Red Light Cameras in Chicago: Follow up (August 31, 2014)
Chicago leads the nation with over 380 red light cameras (RLCs) at 190 intersections. Four years ago, I did a short study that showed accidents in Chicago decreased between 2001 and 2008. In contrast, accidents at intersections with RLCs increased. This result threw doubt on the city’s claims of significant accident reductions because of RLCs. I am now conducting a follow-up study by analyzing the accident data and ticket data between 2009 and 2012.
The results showed that accidents have gone down overall about 6%. This is expected, since there has been a reduction of 8% in the miles driven in Chicago. Consequently, for a safety intervention to be effective, it would need to exceed the overall accident decrease of 6%.
Analyzing RLCs that were placed during the study, it was found that in the first year after a camera was introduced, there was actually an increase of 5% for accidents at the intersection. This result was based on a very small sample size of 11 cameras and should not be considered definitive.
A second interesting finding about RLCs was that accidents dropped at RLC intersections at a rate of 11% between 2009-2012. This translates into 250 fewer accidents at RLC intersections between 2009-2012. The drop is largely due to a one-year 7% drop in accidents at RLC intersections during 2012. It’s not clear what the cause of the drop is or whether it will persist. Nevertheless, in 2012 RLC intersections became safer than other traffic signal intersections.
The third point focused on using ticket data to explain why they are not reducing accidents. I offered two explanations for this: the inconsistent ticketing, e.g., spikes, and the rarity of accidents.
The results here mirror my earlier study. The findings here are mixed and at best, the RLCs have a slight reduction on accidents. Despite the million of dollars invested in RLCs and half a billion dollars in tickets, there is no evidence that the RLCs have had a significant safety benefit.
Red Light Cameras in Chicago: Follow up (August 31, 2014)
Chicago leads the nation in red light cameras (RLCs). Four years ago, I did a short study that showed accidents in Chicago fell between 2001 and 2008. In contrast, accidents at intersections with RLCs in our study actually increased. This result threw doubt on the city’s claims of significant accident reductions because of RLCs.
I am now doing a follow-up that analyzes the accident data and ticket data between 2009 and 2012. I consider this study as exploratory in that it highlights overall trends, rather than a detailed investigation of every facet of the RLCs. This study starts in 2009, because Illinois changed the definition of an accident in 2009 from $500 to $1500 in damage. As a result, you can’t compare the raw numbers after 2009 to prior years.
Accidents Across Chicago
Lets start with some general trends for all accidents in Chicago. Here is a graph of accidents in Chicago between 2009 and 2013:
Clearly there has been a decline in accidents (over 6% between 2009 and 2012). There are a lot of factors affecting accidents, from gas prices to the weather. One factor that can be taken into account is how much people drive. The IDOT provides an Annual Vehicle Miles of Travel statistic for Chicago that allows us to account for how much people drive.
The graph shows vehicle miles dropping from 12 billion miles in 2009 to 11 billion miles in 2012 for Chicago, a drop of 8%. If people drive fewer miles, it seems reasonable there will be fewer accidents. We can analyze this by looking at the accident rate for each mile traveled.
This graph shows that when we account for the fewer miles driven, there is no longer a steady drop in accidents. Instead, there have been more accidents per mile driven over the last few years. So once we account for people driving less, we actually see that the rate of accidents are not dropping.
This step of considering the amount of miles driven is important and often neglected. If you don’t recognize people are driving less, statistics will show fewer accidents at construction zones, school zones, highways, traffic signals, and everywhere else. It can then seem that other variables, e.g., RLCs, are leading to a drop in accidents. The best research accounts for these factors by using a control group, so it’s possible to identify only the contribution of the safety device or program. Otherwise it will appear that RLCs have led to a drop of 6% in accidents, even though the drop is due to other factors.
Accidents at Traffic Signals
So while accidents have not dropped throughout Chicago, are traffic intersections safer now? Advocates of RLCs argue for a “halo effect” where the threat of tickets will lead to safer driving at other intersections. To consider this, we analyze accidents at traffic signals in Chicago. The IDOT data contains a separate variable for accidents at traffic signals that I use.
A quick look at this graph shows that it is very similar to the overall drop in accidents. In this case, accidents have fallen 4% at traffic signals (compared to the overall 6% drop in accidents).
To better understand the drop at traffic signals, I next looked at the percentage of accidents that occur at traffic signals in Chicago between 2009 and 2012. If people were driving more carefully at traffic signals, it would be expected that accidents at traffic signals would become more rare. For example, dropping from a share of 25% of all accidents to 20% of all accidents.
This graph shows that the percentage of accidents at traffic signals hasn’t changed appreciably between 2009 and 2012. The results here suggest that traffic signal accidents are holding relatively constant (accidents have increased by 0.5%). This also suggests the RLCs are not having a “halo effect” (making all intersections safer), because accidents are not dropping throughout the city at traffic signals.
I next looked at the types of accidents occurring at traffic signals in Chicago between 2009 and 2012. My previous research suggested the introductions of RLCs might lead to a rise in rear-end type of accidents.
This graph shows the percentage of each type of accident. For example, in 2012 at traffic signals, rear-end accidents made up 38% of all accidents, while turning accidents account for 22% and angle accidents account for 14%. In looking at the trends here, the biggest finding is the drop in turning crashes of about 3%. While this is noticeable, it will be important to see if this is a temporary drop or one that keeps increasing. The decrease in the turning crashes and the rise in rear-end are consistent with behaviors affected by RLCs. It’s too early to tell, but there may be a slight change in the type of accidents occurring at traffic signals.
I also took a look at changes in crash severity. I found that property crashes dropped 4%, injury crashes dropped 3.6%, and fatal accidents dropped from 42 to 29, a drop of 31%. I will use these numbers in the next section, when I look specifically at RLC intersections.
Analyzing Accidents at RLC Intersections
This part of the analysis focuses on intersections with RLCs. One way to examine the effectiveness of the RLCs is to compare accidents before and after the camera was installed. I did this analysis in my earlier study. For this study, I selected all the RLCs that would provide a year of before and after data. This gave me 11 cameras with installation dates in 2010 and 2011. This is a very small sample. The results found accidents increased from 109 to 115 accidents at these 11 intersections, so an increase of 5%. This increase is not statistically significant. The increase matches earlier studies for RLC intersections.
The next step was looking at 190 RLC intersections, with cameras installed prior to 2013. The criteria for accidents were that they acknowledged a traffic signal in the accident report and occurred within a 100 feet of the intersection. I did this analysis in two ways, the first included all cameras active prior to 2013 and the second was only cameras that were active during the 2009-2012 period (therefore excluding cameras installed during 2009-2011). The results were roughly similar, but I will present data only on cameras that were active during the entire 2009-2012 period.
The result was a drop of 11% at intersections with red light cameras from 2264 to 2016 accidents over 4 years. This was a reduction of 248 accidents at the RLC intersections. This is much larger than the overall 4% drop at traffic signal intersections!
Let us next assume this drop at RLC intersections is entirely due to the cameras and not other factors, such as better enforcement by the police or changes to the intersections. We could then say that accidents at RLC were reduced by 4% as compared to the overall general trend at traffic signals. 4% of the accidents at the RLC intersections is about 91 accidents. So the cameras reduced 157 accidents!
If you look carefully at the graph, you will notice there is a big drop in the number of accidents in 2012 at RLC intersections. From 2009 to 2011, the drop in accidents at traffic signals was 2.7% and a drop of 4% at RLC intersections. So if we were looking at 2011 data, it would seem the RLC intersections were slightly safer. But between 2011 and 2012, accidents at traffic signals dropped 1.2%, while they dropped 7% at RLC intersections! I don’t know what changed during 2012, but some factor led to a substantial drop in accidents. We will have to watch the 2013 data to see if this was a hiccup or a trend. The 2013 traffic data is due to be released in September 2014.
The types of accidents at RLC intersections also changed as the graph shows:
Between 2009 and 2012, at the active RLC intersections, angle accidents went up 3%, rear-end accidents dropped by 8%, and turning accidents dropped by 18%. For turning accidents, where there is the largest drop, accidents went from 727 to 532 between 2009 and 2012. This shows that the type of accidents are changing at RLC intersections.
I also examined changes in crash severity at RLC intersections. I found that property crashes dropped 13%, injury crashes dropped 2.6%, and fatal accidents stayed even at 5. Comparing this to traffic intersections generally, we see property crashes at RLC intersections dropped much faster (13% versus 4%) and injuries relatively rose (2.6% versus 3.6%). This suggests crashes at intersections that occur at RLC intersections are more severe.
Tickets and RLC Intersections
This sections considers the influence of RLC tickets on accidents. If the purpose of RLCs is to ticket risky behavior that leads to accidents, then you would expect intersections that give out lots of tickets to have lots of accidents. After all, more people are breaking the law, so you would expect that as law breaking increases, so would accidents.
A scatter plot is often used to investigate relationships between two variables. The graph shows a scatter plot of the number of tickets versus accidents. Looking at the graph, its clear there is no relationship between tickets and accidents. Whether an intersection gives out lots of tickets or few tickets has no relationship to the number of accidents. For more details on this portion, see my blog post on the relationship of RLC accidents to tickets.
So why aren’t tickets affecting behavior?
There are two reasons I can suggest. First, tickets are given out inconsistently. There is not a clear relationship between running a red light and getting a ticket. The spikes found in red light camera tickets, first identified by the Chicago Tribune in 2014, best exemplify this. I went through the ticket data and used a statistical technique, which looks for outliers in a time series. I ran this on both a daily summation and a weekly summation of the tickets.
The results were revealing. Over 100 intersections had a spike. The spikes affect at least 20,000 to 40,000 tickets. Spikes are largely pre-2010. I found that 2007 and 2008 accounted for 68% of all the spikes. By 2010, the amount of spikes dropped considerably. This suggests that whatever the cause of the spikes was, it was largely corrected several years ago. You can find more details about this analysis on my blog post on ticket spikes.
The second reason is that accidents are a very rare phenomenon. Consider the traffic volume for Ashland & 63rd with about 8.6 million vehicles per year (based on daily volume of 23,500) and an average of 30 crashes per year. This is about one crash for every 287,000 cars that pass through the interesection. This is how rare crashes are and how difficult it is to reduce them.
The average number of tickets at Ashland is about 2,750 a year. If you assume all tickets are related to running red lights (a huge jump in logic), this would still only mean that 1% of red light runners cause a crash.
Thinking about these ratios should help illustrate why cameras can reduce the number of people running red lights, but not have a significant impact on the number of accidents at an intersection. In the end, distractions, alcohol, driving conditions, and other factors can play a much more important factor than the presence of a red light camera.
In Chicago, between 2009-2102 accidents have gone down overall about 6%. This is expected, since there has been a reduction of 8% in the miles driven in Chicago.
In looking at intersections with traffic signals, we find that the drop in accidents is only 4% and the percentage of accidents at traffic signals hasn’t changed appreciably. However, there is a slight change in the types of accidents with a slight increase in rear end accidents and a drop of 3% in turning accidents.
In examining intersections that had RLCs introduced during this study, there is a 5% increase for accidents (even though accidents generally were falling due to decreased driving). In contrast, accidents are down 11% at cameras with RLC intersections.
One very interesting finding is the large drop in accidents at RLCs during 2012. There was a one-year drop of 7%. It’s hard to interpret this finding right now; we will have to look at the 2013 data to see if this is a trend or an anomaly. If this trend is sustained, it could support the notion of RLCs as a safety measure. I can’t explain why the camera effect would start after more than 5 years for some of these intersections. Perhaps it is due to increase attention from the media on RLC in 2012 that led to a greater awareness which in turn affected driving. Or maybe there were structural changes at some of these intersections that made them safer.
Finally, the study considered the role of tickets. It is clear that there is no relationship between tickets and accidents. Whether an intersection gives out lots of tickets or few tickets has no relationship to the number of accidents. I offered two explanations for this, the inconsistent ticketing, e.g., spikes, and the unique factors that go into the very rare occurrence of an accident.
The results here mirror my earlier study. The findings here are mixed, and at best, the RLCs have a slight reduction on accidents. Despite the million of dollars invested in RLCs and half a billion dollars in tickets, there is no evidence that the RLC have had a significant safety benefit.
Limitations: This study relies on data provided by the Illinois Department of Transportation. I have not accounted for any changes that have occurred at the intersections or increased enforcement. Both of these are likely to intersections that have high accident rates. The analysis here is also very simplistic, as it doesn’t account signal timing, lane geometry, and traffic volume. I also have not explored every interesting finding in depth, such as large drop in 2012 RLC intersections in detail.
Data: The data analysis here encompassed the entire year of 2009 through all of 2012. I have created a research tool for red light camera analysis at MiningChi.com. The site allows anyone to find the accident and ticket statistics for any RLC intersection in Chicago.
Future Work: I hope to update this report during the winter with 2013 data. I would also like to extend the data in several ways. First, take into account injuries in accidents. Second, use that information along with the type of accidents to generate an economic impact of the red light cameras. Finally, take into account traffic volume in these calculations.
Acknowledgements: The data in this study was collected with the help of several parties. First, I would acknowledge the wonderful help of the IDOT in providing accident data. The Chicago Tribune’s publication of ticket data acquired through a FOIA was also valuable for my analysis.