Chicago Red Light Cameras and Traffic Safety
Do red light cameras reduce traffic accidents?
I recently received a $100 red light ticket from the city of Chicago. Along with the ticket was a link to a picture and video of my vehicle not coming to a full stop when turning right at a red light. I fell victim to the dreaded Chicago red light camera. A friend told me about a study showing red light cameras made intersections more dangerous. As a data scientist, I was intrigued. This sounded like an answerable question in Chicago, so I built a project to attempt to answer it.
I also created a web app to go along with this project. With it, you can explore detailed Chicago red light camera intersections and crash data. (click image below)
Chicago Red Light Enforcement
In 2003, Chicago began its red light camera enforcement with two cameras. Since that time, the program has grown to over 350 cameras at more than 150 Chicago intersections.
The program has been rocked by scandals including: bribery schemes, corruption charges, alleged targeting of underserved communities, and software tinkering to increase revenue. Red light cameras viewed by many Chicagoans as no more than a revenue generator, and an invisible tax on its residents. The program has generated a staggering $750 million dollars since its inception, with several notable intersections racking up millions of dollars a year in fines.
The origin and intent of the program (as stated on their website) is “to increase safety on Chicago streets by reducing the most dangerous types of crashes at intersections. While all crashes are potentially hazardous, Red Light cameras are designed to reduce angle (or “t-bone”) crashes because of their extreme danger to those involved. As individuals comply with traffic signals, serious crashes and injuries reduce significantly.”
Do Cameras Improve Safety?
A national study of red light camera safety was conducted at Case Western University in 2018. Using red light cameras in and around Houston, it concluded that red light cameras reduce the number of dangerous angle crashes, but may also increase the number of rear-end and other types of accidents, thus not improving overall safety. This and other studies, point to cameras actually shifting traffic patterns: drivers tend to brake harder and more abruptly, causing different types of accidents, but not reducing the overall number.
I decided to look at the data for Chicago, and see if a conclusion could be made for my city; do red light cameras actually improve traffic safety?
My Data
Since September 01, 2017, all Chicago police precincts have had mandatory uniform crash reporting citywide. More than half a million crash reports are made available at the Chicago Data Portal and are updated daily. Included in the detailed crash reports are the accident type, whether they occurred at a red light intersection, and the latitude/longitude of each accident.
The data portal also contains daily red light camera violations for every camera, and includes the name of the intersection and the latitude/longitude location.
After filtering my data to only include crashes occurring between September 2017 and January 2021, I was left with over 60,000 intersection related accidents that occurred at one of Chicago’s 3000+ red light intersections.
By using the latitude/longitude, I designated accidents that happened within 30 meters of the center of a red light camera intersection. More than 8000 accidents labeled as red light crashes at a red light camera enforced intersection. This was the primary dataset for the project.
Accident Types
As mentioned, Chicago is looking to reduce angle (‘t-bone’) type accidents. The chart below is generated using only red light intersections in Chicago. Angle type accidents account for 16% of all accidents, but are responsible for 37% of the deaths and 31% of the injuries. Turning accidents (which could also be reduced by red light cameras) are the most numerous, and are the second leading cause of deaths and injuries at red lights behind angle crashes.
Rear end accidents, which the Houston study showed to increase when cameras were removed, were numerous but resulted in fewer deaths and injuries. An increase in rear end accidents might be acceptable for a city if there was a corresponding decrease in the more dangerous crash types.
Methodology
The Houston study used a natural experiment that occurred when red light cameras were removed by a voter referendum in that city. Crash data from both before and after their removal could be compared to see if the presence of the camera had an effect on accidents.
I sought to do the same in Chicago. Since 2017, a handful of cameras have been turned on/off, or have been moved to other intersections. These removals and additions create a small scale (7 intersection) natural experiment similar to the Houston study. Additionally, an Illinois House bill is currently in legislation to remove all red light cameras in Chicago, which would create an incredible opportunity to evaluate this experiment at a much larger scale.
Seven Intersections of Interest
Unfortunately, I can only look at red light camera intersections that have been active since September 2017 (when crash data was made available). During that time, seven intersections were either added to the red light camera violations data or removed. Crash data is available both before and after the install/removal.
The seven intersections that were both on and off during that timeframe:
['FOSTER AND NORTHWEST HIGHWAY' 'FULLERTON AND NARRAGANSETT'
'HALSTED AND 103RD' 'LAKE AND UPPER WACKER' 'MICHIGAN AND JACKSON'
'MICHIGAN AND ONTARIO' 'MILWAUKEE AND CENTRAL']I built a dataset containing daily crash numbers for each of my intersections, and split the populations into ‘cameras off’ and the ‘cameras on’ (experiment and control groups) using the earliest and latest violation dates for those intersections.
Dataset from 7 intersections of interestCameras OFF:
Cumulative days: 1733
Total crashes off: 69Cameras ON:
Cumulative days: 6786
Total crashes: 134
Correcting for imbalance
The dataset is imbalanced and focuses heavily on the late part of 2017, and early 2018. Normally, this would not be a problem, but through my study, it became obvious that some intersections have far more crashes, injuries, and even particular types of accidents. I did not want a single intersection (which may have a significant bias) to overwhelm the others. To balance the dataset, I randomly sampled the the overrepresented group for each camera so there were an equal number of on and off days for each camera. Despite the lower than desired test power, there was sufficient data for a reliable t-test to compare these two populations with the results below.
Balanced dataset hypothesis testing: Null hypothesis: red light cameras have no effect on the number of crashes.alpha: 0.05
Power: 0.62n (cam on): 628 days
n (cam off): 628 daysmean (cam on): 0.03185 crash/day
mean (cam off): 0.04777 crash/day
Actual crashes with cams off: 30
Expected crashes with cams off: 20
Standard deviation: 4.40
z score: 2.27
P value: 0.0115REJECT the null hypothesis
There is a 1.2% probability of observing a sample mean as large as 0.048 when the true population mean is 0.032. We reject the null hypothesis, and accept the alternative hypothesis that the presence of a red light camera reduces the number of accidents at these seven intersections.
These results are counter to the Houston study which was inconclusive about cameras reducing overall accidents.
Additional Test Results
With the same technique and balanced populations, we can investigate the number of injuries.
Balanced dataset hypothesis testing:Null hypothesis: red light cameras have no effect on the number of injuries.alpha: 0.05
Power: 0.99n (cam on): 628 days
n (cam off): 628 daysmean (cam on): 0.00159 injuries/day
mean (cam off): 0.01752 injuries/day
Actual injuries with cams off: 11
Expected crashes with cams off: 1.0
Standard deviation: 1.00
z score: 10.01
P value: 7.030e-24REJECT the null hypothesis
There is an extremely small probability of observing a sample mean as large as 0.01752 when the true population mean is 0.00159. We reject the null hypothesis, and accept the alternative hypothesis that the presence of a red light camera reduces the number of injuries at these seven intersections.
When performing the same balanced approach and filtering for accident type, the results were mixed and inconclusive. The test power was too small to make a determination if angled accidents were decreased, or rear end accidents were increased. However, all seven intersections had the same or fewer of all types of accidents when red light cameras were present.
Shortcomings of This Study
This study is relatively limited in size and scope when compared to the national study at Case Western. Their study used nearly 10 times more intersections, and also used assumptions about driver decision making, psychological aspects, and citywide safety calculations that were not considered here. I strictly relied on the limited data available. When more data becomes available, perhaps this year, the data may be sufficient to make more and more reliable tests for the different accident types and overall safety.
With such a limited number of cameras used, it is possible that factors other than just the presence of red lights may have affected the outcomes. Potential biases could be: construction dates, traffic pattern changes, intersection modifications, or time of year sampled for the minority group. This again could be improved with additional data and intersections.
Conclusion
For the intersections studied, we accept the hypotheses that the presence of red light cameras reduces the number of accidents and injuries at those intersections.
This supports the city of Chicago’s reasoning for installing the cameras to improve safety.
