SWITRS: Accidents After Daylight Saving Time
The daylight saving time (DST) change is awful—we get less sleep and it might not even save energy as was intended! Worse, studies by Varughese & Allen and Smith have shown that the time change increases the number of automobile accidents! Let’s look for a similar trend in the SWITRS data that I’ve collected.
The Jupyter notebook used to perform this analysis can be found here (rendered on Github).
Accident Ratio
The analysis is relatively simple. I start with the number of accidents that happen on the days following the start of DST in California. I divide the amount of accidents on each day by the number of accidents on the same day of the week but two weeks later.^{1} Taking the ratio cancels out most of the effects that are unrelated to the time change—like the fact that accident rates vary by 30% depending on the year. Two weeks after is a good choice for normalization because:
 The weeks after the time change have similar daylight hours to the week of the time change.
 The accident rate is still slightly elevated a week later, so normalizing by the very next week hides some of the increase that is due to the start of DST.^{2}
The violin plots below show the distribution of these ratios from the years 2001 to 2016. A value greater than 1 means that there are more accidents during the week when DST starts than two weeks after.
Except for Sunday, every day of the week following the time change has on average a higher rate of accidents! I am surprised that the accident rate stays high the entire week. This indicates that it takes even longer than a week for people to catch up on sleep and for the accident rate to go back to normal.
tTest
So the “chibyeye” plot is suggestive, but I can quantify whether the results are significant using a twotailed paired ttest. This is the method Varughese & Allen use. Doing so gives the following results.
Day  tvalue  pvalue 

Monday  2.7  0.017 
Tuesday  2.5  0.023 
Wednesday  1.6  0.122 
Thursday  1.4  0.191 
Friday  0.9  0.361 
Saturday  2.6  0.019 
Sunday (DST)  1.6  0.121 
The increase in accidents on Monday is significant, as is Tuesday and Saturday. Sunday is the only day that trends lower (matching the plot), but not significantly.
So daylight savings time causes more accidents, but those of us in California might be in luck! State Assembly member Kansen Chu has introduced a bill to finally do away with DST! Hopefully it will pass and let us all get that hour of sleep we deserve.
Update: I have rewritten part of this article to make my methodology clearer and add a ttest. The changes can be found in git; the first set and the second set.

It is also possible to use the week before or the week directly after the DST change to normalize. For the curious, I have also made a plot using the week before for normalization and the week after. They both show the same trend. ↩

I assume that people are back to normal after three weeks, and so I use that week as a control. I then compare that ratios of the control week with one week after the DST change and two weeks after the DST change to see which is more normal. One week after has Monday and Thursday high, indicating people are still having more accidents than we expect. Two weeks after the ratios are near one, and so I conclude people are back to normal by then. ↩