Maps and Charts: Crashes around I-95 Express

In collaboration with WLRN’s Kenny Malone on the The End of the Road series about I-95 in South Florida, we worked with experts in geography and data science to analyze and map crashes on one of Miami’s most-traveled stretches of roadway using data from the Florida Department of Transportation.

Since they were added in 2008, millions of South Florida drivers have used the I-95 Express Lanes and their adjacent general-purpose lanes. As of FDOT’s most recent report, approximately 107.5 million express lane trips have been taken, bringing in $92.1 million in revenue. The express lanes, which carry a toll that fluctuates from $0.50 to $10.50 based on traffic patterns and are separated with plastic poles from the general-use lanes of the highway, are also a key safety concern for many I-95 drivers. Drivers often cite “lane-diving,” when drivers dart through the poles to get into or out of the express lanes, and narrow shoulders along the express lanes as dangers on the highway.

A geospatial analysis of FDOT data led by Jeff Onsted, an associate professor of geography at Florida International University, and a follow-up by Malone on fatal crashes in 2013 and 2014, revealed that the deadliest crash on I-95 in Miami in the past decade killed five people, and occurred on one of those I-95 Express shoulders. Malone investigates that horrific crash and other key safety issues along I-95 Express as part of an hour-long “End of the Road” special that airs at 9 a.m. and 7 p.m. today on WLRN, 91.3 FM.

Fatal and Incapacitating Injury Crashes

An interactive map shows how incapacitating injury and fatal crashes are dispersed along the stretch of I-95 with express lanes and their entry and exit points from 2005 to 2012.


Crashes by time of day

Data scientist Christopher Peters found most crashes on I-95 where express lanes are present occur at times of peak traffic, southbound in the morning rush hours, and northbound in the evening rush hours. Incapacitating injury crashes are more evenly distributed throughout the day. Fatal crashes are most likely to occur between 10 p.m. and 5 a.m., and are more common in Southbound lanes.

Total crashes by time of day

Incapacitating injury crashes by time of day

Fatal crashes by time of day

Crashes by year

Peters also examined crashes by year. There was a slight downward trend in crashes from 2005 to 2009. In 2010, crashes spiked to their highest levels, but decreased in 2011 and 2012. (Note: We did find a discrepancy in 2011 counts between this data and an FDOT report for the same time, which FDOT explains below.)

Total crashes by year

Incapacitating injury crashes by year

Fatal crashes by year

About the analysis

We analyzed geographical records of crashes throughout Miami-Dade county from the FDOT Crash Analysis Reporting System (CARS), a massive database that FDOT uses for highway safety analysis. The most recent dataset available was from 2005 to 2012. FDOT counts crashes as fatal when a person in the crash dies within 30 days of the incident from injuries sustained in the crash, and they also count crashes with “incapacitating injuries.” We focused on these two identifiers of severe crashes.

Onsted narrowed the crash data using a special code FDOT uses to identify crashes on I-95. He then narrowed it further to reflect only a 12.6 mile stretch of the roadway containing the actual I-95 Express Lanes – the roughly 7 miles in each direction with plastic poles – as well as the system’s entrance and exit points that could affect driver behavior. For example, the signs warning drivers that the express lanes are coming up could prompt some vehicles to start merging left to enter the express lanes or merging right to avoid the express lanes. And as drivers leave the express lane poles, some will need to cross all lanes of I-95 traffic to reach their exit. There were more than 13,900 crashes reported within this section of I-95 from 2005 to 2012 in the FDOT data.

Onsted found a discrepancy in the year 2011, in which our count of crashes was lower than those in a separate FDOT report issued for the same time. FDOT responded to this issue in an email:

There is always an expected difference between the two file types because at times there are missing fields in the raw data report extracted from CARS that prevent some crashes from being reflected in the shapefiles. For this reason, it is necessary to manually review each police report to identify all the crashes that occurred on the facility within the desired limits. However, the difference for the year 2011 was greater than expected. District Six reached out to Mr. Benjamin Jacobs, Crash Records and Research Administrator in Central Office, to further analyze this discrepancy. He explained that the CARS report used for the analysis was extracted on January 11, 2013 and the file for the 2011 shapefile was extracted on February 21, 2013.  Both datasets ideally would not show much difference since they are less than 2 months apart.  Mr. Jacobs also verified that in a CARS database extract run Friday, January 09, 2015 each of the missing crashes are still present.  What this means is that there was some condition at the time of the shapefile generation that prevented the crashes in question from being included. There are a number of potential causes for this.  Assuming that the missing crashes were there in the extract file at the beginning of the processing (which cannot be confirmed at this time) the Central Office staff are not certain at what point these crashes may have fallen out of the process. Central Office has confirmed that the missing crashes are in the extract file that will source the shapefile updates, which are scheduled to be available starting in mid-February.

Using the narrowed dataset, Peters set out to determine when crashes are most likely to occur on and around I-95 Express, and how the number of crashes were changing over time. Using the software R, Peters sorted crashes by time of day, sorted crashes by year, and sorted fatal crashes and incapacitating injury crashes over time for our interactive map and charts. The narrowed dataset and Peters’s R analysis is available on Github.