SWITRS: Car Crashes After Daylight Saving Time Ends

#california-traffic-data #data-science
A blonde woman adjusts the time on a row of clocks at a store.

We all hate the change to daylight saving time (DST) in the spring; it makes us tired, grumpy, but worst of all it causes us to crash our cars at a higher rate! The end of DST is not as universally reviled, probably because we get back the hour of sleep we lost earlier in the year, but Varughese & Allen found that there was still a “significant increase in number of crashes on the Sunday of the fall shift from DST”.1

With the SWIRTS data that I collected, and the analysis code I developed for my post last year looking at car crashes after the DST change, it should be pretty easy to check if I see the same trend as Varughese & Allen.

The Jupyter notebook used to perform this analysis can be found here (rendered on Github).

Crash Ratio

Just like last time, I will look at the number of crashes on the days following the end of DST. In order to help cancel out effects other than the time change—like the fact that crash rates vary by 30% depending on the year—I will divide each day’s total by the number of crashes a week later, when people are presumably back to normal.

Unlike last time, I am not normalizing by the number of crashes two weeks after the change. The reason for this is simple: that’s Thanksgiving week, and as I showed before the number of crashes is greatly reduced during the holidays.

Just as last time, the violin plots below show the distribution of these ratios from the years 2001 to 2017. A value greater than 1 means that there are more crashes during the week when DST ends than the week after.

Violin plot showing the ratio of crashes per day of the week for the week daylight saving time ends, divided by the week after.

There is, on average, a larger number of crashes on Sunday when the time changes as seen by Varughese & Allen. However, the same excess is not seen when a different normalization is chosen, like using the week before or two weeks after.2 The week before has different lighting during commute times and so it is easier to dismiss, but two weeks after has similar lighting.

t-Test

Instead, we turn away from our “chi-by-eye” test and do an actual statistical test: a two-tailed paired t-test, the same test used by Varughese & Allen. They find a significant (p < 0.002) increase in the number of deadly crashes on the Sunday that DST ends, but I do not (p = 0.082).

Our methods are different in a few key ways:

If I reproduce their methods with my dataset, I still do not find a significant result (p = 0.158).

As for California, Kansen Chu has once again given us a chance to get rid of the time change with Prop 7. His earlier bill failed, so he has gone directly to the voters this time. Although permanent DST is not the ideal solution, I’m still for getting rid of the time change itself!

  1. Varughese, J. and Allen, R., Fatal accidents following changes in daylight savings time: the American experience, Sleep Medicine, Volume 2, Issue 1, p. 31 - 36, doi: https://doi.org/10.1016/S1389-9457(00)00032-0 ↩

  2. The large deviations on the two week plot for Thursday and Friday are explained by Thanksgiving and Black Friday. ↩