Content Is Safety-In-Numbers (SIN) for fatal bicycle Background - - PDF document

Is Safety-In-Numbers (SIN) for fatal bicycle crashes confounded by the general level of road safety?

Paul Schepers Ministry of Infrastructure and the Environment

Content

• Background
• Study on cycling safety in countries and Dutch municipalities
• Conclusions and discussion

Jacobsen who coined the term Safety in Numbers (SIN) in 2003

• Cycling in 14 European countries
• SIN: more cycling lower risk for cyclists

Dataset for 18 countries used in the current study: similar picture

Alternative explanation via road safety?

Bicycle use Bicycle accidents Road safety

Alternative explanation via road safety?

High level of road safety and accordingly:

• more cycling
• lower risk for

cyclists Poor road safety and accordingly

• less cycling
• higher risk for

cyclists

How to control for the general level of road safety?

• Often used proxy is traffic mortality: total road deaths per 100,000

population

Models used to study SIN

Number of cyclist injuries = constant CYCLβ1 MVβ2 eβ3X

CYCL Distance cycled or volume of cyclists MV Distance travelled by or volume of motor vehicles X Traffic mortality

β1<1 is indicative of SIN; risk is linear if β1=1 Injuries CYCL β1<1 β1=1

Model generally used to study Safety in Numbers, expanded with road safety

CYCL Cyclist injuries X (mortality) MV

Outcome meta-analysis Elvik and Bjornskau, 2017

Number of cyclist injuries = constant CYCL0.43 MV0.5 If everything else remains the same, doubling the amount of cycling results in 35% more cycling injuries: 20.43 = 1.35

Methodology for area level modelling

• Using Negative Binomial regression, the same model is fit on

cyclist deaths in: – 18 countries – 366 Dutch municipalities

• Two models:
• I. distance travelled by cyclists and motor vehicles

II.distance travelled by cyclists and motor vehicles and mortality

• A change in the in the exponent for distance travelled by

bicycle (β1 in CYCLβ1) after adding mortality would indicate SIN is (partially) confounded by the general level of road safety

Results for regression for 18 countries; 7,752 cyclist deaths

Regression coefficients (95% CI) Parameter Model I Model II Bicycle km (β1) 0.40 (0.20 to 0.60) 0.66 (0.52 to 0.80) Motor vehicle km (β2) 0.55 (0.29 to 0.81) 0.31 (0.16 to 0.46) Traffic mortality 0.12 (0.08 to 0.16) AIC 225 204

Results for regression for 366 countries; 1,683 cyclist deaths

Regression coefficients (95% CI) Parameter Model I Model II Bicycle km (β1) 0.33 (0.14 to 0.53) 0.58 (0.39 to 0.77) Motor vehicle km (β2) 0.47 (0.27 to 0.68) 0.31 (0.12 to 0.51) Traffic mortality 0.09 (0.07 to 0.12) AIC 1587 1534

Pooled result of both analyses

Regression coefficients (95% CI) Parameter Model I Model II Elvik & Bjørnskau, 2017 Bicycle km (β1) 0.36 (0.22 to 0.50) 0.63 (0.52 to 0.75) 0.43 (0.33 to 0.53) Motor vehicle km (β2) 0.50 (0.34 to 0.68) 0.31 (0.19 to 0.43) 0.50 (0.38 to 0.62) Traffic mortality 0.10 (0.08 to 0.12)

Conclusions

• SIN in area-level models is partially confounded by the general level
• f road safety
• The higher level of cycling safety in areas with higher cycling

participation is partly due to the higher general level of road safety in these areas

• After controlling for traffic mortality, there is still a clear indication of

SIN with cyclist deaths increasing less than proportional in response to more cycling

• We cannot conclude whether or to what degree the remaining SIN

effect is causal as control variables for dedicated bicycle infrastructure are lacking in the models

Main practical implication

• Cycling policies to encourage cycling are most effective in

combination with road safety policies if the aim is to increase public health