Traffic safety is a critical concern for any city, and speed cameras have long been touted as a solution to protect vulnerable road users. But do these electronic watchdogs actually make our streets safer? In this deep dive into Toronto's speed camera program, I uncovered some surprising—and disappointing—findings.
Proponents of speed cameras argue a simple, intuitive point: slower-moving vehicles mean fewer serious pedestrian injuries and fatalities. It sounds logical, right? Who wouldn't want to slam the brakes on road accidents? Armed with curiosity and a stack of data, I set out to test this hypothesis using Toronto's traffic information.
I assembled a comprehensive dataset from the Government of Ontario and the City of Toronto, pulling together:
- Detailed collision records
- Speed camera locations
- Extensive road network information
My goal was straightforward: examine regions around traffic cameras and track how accident numbers changed between 2014 and late 2024.
Note that the cameras were moved throughout the city. So, they would be active for a period of time and then moved to another location. Although a few cameras were returned to the same locations.
The Statistical Toolkit
Measuring rare events like traffic accidents requires some statistical finesse. Most days on most roads, zero collisions occur—which makes traditional statistical models less effective. Enter the negative binomial model with fixed effects, a statistical model designed to handle count data with more zeros. For good measure, I also included a fixed-effects Poisson count model for comparison.
Key Assumptions
To keep the analysis grounded, I made two crucial assumptions:
1. Speed cameras might influence traffic patterns within 250 meters on either side (a total 500-meter zone)
2. Camera effects would likely be limited to the street where they're installed. I’m not assuming that cameras create a broader traffic-calming effect so I omit collisions on nearby streets.
A close-up of a speed camera location in Toronto
The image is a close-up of a speed camera location in Toronto. The camera is the black dot in the centre. The red dots indicate accidents involving pedestrians between 2014 to late 2024. The translucent blue circle indicates the 250 meters around the camera.
Understanding collision rates isn't just about cameras—it's about context. The models controlled for factors that could skew the results:
- Seasonal Variations: Winter months and high-precipitation periods typically increase accident risks compared to other times
- Day of the Week: Weekend pedestrian traffic differs from weekday patterns
- Pandemic Impact: Pandemic years like 2021 and 2022 are likely to see dramatically reduced road and foot traffic
The Critical Variable: Camera Activation
The most important metric was simple: Did the presence of a speed camera correlate with fewer pedestrian collisions? I created a dummy variable tracking when cameras were active versus inactive, hoping to see a clear safety improvement.
To ensure research integrity, I employed the Hausman statistical test to verify that the fixed effects models were the best (In other words, to verify that random effects models were not warranted). The results confirmed that fixed effects were the optimal, choice.
Both statistical models told the same story. There was no statistically significant difference in pedestrian collision rates between areas with active cameras and those without.
The models did reveal some insights:
- Pedestrian collision risks increased during the early winter months
- Pandemic restrictions substantially reduced collision likelihood
- Weekday and Saturday traffic showed higher accident probabilities
Additional modeling techniques, including pooled regression, consistently showed no meaningful impact from speed cameras. Intriguingly, these models revealed that:
- Some city wards experienced significantly fewer collisions, some wards had many more
- Roads with more lanes correlated with higher pedestrian collision risks
Speed cameras might serve various municipal goals, but based on this analysis, pedestrian safety isn't demonstrably one of them. Regional and city councils should critically re-evaluate speed camera implementations if pedestrian collision reduction is the primary motivation.
Tools used
* The data processing of maps (shp files) and open data was done in R. An early version was done in Python.
* The interactive visualizations of the road networks were also done in R.
* The modelling was done using Stata and R: xtnbreg, xtpoisson using fixed effects, nbreg and poisson using clustered standard errors.