SWITRS: On What Days Do Cyclists Crash?

An 1890s advertisement for Wilhelmina Cycle Co. Ltd. showing a family on bicycles.

It is time to use SWITRS data to look at traffic accidents in California again. I have previously used the data to look at when cars crash—during holidays when people both drive to work and to parties after—and when motorcycles crash—during the summer when its good riding weather. Today I want to look at something a little closer to my heart: bicycles.

I have been commuting on my bike for years now, and when I was younger I used to put in thousands of miles a year for fun. So knowing more about when accidents happen is something I am very interested in.

As per usual, the Jupyter notebook used to perform this analysis can be found here (rendered on Github).

A Simple Model

Before we dig into the data, I have a simple model for how many bicycle accidents there are. It is:

That is, the number of accidents involving bicycles () is the probability of an accident happening when a bike encounters a car () times the number of cars encountered (). This ignores some accidents, like solo crashes and those that do not involve a car, but these are rare.1

We won’t be able to test the validity of this model with the SWITRS data alone, but we can use it to reason about what is happening. For example, if the number of accidents increases, that could be because there are more cars or bikes on the road, or because the probability of collision increased (perhaps due to distracted drivers or worse average weather).

Data Selection

I selected accidents involving bicycles from the SQLite database (discussed previously) with the following query:

SELECT Collision_Date FROM Collision
WHERE Collision_Date IS NOT NULL
AND Bicycle_Collision == 1          -- Involves a bicycle
AND Collision_Date <= '2017-12-31'  -- 2018 is incomplete

This gave me 223,772 data points to examine spanning 2001 to 2017. Just as before, accidents from the most recent year are rejected because the database dump comes from September 2018, and so the year is incomplete.

Accidents per Week

For car crashes, I found that there was a large dip in 2008 as people stopped driving to work during the Great Recession. For motorcycle crashes, I found strong seasonality as people hung up their helmets during the winter. For bicycles, we have the following pattern:

Line plot showing bicycles accidents per week from 2001 through 2017

It shows features similar to both cars and motorcycles:

Thinking back to the model we can try to reason about the trend. We know the number of cars increased, so the decrease in accidents in the last few years is either due to a decrease in the number of cyclists—possibly because they traded their bikes for cars as they found employment—or a decrease in the likelihood of an accident—perhaps because drivers are more used to cyclists and look out for them.

Day-by-Day

Car are involved in accidents on holidays during which the drivers also work, like Halloween. Motorcycles are in accidents during summer holidays. Bicycles, on the other hand, have no holidays with a large excess in the number of accidents. Some holidays, like Christmas and Thanksgiving, keep people from getting on their bikes, but none seem to motivate to get out and ride.

Line plot showing average motorcycle accidents by day of the year

New Year’s Day, St. Patrick’s Day, and the 4th of July are all higher than they would be if they were not holidays, although you can’t tell from this plot. On those days, people tend to go out and celebrate with alcohol, which leads to solo crashes. I will examine that in a future post.

Day of the Week

For cars, weekends show a decrease in the number of accidents as people stop commuting. For motorcycles, weekends show an increase in the number of accidents as people use their time off to ride. As a recreational cyclist, I expected accidents to increase on the weekend as people put on their Lycra and take to the back roads for fun. But this is not the case:

Violin plot showing the number of bicycle accidents by day of the week

These violin plots show the distribution of accidents by day of the week over the 17 year period. There is a large drop in the number of accidents on weekends. This is surprising to me. I would have expected a lot more cyclists to be out on the weekend, leading to more interactions with cars.

It’s possible that this (an increased number of cyclists) is true but there are just enough fewer cars that the accident rate still goes down. Or perhaps the riders are better at avoiding accidents. Or maybe the cyclists are out in the countryside away from the cars. Or perhaps weekend drivers are better at avoiding cyclists. Without more data, we can’t tell.

Conclusion

This analysis of bicycle accidents surprised me a little. I expected bikes to show a similar pattern to motorcycles, since both are used to both commute and for fun. However, bikes show a greatly reduced accident rate on the weekend while motorcycles show an increase. Bikes and cars also seem to trade off, with car accidents increasing in recent years while bike accidents fall off. Further study and additional data is necessary before I can determine the reasons behind this trend.


  1. Of the 223,772 recorded accidents with bicycles, 89% involve a car. There is a bias though: SWITRS reports are filled out when the police or CHP are called to the scene. As such, they skew towards worse accidents.