SWITRS: Differences in Vehicle Collision Rates by Manufacturer During COVID-19

As I prepared to write my post on the increase in traffic fatalities during COVID-19, I made some exploratory plots. One plot made me stop and stare. Here it is:

The number of traffic collisions involving Fords compared to those involving Toyotas before and after the COVID-19 stay at home order in California

This plot isn’t perfect—I will fix it below—but even so it is striking. Before the stay-at-home order the number of collisions involving Toyotas was much higher than those involving Fords. After the order, the trend flips. Fords have more collisions. I had to figure out why.

The code for this analysis can be found here (rendered on Github). The data is available on Kaggle or Zenodo. There is a hosted Kaggle Notebook version of this post as well to help you dive right in.

Data

I select all collisions between 2019 and November 30, 2020 that involve a Toyota or a Ford, with this query:

SELECT c.collision_date
    , p.vehicle_make
    , count(1) as total
FROM collisions AS c
LEFT JOIN parties as p
ON p.case_id = c.case_id
WHERE c.collision_date IS NOT NULL 
AND c.collision_date BETWEEN '2019-01-01' AND '2020-11-30'
AND p.vehicle_make IN ('ford', 'toyota')
GROUP BY 1, 2;

I start the data in 2019 because I need a sample from before the pandemic changed behavior, but I didn’t want to go too far back because collision rates vary drastically year-to-year. I cut off the data in November because the reporting is not yet complete for December.

Normalized Collision Rate

The number of collisions depends on many factors, primary among them is vehicle miles traveled.1 To help control for VMT, I normalize the mean number of collisions for each make of vehicle from January through June of 2019. This gives me a baseline to compare against. Here is the normalized plot:

The collision rate for Fords compared to Toyotas before and after the COVID-19 stay at home order in California, with mean normalized from January 2019 through June 2019.

Interpretation

The normalized rates match up well through the Christmas and New Year holidays, which is the two-week dip caused by people taking time off work and hence not commuting. But right after, the series diverge:

Taken together, I think these observations suggest the difference is due to a white-collarblue-collar divide. White-collar workers generally have more flexible work arrangements and their jobs are easier to do from home, whereas blue-collar workers have to travel to a job site to perform their work. Blue-collar workers are more conservative than white-collar workers and more likely to buy American branded cars like Fords.2

Initially I thought this difference would be driven purely by the prevalence of Ford trucks, but as we shall see it is not just trucks versus cars.

Trucks

Is it just that there are more Ford trucks? No.

The collision rate for Ford trucks compared to Toyota trucks before and after the COVID-19 stay at home order in California, with mean normalized from January 2019 through June 2019.

The same pattern holds, although both makes recover faster, with Fords returning to pre-pandemic levels and Toyota getting to 80%, which is much higher than the 50% Toyota reached when including non-trucks.

Location

Perhaps Ford owners just live in areas with looser restrictions, like the Central Valley? No. Here is data from Contra Costa County, part of the Bay Area:

The collision rate for Fords compared to Toyotas in Contra Costa County before and after the COVID-19 stay at home order in California, with mean normalized from January 2019 through June 2019.

It is the same pattern, but with a lot more noise due to the smaller population.

Age

Young drivers get in more accidents. Perhaps there is a strong age difference driving the trend? There is an age difference, see:

Area normalized distribution of Toyota and Ford driver ages during the COVID-19 stay at home order in California.

But that alone doesn’t account for the pattern:

The collision rate for Fords compared to Toyotas for drivers aged 30 to 50 before and after the COVID-19 stay at home order in California, with mean normalized from January 2019 through June 2019.

Putting It All Together

A person’s identity is made up of many traits: their age, their politics, where they live, what job they do, and yes, what car they drive. I looked at three different traits—vehicle type, location, and age—and none of them explain the entirety of the collision rate difference between Toyotas and Fords after the COVID-19 stay-at-home order. My conclusion is that Ford drivers are just different from Toyota drivers, in multiple ways, each of which contributes to the trend.


  1. From the Minnesota Department of Public Safety:

    Volume of traffic, or vehicle miles traveled (VMT), is a predictor of crash incidence. All other things being equal, as VMT increases, so will traffic crashes. The relationship may not be simple, however; after a point, increasing congestion leads to reduced speeds, changing the proportion of crashes that occur at different severity levels.

    Minnesota Department of Public Safety, Office of Traffic Safety (2014). Minnesota Traffic Crashes in 2014, Page 2 

  2. The type of car and brand both are driven by political leaning:

    The most left-leaning models with at least a dozen sightings in Mr. MacMichael’s project were the Honda Civic (80-20 left-leaning), Toyota Corolla (78-19) and Toyota Camry (74-26). The list of most right-leaning was led by another Toyota, but a midsize SUV, the Toyota 4Runner (86-14), followed by the Ford Expedition (76-24) and Ford F-150 (75-25).

    Tierney, John. (April 1, 2005). Your Car: Politics on Wheels, The New York Times.