has been developed (Robertson and Drummer, 1994) to examine driver culpability, and applied in other jurisdictions (Australia and Europe), it does not adequately address Canadian driving conditions. Moreover, the traditional scale relies on subjective scoring by individual assessor drawing on collision reports, which has resulted in most culpability studies employing small samples, with more limited statistical power. The purpose of the current project is to develop and validate an automated, “rule-based” Canadian culpability scale.

Method: Scale development and validation followed four stages. First, the original culpability scale developed by Robertson and Drummer was modified to incorporate Canadian driving conditions, such as a snow or ice covered road surfaces due to wintery weather as well as the extensive police report data collected in the British Columbia Traffic Accident System. This process was done after consultation with Canadian traffic safety experts. Second, the scale was automated, employing a rule-based decision model that avoids subjective interpretation of police/insurance reports. Third, the automated scale was applied to a random sample of 73 collisions involving 134 drivers, drawn from the BC Traffic Accident System (TAS) data. Fourth, two traffic safety experts (former Transport Canada collision investigators) independently determined the culpability of each driver based on collision reports. Discrepant cases were discussed to understand why the scale differed from the expert assessment. The scale was then modified and the revised scale was again compared with the expert assessment.

Results: The final scale included seven categories. Traffic safety experts confirmed that the scoring of items within each category had face validity. Reliability was assessed by comparing the automated culpability scale with collision experts’ assessment, treating the expert assessment as the “gold standard”. There was good agreement between the original culpability scale and expert assessors with a kappa of 0.78 and 0.67, while inter-rater agreement among experts was also high, with a kappa of 0.73. Sensitivity analyses revealed strong agreement on culpability for collisions at either high and low levels of culpability, with less agreement in the middle range of the scale. Compared with expert assessment, the original scale gave more weight to environmental conditions and less weight to the collision event itself (collision type, pre-collision action of each vehicle, and damage location) with striking versus struck vehicle status. We held discussions with experts to address these discrepancies and we modified the scale accordingly. The final scale had very good agreement with the expert consensus agreement on the same set of collisions (kappa = 0.83). Finally, we applied the scale to a set of 65 drivers with known blood alcohol concentration and found that it behaved as expected: the odds of culpability was higher in drivers with blood alcohol concentration above 0.08% (OR = 9.6, 95%CI: 1.17-78.6).

Conclusions: Development of an automated culpability scale contextualized to Canadian driving conditions is feasible. An automated culpability scale will allow road safety researchers to assess collision responsibility in large administrative data sets derived from police reports or insurance data, and to apply culpability analysis to examine collision risk in varying populations (novice drivers, elderly drivers) across a broad range of exposures (alcohol consumption, drug use, fatigue, cell phone use).

Jeff Brubacher, Herbert Chan, Mark Asbridge